The Philosophy of Biological Systematics - Natural History Museum of ... [PDF]

PHILOSOPHY OF BIOLOGICAL SYSTEMATICS Kirk Fitzhugh, [email protected]

Table of Contents Introduction ......................................................................................................................... 2 The Goal of Science. The goal of Biological Systematics ....................................................... 10 Causal Relationships in Systematics .................................................................................... 52 The Nature of Why-Questions ............................................................................................. 63 The Three Forms of Inference: Deduction, Induction, Abduction ......................................... 83 The Uses of Deduction, Induction, and Abduction in Science ..............................................110 Systematics Involves Abductive Inference ..........................................................................155 Inferences of Systematics Hypotheses, i.e. Taxa ..................................................................176 Some Implications for “Phylogenetic” Methods .................................................................228 The Requirement of Total Evidence ....................................................................................336 Homology & Homogeny & Homoplasy ...............................................................................403 Character Coding ...............................................................................................................466 The Mechanics of Hypothesis Testing in Biological Systematics ..........................................500 Implications for Nomenclature ...........................................................................................627 Defining Biodiversity and Conservation ..............................................................................683

THE PHILOSOPHY OF BIOLOGICAL SYSTEMATICS

Kirk Fitzhugh [email protected] Natural History Museum of Los Angeles County

This course differs from most courses on biological systematics in that the emphasis will not be on instructing you on how to use the variety of methods available to researchers. Instead, the emphasis will be on examining what is required to ensure that systematics, as a field of science, has an overarching framework that is consistent with all fields of scientific inquiry. It is from this framework that one can readily decide which methods are scientifically acceptable.

The Philosophy of Biological Systematics The contrast with a systematics course A TYPICAL ‘PHYLOGENETICS’ COURSE: • Phylogenetic theory • Characters and character coding • Tree building techniques • Tree statistics and tree support • Bayesian inference • Maximum Likelihood • Alignment

P Systematics courses usually focus on how to use methods. P The present course will focus on what is required to treat systematics as a science. P The goal is to give you the ability to determine which methods are scientifically acceptable.

• Molecular Dating • Various tree building programs (e.g., MrBayes, POY, and TNT)

The outline of topics shown here formed the basis for a recent 'phylogenetics' course in Denmark. The topics exemplify how the present course differs from what is typically covered in systematics courses.

“If science is not to degenerate into a medley of ad hoc hypotheses, it must become philosophical and must enter into a thorough criticism of its own foundations.” Alfred North Whitehead (1925: 25), Science and the Modern World.

One of the interesting phenomena surrounding the practice of systematics for the past 40 years is that distinct schools of thought have arisen regarding what it means to infer systematics hypotheses and to evaluate them. For instance, with regard to phylogenetic [sic] inference, the two most recognized schools of thought are what are said to be 'parsimony' and 'maximum likelihood.' Or, this dichotomy is sometimes seen as a distinction between hypothetico-deductive and statistical points of view. One of the hallmarks of these different opinions is that no critical assessment of the formal inferential structure of systematics is ever considered, such that neither the concept of parsimony nor likelihood are correctly justified. This lack of critical examination then extends to the matter of how one tests, evaluates, determines support, etc., for systematics hypotheses. As indicated in the quote shown here, in order to address the matter of how we are to assess whether or not one systematics hypothesis is better or worse than another requires that we carefully examine the philosophical foundations of hypothesis inference and testing.

The Philosophy of Biological Systematics Course Outline – Part 1 1.

The goal of Science. The goal of biological systematics.

2.

Causal relationships in systematics.

3.

The nature of why-questions.

4.

The three forms of inference: deduction, induction, abduction.

5.

The uses of deduction, induction, and abduction in science.

This course is arranged in four parts. Part 1 has as its focus identifying the goals of scientific inquiry and biological systematics, followed by some of the consequences of those goals. The three recognized forms of reasoning used throughout the sciences are then described in detail.

The Philosophy of Biological Systematics Course Outline – Part 2

1.

Systematics involves abductive inference.

2.

Inferences of systematics hypotheses, i.e. taxa.

3.

Some implications for “phylogenetic” methods.

In Part 2, we will identify the type of reasoning used in biological systematics to infer hypotheses. We will see that all taxa have the form of explanatory hypotheses, directed at giving us at least initial causal understanding of some of the characters we observe among organisms. Significant implications are then identified for some of the methods commonly used in phylogenetic [sic] systematics.

The Philosophy of Biological Systematics Course Outline – Part 3

1.

The requirement of total evidence.

2.

Homology & homogeny & homoplasy.

3.

Character coding.

4.

The mechanics of hypothesis testing in biological systematics.

Part 3 will address four different issues, all of which have received considerable attention in biological systmeatics, but also have been misrepresented. We will examine the correct interpretation of the requirement of total evidence, which has significant implications for the common approach of inferring systematics hypotheses from partitioned data, as well as attempts to compare cladograms inferred from different data sets. Next we will examine the definition of the term homology (sensu Owen) in relation to E. Ray Lankester's (1870) suggested replacement of that term with the two terms, homogeny and homoplasy. A general overview of character coding will then be presented as it relates to the nature of our observation statements, why-questions, and the goal of biological systematics reasoning. Finally, the nature of hypothesis testing will be carefully examined, showing that traditional attempts to characterize testing in systematics have been incorrect. The proper approach to testing systematics hypotheses will be examined.

The Philosophy of Biological Systematics Course Outline – Part 4

1.

Implications for nomenclature.

2.

Defining biodiversity and conservation.

In the final part of the course, we first will examine the implications of the inferential framework for biological systematics on our nomenclatural practices. The main focus here will be on the 'Linnean' system and the PhyloCode, to show that neither approach correctly takes into consideration the nature of our inferences. For any nomenclatural system to be successful, it must be consistent with the fact that biological systematics is about inferring explanatory hypotheses, referred to as taxa, and formal names must refer to those hypotheses, not just organisms. The last talk will be an opportunity to tie our systematics practices, as presented in this course, to new formal definitions of biodiversity and conservation. Interestingly, the outcome will be to show that the term biodiversity is largely useless and potentially deceptive.

“Science depends on judgments of the bearing of evidence on theory.... One of the central aims of the philosophy of science is to give a principled account of those judgments and inferences connecting evidence to theory.” Peter Lipton (2001: 184, Inference to the best explanation). In: A Companion to the Philosophy of Science.

The quote shown here epitomizes what will be most fundamental throughout this course. Our emphasis will be on recognizing the relations between evidence and biological systematics hypotheses. As we will see, these relations occur in different ways, depending on what we mean by 'evidence,' as well as our objectives in maintaining particular relations. It is by applying the principles of philosophy of science to biological systematics that we can clearly understand why these relations exist between evidence and hypotheses, and recognize the forms evidence takes with respect to hypotheses.

The Philosophy of Biological Systematics Course Outline – Part 1 1.

The goal of Science. The goal of biological systematics.

2.

Causal relationships in systematics.

3.

The nature of why-questions.

4.

The three forms of inference: deduction, induction, abduction.

5.

The uses of deduction, induction, and abduction in science.

Let's start by looking at the goals of science and biological systematics.

The Confusing Variety of Systematics Methods

‘What are species?’

‘What are taxa?’ total evidence?

character weighting?

hypothesis testing?

measures of support?

What is the philosophical basis for choosing a method? One of the greatest difficulties in biological systematics is that we have available a variety of methods. But, there is no clear consensus among systematists as to which methods to use. Similarly, there are fundamental questions regarding what we mean by terms like 'species' or 'taxon.' The only real way to resolve these problems is to have a philosophical basis for choosing among methods. That basis can only come from first ackowledging the goal of doing science, and then applying that goal to systematics.

Basic Criteria for Judging Methods in Biological Systematics • Recognize the goal of Science. • The goal of biological systematics should be consistent with this goal. • Does a particular systematics method satisfy the goal of Science? • Does a particular systematics method accurately represent our perceptions and why-questions?

To determine whether or not an approach to biological systematics is scientifically appropriate, we must first acknowledge the goal of doing science, as well as understand that the goal of systematics must be consistent with that more general goal. We can then determine whether or not specific methods actually serve to fulfill both the goal of science as well as systematics. In related fashion, we have to ensure that the methods we use do accurately represent our observation statements and why-questions, since these are the issues to which the goal of science inquiry is directed.

The Goal of Science: To Causally Understand What We Observe “Broadly speaking, the vocabulary of science has two basic functions: first, to permit an adequate description of the things and events that are the objects of scientific investigation; second, to permit the establishment of general laws or theories by means of which particular events may be explained and predicted and thus scientifically understood; for to understand a phenomenon scientifically is to show that it occurs in accordance with general laws or theoretical principles.” Hempel (1965: 139, emphasis original), Aspects of Scientific Explanation

An answer to the question of what is the goal of science was nicely described by the philosopher of science, Carl G. Hempel. The goal can be identified as having two parts: (1) describing the objects and events we encounter, and (2) presenting explanations of those objects and events, for the purpose of ever-increasing our understanding as well as having the ability to make predictions into the future. Overall, the goal of science is to enable us to *causally understand* phenomena. As we will see throughout this course, this goal will be the highest priority for biological systematics.

The Goal of Science: To Causally Understand What We Observe Scientific inquiry has two fundamental components: Descriptive:

Theoretical:

observations

inferences of hypotheses and theories

Based on Hempel's characterization of science, we recognize science as having two basic parts: 'descriptive' and 'theoretical.' The descriptive component refers to our communicating observations, as observation statements. The theoretical refers to our applications of theories and hypotheses to those observation statements.

The Goal of Science: To Causally Understand What We Observe Scientific inquiry has two fundamental components:

Descriptive:

Theoretical:

observations

inferences of hypotheses and theories

The descriptive and theoretical aspects of inquiry are interdependent – objects and events cannot be described in the absence of theory, and the basis for theories and hypotheses are the objects and events which are in need of understanding.

But, it is well known that observation statements cannot be made in the absence of theories, and theories and hypotheses have their origins in observations. So, the descriptive and theoretical realms are clearly interdependent.

The Goal of Science: To Causally Understand What We Observe Scientific inquiry has two fundamental components:

Descriptive:

Theoretical:

observations

inferences of hypotheses and theories

Understanding explanation / prediction

It is the interplay between the descriptive and theoretical that leads to scientific understanding.

As the principle goal of scientific inquiry is to acquire causal understanding, and from that understanding we have the ability to explain phenomena as well as make predictions of future phenomena, it is the interplay between the descriptive and theoretical that leads to the acquisition of understanding.

Scientific Understanding, Defined

“A phenomenon P can be understood if a theory T of P exists that is intelligible (and meets the usual logical, methodological and empirical requirements).” de Regt & Dieks (Synthese 2005: 150)

We have been referring to 'understanding' in the previous diagrams, so it will be useful to have a formal definition of the term. The definition shown here offers the view that to understand a phenomenon is to associate that phenomenon with some theory.

Scientific Understanding, Defined

“...the cognitive achievement realizable by scientists through their ability to coordinate theoretical and embodied knowledge that apply to a specific phenomenon.” Leonelli (2009: 197)

Leonelli (2009) offers a similar perspective with regard to biological understanding. We apply not only our theories but also our previous established knowledge to a phenomenon to provide us understanding of the latter.

“...biology can be divided into the study of proximate causes, the subject of the physiological sciences (broadly conceived), and into the study of ultimate (evolutionary) causes, the subject matter of natural history....” Mayr (1982: 67)

1904-2005

Even within biology, there have been attempts to characterize the nature of the understanding we seek regarding organisms. An excellent and very useful characterization of biological understanding was developed by evolutionary biologist, Ernst Mayr. Mayr suggested that biological inquiry seeks to acquire understanding that is causal, and that such causal understanding can be separated into 'proximate' and 'ultimate' causes. While Mayr distinguishes proximate and ultimate causes as related to 'physiological sciences' and 'natural history,' we will need to be more precise.

“...proximate causes govern the responses of the individual (and his organs) to immediate factors of the environment while ultimate causes are responsible for the evolution of the particular DNA code of information with which every individual of every species is endowed.” Mayr (1961: 1503)

1904-2005

Mayr originally published his idea of proximate and ultimate causes in biology in 1962. What might be noticed is that proximate causes refer to those causes that only occur within an organism during its lifetime. Ultimate causes, on the other hand, transcend lifetimes.

“The proximate causes of an organism’s traits occur within the lifetime of the organism.... The ultimate causes occur prior to the lifetime of the organism, within the evolutionary history of the organism’s species.” Beatty (1994: 334)

In his analysis of Mayr's proximate/ultimate distinction, Beatty (1994) offers a very good characterization, shown here.

Biological Understanding sensu Mayr proximate

ultimate

ontogenetic / functional

evolutionary

We can now begin to summarize Mayr's view of causal understanding in biology with the more general goal of science we examined earlier. We can see that proximate understanding refers to ontogenetic and functional aspects during the lifetime of an individual organism. Ultimate understanding refers to evolutionary causes that can apply to groups of organisms over time.

Biological Understanding sensu Mayr descriptive biology

ultimate

proximate

(observation statements) “It is sometimes overlooked how essential a component in the methodology of evolutionary biology the underlying descriptive work is. ”

Mayr (1982: 70)

ontogenetic / functional

evolutionary

Goal of Science – acquire ever-increasing understanding: • descriptive • causal - proximate / ultimate • predictive

But in addition to proximate and ultimate understanding, Mayr was very clear in his writings on the subject that there is a third dimension to understanding, what he referred to as 'descriptive biology.' Mayr was correct that in order to pursue either proximate or ultimate understanding, one must already have observations of effects that are in need of explanation. These effects are in the form of the properties, features, characters, etc., of organisms, that we communicate by way of our observation statements. Notice that Mayr's descriptive, proximate, and ultimate understanding are consistent with the goal of science presented earlier. To acquire ever-increasing understanding we see that it must be descriptive as well as causal, and also predictive. We seek descriptive understanding of what we perceive, as well as offering possible past causes that explain what we observe in the present. And we attempt to apply our understanding into the future with predictions of effects due to causal conditions that exist in the present.

Biological Understanding sensu Mayr descriptive biology

proximate

ultimate

(observation statements) “It is sometimes overlooked how essential a component in the methodology of evolutionary biology the underlying descriptive work is. ”

Mayr (1982: 70)

ontogenetic / functional

evolutionary

To what extent is biological systematics successful at acquiring ever-increasing understanding that is descriptive, proximate, and especially ultimate? An important part of this course will be to examine the extent to which descriptive, proximate, and ultimate understanding is acquired in biological systematics. These will be issues that need to be addressed both in terms of knowing the nature of our reasoning from observations to the variety of hypotheses used in systematics, as well as the correct approach to testing any of those hypotheses. It is especially the act of testing that accomplishes the task of increasing our causal understanding, which is the most fundamental goal of scientific inquiry.

SCIENCE: General Principles and Specialized Techniques Specialized Techniques

Specialized Techniques

Specialized Techniques

Astronomy

Psychology

Chemistry

Principles of Scientific Method

Biology

Specialized Techniques

Physics

Geology

Specialized Techniques

Specialized Techniques

While the goal in all fields of science is the acquisition of causal understanding, and that must be regulated by our general rules and methods in science, and more generally by philosophy of science, each field of science must adopt its own specialized techniques for the purpose of acquiring that understanding. The problem we will identify in biological systematics, however, is that the specialized techniques are too often divorced from the more general principles of scientific inquiry and philosophy. We will attempt in this course to correct that problem.

Four Fundamental Criteria Applied in Science

1. Rationality Beliefs and actions should be rational, i.e. they should make sense. A rational belief or action is one based on all evidence that is relevant to the formation of that belief or action.

In order to correctly characterize the nature of biological systematics as a field within the broader realm of science, we need to recognize four fundamental criteria that are applied throughout the sciences. The first criterion is rationality.

Four Fundamental Criteria Applied in Science

2. Truth Truth is a property of statements. The correspondence theory of truth is the most common concept of truth applied in Science: true statements correspond with reality. Facts about the world determine the truth of statements. correspondence External physical world of objects and events

Internal mental world of perceptions and beliefs

The second criterion is truth. As noted in this slide, the 'correspondence theory' is typically used in the sciences, although there are about six theories of truth available. With regard to systematics, in which there has developed a popular culture of thinking in terms of 'true phylogenetic trees' as a basis for judging methods of cladogram inference, it should be apparent that truth cannot be asserted separate from some empirical basis for the truth of statements.

Four Fundamental Criteria Applied in Science

3. Objectivity The existence of objects and events apart from human minds. Objective knowlege is concerned with physical objects and events.

The third criterion is objectivity, which should be apparent.

Four Fundamental Criteria Applied in Science

4. Realism The correspondence of human perceptions with the external and independent (and possibly unobservalbe) realities of physical objects and events.

And finally there is the criterion of realism.

The Foundations of Science Common Sense

Common Sense:

The assumption that physical reality exists.

As we have already noted, the ultimate goal of science is to acquire ever-increasing causal understanding of the phenomena (objects and events) we encounter. To successfully achieve that goal, we have to recognize the hierarchical structure within which science resides as part of human reasoning. The most general rule we have is that of common sense. In other words, we operate under the assumption that physical reality does exist - that all that we perceive around us are not just hallucinations. Without this assumption, empirical inquiry of any kind would not be possible.

The Foundations of Science Common Sense Philosophy

Philosophy:

The study of the way humans think and reason. Composed of four main branches: • logic, the study of reasoning • epistemology, the study of knowledge • metaphysics, the study of concepts and their relations • ethics, the study of moral evaluation

Within the realm of common sense, we have the field known as philosophy - the study of the way humans think and reason. And within philosophy there are four branches.

The Foundations of Science Common Sense Philosophy Philosophy of Science

Philosophy of Science:

The study of the principles and methods applied in all fields of science.

The four branches of philosophy presented in the previous slide are often applied to the subfield of philosophy, known as philosophy of science, which studies the principles and methods used throughout the sciences.

The Foundations of Science Common Sense Philosophy Philosophy of Science Scientific Methods

Scientific Methods: The processes of hypothesis and theory formation, testing, and revision, for the purpose of acquiring understanding of physical reality.

It is by way of philosophy of science that scientific methods are developed. It is those methods that are intended to enable us to achieve the goal of scientific inquiry, i.e. causal understanding.

The Foundations of Science Common Sense Philosophy Philosophy of Science Scientific Methods Scientific Specialties

Scientific Specialties: The fields of study that address specific aspects of physical reality, e.g., physics, chemistry, paleontology, systematics.

By way of particular scientific methods there are the applications of scientific specialities.

The Foundations of Science Common Sense Philosophy Philosophy of Science Scientific Methods Scientific Specialties Technology

Technology:

The specialized techniques applied in a specific field of study.

The applications of scientific methods within scientific specialities are often only possible because of technology, e.g. computers, microscopes, etc.

The Foundations of Science Common Sense Philosophy Philosophy of Science Scientific Methods Scientific Specialties Technology

Scientific methods are constrained by the principles of philosophy, as well as the philosophy of science. The problem in systematics is that methods are too often developed and considered in isolation of philosophy.

Finally, it is important to notice that if we are going to critically evaluate our scientific methods, then this must be done in the context of philosophy of science as well as philosophy in general. Scientific methods cannot operate independent of philosophical principles. Unfortunately, this is exactly what has too often occurred in biological systematics. This course is intended to help correct that error.

What is the Goal of Biological Systematics? Some common answers C “To explain shared similarities among a group of organisms.”

We have seen that the goal of scientific inquiry is to not only describe the objects and events we encounter (observation statements), but more importantly to causally understand those phenomena. We now need to determine if the goal of biological systematics is consistent with the goal of science. When we ask the question, 'What is the goal of systematics?', there are at least three general answers given. What you will notice is that most of these answers are not consistent with the goal of science. And this is a serious problem. One answer to the question we sometimes encounter is that systematics is intended to explain shared similarities.

Parsimony [sic] “A genealogy is able to explain observed points of similarity among organisms just when it can account for them as identical by virtue of inheritance from a common ancestor.” Farris (1983: 18), The logical basis of phylogenetic analysis

The idea of explaining shared similarities has been especially common in the context of cladograms. Unfortunately, this notion is not usually extended to other aspects of systematics, as we will see later in this course. The idea of explanation in systematics has been common in the cladistics literature, especially in connection with the principle of parsimony.

Likelihood [sic] “The concept of likelihood refers to situations that typically arise in natural sciences in which given some data D, a decision must be made about an adequate explanation of the data.” Schmidt & von Haeseler (2010: 181), Phylogenetic inference using maximum likelihood methods.

But we also find claims that explanation is important when 'likelihood' methods are used. But as we will see later in this course, these claims of importance of explanation are too often poorly formulated and usually insufficient.

What is the Goal of Biological Systematics? Some common answers C “To explain shared similarities among a group of organisms.” C “To show the phylogeny / evolutionary history of a group of organisms.”

A much more vague reference to explanation being the goal of systematics comes from the popular view that we want to present 'phylogeny' or 'evolutionary history.'

“Systematics is the study of organic diversity as that diversity is relevant to some specified pattern of evolutionary relationship thought to exist among the entities [sic] studied.” Wiley & Lieberman (2011: 8), Phylogenetics: Theory and Practice of Phylogenetic Systematics

And as we see in this quote, even the explain-as-phylogeny point of view can be taken to a point of being uninformative.

“A phylogenetic tree [cladogram]... is a graphic representation of the historical course of speciation.” Wiley & Lieberman (2011: 4), Phylogenetics: Theory and Practice of Phylogenetic Systematics

And the explanatory nature of cladograms is often inconsistent.

What is the Goal of Biological Systematics? Some common answers C “To explain shared similarities among a group of organisms.” C “To show the phylogeny / evolutionary history of a group of organisms.” C “To discover natural, hierarchical order, then reflect that order in classifications.”

Rather than having a direct or indirect goal of explanation of systematics, there is the still-popular school of thought that causality should be removed from systematics. In this instance, diagrams such as cladograms have no explanatory interpretation, but instead either summarize character distributions, or convey nebulous ideas such as 'natural order' or 'natural hierarchies.' The general phrase commonly used to identify this less-than-scientific perspective is 'pattern cladistics.'

“Systematics is primarily concerned with problem solving. This might seem an obvious statement, yet the majority of those interested in systematics and phylogeny approach the subject as being concerned with ‘inferences’, ‘reconstructions’, or ‘estimations’.... The general problem may be phrased as follows: ‘What are the interrelationships among organisms?’” Williams & Ebach (2008: 21)

The pattern cladistic approach has serious problems in that it is inconsistent with the goal of scientific inquiry.

What is the Goal of Biological Systematics? Some common answers C “To explain shared similarities among a group of organisms.” C “To show the phylogeny / evolutionary history of a group of organisms.” C “To discover natural, hierarchical order, then reflect that order in classifications.”

ARE ANY OF THESE GOALS CONSISTENT WITH THE OVERALL GOAL OF SCIENCE?

Based on what we have already seen regarding the goal of scientific inquiry, the commonly identified goals of biological systematics are not sufficiently consistent with the goal required of all sciences.

What is the Goal of Biological Systematics? A Formal Definition of Biological Systematics

The actions of biological systematization. The goal of which is to obtain causal understanding of the properties or characters of organisms exhibited at different stages of their life history or shared among some set of individuals. The term taxonomy is unnecessary because it is a synonym of systematics.

To correctly and effectively identify the goal of biological systematics requires that this goal be fully consistent with the more general goal of scientific inquiry. Shown here is a formal definition of biological systematics that not only places it squarely in the realm of science, but also establishes the field as having the same goal as all fields of science: to acquire causal understanding of the features we observe of organisms. Notice that with this formal definition, we no longer need to make a distinction between the terms 'systematics' and 'taxonomy.' While some might think that taxonomy refers only to 'species descriptions,' as we will see during this course, all facets of systematics have the same inferential framework, wherein all actions in systematics are directed at achieving the goal of causal understanding. Taxonomy is a term that should be regarded as a synonym of systematics. Systematics is the more accurate term to use.

What is the Goal of Biological Systematics? A Formal Definition of Taxon

Any of a set of classes of hypotheses used in biological systematics for the purpose of explaining particular characters of observed organisms.

With the formal definition of biological systematics accurately manifesting the goal of scientific inquiry, it is crucial that our reference to a taxon also be consistent with that goal. Clearly, as the goal of systematics is to present explanatory hypotheses that give us an opportunity to understand the occurrences of features among organisms, then taxa can only be regarded as synonymous with those hypotheses. Of course, this will have profound consequences, because too often taxa are thought of as being either individuals or things that exist in time and space, much like organisms. As we will see throughout this course, taxa can only be regarded as explanatory hypotheses, not as things or individuals. Indeed, our use of the term taxon or taxa is entirely unnecessary. It would be more appropriate to simply refer to hypotheses.

“...the semaphoront [‘character bearer’] corresponds to the individual in a certain, theoretically infinitely small, time span of its life, during which it can be considered unchangeable.” W. Hennig (1966: 65)

The definition of taxon presented in the previous slide refers to individual organisms and the characters we observe. In his book, "Phylogenetic Systematics" (1966), Willi Hennig correctly stressed that we make observations of individuals at particular moments in time during their entire ontogeny or life history. Hennig suggested that the appropriate term for the individuals we observe should be 'semaphoront.'

“...it follows that we should not regard the organism or the individual (not to speak of the species) as the ultimate element of the biological system. Rather it should be the organism or the individual at a particular point of time, or even better, during a certain, theoretically infinitely small, period of its life. We will call this element of all biological systematics... the character-bearing semaphoront.” W. Hennig (1966: 6, emphasis original)

Hennig's use of the term semaphoront to indicate our observations of organisms at specific times during their life history is especially significant because it better emphasizes that the fundamental units in biological systematics are individual organisms. Indeed, notice that Hennig understands that species are not the fundamental units in systematics.

What is the Goal of Biological Systematics? Obtain causal understanding of the properties or characters of organisms exhibited at different stages of their life history or shared among some set of individuals. Some Consequences: • Biological systematics involves the non-deductive inference of explanatory hypotheses and, where possible, their subsequent testing. • The goal of biological systematics is to move toward causal understanding of what we observe, not merely to obtain “cladograms,” “trees,” or to “reconstruct phylogeny.” • “Cladograms” are not things in themselves, but are very limited explanatory hypotheses of observed properties of individuals among different taxa.

With a formal definition of biological systematics that is consistent with the goal of scientific inquiry, there are several significant implications. The first is that as systematics is about the inferences of explanatory hypotheses, we will be clearly identifying the type of inference involved, as well as acknowledging that the testing of those hypotheses is very different from what has traditionally been presented by systematists. Second, since the goal of systematics is consistent with the goal of science, i.e. to acquire causal understanding, our goal is *never* to just get trees, cladograms, or reconstruct phylogeny [sic]. And finally, we will see in this course that cladograms are *very vague* explanatory accounts. Indeed, they are so poor as explanations that they offer us very little to serve as vehicles for the goal of doing systematics, much less science.

The Two Realms of Science Present (the realm of Observation)

Past

Future Cause Causal Hypothesis

abduction

‘Historical’ Sciences

prediction

Effect

Effect

‘Experimental’ Sciences

Biological systematics is part of the “historical sciences,” where observations in the present are used to infer explanatory hypotheses about past events to account for those observations. For our purposes in biological systematics, we can think of science as having two broad, operational realms: historical and experimental. The historical sciences include such fields as systematics, evolutionary biology (in part), paleontology, archaeology. The experimental sciences include physics, chemistry, geology (in part). There is, of course, a lot of overlap between these. The main distinction between the historical and experimental sciences is that the historical sciences focus on effects that exist in the present, and our goal is to develop explanatory hypotheses of possible past causal events that can account for those oberved effects. The experimental sciences, on the other hand, begin with known causal, or experimental, conditions in the present, to see if predicted effects occur in the future. Another way to think about this distinction is that the historical sciences are mainly concerned with the inferences and testing of explanatory hypotheses, whereas the experimental sciences are mainly concerned with the testing of theories. But, be cautious about this distinction, since there always are exceptions.

The Philosophy of Biological Systematics Course Outline – Part 1 1.

The goal of Science. The goal of biological systematics.

2.

Causal relationships in systematics.

3.

The nature of why-questions.

4.

The three forms of inference: deduction, induction, abduction.

5.

The uses of deduction, induction, and abduction in science.

We can now take an initial look at the nature of the relationships that are referred to in biological systematics. Since the goal of systematics is to present us with causal understanding of the features of organisms, then the nature of the relationships throughout systematics must be causal in form.

RELATIONSHIPS & BIOLOGICAL SYSTEMATICS

When we speak of ‘relationships’ in systematics, we mean causal relationships. The basic unit to which these causal relationships refer is individual organisms. Taxa – as explanatory hypotheses – indicate particular causal relationships among groups of organisms.

To start with our examination of these issues, we need to understand what we mean when we speak of 'relationships' in biological systematics. We use the term relationship on a regular basis, but, the word is often not clearly understood when it is used in systematics. We need to first recognize that when we speak of relationships, we are speaking of causal relations. For example, we say we are related to our parents, we are related to our sisters or brothers, we are related to our grand parents. In every instance, the relations we are talking about are causal relations, because it is that type of relationship that gives one understanding. And, as we will see in the remainder of this course, the units to which those causal relationships refer are individual organisms. Then, we can specifically look at the way in which we infer each of the types of causal relationships, as explanatory hypotheses, that are used in biological systematics. And again, it is causal relationships that we are interested in, because it is those types of relations that best serve the overall goal of scientific inquiry.

(1913-1976)

Hennig, W. 1966. Phylogenetic Systematics

One of the best examinations of the nature of causal relationships in biological systematics can be found in Willi Hennig's (1966) book, "Phylogenetic Systematics."

Classes of Relationships 1. ontogenetic

6

2. cyclomorphic 3. sexual dimorphic

7 4

2

4. tokogenetic 5. polymorphic 6. specific

1

3 5

Hennig, W. 1966. Phylogenetic Systematics

7. phylogenetic

Each of these classes of relationships refer to the different classes of explanatory hypotheses we call taxa.

Shown here is Hennig's (1966) well known figure 6, which we often see reproduced in other works on the principles of biological systematics. It is in this figure that Hennig identifies the fundamental classes of relationships used in systematics. But too often, what is not recognized is that Hennig pointed out that all of these relationships deal with individual organisms. He discussed in great detail seven classes of relationships involving organisms, all of which are shown in his diagram. Ontogenetic relationships. Where we speak of an individual at a particular point in it's life history. Cyclomorphic relationships. Where there are seasonal phenotypic differences among individuals of different generations. Sexual dimorphic relationships. The phenotypic differences between males and females. Tokogenetic relationships. Parents producing offsrping as a result of reproductive events (tokogeny). Polymorphic relationships. Different phenotypes expressed among individuals in a population. Specific relationships. Refers to species hypotheses, accounting for features among a group of organisms that are reproductively isolated from other groups. Phylogenetic relationships. The most general type of relationship in systematics, accounting for shared features among organisms to which different species hypotheses refer, as well as strictly asexual or strictly self-fertilizing hermphroditic organisms. But as will be noted later in the course, because of what classes of causal events are entailed by phylogenetic hypotheses, such hypotheses are actually not applicable to obligate asexual or self-fertilizing hermaphroditic organisms.

7 Classes of Causal Relationships

Proximate

1. ontogenetic

6

2. cyclomorphic 3. sexual dimorphic

7 4

2

4. tokogenetic Ultimate 5. polymorphic 6. specific (species)

1

3 5

7. phylogenetic Descriptive Biology (observation statements)

Using Hennig's (1966) figure 6, we can clearly identify the three broad classes of causal understanding recognized by Ersnt Mayr, that were referred to earlier.

Kingdom Phylum Class phylogenetic hypotheses Order Family

Ultimate explanations

Genus Species Subspecies Families, demes, populations

Individuals

Semaphoronts (e.g., ‘larva,’ ‘juvenile,’ ‘adult’)

specific hypotheses intraspecific hypotheses tokogenetic hypotheses

the objects we perceive

ontogenetic hypotheses

Descriptive explanations (observation statements)

Proximate explanations

And here are the distributions of classes of hypotheses shown in the previous slide, in a different arrangement.

All taxa/hypotheses in biological systematics are inferred by way of abduction Kingdom Phylum Class phylogenetic hypotheses Order Family Genus Species Subspecies Families, demes, populations

Individuals

Semaphoronts

specific hypotheses intraspecific hypotheses tokogenetic hypotheses

the objects we perceive

(observation statements)

ontogenetic hypotheses

(e.g., ‘larva,’ ‘juvenile,’ ‘adult’)

As we saw earlier with the definition of the term 'taxon,' all taxa are explanatory hypotheses. All of the different classes of hypotheses-as-taxa shown here, indicated by the red arrows, are the products of a type of reasoning known as 'abduction,' which we will examine in depth later in the course. And abduction, or abductive inference will form the foundation for the remainder of our examination of systematics in this course.

Examples of Causal Relationships in Systematics Ontogenetic, Specific, Phylogenetic

Semaphoront: an explanatory hypothesis of ontogenetic relationships, derived from ontogenetic theories applied to a particular organism. The hypothesis accounts for features of an organism at a particular age relative to features at another age, by way of ontogeny.

individual ‘larva,’ ‘juvenile,’ ‘adult’

We can briefly look at three of the most common classes of relationships referred to in biological systematics, and discussed by Hennig (1966): semaphoront, specific (species) relationships, and phylogenetic relationships. A semaphoront is an individual at a specific point in time during its life history. In other words, it is a hypothesis that gives us an explanatory account relative to the ontogenetic history of the individual.

Examples of Causal Relationships in Systematics Ontogenetic, Specific, Phylogenetic

Species: an explanatory hypothesis of specific relationships derived from theories of character origin/fixation during tokogeny, applied to a set of semaphorants. A ‘lineage,’ accounting for features of a group of semaphoronts relative to different features in other semaphoronts (in other species).

species a-us

individual

A species is an explanatory hypothesis that refers to specific relationships.

Examples of Causal Relationships in Systematics Ontogenetic, Specific, Phylogenetic

species b-us

species c-us

‘Supraspecific’ Taxon: an explanatory hypothesis of phylogentic relationships, derived from tokogenetic, evolutionary, and population splitting theories, applied to particular semaphoronts. Accounting, by way of phylogeny, for the same features shared by semaphoronts among two or more species relative to different features in semaphoronts of other species.

b-us c-us

a-us phylogenetic relationships

species a-us

individual

And a phylogenetic hypothesis refers to phylogenetic relationships. Notice that it is more accurate to refer to such relationships as hypotheses as opposed to 'taxa.'

Not All of Systematics is ‘Phylogenetic’

‘Phylogenetic systematics’ sensu Hennig (1966) only provides causal understanding of the properties of groups of organisms to which two or more species hypotheses also refer. Explanatory hypotheses of ontogeny, tokogeny, species, etc., represent other levels at which causal understanding can also be achieved, but by using theories different from those applied in phylogeny.

As we will see, distinguishing the different classes of explanatory hypotheses used in systematics is fundamental to identifying the appropriate levels at which our why-questions should be asked and answered.

Historically, there has been confusion regarding what is meant by the phrase 'phylogenetic systematics.' Yet, when one carefully reads Hennig's (1966) book, it is clear that he understood biological ('hologenetic') systematics to refer to the variety of explanatory hypotheses, with phylogenetic systematics only referring to one of those classes of relationships. As we will see in this course, as the goal of systematics is the same as the goal in all fields of science, to acquire causal understanding, to achieve such understanding comes from different classes of hypotheses used to answer our different whyquestions.

The Philosophy of Biological Systematics Course Outline – Part 1 1.

The goal of Science. The goal of biological systematics.

2.

Causal relationships in systematics.

3.

The nature of why-questions.

4.

The three forms of inference: deduction, induction, abduction.

5.

The uses of deduction, induction, and abduction in science.

All of understanding in science begins with observations and our questions associated with those observations in need of being explained. The type of questions most commonly asked in systematics are known as why-questions. It is our why-questions that form the basis for all aspects of biological systematics.

“The scientist is not a person who gives the right answers, he's the one who asks the right questions.” Claude Levi-Strauss (1964) Le Cru et le Cuit

Unfortunately, when we speak of science, we almost always neglect to consider the questions we are actually asking, for which we seek hypotheses and theories to give us answers.

“...a logic in which the answers are attended to and the questions neglected is a false logic.” R.G. Collingwood (1938: 31), An Autobiography

Like any endeavor, science is one we perform to achieve particular goals. And as with any action carried out among a group of people, science has its social component, such that scientific procedures tend to become standardized to the point where we stop examining the bases for what we do, and we just go through the motions. Biological systematics suffers from a perspective where practitioners are seeking answers, yet they either don't know the questions they are asking, or they are asking inappropriate questions. This neglect is what has allowed for the rapid development of systematics methods and computer algorithms that offer contradictory approaches, and with no jusification based on the goal of using those methods according to the why-questions we should be asking.

Why Ask Why!Questions? Look to the goal of scientific inquiry

“To explain the phenomena in the world of our experience, to answer the question ‘why?’ rather than only the question ‘what?’, is one of the foremost objectives of all rational inquiry; and especially [for science]... to go beyond a mere description of its subject matter by providing an explanation of the phenomena it investigates.” From: Hempel & Oppenheim (1948: 135), The logic of explanation. Philosophy of Science 15: 135-175.

By this time it is probably obvious why we ask why-questions -- because we seek causal understanding of the phenomena we encounter. The quote shown here, by Carl Hempel, exemplifies the reason we ask why-questions, and the fact that such questions are a fundamental part of the goal of scientific inquiry.

The Foundation for All of Systematics The Nature of Our Why-Questions

If the goal of biological systematics is to provide causal understanding of the properties of organisms, then we must first recognize the nature of our whyquestions, to which evolutionary theories and systematics hypotheses provide answers.

We now need to examine the specific properties of why-questions, without which any treatment of biological systematics would be incomplete. As we will see throughout much of this course, our why-questions are fundamental components.

Why-Questions How we usually ask them

“Why P?” Example: “Why do these specimens have lateral body wall extensions called ‘appendages’?”

It is essential to know the formal structure of the why-questions we ask. We usually think of why-questions as simply having the form, "Why P?", or "Why is it the case that x is P?" This form is, however, incomplete and thus does not fully represent the basis for such questions.

Why-Questions The proper form: Contrastive questions

“Why P in contrast to X?” Example: “Why do these specimens have lateral body wall extensions (= appendages) in contrast to other specimens with convex body walls?”

P

X

The correct form of why-questions is that they are 'contrastive.' In other words, we ask questions that contrast the surprising or unexpected condition in need of being explained with the expected condition(s) that has already been explained. In the case of systematics-based observations, our contrastive why-questions are of the form shown here.

Why-Questions Three parts: ‘why’

“Why P in contrast to X?” Example: “Why do these specimens have lateral body wall extensions (= appendages) in contrast to other specimens with convex body walls?”

P

X

There are three components to why-questions. First is that such questions are prefaced with 'why.'

Why-Questions Three parts: ‘why’ + fact(s) + foil

‘fact’

‘foil’

“Why P in contrast to X?” Example: “Why do these specimens have lateral body wall extensions (= appendages) in contrast to other specimens with convex body walls?”

P

X

The other two components of why-questions are known as 'fact' and 'foil.' The 'fact' is what is in need of being explained, in contrast to the 'foil.'

Why-Questions Three parts: ‘why’ + fact(s) + foil

‘fact’

‘foil’

“Why P in contrast to X?” contrast class

Example: “Why do these specimens have lateral body wall extensions (= appendages) in contrast to other specimens with convex body walls?”

Three parts: ‘why’ + fact(s) + foil

P

X

The 'fact' and 'foil' together comprise the 'contrast class' of a contrastive whyquestion.

Why-Questions Three parts: ‘why’ + facts-as-presuppositions + foil

Question: “Why P in contrast to X?” Presupposition: it is true that P is the case. Example: “Why do these specimens have lateral body wall extensions (= appendages) in contrast to other specimens with convex body walls?”

Another important condition is that we assume the truth of the observation statement(s) that comprise the 'fact.' These facts are then said to be presuppositions.

Why-Questions Criterion for sensibility

‘fact’

‘foil’

Question: “Why P in contrast to X?”

“...we evaluate the sensibility of a why question by considering whether the fact and foil can be viewed as [alternative] culminating outcomes of some single type of natural causal process.” Barnes, E. 1994. Why P rather than Q? The curiosities of fact and foil. Philosophical Studies 73: 35–53.

The choice of foil for why-questions is not arbitrary. Instead, correctly choosing a foil requires that fact and foil are alternative effects from a single type of causal process. The following examples exemplify this requirement.

Why-Questions Criterion for sensibility

Question: “Why did the match not ignite in contrast to igniting?”

fact

foil

Common causal process: frictional surface

In this example we have the why-question, "Why did the match not ignite in contrast to igniting?" Notice that fact and foil trace back to the common causal process of rubbing a match along a frictional or rough surface. The why-question seeks an explanation for why the match did not ignite given that under the conditions we would have expected it to ignite. The question has proper form regarding an appropriate foil for the fact.

Why-Questions Criterion for sensibility

Incorrect Question: “Why did the match not ignite in contrast to breaking?”

fact

foil

frictional surface

thumb pressure

Separate causal processes

Here is a why-question of incorrect form. Notice that the fact and foil would trace back to separate and different causal processes. Explaining why the match did not ignite cannot be contrasted with why the match broke. The two conditions refer to entirely different causal processes.

Complete Why-Questions Common cause versus separate causes

Question: “Why are these matches burned, in contrast to unburned?”

fact

foil

There is an additional issue that we need to consider with regard to why-questions. This is an issue that is of importance in systematics because we observe shared features or characters among groups of organisms. When we observe multiple effects that have the appearance of being correlated, we have to decide how to explain such correlations.

Complete Why-Questions Common cause versus separate causes

Question: “Why are these matches burned, in contrast to unburned?”

Common cause explanation

Separate cause explanation

In the example shown here, the correlation of finding a group of burned matches requires that we decide whether to answer the why-question, "Why are these matches burned, in contrast to unburned?", by either a common cause explanation or by way of separate cause explanations.

Complete Why-Questions Common cause versus separate causes

Question: “Why are these matches burned, in contrast to unburned?”

How to decide? – background knowledge

Making a decision to provide a common cause explanation or separate cause explanations requires that one take into consideration their background knowledge regarding such effects. What is important to recognize is that offereng separate cause answers will be based on a set of questions that will be different from what will be required for a common cause answer.

Complete Why-Questions Separate causes

Q1

Q2

Q3

A1

A2

A3

In the case of treating the observation of the burned matches as explainable by way of separate causes, we would treat each effect (i.e. burned match) as leading to separate why-questions and separate respective answers.

Complete Why-Questions Common cause

Q A For a common cause explanation, we regard the correlation would be far less surprising if explained by a single, commmon cause. Hence the single question shown earlier, "Why are these matches burned, in contrast to unburned?"

Complete Why-Questions All Questions Have a Contrastive Form

The contrastive nature of why-questions, plus the reasoning used to answer to those questions, provide the strongest criteria for critically evaluating the methods and procedures used in systematics.

Any critical appraisal of biological systematics must stand on two issues. The first being the form of contrastive why-questions, as we have just seen. The second is that the proper form of our why-questions then lead to inferences of answers to those questions.

The Philosophy of Biological Systematics Course Outline – Part 1

1.

The goal of Science. The goal of biological systematics.

2.

Causal relationships in systematics.

3.

The nature of why-questions.

4.

The three forms of inference: deduction, induction, abduction.

5.

The uses of deduction, induction, and abduction in science.

Now that we have identified that the goals of science and biological systematics are both to acquire causal understanding of the phenomena we encounter, we next need to carefully examine the types of reasoning used in the sciences to achieve our goal. As we will see later in the course, these types of reasoning will play critical roles in attempting to correctly characterize the tasks of systematics.

The Fundamentals of Inference Inference: The act of reasoning from a statement (premise) or statements (premises), to a conclusion or set of conclusions.

This section of the course will focus on identifying the types of reasoning, known as inference, we use every day as well as in the sciences.

Two Types of Inference Have Traditionally Been Recognized Deduction: Inferences in which a conclusion drawn from a set of (true) premises cannot contradict those premises, and therefore must also be true. • All humans are mortal • Kirk is human • Kirk is mortal

Traditionally, when people speak of logic as the study of reasoning, they only make a distinction between two types of reasoning: deductive and inductive. Let's first look at this distinction, before more accurately segregating reasoning. In this example of deduction, notice that the premises, 'All humans are mortal' and 'Kirk is human,' is separated from the conclusion, 'Kirk is mortal,' by a single line.

Two Types of Inference Have Traditionally Been Recognized Induction: Inferences in which similarities are identified between observed objects or events of a given class, and hypothetically extended to unobserved objects or future events of that class. • Kirk is human • Kirk is mortal • All humans are mortal

In the case of an induction (or any non-deductive inference), the premises are separated from the conclusion, or conclusions, by a double line.

Two Types of Inference Have Traditionally Been Recognized Deduction: Inferences in which a conclusion drawn from a set of (true) premises cannot contradict those premises, and therefore must also be true.

Induction: Inferences in which similarities are identified between observed objects or events of a given class, and hypothetically extended to unobserved objects or future events of that class.

Deduction & Induction The Popular View of Their Relations

data

d

deduction

h

induction

deduction

h hypothesis

d

þ

induction

h

(1) Deduction: predictions of potential test consequences derived from the hypothesis to be tested. (2) Induction:

performing the test; observations of test consequences, providing either confirming/ corroborating or disconfirming/falsifying evidence.

This diagram illustrates how people often speak of the relations between deduction and induction in science. Starting with a hypothesis or theory, inferred by way of induction, one uses deduction to predict potential test evidence, then induction is used in the process of testing. The view is that there are cycles of deduction and induction in a continual process of evaluating theories and hypotheses.

Deduction & Induction The Popular View of Their Relations

deduction

Given Hypothesis

Expected Data

induction

Inferred Hypothesis

Actual Data

“Deduction is reasoning from what is in the mind to what is in the world.” “Induction is reasoning from what is in the world to what is in the mind.” H.G. Gauch, Jr. (2003: 160), Scientific Method in Practice

This is another, common view of the relation between deduction and induction in science.

The Structure of Inferences The Basic Components The premises and conclusion(s) of an inference contain statements that can be categorized as three possible forms:

Rule:

a law, empirical generalization, or theory, often stating a relation between cause and effect;

Case:

a statement about a thing(s), or event(s), in the form of causal or initial conditions;

Result:

a statement of a consequence or effect that is related to the ‘Case.’

For our purposes of examining the nature of reasoning that exists throughout biological systematics, we need to make more precise distinctions between the types of reasoning used in science. To compare and contrast the different types of reasoning, we will use a set of statements that can be used as either premises or conclusions. These statements are referred to as Rule, Case, and Result. By identifying premises or conclusions as Rule, Case, and Reult, we will find that in addition to deduction and induction (sensu stricto), we will also have to recognize a third type of non-deductive reasoning, called abduction.

Deduction A Simple Example

Rule:

All marbles in this bag [M] are red [P].

S = subject P = predicate ‘end terms’ M = ‘middle term’

In this example of deduction, as well as in following examples, the components in each of the statements comprising the premises and conclusions are identified as subject, predicate, or 'middle term.' The subject and predicate are sometimes referred to as 'end terms' since in a deductive arrangement they are present in the premises and conclusion. The 'middle term,' which functions as a predicate, then joins together the end terms in the conclusion.

Deduction A Simple Example

Rule:

All marbles in this bag [M] are red [P].

Case:

This marble [S] is from this bag [M & P].

S = subject P = predicate ‘end terms’ M = ‘middle term’

Notice that 'this bag' functions as both the middle term and predicate for the Case.

Deduction A Simple Example

Rule:

All marbles in this bag [M] are red [P].

Case:

This marble [S] is from this bag [M & P].

Result: This marble [S] is red [P].

S = subject P = predicate ‘end terms’ M = ‘middle term’

The predicate 'red' in the Rule, and the subject 'marble' in th Case are brought together in the Result. The middle term, 'this bag,' is only referred to in the premises. In deduction, the middle term serves to bring together the end terms in the conclusion.

Deduction A Simple Example

Rule:

All marbles in this bag [M] are red [P].

TRUE

Case:

This marble [S] is from this bag [M & P].

TRUE

Result: This marble [S] is red [P].

TRUE

S = subject P = predicate ‘end terms’ M = ‘middle term’

Because of the form required of the premises in deduction, if the premises are true, then the conclusion must also be true. In other words, the conclusion is certain.

Deduction

Rule:

The marbles in this bag [M] are red [P].

Case:

This marble [S] is from this bag [M & P].

Result: This marble [S] is red [P].

P S M P

M

S complete inclusion

(a)

(b)

Deduction has a structure wherein the 'middle term' [M] serves to bring together the subject [S] and predicate [P] in the conclusion. This relationship is illustrated here in (a), where the solid lines indicate relations stated in the premises, and the dashed line denotes the relation provided by the conclusion. The Euler diagram in (b) provides another representation of these relations, where deduction is characterized by 'complete inclusion:' the subject [S] is a subset of the middle term [M], and the latter is a subset of the predicate [P].

Induction A Simple Example

Case:

These marbles [S] are from this bag [M & P].

Result: These marbles [S] are red [P].

With induction, the premises are comprised of the Case and Result. Notice that the subject [S] is present in both premises.

Induction A Simple Example

Case:

These marbles [S] are from this bag [M & P].

Result: These marbles [S] are red [P].

Rule:

All marbles in this bag [M] are red [P].

From the premises is concluded the Rule. You might notice that the premises state a limted set of observations, from which a general statement is inferred. In fact, the example looks very similar to a statistical inference, proceeding from observations of a sample to a conclusion about the population from which the sample was taken. As we will see later in the course, induction is the principle mode of reasoning used in statistics. And, since statistics is about testing statistical hypotheses, we will find that induction is the approach taken for testing in general.

Induction A Simple Example

Case:

These marbles [S] are from this bag [M & P].

Result: These marbles [S] are red [P]. Rule:

All marbles in this bag [M] are red [P].

TRUE TRUE TRUE / FALSE

In contrast to deduction, where true premises always guarantee a true conclusion, an inductive conclusion from true premises cannot guarantee a true conclusion. The conclusion is not certain; it is only probable, as determined by the premises. Notice that the conclusion thus makes a claim that goes beyond what is offered by the premises.

Induction

Case:

P

This marble [S] is from this bag [M & P].

M S P

Result: This marble [S] is red [P]. Rule:

The marbles in this bag [M] are red [P].

M

S partial inclusion

(a)

(b)

As shown in (a), induction differs from deduction in bringing together the predicate [P] and middle term [M] in the conclusion by the presence of the subject [S] in both premises. The Euler diagram (b) shows induction to be a matter of 'partial inclusion.'

A Third Type of Inference is Often Recognized

Abduction: Reasoning from observed effects in the present (consequents) to a conclusion(s) of possible cause (or causes) in the past (antecedent). Abduction is also the form of inference used to develop our observation statements. As a result, abductive inference is the most common type of reasoning we use on a daily basis.

In addition to deduction and induction, there is a third type of non-deductive inference that is often recognized, called abduction. Abduction is a form of reasoning we use on a daily basis to infer from observed effects to a possible cause or causes.

A Third Type of Inference is Often Recognized Abduction “[A] hypothesis cannot be admitted, even as a hypothesis, unless it be supposed that it would account for the facts or some of them. The form of inference, therefore, is this:

The surprising fact, C, is observed; But if A were true, C would be a matter of course, Hence, there is reason to suspect that A is true. ”

Charles Sanders Peirce (1839-1914)

While abduction was recognized by Aristotle, it was not until the 19th century that the importance of this type of reasoning was recognized. The most prominent proponent to study the relations of abduction to deduction and induction was Charles Sanders Peirce (pronounced 'Purse'). But, it was not until the second half of the 20th century that philosophers and scientists started to take seriously the importance of abduction.

Abduction A Simple Example

Rule:

All marbles in this bag [M] are red [P].

In abduction, the major premise is the Rule.

Abduction A Simple Example

Rule:

All marbles in this bag [M] are red [P].

Result: This marble [S] is red [P].

The minor premise is the Result. Notice that the predicate, 'red,' appears in both premises.

Abduction A Simple Example

Rule:

All marbles in this bag [M] are red [P].

Result: This marble [S] is red [P].

Case:

This marble [S] is from this bag [M & P].

What you should notice is that the Rule, as a theory, is applied to the Result, where the Result can be regarded as an effect. The conclusion, Case, then has the quality of an explanatory account. In this example, we explain why 'this marble' is red because it came from 'this bag' of red marbles.

Abduction A Simple Example

Rule:

All marbles in this bag [M] are red [P].

Result: This marble [S] is red [P]. Case:

This marble [S] is from this bag [M].

TRUE TRUE TRUE / FALSE

As with any non-deductive inference, the true premises of an abduction do not guarantee the truth of the conclusion.

Abduction

Rule:

P

The marbles in this bag [M] are red [P].

Case:

This marble [S] is from this bag [M].

S

M

Result: This marble [S] is red [P].

M

S

P exclusion

(a)

(b)

The structure of abductive inference is the conjunction of some theory or law-like statement (Rule) and observed effects (Result) to conclude a possible cause (Case). Abduction is sometimes referred to as 'reverse deduction' in that the Case (cause) is concluded from the Rule (theory) and Result (effect), rather than the Result being concluded from the Rule and Case as in deduction. As a result (a), it is the presence of the predicate (P) in both premises which suggests the relation between the subject (S) and middle term (M) in the conclusion. Unlike deduction, which shows 'inclusion,' and induction, which shows 'partial inclusion,' abduction is characterized by 'exclusion' (b).

Relations Between Non-Deductive and Deductive Inference Induction & Abduction

Deduction

• Ampliative: conclusion can imply things not stated in premises.

• Not ampliative: conclusion cannot go beyond what is stated in premises.

• Not necessarily truth preserving: truth of conclusion not guaranteed.

• Truth preserving: conclusion is true if premises are true.

• Support for conclusion by premises can vary in strength.

• Degree of support for conclusion irrelevant - conclusion is either true or false.

• Requirement of total evidence must be considered.

• Requirement of total evidence is satisfied automatically.

There are some fundamental distinctions we need to be aware of between deductive and non-deductive (induction & abduction) reasoning. As we will see, these characteristics are significantly important when examining the types of reasoning used in biological systematics.

AMPLIATIVE REASONING Requirements C Non-monotonic:

allow a certain conclusion to be defeated by inclusion of additional information in premises.

C Cut-off Point Problem:

show that generalizations from observations are justified.

C Vertical Extrapolation:

support conclusions that make reference to entities not referred to in premises.

C Eliminative Dimension:

allow multiple conclusions consistent with premises.

Since non-deductive reasoning is ampliative (see previous slide), there are four characteristics that need to be recognized.

AMPLIATIVE REASONING Requirements

C Non-monotonic:

allow a certain conclusion to be defeated by inclusion of additional information in premises.

Induction

C Cut-off Point Problem: show that generalizations from observations are justified.

Induction / Abduction

C Vertical Extrapolation: support conclusions that make reference to entities not referred to in premises.

Induction / Abduction

C Eliminative Dimension: allow multiple conclusions consistent with premises.

Induction / Abduction

Most of these characteristics apply to both induction and abduction.

The Philosophy of Biological Systematics Course Outline – Part 1 1.

The goal of Science. The goal of biological systematics.

2.

Causal relationships in systematics.

3.

The nature of why-questions.

4.

The three forms of inference: deduction, induction, abduction.

5.

The uses of deduction, induction, and abduction in science.

We are now in a position to examine the specific ways in which deduction, induction, and abduction are used in our processes of scientific inquiry.

Inferences in Science

Deduction & Induction The Popular View of Their Relations

data

d

deduction

h

induction

deduction

h hypothesis

d

þ

induction

h

(1) Deduction: predictions of potential test consequences derived from the hypothesis to be tested. (2) Induction:

performing the test; observations of test consequences, providing either confirming/ corroborating or disconfirming/falsifying evidence.

Recall that earlier we noted that people often speak of the relations of deduction and induction in science, where science is only seen as cycles of deduction and induction in a continual process of inferring and evaluating theories/hypotheses. But in fact, abduction is a fundamental component that we need to take into consideration as completely separate from induction (sensu stricto).

Operational Relations Between Types of Inference in Science

abduction

Inferences of Hypotheses & Theories

deduction

Inferences of Tests

induction

Conducting Tests: Hypothesis Acceptance or Rejection

The actual relations between abduction, deduction, and induction are summarized here. Abduction involves our reasoning process for inferring hypotheses and theories. Deduction is used to derive potential consequences from our hypotheses and theories that might serve as test evidence when the act of testing occurs. Induction is the process of testing that leads to our concluding that a theory or hypothesis is confirmed or disconfirmed.

Fact Hypothesis Theory

What do these terms mean? In order to clearly understand the different types of reasoning we use in science, including biological systematics, we need to first understand the meanings of three words that are commonly used, but too often misunderstood.

Fact

The facts Facts are objects and events. The conditions of truth or falsity do not apply to facts.

A 'fact' is nothing more than an object or event that exists, whether we perceive it or not. It is important not to confuse the observation statement, 'This is a glass of ice water,' with the facts you perceive. The facts exist independent of you. Your observation statement is a conclusion from your inference (abduction!) used to explain the facts. Also, keep in mind that the conditions of truth or falsity cannot be applied to facts. Facts simply are! What can be true or false are your statements regarding those facts.

Fact “...a fact is either the being of a thing in a given state, or an event occurring in a thing. Constructs do not qualify as facts since they are not objects that can be in a certain state, let alone undergo changes of state.... Similarly, there are no 'scientific facts': only a procedure to attain knowledge can be scientific (or not), not the object of our investigation. Accordingly, scientists neither 'collect' facts nor do they come up with or, worse, 'construct' facts, but advance hypotheses and theories referring to or representing facts.” Mahner & Bunge (1997: 34), Foundations of Biophilosophy

The quote shown here is an excellent definition of 'fact,' and corrects a longstanding misconception that we have 'scientific' facts as opposed to 'non-scientific' facts.

Inference of a Theory

Now that we know what facts are, we need to understand the meaning of the term 'theory' and how they are inferred.

Theory An explanatory concept(s), stating cause-effect relations, that we can apply to our sense perceptions, to give us understanding. • theories are spatio-temporally unrestricted. • theories are not limited to the realm of Science.

What is important to notice in this definition is that a theory is a spatio-temporally unrestricted concept. In other words, a theory can be applied to the past, present, and future. It does not refer to a specific instance. And, as you will recall that the goal of science is to increase our causal understanding, theories are the fundamentally important conceptual tools that allow us to pursue that understanding, because theories enable us to infer explanatory hypotheses.

Abductive Inference as the Mechanism for Theory Formation • background knowledge (theories, laws, etc.) • tentative theory of cause-effect relations (adapted from an analogous theory) • observed effects in need of being explained • explanatory hypothesis

The inference of a theory is by way of abduction, and often as a matter of analogy. One takes a previously established theory, and uses it as an analogy for a new theory, where that analogous application serves to explain some set of surprising or unexpected phenomena.

Abductive Inference as the Mechanism for Theory Formation • Background knowledge: variation / inheritance / differential survival and reproduction

• Tentative theory: Based on what is known of the actions of artificial selection, in conjunction with the above background knowledge, maybe an analogous system of cause and effect relations exists in nature: Natural selection - organisms with traits that enhance survival and reproduction will leave offspring with those traits .

• Observations: There are differentially shared traits among these observed organisms.

• Hypothesis: Variation arose in an ancestral population, subsequent to which the traits in question allowed for enhanced survival and reproduction.

A classic example of the combined use of analogy and abductive inference can be found in the development of Charles Darwin's (1859) theory of natural selection.

Inference of a Hypothesis

Hypothesis An explanation of some set of facts, giving us at least initial understanding of what we perceive. • hypotheses are spatio-temporally restricted. • hypotheses are not limited to the realm of Science.

Notice that unlike a theory, which does not refer to specific instances, a hypothesis does present a narrow set of conditions for a particular time and location. In the context of science, the most useful way to characterize hypotheses is as explanatory accounts, the purpose of which is to provide us with causal understanding of an observed effect or set of effects.

Abductive Inference of a Hypothesis • background knowledge • theory (cause-effect relations) • observed effects in need of being explained

• explanatory hypothesis

The schematic example shown here illustrates the most basic components of the abductive inference of a hypothesis. The premises comprise at least one theory that is applied to the effect(s) we wish to explain, from which we conclude an explanatory hypothesis that suggests that the effect(s) is/are the product of particular past causal events that are consistent with the theory.

TESTING: a definition

The inferential process of critically and empirically assessing the ability of theories and hypotheses to give us understanding.

Now that we have examined the basics of inference, including the inferences of hypotheses and theories by way of abduction, we can briefly look at the process of testing. We will address testing in greater detail later in the course, as it applies to the testing of biological systematics hypotheses.

The Two Realms of Science Present (the realm of Observation)

Past

Future Cause

Explanatory Hypothesis

abduction

prediction

Effect

Effect

‘Historical’ Sciences

‘Experimental’ Sciences

Hypothesis testing

Theory testing

Recall the distinction we made earlier between 'historical' and 'experimental' sciences. This will serve to illustrate the difference between the testing of hypotheses and theories.

Testing: Experimental vs. Historical Sciences Present (the realm of Observation)

Past

Future Known Cause (experiment)

Unknown Cause (not observable)

explanation

prediction (deduction via theory)

Effect (potentially observable)

Known Effect

‘HISTORICAL’

‘EXPERIMENTAL’

Hypothesis testing

Theory testing

While the focus of this course will be on explanatory hypotheses in the historical sciences, most discussions about testing use examples from the experimental sciences. There are some important differences between these fields regarding the nature of testing, that need to be mentioned. What is of principle interest in the experimental sciences is testing by way of controlled experiments. A theory is tested by providing controlled (e.g. experimental) causal conditions in the present. In other words, the causal conditions are known to us. It is then a matter of observing whether or not a predicted effect occurs. What you will notice is that both cause and effect can be observed. We have the opportunity to know both. But, in the case of the historical sciences, what we know in the present are effects that are in need of being explained. The difficulty is that the cause that explains observed effects occurred in the past so no longer exists in the present. As a result, the cause is often unknown and unobservable.

Testing: Experimental vs. Historical Sciences Present (the realm of Observation)

Past

Future Known Cause (experiment)

Explanatory Hypothesis

explanation

prediction (deduction via theory)

Effect (potentially observable)

Known Effect

‘HISTORICAL’

‘EXPERIMENTAL’

Hypothesis testing

Theory testing

Thus, we infer an explanatory hypothesis to account for the observed effects. It is this hypothesis that we then want to test. But, in comparison to the experimental sciences, where the relations between cause and effect can both be known, the fact that a past causal event is usually not known can make it very difficult to test explanatory hypotheses since the relevant effects needed for a test might not be available. We will see in this course that this limitation certainly applies to the testing of many biological systematics hypotheses.

Testing Explanatory Hypotheses

Present (the realm of Observation)

Past Explanatory Hypothesis

Specific Causal Condition(s)

Future abduction

deduction

Known Effect

test that should be performed

induction

testing of hypothesis by observations of effects

We can now summarize the relations between the abductive inference of an explanatory hypothesis and the subsequent testing of that hypothesis. It is from effects observed in the present that we infer by way of abduction an explanatory hypothesis. From the specific causal conditions stated in that hypothesis we deduce effects that should be observed that are only possible because the specified causal conditions that occurred in the past would allow for those effects. The deduction of such effects provides the basis for the tests that need to be performed. The act of testing the hypothesis is, however, a matter of induction, where the hypothesis is either accepted or rejected on the basis of searching for the specified test evidence. Since no test can guarantee the truth of a hypothesis, and a disconfirmed hypothesis simply leaves us with alternative hypotheses to consider, testing is always inductive.

I. ABDUCTION The Inference of Hypotheses

Hypotheses

Abduction: causal theory + observed effects

background knowledge + causal theory

‘Why...?’

Observed Effects

Let's now look a very simple summary of the relations between abduction, deduction, and induction. In this slide, the abductive inference of hypotheses is presented.

II. ABDUCTION The Inference of New Hypotheses: additional abductive inferences are required when new effects are observed

Hypotheses

Additional Effects

Very often, subsequent to inferring a hypothesis (or hypotheses), we encounter additional effects or observations that also need to be explained in the same manner as the previous effects.

II. ABDUCTION The Inference of New Hypotheses: additional abductive inferences are required when new effects are observed

Hypotheses

Additional Effects

‘Why...?’

New Set of Observed Effects (old + new)

The result is that these additional effects/observations need to be included with previous observations.

II. ABDUCTION The Inference of New Hypotheses: additional abductive inferences are required when new effects are observed

New Hypotheses Additional Effects

Abduction: causal theory + observed effects

background knowledge + causal theory

‘Why...?’

New Set of Observed Effects (old + new)

Then, a new abductive inference is performed that leads to new hypotheses, that replace the previous hypotheses, and provide us with updated explanations of our observations.

III. Ranking Hypotheses

Determine Which Hypotheses are to be Tested

ranking

Hypotheses

In some fields of science, these hypotheses might be ranked by way of some criterion, in order to determine which hypotheses will be tested first. For example, the time and/or expense involved with testing might make it more feasible to test some hypothesese as opposed to others.

IV. DEDUCTION: Predicte Possible Test Consequences

If a hypothesis provides a sufficiently detailed account of past causal conditions, then it should be possible to predicte consequences, as effects, that are related as specifically as possible to those causal conditions stated in the hypothesis.

ranking

Hypotheses

Deduction: predicted test consequences

Depending on which hypothesis or hypotheses are to be tested, the next step is to deduce from each hypothesis potential test consequences. Recall that an explanatory hypothesis presents claims that specific causal conditions existed in the past that account for the effects in the present that are in need of being explained. Thus, the potential test consequences that we might deduce from the stated causal conditions would have to be evidence as closely associated as possible with those conditions. In other words, we would want to seek effects with the lowest probability of occurrence if the stated causal conditions did not occur. We will return to this issue several times during this course.

V. INDUCTION: Hypothesis Testing Determine whether or not predicted consequences are the case.

ranking

Hypotheses

Abduction:

Deduction:

causal theory + observed effects

predicted test consequences

background knowledge + causal theory

Induction: observed test consequences, leading to one of the following conclusions –

‘Why...?’

Observed Effects

(a) hypothesis confirmation, (b) hypothesis revision, or (c) hypothesis rejection.

Finally, there is the actual act of testing. This entails putting oneself in a position to witness the necessary conditions for determining whether or not the predicted test consequences are the case. If the predicted test consequences are observed, then the hypothesis is said to be confirmed. If consequences other than what were predicted are observed, or only some test evidence is observed, then the hypothesis might be disconfirmed or in need of revision, respectively. In these cases, these observations might need to be integrated into the set of previous observed effects and a new abductive inference to a new or revised hypothesis is accomplished. Then, the process of testing can be performed again.

An Example – Hypothesis Confirmation Hypothesis: John was in my yard, throwing the ball, and it broke the window.

Abduction:

Deduction:

(a) John sometimes practices pitching in my yard, and he was there earlier today, (b) broken window + ball on the floor.

background knowledge + causal theory

John’s finger prints should be on the ball.

Induction: Observed Consequences – Finger prints are on the ball, that match John.

‘Why...?’

Therefore, the hypothesis has been confirmed.

Observed Effects: (a) my window is broken, (b) there is a ball on the floor.

Here is a simple, everyday example that illustrates how we use abductive inference to develop an explanatory hypothesis, deduce potential test evidence to evaluate that hypothesis, and then carry out testing the hypothesis. In this instance, the observed test evidence confirms the hypothesis.

An Example – Hypothesis Confirmation Hypothesis: John was in my yard, throwing the ball, and it broke the window.

Deduction: John’s finger prints should be on the ball.

* Notice that the evidence (observed consequences) confirming/ supporting the hypothesis suggests, but does not guarantee, the hypothesis is true. Acceptance of the hypothesis as confirmed is inductive.

Induction: Observed Consequences – Finger prints are on the ball, that match John.

Therefore, the hypothesis has been confirmed.

An Example – Hypothesis Revision, Part I Hypothesis: John was in my yard, throwing the ball, when it broke the window.

Abduction: (a) John sometimes practices pitching in my yard, and he was there earlier today, (b) broken window + ball on the floor.

background knowledge + causal theory

‘Why...?’

Deduction: John’s finger prints should be on the ball.

Induction: Observed Consequences – Finger prints are on the ball, that match John, but there is also red dirt on the ball. Therefore, the hypothesis has been partially confirmed, but needs to be revised.

Observed Effects: (a) my window is broken, (b) there is a ball on the floor.

In this case, however, the observed test evidence is not entirely what was predicted. There are additional observations made during the test that suggest that the hypothesis needs to be revised.

An Example – Hypothesis Revision, Part I Hypothesis: John was in my yard, throwing the ball, when it broke the window.

* Notice that while the evidence (i.e. observed consequences that John’s finger prints are on the ball) does confirm/ support the hypothesis, observing red dirt on the ball during the test suggests that the hypothesis must be revised. Red dirt only occurs at the baseball field near my house, suggesting that John was not throwing the ball in my yard, but rather was at the baseball field, which slightly alters the causal conditions stated in the hypothesis. A revision of the hypothesis then occurs through the subsequent inference of a new abduction based on a new set of observed effects, as is shown in the next diagram (Hypothesis Revision, part II).

Deduction: John’s finger prints should be on the ball.

Induction: Observed Consequences – Finger prints are on the ball, that match John, but there is also red dirt on the ball. Therefore, the hypothesis has been partially confirmed, but needs to be revised.

An Example – Hypothesis Revision, Part II Revised Hypothesis: John was at the baseball field next door, when he threw a wild pitch, and it broke the window.

New Abduction: (a) John sometimes practices pitching at the baseball field next door, (b) broken window + ball on the floor + John’s finger prints + red dirt on ball.

additional background knowledge + causal theory

‘Why...?’

New Observed Effects: (a) my window is broken, (b) there is a ball on the floor, (c) John’s fingerprints are on the ball, (d) there is red dirt on the ball.

Induction: Observed Consequences – Finger prints are on the ball, that match John, but there is also red dirt on the ball. Therefore, the hypothesis has been partially confirmed, but needs to be revised.

Thus, we need to integrate the new observations obtained during the previous test with our previous observations, and then infer a revised hypothesis.

An Example – Hypothesis Revision, Part II Revised Hypothesis: John was at the baseball field next door, when he threw a wild pitch, and it broke the window.

New Abduction: (a) John sometimes practices pitching at the baseball field next door, (b) broken window + ball on the floor + John’s finger prints + red dirt on ball.

additional background knowledge + causal theory

‘Why...?’

New Observed Effects: (a) my window is broken, (b) there is a ball on the floor, (c) John’s fingerprints are on the ball, (d) there is red dirt on the ball.

* The test of the hypothesis, presented in part I, resulted in the observation of effects that confirm/support the claim that John was a causal factor in the breaking of my window. But, there was also the additional observed effect that red dirt is on the ball, suggesting that, contrary to my initial hypothesis that John was not in my yard, he was at the nearby baseball field. This new observed effect needs to be considered in a new abductive inference to a revised explanatory hypothesis. What is important to notice is that the test that lead to the confirmation that John was responsible for the window being broken, is still inductive for it continues to suggest, but not guarantee, the truth of past events. Induction: Observed Consequences – Finger prints are on the ball, that match John, but there is also red dirt on the ball. Therefore, the hypothesis has been partially confirmed, but needs to be revised.

An Example – Hypothesis Rejection, Part I Hypothesis: John was in my yard, throwing the ball, when it broke the window.

Abduction: (a) John sometimes practices pitching in my yard, and he was there earlier today, (b) broken window + ball on the floor.

background knowledge + causal theory

‘Why...?’

Deduction: John’s finger prints should be on the ball.

Induction : Observed Consequences – Finger prints are on the ball are from Bob, not John.

Therefore, the hypothesis has been disconfirmed.

Observed Effects: (a) my window is broken, (b) there is a ball on the floor.

Here is an alternative outcome to the test, showing an instance of disconfirmation of a hypothesis. While it was predicted that John's finger prings would be found on the ball, the actual examination of finger prints reveals that they are from Bob. The evidence observed is different from what was predicted, thus the hypothesis has been disconfirmed.

An Example – Hypothesis Rejection, Part I Hypothesis: John was in my yard, throwing the ball, when it broke the window.

* Notice that the observation of consequences that were not predicted has the form of modus tollens: ‘If p, then q; not-q; therefore not-p.’ While this deductive form allows for disconfirming the hypothesis, the observed effects during the test provide the basis for the next abductive inference. The act of testing suggests that different causal events occurred. But while the original hypothesis has been rejected, a significant part of the test is that an alternative hypothesis needs to be considered. And again, this indicates that the test is inductive.

Deduction: John’s finger prints should be on the ball.

Induction : Observed Consequences – Finger prints are on the ball are from Bob, not John.

Therefore, the hypothesis has been disconfirmed.

An Example – Hypothesis Rejection, Part II New Hypothesis: Bob was in my back yard, throwing the ball, when it broke the window.

New Abduction: (a) Bob sometimes practices pitching in my yard, (b) broken window + ball on the floor + Bob’s finger prints.

additional background knowledge + causal theory

‘Why...?’

New Observed Effects: (a) my window is broken, (b) there is a ball on the floor, (c) Bob’s fingerprints are on the ball.

Induction: Observed Consequences – Finger prints are on the ball are from Bob, not John. Therefore, the hypothesis has been disconfirmed.

The test observations provide the basis for another abductive inference.

The new evidence obtained during the test of the old hypothesis must now be considered as part of the total evidence, and it is from this revised set of observations that we would then engage in a new abductive inference to an entirely new hypothesis. This new hypothesis would then be available for a new process of testing.

An Example – Hypothesis Rejection, Part II New Hypothesis: Bob was in my back yard, throwing the ball, when it broke the window.

New Abduction: (a) Bob sometimes practices pitching in my yard, (b) broken window + ball on the floor + Bob’s finger prints.

additional background knowledge + causal theory

‘Why...?’

New Observed Effects: (a) my window is broken, (b) there is a ball on the floor, (c) Bob’s fingerprints are on the ball.

* Subsequent to the rejection of the hypothesis (see previous diagram) as an explanation of the broken window, we need to consider the new observations that were made during testing, i.e. the presence of Bob’s finger prints on the ball. These new observations are combined with previous observations to give us a new set of effects in need of explanation. These effects are then conjoined with a causal theory regarding Bob, and pitching, to abductively infer a new explanatory hypothesis.

Induction: Observed Consequences – Finger prints are on the ball are from Bob, not John. Therefore, the hypothesis has been disconfirmed.

The test observations provide the basis for another abductive inference.

Deduction & Induction Traditional Relations data

d

deduction

h

induction

d

deduction

h

induction

h

hypothesis

(1) Deduction: predictions of potential test consequences derived from the hypothesis to be tested. (2) Induction:

performing the test; observations of test consequences, providing either confirming/corroborating or disconfirming/ falsifying evidence.

So, the standard model that we discussed earlier, where the activity of scientific inquiry is one that alternates between deduction and induction, is not accurate.

Abduction, Deduction & Induction A More Accurate View of Their Relations d

data

abduction 1

h

h

d abduction 1

h

hypothesis

(1) Abduction1: observed, unexpected or surprising effects provide part of the premises for abductions to new hypotheses.

As we have seen, abduction is a fundamentally important part of scientific inquiry. We are continually confronted with surprising or unexpected observatoins that are in need of being explained. Abductive inference is the process of providing hypotheses that give us at least inital understanding of what we observe. In other words, we attempt to make the surprising not so surprising by bringing it into the realm of what is already familiar to us, or what we already understand, via our established theories.

Abduction, Deduction & Induction A More Accurate View of Their Relations d

data

d

abduction 1 deduction

h

abduction 1

deduction

h

h

hypothesis

(1) Abduction1: observed, unexpected or surprising effects provide part of the premises for abductions to new hypotheses. (2) Deduction: predictions of potential test consequences derived from the hypothesis to be tested.

Deduction then serves to predict potential test evidence that can serve to evaluate the hypothesis. Recall, however, that deduction is not the actual act of testing - that is a process that is inductive.

Abduction, Deduction & Induction A More Accurate View of Their Relations data

d

d

abduction 1 deduction

h

induction

deduction

h

abduction 1 induction

h

hypothesis

(1) Abduction1: observed, unexpected or surprising effects provide part of the premises for abductions to new hypotheses. (2) Deduction: predictions of potential test consequences derived from the hypothesis to be tested. (3) Induction:

performing the test; observations of test consequences, providing either confirming or disconfirming evidence.

Then, the act of seeking the predicted test evidence is a matter of induction, where the observations of the consequences of the test serve as part of the premises from which one concludes that the hypothesis being tested has either been confirmed or disconfirmed, or is in need of revision.

Abduction, Deduction & Induction A More Accurate View of Their Relations data

d

d

abduction 1 deduction

induction

deduction

induction

abduction 2

abduction 2

h

abduction 1

h

h

hypothesis (1) Abduction1: observed, unexpected or surprising effects provide part of the premises for abductions to new hypotheses. (2) Deduction: predictions of potential test consequences derived from the hypothesis to be tested. (3) Induction:

performing the test; observations of test consequences, providing either confirming or disconfirming evidence.

(4) Abduction2: in the case that test evidence disconfirms a hypothesis, then this new information could provide part of the premises for abductions to new or revised hypotheses.

And finally, in the case of a hypothesis being disconfirmed, then we have new observations from the test that will then need to be considered as part of the premises for a new abductive inference to a new or revised hypothesis.

Beware of ‘Normal Science’

“...‘normal science’ means research firmly based upon one or more past scientific achievements, achievements that some particular scientific community acknowledges for a time as supplying the foundation for its further practice.” T.S. Kuhn (1970: 10), The Structure of Scientific Revolutions

Now that we have examined the goal of scientific inquiry, and established that biological systematics must be consistent with that goal, and that there are three recognized classes of reasoning used in the sciences, we need to be cautious of what is known as 'normal science.' 'Normal science' is a phrase coined by philosopher of science, Thomas Kuhn, to describe the day-to-day activities of scientists. This involves the routine and accepted protocols within a particular field of science that practioners use in the course of inquiry.

‘Normal Science’

‘Scientific Revolution’

factual discoveries

paradigm shift(s) conceptual

methodological

What might be regarded as 'normal science' in the present was at some point in the past part of some 'scientific revolution' that replaced what was earlier the standard of 'normal science.' Kuhn referred to these moments of scientific revolution as paradigm shifts, where a scientific community comes to a point of being faced with conceptual and/or methodological changes, that result in a new and somewhat different stage of 'normal science.' A classic example of a conceptual paradigm shift was Charles Darwin's introduction of his theory of natural selection. More specifically, within biological systematics, we might regard Willi Hennig's emphasis on distinguishing para- from monophyly, and the sole advocation of the latter, as a conceptual paradigm shift that especially started after 1966. And with the advent of computer algorithms we have seen methodological paradigm shifts with respect to the inferences of phylogenetic hypotheses, as cladograms.

“‘Normal’ science... is the activity of the non-revolutionary, or more precisely, the not-too-critical professional: of the science student who accepts the ruling dogma of the day; who does not wish to challenge it; and who accepts a new revolutionary theory only if almost everybody else is ready to accept it – if it becomes fashionable by a kind of bandwagon effect. To resist a new fashion needs perhaps as much courage as was needed to bring it about.” K. Popper (1970: 52), Normal science and its dangers. In: Criticism and the Growth of Knowledge

But it is instructive to consider the warning offered by the philosopher of science, Karl Popper, who tells us that 'normal science' can be dangerous. This danger derives from workers passively and uncritically accepting conceptual or methodological approaches. The danger is increased within a scientific community if practitioners have little or no understanding of the more general foundations by which scientific inquiry is supposed to follow. The result can be the development of conceptual and methological protocols that are inconsistent and maybe even contradictory to established scientific inquiry. Ineed, what we will see in much of this course is that while biological systematics has developed its own community of 'normal science,' most of it has been developed in a vacuum, largely isolated from the basic principles of inference and testing that are required in the sciences.

Systematics as a ‘Normal Science’

Finding answers without consideration of questions.

One of the most significant dangers of the 'normal science' of systematics that we see today is that it is a field of study in which the routine is to seek answers, e.g. cladograms and taxa, without actually considering the questions we are supposed to be asking. Throughout this course we will attempt to link our why-questions with the inferences of answers to those questions. It is this relation between questions and answers, between evidence and hypotheses, that forms the foundation for critically assessing biological systematics.

The Philosophy of Biological Systematics Course Outline – Part 2

1.

Systematics involves abductive inference.

2.

Inferences of systematics hypotheses, i.e. taxa.

3.

Some implications for “phylogenetic” methods.

In Part 1 of this course, we first identified that the goal of all the sciences, including biological systematics, is to continually acquire causal understanding of the objects and events we encounter. We then examined the three fundamental classes of reasoning used in the sciences, as well as everyday life. Among these classes of reasoning, we found that abduction is the most common. What we will now find throughout the remainder of this course is that abduction is the mode of reasoning that binds together all of biological systematics. This means we will need to carefully examine how abductive inference is used in all facets of systematics, and from this we can discover some significant implications for nearly all systematics methods.

SOME OF THE RELATIONSHIPS WITHIN BIOLOGICAL SYSTEMATICS

When we speak of ‘relationships,’ we mean causal relationships. Determining the status of species, and all taxa, can best be accomplished by first recognizing the basic unit to which these causal relationships refer. Then, examining the inferential basis for each of these classes of causal relationships.

To begin our examination of these issues, we need to understand what we mean when we speak of 'relationships' in biological systematics. We use the term relationship on a regular basis, but the word is often not clearly understood when it is used in systematics. We first need to recognize that when we speak of relationships, we are speaking of causal relations. For example, we say we are related to our parents, we are related to our sisters or brothers, we are related to our grand parents. In every instance, the relations we are talking about are causal relations, because it is that type of relationship that gives one understanding. And, as we will see in the rest of this course, the units to which those causal relationships refer are individual organisms. Then, we can specifically look at the way in which we infer each of the types of causal relationships that are used in biological systematics. And again, it needs to be emphasized that it is causal relationships that we are interested in, because it is those types of relations that best serve the overall goal of doing science.

(1913-1976)

Hennig, W. 1966. Phylogenetic Systematics

Recall that we saw earlier that one of the best examinations of the nature of causal relationships in systematics can be found in Willi Hennig's (1966) book, Phylogenetic Systematics.

Classes of Relationships 1. ontogenetic

6

2. cyclomorphic 3. sexual dimorphic

7 4

2

4. tokogenetic 5. polymorphic 6. specific

1

3 5

Hennig, W. 1966. Phylogenetic Systematics

7. phylogenetic

Each of these classes of relationships refer to the different classes of explanatory hypotheses we call taxa.

As was noted in Part 1 of this course, Hennig's (1966) well known figure 6 depicts the fundamental classes of relationships used in biological systematics. Clearly, the only way to interpret the relationships represented in this diagram is in a causal context.

Observation statement

Intraspecific hypothesis

Tokogenetic hypothesis

present

Species hypothesis

Phylogenetic hypothesis A-us a-us

adult b-us

(semaphoront)

Cyclomorphic hypothesis

juvenile (semaphoront)

embryo (semaphoront)

%&

Ontogenetic hypothesis

Sexual dimorphic hypothesis

This diagram is redrawn from Hennig's (1966) fig. 6. Notice that when we refer to a cladogram, it summarizes at least two classes of explanatory hypotheses: specific (species) and phylogenetic, as are named here with formal names, A-us, a-us, and b-us. But what is especially clear is that how we obtain such diagrams is not by way of regarding species as classes or individuals. Each of the branches in this diagram are the products of particular inferential actions as reactions to observing in the present particular characters of organisms, and attempting to answer why-questions related to those observations by way of past causal events. What is also important to remember is that in addition to the cladogram in this diagram, there are a wide range of explanatory hypotheses used in biological systematics that assist us in understanding the organisms we observe. What exists in nature are organisms, not taxa, not species. And, what might be obvious at this point is that to say taxa are shown here is to say we have hypotheses addressing particular questions regarding the organisms we have observed.

Causal Relationships (Taxa) in Biological Systematics Some preliminary examples

Let's now look at very simple examples of how we actually engage in the abductive inferences of some of the causal relationships we just examined in the diagrams.

Observation statement

Inference of Observation Statement

First we will consider how we infer an observation statement, which is itself an explanatory hypothesis.

Inference of Observation Statement

“Why do I have these sense data about this object?” sense data

Let's say we observe the organisms shown here. At each instance that we perceive an object, such as the individual circled here. We might then ask why we have these particular sense perceptions or sense data in our brains? In other words, we are asking for an explanation that provides the cause of our sense data.

Inference of Observation Statement Observation statement (=Perceptual hypothesis)

“Why do I have these sense data about this object?” sense data theory of perception + sense data

ˆ “Because this object exists separate from me.”

We therefore infer, by way of abduction, an answer to our why-question that says the object exists separate from ourselves. In other words, we explain the sense data in our brains by our observation statement that this object does in fact exist, rather than giving an alternative explanation. For example, that we might be hallucinating.

Inference of Ontogenetic Hypothesis (Semaphoront)

present

“Why is this individual an adult?”

Another type of question that might be asked regarding this individual is why it shows the characters of being an adult.

Inference of Ontogenetic Hypothesis (Semaphoront)

present

adult (semaphoront)

juvenile (semaphoront)

embryo (semaphoront)

“Why is this individual an adult?” theory of ontogeny + observation statement

ˆ “Because it is the product of ontogeny.”

The explanation would be that this individual is the product of an ontogenetic process, that could be illustrated as shown here. We place our observations of this individual into the context of what we know of the life history of these types of organisms, based on our application of some ontogenetic theory to our observations. And this ontogenetic hypothesis then gives us understanding of some of the particular characters we observe of this individual. The individual is what Willi Hennig (1966) referred to as a semaphoront.

Inference of Tokogenetic Hypothesis

present

“Why does this individual have black spots?”

A third type of why-question we often ask is in relation to unique characters of a particular individual. For instance, the black spots we only observe on this individual, as opposed to no spots on other individuals.

Inference of Tokogenetic Hypothesis

Tokogenetic hypothesis

present

“Why does this individual have black spots?”

theory of tokogeny + observation statement

ˆ “This individual has black spots because the condition was inherited, as the product of past interbreeding events.”

The relevant explanation could then be one pointing out the inheritance of the feature of black dots as the result of past reproductive (interbreeding) events (= tokogeny). For example, the type of relations are tokogenetic, that exist between parents and the offspring that are produced as a result of reproductive events.

Inference of a Specific (Species) Hypothesis present

“Why do these individuals have antennae in contrast to a smooth dorsum?”

A more general question we might ask is in regard to characters shared among a group of individuals. For instance, why do these individuals have antennae in contrast to a smooth dorsum, as is the case in other individuals we have observed? Notice again that we are asking this why-question because it is surprising or unexpected to find specimens with antennae; because from our past experience, we would expect to see more individuals with a smooth dorsum. So we have new observations that are in need of being explained.

Inference of a Specific (Species) Hypothesis present

Species hypothesis (b-us)

“Why do these individuals have antennae in contrast to a smooth dorsum?” species theory (mutation + tokogeny + fixation)

ˆ “These individuals have antennae because the character originated in the population in the past, and became fixed in the population.”

The answer to this question could be in the form of a hypothesis that antennae arose in the past in a reproductively isolated population and as a consequence of natural selection the feature eventually became fixed throughout the population. What is important to recognize here is that we inferred this answer to the question by applying at least three different theories: mutation, tokogeny, and natural selection. And it is this combination of theories that we would refer to by the shorthand phrase of specific or species theory. This hypothesis is illustrated in the diagram shown here, and is the class of hypothesis we typically call species. Notice that this diagram would traditionally be referred to as a lineage. But, all we are doing is providing a summary of past causal events that explain our present observations. Once again, it is important to stress that what we here call a species, is nothing more than a particular class of explanatory hypothesis.

Inference of a Polymorphism Hypothesis

present

Species hypothesis (b-us)

“Why are individuals in this population polymorphic for body color?”

A somewhat less general question we might ask is in regard to the variations among characters shared among a group of individuals. Here we might ask why members of this population are polymorphic for body color?

Inference of a Polymorphism Hypothesis Intraspecific hypothesis present

Species hypothesis (b-us)

“Why are individuals in this population polymorphic for body color?” polymorphism theory (mutation/pleiotropy + tokogenetic)

ˆ “The polymorphism originated in the past in the population with the introduction of red, and the two conditions have been maintained during tokogeny.” The answer to this why-question would be by way of another class of hypothesis, pointing to the fact that subsequent to the introduction of the red condition because of some mutation, both red and blue forms have been maintained by way of tokogeny. What we might call a theory of polymorphism actually means we are applying at least two fundamental theories to our observations in order to answer this question. The hypothesis would then be illustrated as shown in the diagram, and is the class of hypothesis we often call intraspecific.

Inference of a Phylogenetic Hypothesis a-us present

b-us

“Why do these individuals have appendages as opposed to a smooth ventrum?”

And finally, we have the most general class of hypothesis we use in systematics: phylogenetic. We notice that all of the individuals we observe here have ventral appendages, whereas all other types of individuals we have seen have a smooth ventrum.

Inference of a Phylogenetic Hypothesis a-us present

b-us

“Why do these individuals have appendages as opposed to a smooth ventrum?” phylogenetic theory (mutation + tokogeny + fixation + population splitting)

Phylogenetic hypothesis (A-us)

ˆ “These individuals have appendages because the character originated in the past and became fixed in the ancestral population, followed by a splitting of the population .”

Answering the question of why these individuals have appendages in contrast to a smooth ventrum could be in the form of what we usually refer to as 'descent with modification,' followed by the process of population splitting. Unlike a species theory, which at a minimum relies upon the conjunction of theories of mutation, tokogeny, and character fixation, a phylogenetic-based inference also applies a theory of population splitting. This splitting event is traditionally referred to as 'speciation,' but this term is not an accurate description because it implies that species are things that come into existence, which as we will see, is not the case. The hypothesis that is our answer to the why-question is represented by the illustration. And it would be this type of hypothesis we typically refer to as being a phylogenetic hypothesis.

Inference of a Phylogenetic Hypothesis a-us present

b-us

“Why do these individuals have appendages as opposed to a smooth ventrum?” phylogenetic theory (mutation + tokogeny + fixation + population splitting)

Phylogenetic hypothesis (A-us)

ˆ “These individuals have appendages because the character originated in the past and became fixed in the ancestral population, followed by a splitting of the population .” A-us a-us

b-us

This phylogenetic hypothesis is just one part of the diagrams we refer to as cladograms.

Causal Relationships (Taxa) in Biological Systematics If the goal of biological systematics is to provide causal explanations for the phenomena of differentially shared characters among organisms, then... the inferential structure of almost all of systematics is ABDUCTIVE.

From the simple examples we have just seen, it should be apparent that since the goal of biological systematics is consistent with the goal of all scientific inquiry, then the most important class of reasoning throughout all systematics is going to be abduction.

The Philosophy of Biological Systematics Course Outline – Part 2

1.

Systematics involves is abductive inference.

2.

Inferences of systematics hypotheses, i.e. taxa.

3.

Some implications for “phylogenetic” methods.

We are now at a point that we can examine in greater detail the actual structure of abduction used in biological systematics to infer the hypotheses that we typically refer to as taxa. The main focus here will be on the inferences of specific (species) and phylogenetic hypotheses.

A Formal Definition of TAXON

Any of a set of classes of hypotheses used in biological systematics for the purpose of explaining particular characters of observed organisms.

First, a formal definition of the word taxon. This definition is consistent with the goal in any field of science, which is that we attempt to acquire causal understanding by way of our hypotheses explaining our observations of phenomena. The observations we make relative to systematics are of organisms, and our whyquestions are in regard to the properties of those organisms. A taxon is just a term referring to a particular class of explanatory hypotheses, and our systematics hypotheses are our attempts at answering certain of our why-questions.

Abduction: The Inference of Explanatory Hypotheses The inferential structure that leads to all taxa (including species)

• background knowledge • causal theory, stating relations between particular causes and effects • observed effect(s) in need of explanation

• hypothesis of possible past causal condition(s)

= a taxon All of the classes of relationships Hennig (1966) referred to in the diagram presented earlier comprise hypotheses that are derived by way of abduction. Abductive inference begins when we have surprising or unexpected observations that are in need of explanation. We then ask why-questions as the start to understanding what we encounter. What is shown here is what we have already discussed, that an abductive inference consists of observed effects in need of explanation. We apply a particular theory, or theories, to those effects. It is here that we assume that a theory that has been successful in the past at giving us causal understanding will be useful in this instance. It is from these premises that we conclude an explanatory hypothesis that states at least tentative past causal conditions that account for the observed effects. In other words, we have an initial answer to the why-question regarding the observed effects. In the case of biological systematics, we often refer to these hypotheses as 'taxa.'

Inferences of Taxa

Specific (Species) Hypotheses

Let's now look in greater depth at the point of view that, like all taxa, species are nothing more than abductively-inferred explanatory hypotheses.

What are Species? species A

species B

The Solution to the ‘Species Problem’ (actually, the problem is worse than you think!)

Species... We speak of them all the time. Whether as scientists or in our everyday lives, we and most of society refer to species. We talk of endangered species, species diversity, and the conservation of species. But there is a little problem that we too often ignore or do not want face. The problem is that we do not all mean the same thing when we use the word species. For several years now, part of my research has focused on the question of what are species, and can a formal definition be provided that solves what is often referred to as 'the species problem.'

1928

2003

1957

2010

For much of the 20th century, and now into the 21st century, we have seen books and other publications trying to determine the nature of species. Here are four examples from the past 80+ years, each with the same title, from the large literature on the subject, all attempting to solve 'the species problem.' Yet, at the present time, biologists do not have a consensus on what they mean by the term species.

My own attempt to deal with the issue of species began in 2005, when I published a small paper in the journal Marine Ecology of what I thought was a novel solution. In this paper, I pointed out that species are actually just one of the many hypotheses we infer to help understand particular properties or characters of organisms.

And later I published a larger paper in the journal Acta Biotheoretica that provides more details on how my definition of species is related to the overall goal of doing biological systematics. It was in this paper that I showed that all aspects of biological systematics are derived from the same type of reasoning - all with the purpose of meeting our goal as scientists, which is to acquire causal understanding of observations of organisms.

In: The Species Problem: Ongoing Issues (in press) And this chapter, to appear in an edited book in early 2013, expands further on the subject of species being explanatory hypotheses. I point out that the one term species entails at least five separate classes of explanatory hypotheses, which we will examine later. Interestingly, because of the nature of some of these hypotheses, phylogenetic-level hypotheses cannot also be applied to the same organisms.

“[DNA barcoding] provides a way to identify the species to which a plant, animal or fungus belongs.”

Especially with the growing interest in the procedure known as DNA barcoding, which claims that with nucleotide sequences we can place specimens in the appropriate species, it is obvious that the biological community needs to seriously address the question of what we mean when we use the term species. If we do not agree on what we mean by species, then there are no scientific advantages to spending large amounts of money and time using technology that will not satisfy our goals as scientists. Unfortunately, we seem to be more excited about applying technology to provide us with what we think are answers. But, the reality is that we too often do not have a clear idea of what why-questions we are actually asking.

Indeed, I pointed out a few years ago that DNA barcoding is lacking in clear scientific justification because of the very fact that biologists have not yet agreed on what they mean when they use the word species. Once again, the consequence is that we are producing a lot of results, but we do not talk about what why-questions those results are supposed to be answering, and if we are even asking the right questions.

Problem One Class versus individual

• If species are classes – organisms are members based on their characters • If species are individuals – organisms are just the parts that make the whole

To consider a definition of species, there are two fundamental problems we first need to recognize, as these have contributed to misunderstandings of the nature of species and other taxa. The first problem is that discussions about species have mainly focused on deciding whether species are classes or individuals. The arguments that are usually presented state that... ...if species are classes, then organisms are assigned as members of a class according to particular characters; ...alternatively, if species are individuals or things that exist in nature, then organisms form the parts that make up a species. The view that species are individuals or things is probably the most common opinion among systematists. We very often hear people speak of species as though they are objects that exist in time and space, beyond the organisms we observe on a daily basis.

Problem Two: Species -Take Your Pick! Species ‘concepts’ rather than definition 1.

Agamospecies

12.

Hennigian

2.

Biological

13.

Internodal

3.

Cohesion

14.

Morphological

4.

Cladistic

15.

Non-dimensional

5.

Composite

16.

Phenetic

6.

Ecological

17.

Phylogenetic

7.

Evolutionary Significant Unit

18.

Polythetic

8.

Evolutionary

19.

Recognition

9.

Genealogical Concordance

20.

Reproductive Competition

10. Genetic

21.

Successional

11. Genotypic Cluster Definition

22.

Taxonomic

1997

The second problem that has contributed to misunderstanding the nature of species is that emphasis is almost always placed on species 'concepts' rather than a formal definition of the term species in terms of our reactions to observations of organisms. The consequence is that we have over 20 'concepts' that tell us what species are supposed to be, but they do not give us a formal definition.

Species - Take Your Pick! Species ‘concepts’ rather than definition

26 species ‘concepts’

Wilkins (2009)

In the recent book by Wilkins, the number of species 'concepts' has increased to 26.

Problem Two: Species -Take Your Pick! Species ‘concepts’ rather than definition 1.

Agamospecies

12.

Hennigian

2.

Biological

13.

Internodal

3.

Cohesion

14.

Morphological

4.

Cladistic

15.

Non-dimensional

5.

Composite

16.

Phenetic

6.

Ecological

17.

Phylogenetic

7.

Evolutionary Significant Unit

18.

Polythetic

8.

Evolutionary

19.

Recognition

9.

Genealogical Concordance

20.

Reproductive Competition

10. Genetic

21.

Successional

11. Genotypic Cluster Definition

22.

Taxonomic

1997

During much of the 20th century, Enst Mayr's 'biological species concept' has been popular among biologists. But more recently, especially with the development of cladistics, there has been the view that the 'evolutionary species concept' is more appropriate.

Evolutionary Species !

‘... a lineage (an ancestral-descendant sequence of populations) evolving separately from others and with its own unitary evolutionary role and tendencies.’ (Simpson 1961: 153)

!

‘... a single lineage of ancestor-descendant populations which maintains its identity from other such lineages and which has its own evolutionary tendencies and historical fate.’ (Wiley 1978)

!

‘... an entity composed of organisms which maintains its identity from other such entities through time and over space, and which has its own independent evolutionary fate and historical tendencies.’ (Wiley & Mayden 1997)

The evolutionary species concept promotes the view that species are things or individuals, and organisms are the parts that make up species.

Species as Individuals “...there are reasons to favor what has become known as the species-asindividuals thesis.” Richards (2010: 14)

Can, or should, species be treated as individuals, entities, or things?

As the view that species are individuals, things, or entities that exist in time and space has become a prominent point of view, we need to look more closely at the criteria used to consider species as individuals, which will help us to see why this point of view is incorrect.

Can, or should, species be treated as individuals, entities, or things? Individuals versus Classes: Some Criteria

• Individuals can be experienced, but classes cannot.

• Individuals can change, whereas classes do not.

• Individuals are involved in events, which is not possible for classes.

• Classes are composed of individuals, and represent concepts regarding those individuals. Thus, classes are mental constructs.

Here are some of the criteria commonly used to claim that species are individuals, and not classes: *** We can experience the existence of individuals, but we cannot experience a class. We can only experience members of a class, not the class itself. *** Individuals can change through time, but classes do not change. Individuals are involved in events or phenomena, but this cannot occur with classes. *** Classes are simply composed of individuals and represent some concept related to those individuals. In other words, classes are concepts that only exist in the human mind. But, let's ask this question: Have you ever seen a species? If species are things or individuals, then obviously they must have emergent characters or properties that allow us to perceive them. Have you ever observed a species by way of its characters or properties? The answer has to be 'no.' We need to stop thinking about species as objects. We need to shift our thinking so that it actually reflects our actions as scientists, in relation to our observations of organisms.

Species a-us

Species b-us

present

‘Individuals change over time and can only be described [as opposed to defined]’ (Mayden 1997: 388).

Contrary to what is claimed, species cannot change through time.

past

This is not ‘change in a lineage.’ There are only differences between organisms.

Let's first address the popular myth that species or lineages 'change through time.' If species are in fact individuals then they should be able to change or evolve. It is rather easy to show that that is not the case. In the example shown here, we observe two groups of organisms in the present. Let's say we know two different species hypotheses apply to these organisms because we know the exact histories, shown here, that occurred in the past to result in the observed individuals. Among these past events we can summarize the transformation series of body colors as shown here. The question is, however, can we say that species a-us and species b-us have exhibited changes over time with regard to body color? In other words, do we see a 'change' or 'evolution' from one color to another? No, not at all. What we observe here is not 'change,' since change is a phenomenon that is only possible with an individual. Instead, all we observe are differences in color between individual organisms. While the diagram suggests that there were mutations, reproductive events, and maybe natural selection in the past, these are all events that occurred with regard to individual organisms, not to a lineage or species. Each of the branches we see here are not individuals in themselves. What we are calling branches or lineages in this diagram are simply summaries of a series of past events involving individual organisms. A species, or lineage, cannot change through time, and they cannot change through time for the very fact that they are not individuals.

a-us

b-us

present

Are these really individuals? Or, simply representations of past causal events?

A-us

So when we see a diagram like this, and formal names are applied, in the form of two species and one genus, we should ask are these individuals? Or, are we simply illustrating our hypotheses (i.e. two specific and one phylogenetic) of possible past events that help give us causal understanding of the characters we observe among individual organisms in the present? As we will see in this course, formal names are not referring to classes or individuals. Rather, they refer to our explanatory hypotheses.

Individuals versus Classes: Neither! • Why do systematists refer to species in relation to organisms? • What is the inferential basis for species? • How does the concept of ‘species’ differ from the concept of any ‘subsepecific’ or ‘supraspecific taxon?’ • The real issue is not to ask, “What is the best species concept?”, but rather to ask, “What is the most appropriate definition for the term species?”

An important consequence is that the ‘individual / class’ distinction is not entirely accurate. The real distinction that needs to be considered is between ‘individual’ and ‘explanatory hypothesis.’

The claim in this course, that all taxa are explanatory hypotheses, and not individuals, things, entities, or classes, will be justified by answering the questions shown here. By answering these questions we will readily see that species, indeed all taxa, are not individuals or things that exist in time and space. Rather, they are our explanatory hypotheses that we develop to give us understanding of what we observe, which are organisms, or past traces of organisms, in the form of fossils.

Q1: Why do some individuals have a white spot in contrast to a completely blue body? Q2: Why do some individuals have antennae in contrast to a smooth dorsum?

The inferential basis for species will be outlined in this example. We observe individuals with unexpected or surprising characters. These new observations can be represented by the two why-questions shown here.

Abductive Inference of Species Hypotheses Q1 Species Theory: If character x(1) originates by mechanisms a, b, cJn, among gonochoristic or cross-fertilizing hermaphroditic individuals of a reproductively isolated population with character x(0), and x(1) subsequently becomes fixed throughout the population during tokogeny by mechanisms d, e, f n, then individuals observed in the present will exhibit character x(1). Observations (effects): Individuals have a white spot in contrast to a completely blue body as seen among individuals to which other species hypotheses refer (a-us, b-us, etc.).

Causal Conditions (specific hypothesis x-us) : The white spot condition originated by unspecified mechanisms within a reproductively isolated population with completely blue bodies and eventually became fixed throughout the population during tokogeny by additional unspecified mechanisms.

With regard to why-question Q1, the answer is abductively inferred by applying a 'species theory' to what are observed among individuals with the white spot condition. What is important to recognize here is that the 'theory' refers to at least three theories: mutation, tokogeny, and natural selection. The conclusion, outlining causal conditions accounting for the presence of white spots, is the hypothesis we have formally called 'species x-us.'

Abductive Inference of Species Hypotheses Q2 Species Theory: If character x(1) originates by mechanisms a, b, cJn, among gonochoristic or cross-fertilizing hermaphroditic individuals of a reproductively isolated population with character x(0), and x(1) subsequently becomes fixed throughout the population during tokogeny by mechanisms d, e, f n, then individuals observed in the present will exhibit character x(1). Observations (effects): Individuals have a dorsal margin with antennae in contrast to a smooth dorsal margin as seen among individuals to which other species hypotheses refer (a-us, b-us, etc.).

Causal Conditions (specific hypothesis y-us) : The antennate dorsal margin condition originated by unspecified mechanisms within a reproductively isolated population with smooth dorsal margins and eventually became fixed throughout the population during tokogeny by additional unspecified mechanisms.

With regard to why-question Q2, the same inferential form used for Q1 is applied to explain the presence of antennae.

Abductive Inference of Species Hypotheses

present

Species hypothesis (x-us)

Causal Conditions (specific hypothesis x-us) : The white spot condition originated by unspecified mechanisms within a reproductively isolated population with completely blue bodies and eventually became fixed throughout the population during tokogeny by additional unspecified mechanisms.

Species hypothesis (y-us)

Causal Conditions (specific hypothesis y-us) : The antennate dorsal margin condition originated by unspecified mechanisms within a reproductively isolated population with smooth dorsal margins and eventually became fixed throughout the population during tokogeny by additional unspecified mechanisms.

The two species hypotheses can be illustrated in the form shown here. Notice that this diagram would traditionally be referred to as two 'lineages.' But, all we are doing is providing a summary of past events that explain our present observations. Once again, it is important to stress that what we here call a species, is nothing more than a particular class of explanatory hypothesis among those classes of hypotheses we refer to as taxa.

A (preliminary) Formal Definition of ‘SPECIES’ An explanatory account of the occurrences of the same character or characters among gonochoristic or crossfertilizing hermaphroditic individuals by way of character origin and subsequent fixation during tokogeny. Important consequences: We do not ‘indentify,’ ‘describe,’ or ‘discover’ species. We discover and describe individual organisms. We present formal names to some of our hypotheses (taxa), and those names should be defined in terms of our hypotheses. Species hypotheses do not apply to strictly asexual or self-fertilizing hermaphroditic organisms.

Based on the examples just presented, here is a formal definition of species. The definition is, however, preliminary. This is the case because the definition only considers gonochoristic and cross-fertilizing hermaphroditic organisms. As we will see later, we also need to consider hypotheses regarding organisms with other modes of tokogeny. Given the definition presented here, a species is simply one class of explanatory hypothesis, that uses a set of theories that differs from what are used among the other types of hypotheses we refer to in biological systematics. In the case of species hypotheses, we speak of character origin and subsequent fixation through evolutionary processes during tokogeny. The theories applied to infer phylogenetic hypotheses, on the other hand, not only include those of character origin and fixation, but also population splitting (which will be presented later). From this definition, there are several important practical consequences. One is that we do not 'identify' or 'describe' species. Rather, we describe individual organisms by way of their properties or characters, and we associate a species hypothesis with those individuals. We present formal names to some of our hypotheses, and those names are then defined in terms of those hypotheses. Another very interesting consequence is that species hypotheses cannot be applied to strictly asexual organisms, or organisms that are strictly self-fertilizing hermaphroditic. In these latter cases, character origin/fixation shows a pattern that is somewhat similar to what is seen for phylogenetic hypotheses. Later we will identify these different hypotheses traditionally entailed by the one term, species.

In saying that species are relations, I mean that when biologists correctly delimit species and when the rest of us correctly use species words... we are all in effect referring neither to entities abstract or concrete nor to their members or parts; instead we are referring to the individual organisms and the relations between them that together constitute their reality as species.

Stamos (2003: 25)

The view that species are abductively inferred hypotheses is similar to the view developed by philosopher of science, David Stamos, in his book, 'The Species Problem.' Stamos claims that species refer to causal relations. As he points out, that as relations, species are not classes or individuals, but rather, we are speaking of the past causal events that give us the organisms we observe. Where the development of my ideas differ from those of Stamos is that I have attempted to investigate to a much greater depth the nature of the reasoning we use as systematists to infer all of the types of hypotheses we call taxa, not just species. Just as important is the fact that all taxa, or more correctly hypotheses, are inferred by use of the same type of reasoning process: abduction.

Definitions of SPECIES1–5 Species1 hypothesis: character origin, with subsequent fixation via tokogeny by sexual reproductive events. Species2 hypothesis: simultaneous character origin/fixation via tokogeny by sexual reproductive events, i.e. hybridization, polyploidy. Species3 hypothesis: simultaneous character origin/fixation, with subsequent tokogeny by asexual, apomictic/ parthenogenetic, or self-fertilizing hermaphroditic reproductive events. Species4 hypothesis: character origin, with subsequent fixation via tokogeny by alternations of sexual and asexual reproductive events. Species5 hypothesis: immediate character origin/fixation via horizontal genetic exchange.

Up to this point, we have acknowledged that... (1) all taxa refer to explanatory hypotheses, as answers to why-questions; (2) these hypotheses are inferred by way of abductive reasoning; and, (3) there are a variety of classes of hypotheses/taxa, depending on the combination of theories that are applied to observed properties of organisms. Clearly, based on what we saw earlier in the example of inferring a species hypothesis, the one term 'species' cannot accommodate the different connotations to which the term species have been applied. In addition to the theory used to infer the hypothesis shown earlier, we can recognize at least four additional theories that have been used in conjunction with species, shown here. Ideally, these five classes of hypotheses should be distinguished as separate taxa beyond the one word of 'species.' After we examine the inferential structure leading to phylogenetic hypotheses, we will see that specific hypotheses 3 and 5 cannot be used in conjunction with phylogenetic hypotheses.

The Fundamental Misunderstanding of DNA Barcoding

“[DNA barcoding] provides a way to identify the species to which a plant, animal or fungus belongs.”

Let's return to the topic of DNA barcoding. Since we do not 'identify' or 'describe' species by way of the characters of organisms, but instead, we apply species hypotheses to our observations of organisms, DNA barcoding is not an acceptable approach in biological systematics. We can look a simple example to see why this is the case.

body:

green (as opposed to brown)

dorsum:

antennae (as opposed to smooth)

ventrum:

legs (as opposed to smooth)

DNA sequence:

CCAGAGGCCCAA (as opposed to C-AAAGGCGCAT)

In this example, we observe these new specimens, with characters we have never seen before compared to the characters we have previously observed of this group. The new specimens have green bodies, as opposed to brown. The dorsum has antennae, as opposed to being smooth. The ventrum has legs, as opposed to being smooth. And, we have some DNA sequence data that shows unique differences, shown here in red, compared to what have been previously observed.

a-us

body:

green (as opposed to brown)

dorsum:

antennae (as opposed to smooth)

ventrum:

legs (as opposed to smooth)

DNA sequence:

CCAGAGGCCCAA (as opposed to C-AAAGGCGCAT)

Definition of a-us: A species hypothesis, accounting for the presence of (1) body is green, (2) dorsum with antennae, (3) ventrum with legs, and (4) unique nucleotides for positions 151, 153, 158, and 161, among observed individuals, as consequences of past mutations, tokogenetic events, and selection / drift among past members of the population.

Based on these observations, we decide to explain the new characters with a new species hypothesis, illustrated here, showing origins and fixation of characters during past tokogeny. The formal definition of species a-us would be written out as shown here, which reflects what is shown in the diagram. The name a-us refers to an explanatory hypothesis for the new characters we have observed among these individuals. We can ask this question: Would DNA barcoding be a proper way to apply the hypothesis we call a-us to the individuals we observe? The answer would be 'no.' Since what we call species a-us is a hypothesis, and not a thing, a sequence of DNA cannot represent all of the observations to which that hypothesis serves as an explanation. There are other characters to which the hypothesis also refers, and these must be taken into consideration.

The Fundamental Misunderstanding of DNA Barcoding

a-us Species only defined by sequence data, ignoring all other relevant observations.

DNA sequence:

CCAGAGGCCCAA (as opposed to

C-AAAGGCGCAT )

Here is a simple example to illustrate the problem. We previously defined species a-us as a hypothesis that explains the occurrences of three morphological characters and particular nucleotides. But, what if we only rely on the sequence data, as is claimed is all that is necessary for DNA barcoding of species? Let's say that with future collecting, one finds more specimens that have the same nucleotide sequence.

The Fundamental Misunderstanding of DNA Barcoding

a-us Species only defined by sequence data, ignoring all other relevant observations. Additional specimens observed, with the same ‘barcode.’

DNA sequence:

CCAGAGGCCCAA (as opposed to

C-AAAGGCGCAT )

But, if we actually examine the other characters to which hypothesis a-us applies, we notice that some of these characters are different.

The Fundamental Misunderstanding of DNA Barcoding

a-us - revised

Species only defined by sequence data, ignoring all other relevant observations.

b-us

Additional specimens observed, with the same ‘barcode.’ Sequence data are now explained phylogenetically, not by species hypotheses. DNA sequence:

CCAGAGGCCCAA (as opposed to

C-AAAGGCGCAT )

In fact, we would have to infer a completely new species hypothesis to account for these new observations. And as a consequence, we also would have to completely revise the hypothesis we call a-us. The sequence data would no longer be part of the species hypothesis, but instead would be explained by a new phylogenetic hypothesis. So clearly DNA barcoding is not a proper scientific procedure for applying species hypotheses to the organisms we observe. Since what we call species are hypotheses, and not things, a sequence of DNA alone cannot represent that hypothesis if there are other characters that the hypothesis also refers to. It is when we stop thinking of species as objects we are trying to find in nature, and instead recognize that species are just one type of hypothesis we apply to organisms, we see that DNA barcoding is a technique with very serious problems.

Inferences of Taxa

Phylogenetic Hypotheses

Continuing with the earlier example, let's look at the form abductive inference takes to produce phylogenetic hypotheses.

Q1: Why do some of these individuals have a white spot in contrast to completely black? Q2: Why do some of these individuals have antennae in contrast to a smooth dorsum?

Q3: Why do individuals to which species hypotheses x-us and y-us refer have ventral appendages? Recall that the two previous why-questions were answered by way of abductive inferences to respective species hypotheses. Question Q3, however, addresses the occurrence of ventral appendages among individuals to which species hypotheses xus and y-us refer. Note that this question is still contrastive, in that we are asking why some individuals have ventral appendages in contrast to a smooth ventrum.

The Relation Between Why-Questions and Phylogenetic Hypotheses The most fundamental basis for the inference of phylogenetic hypotheses is the application of a causal theory which is appropriate to the why-questions being asked.

All phylogenetic-level questions (as opposed to ontogenetic, tokogenetic, specific, etc.) have the form, “Why do individuals to which species hypotheses x-us and y-us refer have character x(1), in contrast to members of other species with character x(0)?”. The only appropriate theory, relative to such a question, is one which accounts for shared similarities by way of a common cause. This requirement has distinct implications for most “phylogenetic” methods.

Since biological systematics is about seeking causal understanding by way of the abductive inferences of explanatory hypotheses, we have to be aware of the relations between the forms of why-questions we ask and the hypotheses that serve as answers to those questions. An important component to these relations is knowing what causal theories are appropriate for answering questions. With regard to questions that are to be answered by way of phylogenetic hypotheses, we must clearly understand the form of the questions we ask. Plus, because we are asking questions about our observations of shared characters, the only rational approach is to use theories that provide us with common causes, such that the integrity of our observation statements is maintained as much as possible.

present new observations

Q3: Why do individuals to which species hypotheses x-us and y-us refer have ventral appendages? Phylogenetic Theory: If character x(0) exists among individuals of a reproductively isolated, gonochoristic or cross-fertilizing hermaphroditic population and character x(1) originates by mechanisms a, b, cJ n, and becomes fixed within the population by mechanisms d, e, fJ n (=ancestral species hypothesis), followed by event(s) g, h, iJ n, wherein the population is divided into two or more reproductively isolated populations, then individuals to which descendant species hypotheses refer would exhibit x(1).

To answer question Q3 requires use of a 'descent with modification' theory that will explain the presence of shared characters by way of a common cause. The phylogenetic theory shown here consists of three theories: (1) mutation, (2) fixation, and (3) population splitting. It is important to notice that the phylogenetic theory is very vague - it says nothing specific about the causal mechanisms that might have occurred. We will see later that this vague quality has significant implications.

Abductive Inference of Phylogenetic Hypotheses Phylogenetic Theory: If character x(0) exists among individuals of a reproductively isolated, gonochoristic or cross-fertilizing hermaphroditic population and character x(1) originates by mechanisms a, b, cJ n, and becomes fixed within the population by mechanisms d, e, fJ n (=ancestral species hypothesis), followed by event(s) g, h, iJ n, wherein the population is divided into two or more reproductively isolated populations, then individuals to which descendant species hypotheses refer would exhibit x(1). Observations (effects): Individuals to which specific hypotheses x-us and y-us refer have ventrolateral margins with appendages in contrast to smooth as seen among individuals to which other species hypotheses (a-us, b-us, etc.) refer.

Causal Conditions (phylogenetic hypothesis X-us): Ventrolateral margin appendages originated by some unspecified mechanism(s) within a reproductively isolated population with smooth ventrolateral margins, and the appendage condition became fixed in the population by some unspecified mechanism(s) (= ancestral species hypothesis), followed by an unspecified event(s) that resulted in two or more reproductively isolated populations.

Based on question Q3, we would apply the phylogenetic theory to the observed effects in need of explanation, and abductively infer the hypothesized set of past causal conditions shown here. Notice that because the theory applied is vague, the hypothesis too is lacking in detail.

x-us

y-us

present

Causal Conditions (phylogenetic hypothesis X-us): Ventrolateral margin appendages originated by some unspecified mechanism(s) within a reproductively isolated population with smooth ventrolateral margins, and the appendage condition became fixed in the population by some unspecified mechanism(s) (= ancestral species hypothesis), followed by an unspecified event(s) that resulted in two or more reproductively isolated populations.

X-us

The written form of the hypothesis, formally named X-us, can be illustrated as shown here. You might notice that the two species hypotheses, x-us and y-us, are also shown, even though they were separately inferred from previous abductive inferences.

x-us

y-us

present X-us x-us

y-us

A-us

This detailed illustration of the written form of the hypothesis is what we typically represent in the much more simplified form known as a 'cladogram.' Notice that a cladogram can do no more than imply the already vague causal conditions provided by the written hypothesis. As we will see, this is a problem because systematists commonly do not understand all of the causal aspects implied by cladograms.

Definitions of SPECIES1–5

*Species hypothesis: 1

*Species hypothesis: 2

character origin, with subsequent fixation via tokogeny by sexual reproductive events. simultaneous character origin/fixation via tokogeny by sexual reproductive events, i.e. hybridization, polyploidy.

Species3 hypothesis: simultaneous character origin/fixation, with subsequent tokogeny by asexual, apomictic/ parthenogenetic, or self-fertilizing hermaphroditic reproductive events.

*Species hypothesis: 4

character origin, with subsequent fixation via tokogeny by alternations of sexual and asexual reproductive events.

Species5 hypothesis: immediate character origin/fixation via horizontal genetic exchange.

* Phylogenetic hypotheses can only be applied to individuals to which these hypotheses are applied. Recall in the example just shown for inference of a phylogenetic hypothesis, as a cladogram, that the species hypotheses would have been separately inferred. We also found earlier that the one term 'species' refers to at least five classes of hypotheses, each with distinctly different causal conditions. There is the added implication that phylogenetic hypotheses are not applicable to all organisms to which these species hypotheses apply. In fact, only species1, species2, and species4 hypotheses can be used in conjunction with phylogenetic hypotheses. Since population spitting events are a necessary part of phylogenetic hypotheses for causally accounting for shared characters, such splitting events are simply not applicable to species3- and species5-type hypotheses.

A-us

X-us

a-us b-us x-us y-us z-us

A cladogram. What does it imply? Converting the diagram into words.

Let's look at an example that further illustrates the relations between cladograms and the explanatory hypotheses they imply. Such relations must exist if we are to claim that biological systematics is a field of science, and it is the case that systematics has the goal of acquiring causal understanding of the occurrences of characters among organisms.

Implied specific and phylogenetic hypotheses:

A-us

X-us

a-us b-us x-us y-us z-us 1 2 3 4 5 6 7 8 9 10 a-us 1 0 0 0 0 1 0 0 0 0

6(1)

7(1) 8(1)

b-us 1 0 0 0 0 0 1 0 0 0 x-us

0 1 1 1 0 0 0 1 0 1

y-us

0 1 1 1 1 0 0 0 0 0

z-us

0 0 1 1 1 0 0 0 1 1

2(0)* 9(1) 5(1)

1(1) 10(1) 4(1) 3(1) 2(1)

As will be discussed in a later lecture on character coding, a data matrix is more than just a summary of one's observations of characters among organisms. It also summarizes our why-questions associated with those observations. Indeed, it is a necessity that our why-questions be present in a data matrix if we are to claim that cladograms serve in some capacity as explanations. Thus, when we see the standard cladogram with 'character state changes on branches,' these are actually vaguely implying the series of explanatory hypotheses that are answers for the separate why-questions implied by a data matrix. In the following slides, we can identify each of the explanatory hypotheses that are implied by this cladogram.

a-us Formally named specific hypothesis:

6(1)

Definition of a-us: A specific hypothesis, accounting for the presence of character 6(1) among observed individuals. Character 6(1) originated in a population of individuals with 6(0) by unspecified mechanisms, subsequent to which 6(1) became fixed in the population by unspecified mechanisms, leading to individuals observed in the present, all with 6(1).

First, let's identify each of the species hypotheses on the cladogram, referred to as aus, b-us, x-us, y-us, and z-us. Once again, keep in mind that species hypotheses are inferred separately from phylogenetic hypotheses. For each species hypothesis indicated on the cladogram, we can present the formal definition of each name. Note, however, that these formal definitions are nothing like what is required by the international codes of nomenclature or the PhyloCode. Problems associated with these nomenclatural issues will be addressed in a later lecture. For now, the important thing is that we recognize the variety of explanatory hypotheses that are represented by the cladogram. Notice that the formal definition here provides an explanation for character 6(1) among individuals. Similar definitions are shown next for remaining species hypotheses.

b-us Formally named specific hypothesis:

7(1)

Definition of b-us: A specific hypothesis, accounting for the presence of character 7(1) among observed individuals. Character 7(1) originated in a population of individuals with 7(0) by unspecified mechanisms, subsequent to which 7(1) became fixed in the population by unspecified mechanisms, leading to individuals observed in the present, all with 7(1).

x-us Formally named specific hypothesis:

8(1)

Definition of x-us: A specific hypothesis, accounting for the presence of character 8(1) among observed individuals. Character 8(1) originated in a population of individuals with 8(0) by unspecified mechanisms, subsequent to which 8(1) became fixed in the population by unspecified mechanisms, leading to individuals observed in the present, all with 8(1).

y-us Formally named specific hypothesis:

10(1)

Definition of y-us: A specific hypothesis, accounting for the presence of character 10(0) among observed individuals. Character 10(0) originated in a population of individuals with 10(1) by unspecified mechanisms, subsequent to which 10(0) became fixed in the population by unspecified mechanisms, leading to individuals observed in the present, all with 10(0). Hypothesis y-us is an ad hoc hypothesis, as a consequence of inferring phylogenetic hypothesis X-us.

Species hypothesis y-us might appear unusual for the fact that it is defined as an ad hoc hypothesis of homoplasy relative to the phylogenetic hypotheses implied by the cladogram. While in the context of phylogenetic hypotheses in the cladogram the definition is an instance of homoplasy, hypothesis y-us still would have been defined prior to the phylogenetic inference.

z-us Formally named specific hypothesis:

2(1)

Definition of z-us: A specific hypothesis, accounting for the presence of character 9(1) among observed individuals. Character 9(1) originated in a population of individuals with 9(0) by unspecified mechanisms, subsequent to which 9(1) became fixed in the population by unspecified mechanisms, leading to individuals observed in the present, all with 9(1). Inclusive of ad hoc hypothesis (required as part of definition of phylogenetic hypothesis X-us): Character 2(0) originated by unspecified mechanisms among a poulation of individuals with 2(1), subsequent to which 2(0) became fixed in the population.

A-us a-us b-us

Formally named phylogenetic hypothesis:

1(1)

Definition of A-us: A phylogenetic hypothesis, accounting for the presence of character 1(1) among observed individuals. Character 1(1) originated in a population of individuals with 1(0) by unspecified mechanisms, subsequent to which 1(1) became fixed in the population by unspecified mechanisms, subsequent to which there was an unspecified population splitting event, leading to individuals observed in the present, all with 1(1), and to which specific hypotheses a-us and b-us also apply.

Next, there are three phylogenetic hypotheses implied by the cladogram. As shown in each definition, character origin and fixation within a reproductively isolated ancestral population is indicated, followed by a population splitting event. Once again, notice how vague these hypotheses are with regard to presenting past causal conditions.

X-us y-us z-us 10(0)* 2(0)*

Formally named phylogenetic hypothesis:

10(1) 4(1) 3(1) 2(1)

Definition of X-us: A phylogenetic hypothesis, accounting for the presence of characters 2(1), 3(1), 4(1), and 10(1) among observed individuals. Characters 2(1), 3(1), 4(1), and 10(1) originated by unspecified mechanisms among a population of individuals with 2(0), 3(0), 4(0), and 10(0), subsequent to which 2(1), 3(1), 4(1), and 10(1) became fixed in the population by unspecified mechanisms, followed by a population splitting event, leading to individuals with 2(1), 3(1), 4(1), and 10(1), and to which specific hypotheses x-us, y-us, and z-us also apply. Required ad hoc hypothesis: Character 2(0) originated by unspecified mechanisms among a poulation of individuals with 2(1), subsequent to which 2(0) became fixed in the population. This hypothesis is a subset to the specific hypothesis, z-us. Required ad hoc hypothesis: Character 10(0) originated by unspecified mechanisms among a poulation of individuals with 10(1), subsequent to which 10(0) became fixed in the population. This hypothesis is referred to as specific hypothesis, y-us.

It is important to notice in this example that the phylogenetic hypothesis accounting for the presence of 2(1) and 10(1) also requires two ad hoc hypotheses (homoplasy). Formally, defining X-us should not only include the causal conditions explaining the occurrences of 2(1), 3(1), 4(1), and 10(1), but also include the required instances of homoplasy.

y-us z-us Unnamed phylogenetic hypothesis: 5(1)

Definition:

A phylogenetic hypothesis, accounting for the presence of character 5(1) among observed individuals. Character 5(1) originated by unspecified mechanisms among a population of individuals with 5(0), subsequent to which 5(1) became fixed in the population by unspecified mechanisms, followed by a population splitting event, leading to individuals with 5(1), and to which specific hypotheses y-us and z-us also apply.

The Philosophy of Biological Systematics Course Outline – Part 2

1.

Systematics involves is abductive inference.

2.

Inferences of systematics hypotheses, i.e. taxa.

3.

Some implications for “phylogenetic” methods.

At this point, we have examined relations between the goals of scientific inquiry and biological systematics, finding that in both instances the acquisition of causal understanding is fundamentally important. We then looked at the nature of the reasoning process, known as abduction, used to go from our why-questions regarding observations, to explanatory hypotheses that provide some degree of initial understanding. The structure of abductive inference was then examined in relation to systematics. Given that most of what is done in biological systematics in terms of methods is related to the inferences of phylogenetic hypotheses, as cladograms, it will be useful to consider some significant implications for these methods in relation to abductive reasoning.

The Limits of Phylogenetic Hypotheses Phylogenetic hypotheses are ‘explanation sketches,’ not complete, formal explanations.

“What the explanatory analyses of historical events offer is, then, in most cases not an explanation..., but something that might be called an explanation sketch. Such a sketch consists of a more or less vague indication of the laws and initial conditions considered as relevant, and it needs ‘filling out’ in order to turn into a full-fledged explanation.” Hempel (1965: 238), Aspects of Scientific Explanation

One fundamental implication that we need to consider is the basic limitation of cladograms as explanatory vehicles. You will recall that earlier we found that the abductive inferences of phylogenetic hypotheses, in the form of cladograms, only allow one to present extremely vague causal accounts of the distributions of observed characters among individuals. The vague nature of such explanations are certainly characterized by what little is conveyed in cladograms. This means that cladograms, as implying a set of phylogenetic hypotheses, are nothing more than what philosopher of science Carl G. Hempel referred to as 'explanation sketches.'

x-us

y-us

present X-us x-us

X-us

y-us

Phylogenetic hypotheses offer very vague explanations of our observations.

Obviously, when we examine either a cladogram (right) or a slightly more detailed causal account (left), we find that nothing is indicated with regard to the specific past events of how a character originated in an ancestral population, how that character subsequently became fixed in that population, or the cause of the later population splitting event.

The Limits of Phylogenetic Hypotheses Phylogenetic hypotheses present very limited causal events.

Phylogenetic hypotheses, as graphically represented by ‘cladograms,’ are explanation sketches consisting of two classes of causal conditions: 1.

character origin and fixation by unspecified causal events among members of an ancestral population/species, and...

2.

subsequent population splitting events by unspecified causal events.

The explanatory depth of cladograms is extremely limited. Cladograms do not provide specific information regarding causal conditions which can serve as complete explanations.

Indeed, it is infrequent that we see hypothesized causal events actually referred to in the context of cladograms. Unfortunately, cladograms are too often only referred to in terms of being branching diagrams, composed of 'terminal branches,' 'nodes,' and 'internal branches' or 'internodes.' None of these terms are appropriate from an explanatory perspective. A 'terminal branch' is actually a previously inferred species hypothesis. A 'node' is a hypothesized population splitting event. An 'internal branch' denotes the events of character origin and fixation. But in all instances, the explanatory depth provided by a cladogram is extremely limited. This is why cladograms are no more than explanation sketches. From a scientific perspective, they actually are not particularly significant, and certainly not worthy of the importance too often given to them by systematists. As we shall see later in the course, such vague explanatory accounts offer no means to effectively engage in hypothesis testing.

present

x-us

y-us h1, 2: origin/fixation X-us

h3b: population splitting

X-us

h3a: origin/fixation

x-us

y-us

Phylogenetic hypotheses, as cladograms, only state two classes of vague causal events: (1) character origin/fixation, and (2) population splitting.

The diagram presented here summarizes what was discussed in the previous two slides, indicating the very vague causal events typically implied by cladograms.

The Explanatory Components of Cladograms

0 a-us

1 x-us

1 y-us Hypothesis accounting for origin/fixation of population-level character(s)

Population splitting hypothesis Hypothesis accounting for origin/fixation of character 1

‘Phylogenetic hypothesis’

To solve the problem of systematics hypotheses being explanation sketches, rather than full explanations that can be empirically tested, we first need to identify what these sketches actually have to offer. Using the cladogram shown here, we can identify three classes of hypotheses. One of these is said to be a species hypothesis, e.g. y-us. The other two classes of hypotheses comprise what would be more generally called a phylogenetic hypothesis. In the example shown here, the phylogenetic hypothesis is summarized by the cladogram, but two hypotheses are indicated - what are commonly known as the 'internode' and 'node.' The species hypotheses would have been inferred separately from the phylogenetic hypothesis. What you should notice for the three hypotheses shown on the cladogram is that all of them are extremely vague. None of them provides full explanatory accounts. As such, there would be no way to proceed with testing any of these hypotheses. It first would be necessary to fill out the causal conditions within each. This is a fundamental condition that is usually ignored in systematics.

‘Phylogenetic trees’

Unfortunately, the lack of emphasis on the explanatory components of cladograms have led to the inappropriate focus on what are called 'phylogenetic trees.'

“Evidence [sic] from morphological, biochemical, and gene sequence data suggests that all organisms on Earth are genetically related, and... can be represented by a vast evolutionary tree, the Tree of Life. The Tree of Life then represents the phylogeny of organisms, i. e., the history of organismal lineages as they change through time.”

And this emphasis on seeking 'the tree of life' has contributed to his misunderstanding of our real goal in biological systematics.

‘Phylogenetic trees’ ~ Fallacy of reification ~ “Constructing the Tree of Life (ToL), a diagrammatic depiction of the evolutionary relationships among all extinct and extant taxa, is one of biology’s most important tasks.” Novick et al. (2012: 757) BioScience

The result has been what is called the fallacy of reification. Phylogenetic trees are treated as if they are the objects we seek, when in fact they are supposed to simply represent components of only some of our explanatory hypotheses.

‘Phylogenetic trees’ ~ Fallacy of reification ~ • tree topology comparisons (violation of requirement of total evidence)

• branch lengths (misinterpreting causal hypotheses)

• statistical consistency (ignoring nature of abductive inference)

Some of the unfortunate, and quite significant, consequences of this reification has been the incorrect view that (1) phylogenetic trees can be compared with one another simply as a matter of comparing 'tree topologies; (2) that 'branch lengths' are of some importance, when in fact they are not; and (3) that statistical consistency is important when it comes to comaring tree topologies. As we will see when we examine the 'requirement of total evidence,' tree topologies cannot be meaningfully compared. And this has consequences for the incorrect view that statistical consistency matters. We will address this misconception when we look at 'maximum likelihood' as a method of inferring phylogenetic hypotheses. Related to likelihood, we will see that 'branch lengths' are misinterpretations of the causal components implied by cladograms. In all, these are problems directly related to the reification of cladograms as abstract things that have lost contact with our real goal in systematics: to causally understand our observations of the properties of organisms.

‘Phylogenetic trees’ ~ Fallacy of reification ~

“Hypothesis of the protostome tree of life, placing Arthropoda within the ecdysozoan phyla. This tree is a summary of diverse sources, with emphasis on groups recognized in phylogenomic analyses.” Giribet, G. & G.D. Edgecombe. 2012. Reevaluating the arthropod tree of life. Annual Review of Entomology 57: 167186.

Not only the common practice of tree/cladogram comparisons is mistaken, but also diagrams like that shown here. Such 'summaries' are diagrams that are empirically meaningless. There are no explanatory components to be found because the requirement of total evidence is violated. As a field of science, biological systematics should not tolerate such illustrations for the fact that the illustrations show no relation to our goal of acquiring causal understanding.

Phylogenetic Inference is Not Deductive Causes cannot be deduced from effects

1.

The inference from effects to cause(s) in phylogenetics is ampliative. The non-ampliative nature of deduction only provides for inferences from cause to effect(s).

2.

Multiple conclusions are possible in phylogenetic inference – it is impossible to have multiple conclusions in deduction.

3.

The requirement of total evidence must be specifically considered, which is not necessary in deduction.

Thus far in this course we have seen that the inference from observed effects to an hypothesis offering an explanation of those effects is not a matter of deductive inference. It is instead, abductive. You will, however, sometimes encounter biological systematics publications in which authors claim phylogenetic inference is deductive. The rules of deductive reasoning would not allow this. The three main reasons are shown in this slide.

Evidence and Inference For any inference, the evidence for a conclusion will be the premises.

evidence

1o Premise...

1o Premise... 2o Premise... Conclusion

or

2o Premise... Conclusion

Another common misunderstanding lies in use of the term 'evidence.' Recall from our examination of the different classes of reasoning, that for any inference, the premises serve as evidence for their respective conclusions. Regardless of the class of reasoning being deductive or non-deductive (abductive or inductive), the relation between evidence and conclusion remains the same. It is the relation between premises and allow for a particular conclusion or set of conclusions. Systematists often speak of evidence. They speak of 'evidence supporting' particular cladograms or taxa. But as we will see, especially when we examine the nature of hypothesis testing, systematists are routinely confused when it comes to correctly referring to 'evidence' in the proper context.

Relations Between the Types of ‘Evidence’ in Biological Systematics Evidence 1: the basis for initially suggesting a hypothesis . Since a hypothesis is abductively inferred, the evidence consists of character data and the causal theory. The relation of evidence 1 to the hypothesis is that of premises to conclusion. Evidence 2: the basis for judging a hypothesis to be true . This is the evidence obtained during the actual test of the hypothesis. The relation of evidence 2 to the hypothesis is that of premises to conclusion. The evidence suggesting a hypothesis is not the same as the evidence used to test the hypothesis.

We can broadly identify notions of evidence as follows: (1) The evidence that initially suggests a hypothesis, and (2) the evidence for judging a hypothesis to be true. The evidence in (1) is that used in abduction from observed effects to hypothesized past causal events. The evidence in (2) is that used in induction, thus consists of observations or outcomes of actual test conditions. Considerations of (1) and (2) in biological systematics are too often confused - it is common to see claims that character data, i.e. evidence in (1), can serve as evidence used to test phylogenetic hypotheses, (2). As we will see later when we examine the testing of phylogenetic hypotheses, evidence in the form of shared characters, i.e. (1), cannot serve as test evidence, i.e. (2). To claim character data can be used to both infer hypotheses explaining those data as well as serve as test evidence for those hypotheses would be circular. More technically, any given phylogenetic hypothesis cannot be used to deduce the occurrences of novel characters. Thus, what becomes important to recognize is that while shared characters are evidence used to abductively infer phylogenetic hypotheses, as explanatory accounts for the occurrences of those characters, new character data cannot be used as premises to serve as test evidence for those hypotheses.

Example: Relations Between Phylogenetic Inference, Deduction, and Induction Abduction Causal theory: character origin/fixation, with subsequent population splitting events.

observations

I.

why-questions

Observations (effects): individuals to which species hypotheses b-us and c-us refer have character x(1) in contrast to x(0) as seen among individuals to which other species hypotheses refer.

a-us b-us c-us 0 1 1

Explanatory hypothesis: character x(1) origin/fixation, followed by a population splitting event.

a-us b-us c-us 0 1 1

abduction

The next five slides will summarize the relations between evidence and conclusions. What is important to notice is that while character data are used as evidence to abductively infer phylogenetic hypotheses, such data cannot play any part in the subsequent evaluation or testing of those hypotheses. The relations between inferring hypotheses and their being tested will be further examined later when we address the mechanics of hypothesis testing.

Example: Relations Between Phylogenetic Inference, Deduction, and Induction Explanation sketch (deductive) Causal theory: character origin/fixation, with subsequent population splitting events. Explanatory hypothesis: character x(1) origin/fixation, followed by a population splitting event.

a-us b-us c-us 0 1 1

II. Observations (effects): individuals to which species hypotheses b-us and c-us refer have character x(1) in contrast to x(0) as seen among individuals to which other species hypotheses refer.

a-us b-us c-us 0 1 1

Subsequent to our abductive inferences, we can characterize phylogenetic hypotheses (and the same applies to most biological systematics hypotheses) as 'explanation sketches.' Characterized here in the classic deductive-nomological form, we see that the explanatory hypothesis, plus relevant theories, serve to provide us with at least a vague explanation of the observed effects of shared characters.

Example: Relations Between Phylogenetic Inference, Deduction, and Induction Formal explanation (deductive) Causal theory: character origin/fixation by mechanisms a, b, c, ... n, with subsequent population splitting events caused by x, y, z, ... n.

III.

Explanatory hypothesis: detailed descriptions of all causal conditions related to character origin/fixation in ancestral population, and detailed descriptions of events leading to splitting of that population.

Observations (effects): individuals to which species hypotheses b-us and c-us refer have character x(1) in contrast to x(0) as seen among individuals to which other species hypotheses refer.

a-us b-us c-us 0 1 1

a-us b-us c-us 0 1 1

If one wishes to move beyond the overly simplistic 'explanation sketches' that are cladograms, they would have to fill out the causal conditions stated in the explanatory hypothesis. Converting cladograms to full explanatory accounts is very rarely performed, but would be necessary to proceed to testing.

Example: Relations Between Phylogenetic Inference, Deduction, and Induction Deduction of test consequences Causal theory: character origin/fixation by mechanisms a, b, c, ... n, with subsequent population splitting events caused by x, y, z, ... n. Explanatory hypothesis: detailed descriptions of all causal conditions related to character origin/fixation in ancestral population, and detailed descriptions of events leading to splitting of that population.

IV.

Observations (effects): individuals to which species hypotheses b-us and c- us refer have character x(1) in contrast to x(0) as seen among individuals to which other species hypotheses refer.

a-us b-us c-us 0 1 1

Test predictions: effects, related as closely as possible to the hypothesized causal conditions – – effect m – effect n – effect o

The reason it is stressed that cladograms are inadequate as explanatory vehicles is the fact that the only way to actually move forward with considering the formal testing of any explanatory hypothesis is that we must present a detailed causal account. Without these details being provided, there is no opportunity to deduce from the hypothesis the potential test evidence or consequences needed to evaluate the hypothesis. What is shown here is the deduction of potential test evidence. This evidence would consist of effects that are direct consequences of the causal conditions stated in the hypothesis. Such effects should preferably have the lowest probability of occurrence if the hypothesized events did not occur.

Example: Relations Between Phylogenetic Inference, Deduction, and Induction Induction: performing hypothesis test Causal theory: character origin/fixation by mechanisms a, b, c, ...n, with subsequent population splitting events caused by x, y, z, ...n. Explanatory hypothesis: detailed descriptions of all causal conditions related to character origin/fixation in ancestral population, and detailed descriptions of events leading to splitting of that population. Observations (effects): individuals to which species hypotheses b-us and c- us refer have character x(1) in contrast to x(0) as seen among individuals to which other species hypotheses refer.

V.

a-us b-us c-us 0 1 1

Test predictions: effects, related as closely as possible to the hypothesized causal conditions – – effect m – effect n – effect o Actual test conditions: the detailed actions taken to attempt to observe predicted effects m, n, and o. Test outcomes:

– effect m is observed – effect n is observed – effect o is observed

ˆ HYPOTHESIS IS CONFIRMED

The act of actually engaging in hypothesis testing is not deductive, contrary to what is too often presented in the biological systematics (especially 'cladistic') literature. Notice that the premises in this example comprise the test. In order to perform the test, we must acknowledge our applications of relevant theories, the hypothesis being tested, as well as the original effects from which the hypothesis was abductively inferred. We also acknowledge the potential test evidence inferred by deduction. The test conditions that are carried out in accordance with attempting to find the potential test evidence are also part of the premises. And the premises include the actual test results. Notice in this example the test outcomes match what was predicted. From these results (premises) we can conclude that the hypothesis is confirmed. In other words, the test results provide positive support for the hypothesis. The conclusion that the hypothesis has received support is derived from an inductive processs. While the test results might give us a sense that the probability of the hypothesis has been increased, this is no guarantee that the hypothesis is certain.

Systematics and Abduction Some Implications for Phylogenetic Inference Once we acknowledge the formal structure of the why-questions we ask in systematics, and the abductive form of inference to hypotheses, there are distinct implications for the following issues:

1.

‘Parsimony’ vs ‘maximum likelihood’ methods.

2.

‘Bayesian’ inference.

There have been three issues stressed thus far in this course: (1) the goal of science, as well as biological systematics, is the acquisition of causal understanding; (2) the process of seeking that understanding starts with acknowledging our why-questions in relation to our observations of the features of organisms; and (3) the inferential process used to provide at least intitial answers to why-questions is known as abduction. If these three issues form the foundation for all of biological systematics, then there are distinct implications for the subfield known as phylogenetic systematics. The three common methods used to infer cladograms, 'parsimony,' 'maximum likelihood,' and 'Bayesianism,' are typically not examined in the context of the issues mentioned above. This has led to the consequence that systematists have not been sufficiently critical of the methods they use, or when they are critical, the logic of the arguments are usually unsound.

Why-Questions and Abductive Reasoning Why-questions abductive inference

hypothesis(es)

plausibility

An issue surrounding the subject of abductive inference is the plausibility of hypotheses.

Why-Questions and Abductive Reasoning Why-questions abductive inference

hypothesis(es)

parsimony

plausibility

likelihood

Such plausibility is initially determined on the basis of the premises used in the inference, while the more in-depth assessment of plausibility relies on the actions of testing. Regarding initial plausibility, two criteria have been suggested as important: parsimony (or simplicity) and likelihood. It is these two criteria that we need to examine in relation to the inferences of phylogenetic hypotheses.

Parsimony vs Likelihood What is Parsimony?

Simplicity Syntactic (‘elegance’)

Ontological (‘parsimony’)

# of hypotheses

# of entities/causes/processes postulated (it is rational to prefer theories/hypotheses with fewer ontological commitments)

With regard to the use of methods referred to as 'parsimony' and 'maximum likelihood,' we first have to understand what these terms mean, and the relations they have to abductive inference. To answer the question of what is parsimony, it's useful to recognize the more general term, simplicity. The concept of simplicity can be divided into what are referred to as syntactic and ontological forms. The ontological form is also known as parsimony. Parsimony is then the act of minimizing the number of entities, causes, or postulated processes. It is the view that it is more rational to prefer theories and hypotheses that require fewer ontological commitments.

Parsimony vs Likelihood What is Parsimony?

Simplicity “Scientists often appeal to a criterion of simplicity as a consideration that helps them decide which hypotheses are most plausible.” E. Sober (2001: 433), Simplicity (in: A Companion to the Philosphy of Science)

Ontological (‘parsimony’)

# of entities/causes/processes postulated (it is rational to prefer theories/hypotheses with fewer ontological commitments)

There are several different ways in which parsimony has been characterized. Sober (2001) suggests that simplicity/parsimony allows for choosing hypotheses that are most plausible. But, there are several different conceptions regarding how to determine plausibility in relation to simplicity/parsimony.

Parsimony vs Likelihood What is Parsimony?

Simplicity Ontological (‘parsimony’)

# of auxiliary hypotheses – a simpler theory is one with fewer auxiliary hypotheses (sensu Thagard 1988)

For instance, Thagard (1988: "Computational Philosophy of Science") suggests that one should choose theories or hypotheses that rely on the fewest number of auxiliary hypotheses.

Parsimony vs Likelihood What is Parsimony? Parsimony / simplicity: “...simplicity is a function of the size and nature of the [auxiliary hypotheses] needed by a theory T to explain facts F.” P. Thagard (1988: 83), Computational Philosophy of Science The fewer the number of auxiliary hypotheses required by a theory to explain the facts in question, the simpler it is. Auxiliary hypothesis – a statement, not part of the original theory, which is assumed in order to help explain certain aspects of effects.

Parsimony vs Likelihood What is Parsimony?

Simplicity Ontological (‘parsimony’)

testability – a simpler theory is more testable (sensu Popper 1959)

Karl Popper (1959: "The Logic of Scientific Discovery") suggested that parsimony should refer to the testability of theories and hypotheses.

Parsimony vs Likelihood What is Parsimony? Parsimony / simplicity: “Simple statements... are to be prized more highly than less simple ones because they tell us more; because their empirical content is greater; and because they are better testable.” K. Popper (1959: 142), The Logic of Scientific Discovery

Parsimony vs Likelihood What is Parsimony?

Simplicity Ontological (‘parsimony’)

informativeness – a simpler theory is more informative (sensu Sober 1975)

Sober (1975: "Simplicity") offered the view that simplicity/parsimony should be determined according to the informativeness of a theory or hypothesis.

Parsimony vs Likelihood What is Parsimony? Parsimony / simplicity: “...the simplicity of a hypothesis can be measured by... how well it answers certain kinds of questions. [T]he more informative a hypothesis is in answering these questions, the simpler it is.” E. Sober (1975: vii), Simplicity Hypothesis H is more informative than HN with respect to question Q if H requires less extra information than HN to answer Q.

Parsimony vs Likelihood What is Parsimony? Parsimony / simplicity: “To justify simplicity is to show why it should be taken into account in judging how plausible a theory is.” E. Sober (2001: 18), What is the problem of simplicity? Common criteria for plausibility: Popperian –

one theory/hypothesis is better corroborated than another.

Likelihoodist –

one theory/hypothesis has more evidential support than another, L(e | h).

Bayesian –

one theory/hypothesis is more probable than another.

Akaike framework – one theory/hypothesis has more predictive accuracy than another.

In his overview of simplicity/parsimony, Sober (2001) states that there are four common criteria for determining the plausibility of theories and hypotheses in terms of simplicity/parsimony: (1) Popperian, (2) Bayesian, (3) Likelihoodist, (4) and the Akaike framework. What is important to notice is that the Popperian, Bayesian, and Akaike criteria apply to theories or hypotheses subsequent to testing. The concern in biological systematics, however, is the relation of parsimony to abductive inference, not testing. As we will see later, the likelihoodist criterion can be considered important, but this must be in the context of both the auxiliary hypotheses used (sensu Thagard 1988) and informativeness (sensu Sober 1975).

Parsimony vs Likelihood What is Parsimony? ‚ Abductive inference is the reasoning process to provide answers (as explanatory hypotheses) to why-questions. ‚ The simplest/most parsimonious answers are those that require the fewest causes, because they are most informative for the questions asked (sensu Sober 1975), as well as rely on fewer auxiliary hypotheses ( sensu Thagard 1988), and thus result in hypotheses with greatest likelihood. ‚ In other words, the integrity of our observation statements, in the form of why-questions, should be maintained as fully as possible in the hypotheses that serve as answers to those questions.

We can now look at the relation between parsimony and the abductive inferences of hypotheses.

Parsimony vs Likelihood What is Parsimony? Why-question Q, leads to... Abductive inference X:

Abductive inference Y:

• auxiliary hypothesis, hx

• auxiliary hypotheses, hx, hy, hz

• theory, T1

• theory, T2

• shared similarities, e

• shared similarities, e

• phylogenetic hypothesis, h1

• phylogenetic hypothesis, h2

Consider this example of two abductive inferences, both attempting to answer the same set of why-questions. Notice that the principle differences between the two inferences are the number of auxiliary hypotheses, types of theories used, and hypotheses inferred.

Parsimony vs Likelihood What is Parsimony? Why-question Q, leads to... Abductive inference X:

Abductive inference Y:

• auxiliary hypothesis, hx

• auxiliary hypotheses, hx, hy, hz

• theory, T1

• theory, T2

• shared similarities, e

• shared similarities, e

• phylogenetic hypothesis, h1

• phylogenetic hypothesis, h2

Parsimony is the relation between a question(s) and answer(s). Hypothesis h1 is more informative than h2 with respect to question Q if h1 requires less extra information than h2 to answer Q.

For the question of which hypothesis is most parsimonious, we would take into consideration the relations between why-questions and the answers offered, as well as the respective number of auxiliary hypotheses used.

“Why are there marks in the sand?”

Let's look at a simple example of how parsimony is used in our process of abductively inferring explanatory hypotheses. You encounter these depressions in the sand, leading to the why-question shown here.

“Why are there marks in the sand?”

H1

H2

As answers to this question, you might consider two alternative hypotheses. One hypothesis explains the patterns as being footprints due to a person walking on the beach. The other hypothesis says that Bigfoot walked on the beach.

S(H1 | Q) > S(H2 | Q) H1 is more informative than H2; i.e. H1 requires less extraneous information beyond the observation to answer the question.

“Why are there marks in the sand?”

H1

H2

Obviously, we would say that hypothesis H1 is the more parsimonious (or simpler) answer to the why-question in contrast to H2. In other words, H1 is more informative than H2 because the former requires us to invoke fewer auxiliary hypotheses or extra information. Another way to say this is that H1 is more consistent with our background knowledge, given that we have no useful evidence to rationally consider Bigfoot as a possible causal factor. What is important to notice in this example is that parsimony or simplicity refers to the relation between the why-question and the answer to that question.

Parsimony vs Likelihood What is Parsimony? ‚ Notice that parsimony is not a criterion applied within the abductive inference of a hypothesis. Rather, it is the premises used in the inference that determines what hypotheses can be inferred. [But there is an exception in the case of computer algorithms, as we will see later.] ‚ The criterion of parsimony only applies to choices among hypotheses, as determined by criteria that determine hypothesis plausibility.

Parsimony vs Likelihood What is Parsimony? Why-question Q, leads to...

Abductive inference X:

Abductive inference Y:

• ‘common ancestry’

• ‘common ancestry’ + rate of character evolution

•abcd 0011 0111

•abcd 0011 0111

• phylogenetic hypothesis,

• phylogenetic hypothesis,

a b c d ‘2 steps’

a d c b ‘3 steps’

Let's now look at an example referring to phylogenetic inference. In the two abductive inferences, the premises differ in that one uses strict 'common ancestry,' while the other uses a combination of 'common ancestry' plus consideration of evolutionary rates of character change.

Parsimony vs Likelihood What is Parsimony? Why-question Q, leads to... Abductive inference X:

Abductive inference Y:

• ‘common ancestry’

• ‘common ancestry’ + rate of character evolution Parsimony: ‘Two-step’ hypothesis is most informative at answering the why-question.

•abcd 0011 0111 • phylogenetic hypothesis,

•abcd 0011 0111 • phylogenetic hypothesis,

a d c b

a b c d ‘2 steps’

‘3 steps’

Which hypothesis is most parsimonious?

In asking which hypothesis is most parsimonious, we have to consider the informativeness of each of the hypotheses relative to the why-questions that were asked. Clearly, the hypothesis requiring two steps provides an explanation that best maintains the integrity of the observations presented in the why-question.

Parsimony vs Likelihood What is Parsimony? Why-question Q, leads to... Abductive inference X:

Abductive inference Y:

• auxiliary hypothesis,

• auxiliary hypotheses,

(1) all observation statements are true

(2) morphology observation statements are true; (3) DNA sequence observations are not necessarily true

• ‘common ancestry’

• ‘common ancestry’ + rate of character evolution

•abcd 0011 0111

•abcd 0011 0111

• phylogenetic hypothesis,

• phylogenetic hypothesis,

a b c d ‘2 steps’

a d c b ‘3 steps’

We can also consider the auxiliary hypotheses, or background knowledge, required by each inference. Notice that the strict 'common ancestry' theory alone carries with it the auxiliary hypothesis that our observation statements are true. This should seem rational given that the goal of the inference is to answer the why-question regarding our observations, and there is the presupposition that our observations of shared characters are true. The other inference, however, must include two auxiliary hypotheses due to the fact that 'common ancestry' is used in conjunction with consideration of rates of character change. The auxiliary hypotheses are interesting because while one assumes our observations of some characters are true, we must include the second auxiliary hypothesis in order to apply the theory of rates of character evolution. You might notice that to introduce these auxiliaries, as well as the theory of rate change, is at odds with our observation statements as well as the why-question.

Parsimony vs Likelihood What is Parsimony? Why-question Q, leads to... Abductive inference X:

Abductive inference Y:

• auxiliary hypothesis,

• auxiliary hypotheses,

(1) all observation statements are true

• ‘common ancestry’ •abcd 0011 0111

(2) morphology observation statements are true; (3) DNA sequence observations are not necessarily true

Auxiliary hypothesis (3) is added to only deal with sequence data, whereas • ‘common ancestry’ + rate (1) addresses all data. Note also that (3) is of character evolution inconsistent with both observations and why•abcd questions.

0011 0111

• phylogenetic hypothesis,

a b c d ‘2 steps’

• phylogenetic hypothesis, The likelihood of a(b(c d)) is higher than a(d(c b)).

a d c b ‘3 steps’

We can see in this example that asking which hypothesis is most parsimonious entails considering the following issues: (a) Informativeness of the hypotheses relative to the why-question(s). Clearly the two-step hypothesis is the more informative; (b) The three-step hypothesis was inferred using two as opposed to one auxiliary hypothesis, where auxiliary hypothesis (3) is especially inconsistent with observation statements and why-questions; (c) We can give consideration to the likelihood of each hypothesis, which is actually related to conditions (a) and (b). The likelihood of hypothesis a(b(c d)) [two-step hypothesis] is higher for the fact that the hypothesis makes the observations in need of explanation more probable (from an abductive standpoint).

Parsimony vs Likelihood ‘Maximum Parsimony’ and Algorithms The inferences of phylogenetic hypotheses by ‘maximum parsimony’ computer algorithms do apply parsimony since the algorithms consider all (or as many as possible) cladograms during the process of finding the ‘shortest trees.’ This means that these computer algorithms have an inferential form more like: • D is a collection of character data. • Hypothesis H1 has x steps (= H1 explains D). • Hypothesis H2 has x+n steps (= H2 does not explain D as simply). • Therefore, H1 is preferable.

Note that it was mentioned earlier that parsimony is not a criterion applied *within* an abductive inference. While this generally is the case, the situation is somewhat different with regard to the implementation of abduction by computer algorithms, where there is a process of searching through 'tree space' using the criterion of minimizing the number of 'evolutionary steps' required to account for the distributions of characters. Recall that an abductive inference usually consists of a theory or set of theories that is/are applied to character data. In the case of computer algorithms, the form of abduction is closer to what is shown here, in which case parsimony is in fact applied within the inference.

Parsimony vs Likelihood ‘Maximum Parsimony’ and Algorithms Why-question Q, leads to... Abductive inference X: • ‘common ancestry’ •abcd 0011 0111

We can, however, apply the criteria of informativeness, number of auxiliary hypotheses, and likelihood to the manner in which computer algorithms apply the criterion of parsimony. Consider this example, in standard abductive form.

Parsimony vs Likelihood ‘Maximum Parsimony’ and Algorithms Why-question Q, leads to... Abductive inference X: • ‘common ancestry’ •abcd 0011 0111

a b c d

a b c d a c b d a d b c

2 ‘steps’ 3 ‘steps’

Which hypothesis is most parsimonious?

A computer algorithm will consider the variety of 'tree topologies' with respect to tree 'length.' In an analogous manner, algorithms consider parsimony in the selection process.

Parsimony vs Likelihood ‘Maximum Parsimony’ and Algorithms Why-question Q, leads to... Abductive inference X: • ‘common ancestry’ Parsimony: Two-step hypothesis is most informative at answering the whyquestion, and has the highest likelihood.

•abcd 0011 0111 • phylogenetic hypothesis,

a b c d

a b c d a c b d a d b c

2 ‘steps’ 3 ‘steps’

Which hypothesis is most informative?

Informativeness is the principle criterion in the selection process, especially given that the goal of searching for trees of minimal length is most consistent with the observations in the data matrix. And, as will be shown later with regard to the coding of characters, a data matrix does imply our why-questions. As well, in terms of likelihood, the selection of trees of minimal length will also be those that make our observations most probable.

Parsimony vs Likelihood What is likelihood? The likelihood L of a hypothesis h, given observation e is: L(h1 | e) o L(h2 | e).

The likelihood of hx is the degree to which e supports (abductively, in the present case) hx over hy. Hypothesis hx makes the occurrence of e more probable than hy: P(e | hx) o P(e | hy).

We now need to examine the nature of likelihood as it relates to abductive inference. The standard characterization of likelihood (L) for a hypothesis (h), given evidence (e), is shown here. Recall, however, that evidence (e) can be conceived in two fundamentally different ways. Similarly, to speak of 'support' for a hypothesis by way of evidence can have two very different meanings. The lack of making this distinction is one of the biggest misconceptions in systematics when it comes to speaking of methods, as well as the process of testing. The support for an abductively inferred hypothesis by evidence will be (in part) the characters explained by that hypothesis, i.e. the premises. Remember that such evidence is NOT the same as the evidence one would seek to actually test that hypothesis. Historically, the reference to evidence (e) in the concept of likelihood has been in regard to test evidence, not abductive evidence. As we will see in this section, this has important consequences for the method called 'maximum likelihood' in biological systematics, especially when one attempts to compare that method with what is called 'maximum parsimony.'

Parsimony vs Likelihood What is likelihood? The likelihood L of a hypothesis h, given observation e is: L(h1 | e) o L(h2 | e).

The likelihood of hx is the degree to which e supports (abductively, in the present case) hx over hy. Hypothesis hx makes the occurrence of e more probable than hy: P(e | hx) o P(e | hy). “‘Support’ is used to express the informal idea that some of our beliefs give us evidence for others.”

Sober (1975: 33), Simplicity

“In an argument, the evidence is given in the statements: the premises.”

Salmon (1984: 9), Logic

As we saw earlier, for any inference, it is the premises that provide the evidential support for a conclusion. Thus, one has to know the type of inference they are referring to when speaking of support: abductive support or inductive (testing) support.

Parsimony vs Likelihood What is Likelihood? Why-question Q, leads to... Abductive inference X:

Abductive inference Y:

• theory, T1

• theory, T2

• shared similarities, e

• shared similarities, e

• phylogenetic hypothesis, h1

• phylogenetic hypothesis, h2

Likelihood is the relation between evidence (premises) and answer(s)

L(h1 | e) = P(e | hx)

Since likelihood refers both to the support for a particular hypothesis, as well as the extent to which a hypothesis makes evidence most probable, likelihood in an abductive context is just the relation between the premises and hypothesis that serves as answer(s) to a why-question(s).

“Why are there marks in the sand?”

L(H1 | e)

L(H2 | e)

We can use the previous example to illustrate likelihood in relation to abductive inference. Based on the observation of the patterns in the beach sand, you have the alternate inferences of causes due to humans or a Bigfoot. We can recognize the likelihood of each hypothesis.

L(H1 | e) = L(H2 | e)

“Why are there marks in the sand?”

L(H1 | e)

L(H2 | e)

Considering just the relations between the evidence (foot prints) and the respective hypotheses, both hypotheses make the presence of the foot prints most probable; both hypotheses are equally supported. The likelihoods are the same.

L(H1 | e, b) > L(H2 | e, b)

“Why are there marks in the sand?”

L(H1 | e)

L(H2 | e)

But we know that we cannot just consider the relation between the foot prints and the alternate hypotheses. We also need to take into consideration our background knowledge. We know there is no legitimate theory of, or evidence for Bigfoots, whereas our experience with humans is well established. It is the inclusion of our background knowledge that enables us to show that hypothesis H1 has higher likelihood over H2.

Parsimony vs Likelihood What is Likelihood? Why-question Q, leads to... Abductive inference X:

Abductive inference Y:

• auxiliary hypothesis, hx

• auxiliary hypotheses, hx, hy, hz

• theory, T1

• theory, T2

• shared similarities, e

likelihood

• phylogenetic hypothesis, h1

• shared similarities, e • phylogenetic hypothesis, h2

Likelihood is the relation between evidence (premises) and answer(s)

L(h1 | e) = P(e | hx)

Implicitely, our background knowledge comprises part of the premises of the abductive inferences.

‘Maximum likelihood’ in Phylogenetic Inference (A) data matrix of observations

The following five slides illustrate the method of 'maximum likelihood' as applied in the inference of phylogenetic hypotheses.

‘Maximum likelihood’ in Phylogenetic Inference (A) data matrix of observations (B-C) map ‘character states’ on rooted tree

‘Maximum likelihood’ in Phylogenetic Inference (A) data matrix of observations (B-C) map ‘character states’ on rooted tree

(D) likelihood values for each ‘character state’ optimization on tree

‘Maximum likelihood’ in Phylogenetic Inference (A) data matrix of observations (B-C) map ‘character states’ on rooted tree

(D) likelihood values for each ‘character state’ optimization on tree

(E) sum of likelihoods for all ‘characters’

‘Maximum likelihood’ in Phylogenetic Inference (A) data matrix of observations (B-C) map ‘character states’ on rooted tree

(D) likelihood values for each ‘character state’ optimization on tree

(E) sum of likelihoods for all ‘characters’ (F) sum of log of likelihoods for all ‘characers’

Parsimony vs Likelihood The relation between parsimony and likelihood

We can now examine the relation between parsimony and maximum likelihood in the context of abductive inference in biological systematics.

Parsimony vs Likelihood The relation between parsimony and likelihood Why-question Q, leads to... Abductive inference X: • strict common ancestry

likelihood [= L(h | e)]

•abcd 0011 0111 • phylogenetic hypothesis,

a b c d ‘2 steps’

First, recall that likelihood is nothing more than the phenomenon of applying a theory as completely as possible to a set of effects.

Parsimony vs Likelihood The relation between parsimony and likelihood Why-question Q, leads to... Abductive inference X: • strict common ancestry

likelihood [= L(h | e)]

parsimony

•abcd 0011 0111 • phylogenetic hypothesis,

a b c d ‘2 steps’ Hypothesis has maximum likelihood, and is most parsimonious.

Parsimony, on the other hand, refers to the relation between why-questions and the hypothesis that serves as an answer. What should be apparent in this example is that parsimony and likelihood are different concepts. In this example, the inference leads to a hypothesis of maximum likelihood, as well as being most parsimonious.

Parsimony vs Likelihood The relation between parsimony and likelihood Why-question Q, leads to... Abductive inference X: • common ancestry + rate of character evolution

likelihood [= L(h | e)]

•abcd 0011 0111 • phylogenetic hypothesis,

a c b d ‘3 steps’

Consider a different example, where two theories are used: common ancestry plus rates of character evolution. Once again, the hypothesis must be of maximum likelihood for the fact that the theories are applied to effects.

Parsimony vs Likelihood The relation between parsimony and likelihood Why-question Q, leads to... Abductive inference X: • common ancestry + rate of character evolution

likelihood [= L(h | e)]

•abcd 0011 0111

not-parsimonious

• phylogenetic hypothesis,

a c b d ‘3 steps’ Hypothesis has maximum likelihood, but is not parsimonious.

But, with regard to parsimony, we see that the conclusion is not most parsimonious relative to our why-questions. One could just rely on the theory of strict common ancestry, and infer a hypothesis that is more parsimonious. And again, keep in mind that the likelihood obtained is automatic, given the premises. No matter what premises one uses, an abductive inference will produce a hypothesis of maximum likelihood. From the perspective of scientific inquiry, parsimony is the more critical issue since we should place a higher value on hypotheses that best answer our why-questions with the most plausible hypotheses. And, in the case of abductive inference, the most plausible hypotheses will be those that best explain our observations while maintaining the integrity of those observations.

Parsimony vs Likelihood The relation between parsimony and likelihood Why-question Q, leads to...

Abductive inference X:

Abductive inference Y:

• theory, T1

• theory, T2

• shared similarities, e

• shared similarities, e

• phylogenetic hypothesis, h1

• phylogenetic hypothesis, h2

a b c d ‘2 steps’

a c b d ‘3 steps’

If parsimony and likelihood conflict, how does one choose?

We can summarize the relation between parsimony and likelihood when it comes to abductive inference. Consider the two inferences shown here. The premises differ in the theory each uses. If it is the case that parsimony and likelihood are not necessarily consistent with each other, as we saw in the previous example, on what basis should one choose between these conflicting hypotheses?

Parsimony vs Likelihood The relation between parsimony and likelihood Why-question Q, leads to...

Abductive inference Y:

Abductive inference X:

• theory, T2

• theory, T1 • shared similarities, e • phylogenetic hypothesis, h1

likelihood

• shared similarities, e • phylogenetic hypothesis, h2

‚ Recall that in any abductive inference, the likelihood of the conclusion(s) is trivially maximum.

It makes no sense to attempt to compare the likelihoods of the two hypotheses, since both inferences must, by default, produce hypotheses with respective maximum likelihoods.

Parsimony vs Likelihood The relation between parsimony and likelihood Why-question Q, leads to...

Abductive inference X:

parsimony

• theory, T2

• theory, T1 • shared similarities, e • phylogenetic hypothesis, h1

Abductive inference Y:

likelihood

• shared similarities, e • phylogenetic hypothesis, h2

‚ Recall that in any abductive inference, the likelihood of the conclusion(s) is trivially maximum. ˆ Likelihood must be considered within the context of parsimony. The two concepts cannot be weighed against one another.

Where comparison becomes critical is with regard to parsimony. Parsimony is a criterion that does not reside *within* an inference, but instead is determined in relation to our why-questions. The consequence is that parsimony and likelihood are not conditions that can be compared to one another, contrary to what is so commonly claimed in the systematics literature.

Parsimony vs Likelihood Why-question Q, leads to... Abductive inference X:

parsimony

Abductive inference Y:

• auxiliary hypothesis, hx

• auxiliary hypotheses, hx, hy, hz

• theory, T1

• theory, T2

• shared similarities, e • phylogenetic hypothesis, h1

likelihood

• shared similarities, e • phylogenetic hypothesis, h2

Why-questions determine the roles of parsimony and likelihood in abductive inference.

Considering the question of which hypothesis is most parsimonious, we would want to take into consideration the relations between why-questions and the answers offered, as well as the respective auxiliary hypotheses used.

Parsimony vs Likelihood

parsimony

likelihood

Incorrect relationship between parsimony and likelihood.

‚ Recall that in any abductive inference, the likelihood of the conclusion(s) is trivially maximum. ˆ Likelihood must be considered within the context of parsimony. The two concepts cannot be weighed against one another.

Within biological systematics, the relationship between parsimony and likelihood has been seen as one in which parsimony and likelihood are to be compared to one another. But as we have seen, in the context of abductive inference, such a relationship is incorrect.

Parsimony vs Likelihood

parsimony likelihood

Correct relationship between parsimony and likelihood.

‚ Choosing between methods such as ‘maximum parsimony’ and ‘maximum likelihood’ requires careful consideration of the auxiliary hypotheses and theories used in each.

The correct relationship between parsimony and likelihood is where parsimony entails likelihood. The consequence is that we cannot compare methods called 'maximum parsimony' and 'maximum likelihood' on the basis of likelihood. The comparison must be made in terms of parsimony, and that requires that we examine the auxiliary hypotheses and theories used with each approach, as well as the ability of the abductively-inferred hypotheses to serve as answers to our why-questions.

Parsimony vs Likelihood wh

y - quest i o ns

parsimony likelihood

Correct relationship between parsimony and likelihood.

But... it is the nature of our why-questions that dictate phylogenetic inference. And why-questions in conjunction with parsimony (not likelihood) determine hypothesis plausibility. But ultimately, our why-questions will need to be considered in relation to parsimony.

Where does this leave us? Why-question Q, leads to... Abductive inference X:

Abductive inference Y:

• auxiliary hypothesis, hx

• auxiliary hypotheses, hx, hy, hz

• theory, T1

• theory, T2

• shared similarities, e • phylogenetic hypothesis, h1

likelihood

• shared similarities, e • phylogenetic hypothesis, h2

1.

Why-questions seek common-cause explanations of fact(s) and foil(s).

2.

Per (1), abductive inferences require common cause theories.

3.

Per (1) & (2), a phylogenetic hypothesis is most parsimonious relative to why-questions.

4.

Per (1) & (2), likelihood is proximately determined by why-questions, distally by premises.

5.

CONCLUSION: The constraint on phylogenetic inference, and thus methods, comes from why-questions, not parsimony or likelihood.

We can summarize the relations between our why-questions with parsimony and likelhood as follows...

Why-Questions and Abductive Reasoning Why-questions abductive inference

parsimony

likelihood

hypothesis(es)

plausibility

As noted in the previous slide, the plausibility of our hypotheses is primarily determined by our why-questions, and secondarily by the relations of those questions to hypotheses in terms of parsimony.

‘Maximum Likelihood’ Problems with likelihood methods in phylogenetic inference

“The main idea behind phylogeny inference with maximum-likelihood is to determine the tree topology, branch lengths, and parameters of the evolutionary model (e.g. transition/transversion ratio, base frequencies, rate variation among sites)... that maximize the probability of observing the sequences at hand.” Schmidt & von Haeseler (2010: 183), Phylogenetic inference using maximum likelihood methods.

Based on what we have now covered with regard to abductive inference, we can identify several problems with the 'maximum likelihood' method.

‘Maximum Likelihood’ Problems with likelihood methods in phylogenetic inference

1: c-us

a-us: 0

b-us: 0

‘parsimony’ tree

1: d-us

Let's look the standard argument that has been used to defend likelihood as preferable to parsimony. We have the phylogenetic hypothesis shown here, inferred using strict 'common ancestry.' [nb: It is incorrect to refer to this as a 'parsimony' tree. While the hypothesis is most parsimonious, that is only because of the theory applied to observations in the abductive inference. The hypothesis is also of maximum likelihood. As we have already seen, the relevant issue is not parsimony versus likelihood, but rather the theories one chooses, in conjunction with background knowledge, in relation to why-questions.]

‘Maximum Likelihood’ Problems with likelihood methods in phylogenetic inference

a-us: 0

1: c-us

‘parsimony’ tree

b-us: 0

1: d-us

c-us: 1 1: d-us ‘long branch’ attraction

a-us: 0

‘true’ tree

0: b-us

The argument is that 'parsimony' can lead to an 'incorrect' answer. The reason being that we have not taken into consideration the phenomenon known as long branch attraction.

‘Maximum Likelihood’ Problems with likelihood methods in phylogenetic inference

a-us: 0

b-us: 0 The method of ‘maximum likelihood’ inference is claimed to be better than ‘parsimony’ because it...

1: c-us

‘parsimony’ tree

1: d-us

c-us: 1 1: d-us

• considers rates of evolutionary change ‘along branches;’

‘long branch’ attraction

a-us: 0

‘true’ tree

0: b-us

Thus, we are told that 'maximum likelihood' is more effective at inferring 'correct' hypotheses for the fact that (1) the method takes into consideration rates of character evolution, and (2) the method has the property of 'statistical consistency.'

‘Maximum Likelihood’ Problems with likelihood methods in phylogenetic inference

“The main idea behind phylogeny inference with maximum-likelihood is to determine the tree topology, branch lengths , and parameters of the evolutionary model (e.g. transition/transversion ratio, base frequencies, rate variation among sites)... that maximize the probability of observing the sequences at hand.” Schmidt & von Haeseler (2010: 183), Phylogenetic inference using maximum likelihood methods.

Let's first address the related issues of branch length and evolutionary rates.

‘Maximum Likelihood’ Problems with likelihood methods in phylogenetic inference

Branch length:

“...the number of substitutions along each branch of the tree...” Schmidt & von Haeseler (2010: 185), Phylogenetic inference using maximum likelihood methods.

An issue noted earlier, the fallacy of reification, is exemplified by the idea that cladogram branches can be assigned lengths. A notion of branch length works if cladograms have some tangible quality as objects in time and space, where their component parts consist of 'branches,' 'nodes,' and 'leaves.' But this is not the case. As we have already seen, cladograms are nothing more than graphic devices that imply at least three classes of causal events by way of (1) species hypotheses, and (2) phylogenetic hypotheses involving the events of (a) character origin/fixation and (b) subsequent population splittings. The concept of branch length not only ignores the totality of these hypotheses, it incorrectly assigns properties to 'branches' that are irrelevant to our goal of scientific inquiry. The separate hypotheses of character origin/fixation that are implied by, and therefore comprise branches, cannot be meaningfully summarized in terms of 'length.'

‘Maximum Likelihood’ Problems with likelihood methods in phylogenetic inference Rate variation violates the common cause requirement of whyquestions for explaining facts and foil in terms of phylogenetic hypotheses. “Why do members of a-us and b-us have A at position 546 in contrast to T, C, or G?”

facts:

foils:

common cause process: character origin/fixation + population splitting event

“The main idea behind phylogeny inference with maximum-likelihood is to determine the tree topology, branch lengths, and parameters of the evolutionary model (e.g. transition/transversion ratio, base frequencies, rate variation among sites)... that maximize the probability of observing the sequences at hand.” Schmidt & von Haeseler (2010: 183), Phylogenetic inference using maximum likelihood methods.

In relation to branch length, the inferences of cladograms using 'maximum likelihood' takes into consideration rates of character evolution. The idea is that such rates need to be acknowledged in order to avoid the 'error of long branch attraction.' The problem with invoking evolutionary rates in the context of abductive inference is that it is not correctly related to our why-questions. Recall that why-questions operate under the presupposition that the facts and foils referred to are true. By extension, we would want to pursue answering such questions by way of common cause events. The use of evolutionary rates, however, goes against the presupposition in why-questions if one is applying those rates to observations of shared similarities. In other words, considering evolutionary rates is part of one's background knowledge that needs to be taken into consideration *prior to* an abductive inference, not *part of* the inference. This means one would need to alter their why-questions, shown in the next slide.

‘Maximum Likelihood’ Problems with likelihood methods in phylogenetic inference

Abduction Theory:

If cause X, then effect Y

TRUE

Effects:

Effects of type Y are observed

TRUE

Hypothesis:

Causal condition(s) X might have occurred

TRUE / FALSE

Recall that for any inference, we should assume that our premises are true.

Including evolutionary rates in the context of the why-questions being answered by way of phylogenetic inference requires that we not assume our observation statements to be necessarily true. In other words, while an observation statement might claim that a group of organisms have the same character, to bring in considerations of rates of character evolution immediately implies that we cannot assume the truth of observation statements. In such an instance, we should not consider moving forward with explanation. Unfortunately, proponents of 'maximum likelihood' ignore this concern or are unaware of it. Recall from our earlier overview of inference that a basic assumption required to ensure any inference is rational is that we assume the truth of the premises. Without such an assumption, the utility of our abductive inferences would be reduced to nothing more than a trivial exercise.

‘Maximum Likelihood’ Problems with likelihood methods in phylogenetic inference

Abduction TRUE

Theory:

If cause X, then effect Y

Effects:

Effects of type Y are observed

TRUE / FALSE

Hypothesis:

Causal condition(s) X might have occurred

TRUE / FALSE

The method of ‘maximum likelihood,’ incorrectly assumes our observation statements of shared similarities are not necessarily true.

Within the context of 'maximum likelihood,' the premises of an abductive inference must be modified, such that we effectively do not trust our observation statements. Under the assumption that we should either not trust our observation statements or that statements should be revised, per some additional theory, we are in no position to then move forward with inferring explanatory hypotheses to account for the observations we do not even regard as true!

‘Maximum Likelihood’ Problems with likelihood methods in phylogenetic inference

Solution to establish common cause...

facts:

foils:

common cause process: character origin/fixation + population splitting event

facts:

foils:

common cause process: character origin/fixation + population splitting event

To correct the problem caused by considering rates of character evolution in relation to why-questions, the solution requires that one alter the presuppositions in their questions. We would have to acknowledge that what might initially appear to be shared characters are actually not the case. Rather, what might have been one why-question regarding shared characters must be divided into several whyquestions that acknowledge that one must actually explain different features that accommodate their background knowlege that characters have evolved at different rates.

‘Maximum Likelihood’ Problems with likelihood methods in phylogenetic inference

a-us: 0

b-us: 0 The method of ‘maximum likelihood’ inference is claimed to be better than ‘parsimony’ because it...

1: c-us

‘parsimony’ tree

c-us: 1 1: d-us

• considers rates of evolutionary change ‘along branches;’ • has the property of ‘statistical consistency’ (= ‘convergence’).

1: d-us

‘long branch’ attraction

a-us: 0

‘true’ tree

0: b-us

The second issue that needs to be addressed with the method of 'maximum likelihood' has to do with the claim that the method has the property of statistical consistency or convergence. But as we will see, while consistency can be applied to statistical hypotheses, as part of testing, it cannot be applied to abductive inferences.

‘Maximum Likelihood’ Statistical consistency as a defense of likelihood methods

“An estimator is consistent if, as the amount of data gets larger and larger (approaching infinity), the estimator converges to the true value of the parameter with probability 1.” “The inconsistency of parsimony has been the strongest challenge to its use. It becomes difficult to argue that parsimony methods have logical and philosophical priority, if one accepts that consistency is a highly desirable property.” J. Felsenstein (2004: 107, 121, respectively) Inferring Phylogenies

The first quote provides an accurate characterization of consistency, insofar as it applies to statistical, not explanatory hypotheses. And the fact that this applies to statistical hypotheses means that the context in which consistency applies is inductive (sensu stricto), not abductive. Unfortunately, attempts have been made to claim that consistency is a relevant arbiter of phylogenetic methods. But as we will see, consistency has no relevance to abductive inference. This means that consistency is also irrelevant to the subject of parsimony. Felsenstein (2004) is incorrect in suggesting that parsimony does not have 'philosophical priority.' As we saw earlier, parsimony transcends abductive inferences, thus does indeed have priority.

Statistical consistency, or convergence, does not apply to abductive reasoning: “[Abduction] is the only kind of reasoning which supplies new ideas, the only kind which is, in this sense, synthetic. Induction is justified as a method which must in the long run lead up to the truth, and that, by gradual modification of the actual conclusion.

C.S. Peirce, Collected Works (1932: 777, emphasis added).

The quote shown here by C.S. Peirce points to the fundamental difference between abduction and induction. As we have seen, abduction produces new explanatory hypotheses. Induction is the process of testing those hypotheses. Peirce, among others, held the view that with the continual introduction of new test evidence, we should come ever closer to having the 'true' hypothesis ("in the long run"). You might notice that Peirce's description of induction is similar to what was described earlier for consistency. This is not surprising, given that statistics deals with the testing of statistical hypotheses via induction. Peirce says, "Induction is justified as a method which must in the long run lead up to the truth." This is the very essence of consistency.

Statistical consistency, or convergence, does not apply to abductive reasoning: “[Abduction] is the only kind of reasoning which supplies new ideas, the only kind which is, in this sense, synthetic. Induction is justified as a method which must in the long run lead up to the truth, and that, by gradual modification of the actual conclusion.

There is no such warrant for [Abduction]. The hypothesis which it problematically concludes is frequently utterly wrong itself, and even the method need not ever lead to the truth; for it may be that the features of the phenomena which it aims to explain have no rational explanation at all. Its only justification is that its method is the only way in which there can be any hope of attaining a rational explanation.”

C.S. Peirce, Collected Works (1932: 2.777, emphasis added).

But, Peirce was very clear, as well as correct, in also pointing out that abduction cannot be justified in the same way as induction. In other words, we cannot justify abduction in terms of consistency. An abductive inference simply leads to an explanatory hypothesis. There is no means within abduction itself to correct the hypotheses it infers. A very clear example of the failure to associate consistency with abduction is presented in the next slide.

Statistical consistency, or convergence, does not apply to abductive reasoning – An example: Time t1: theory x + data ‘a’

H1: A(C(D(BE)))

t2: theory x + data ‘a+b’

H2: A(C(B(DE)))

t3: theory x + data ‘a+b+c’

H3: A(B(D(CE)))

t4: theory x + data ‘a+b+c+d’

H4: A(B(C(DE)))

‚

t5: theory x + data ‘a+b+c+d+e’

H5: A(B(C(DE)))

‚

t6: theory x + data ‘a+b+c+d+e+f’

H6: A(B(C(DE)))

‚

This example shows that the concept of consistency cannot be applied to a series of abductive inferences. Six successive inferences are shown, with new data added in each inference. Notice that in the first four inferences, different hypotheses are produced. But in inferences five and six, it appears that the hypotheses are the same as in inference four. Simplistically, it would be claimed that with the addition of more and more data, we have 'converged' on the 'true tree.' In other words, we have an instance of consistency. But this conclusion is entirely incorrect for several reasons. Hypotheses 4-6 are not identical. While the 'tree topologies' appear the same, the hypotheses implied by the topologies are different. Keep in mind a tree topology only shows a branching structure, not the actual hypotheses implied by that structure. Given that each inference involves different sets of effects, each inferred conclusion will lead to a new hypothesis, regardless of the fact that the topologies appear to be the same. In conclusion, there can be no sense of 'consistency' achieved with successive abductive inferences. The argument that 'maximum likelihood' can achieve consistency is incorrect, and is a product of the lack of understanding the nature of abductive inference, the lack of understanding the nature of the hypotheses implied by cladograms, and confusing explanatory with statistical hypotheses.

Parsimony vs Maximum Likelihood Some Conclusions – I

There can be no distinctly separate phylogenetic methods termed ‘maximum parsimony’ and ‘maximum likelihood.’ Both ‘methods’ do consider parsimony and likelihood. The claim that there can be distinct methods called ‘parsimony’ and ‘likelihood’ is the result of (1) misunderstanding the inferential basis of systematics, (2) a lack of focus on the why-questions being asked, and/or (3) incorrect characterizations of parsimony and likelihood.

With regard to the criteria known as parsimony and maximum likelihood, we can now draw several conclusions. Parsimony and likelihood cannot be treated as separate in the context of abductive inference. By the nature of abduction, likelihood is automatically considered within the inference itself. The parsimony criterion transcends the inference given that we have particular why-questions that need to be answered, and these why-questions impose limitations on the inference.

Parsimony vs Maximum Likelihood Some Conclusions – II ‘Maximum parsimony:’ Contrary to the popular belief that no causal assumptions are made, the formal structure of such inferences show that the theory of ‘descent with modification’ is in fact used. ‘Maximum likelihood.’ The basis for ‘likelihood’ methods is that rates of evolutionary change within lineages must be taken into consideration in addition to ‘descent with modification.’ Such an approach is defective because it does not correctly consider the form of why-questions, theories, and inferences.

More basic conclusions.

Parsimony vs Maximum Likelihood Some Conclusions – III

The relevant criterion for phylogenetic inference is not ‘parsimony versus maximum likelihood.’ The real criterion is the relations between why-questions and the theory/theories used to answer those questions.

This is the most fundamental conclusion. What controls our abductive inferences are our why-questions in conjunction with the specific theories that allow for providing answers to those questions. Likelihood is immaterial to this issue since it is automatically satisfied within the inference, while parsimony is a vital criterion to ensure the plausibility of our hypotheses.

Parsimony vs Maximum Likelihood Some Conclusions – IV The only means by which phylogenetic methods can be evaluated is by way of the observations (effects) and why-questions being asked. The limiting factor for any phylogenetic inference is that the conclusion, as an explanation sketch, should be as consistent as possible with one’s observations and why-questions (parsimony). The consequence is that the imposed distinction between ‘parsimony’ and ‘likelihood’ is both unnecessary and irrelevant to the objective of systematics.

This conclusion is an extension of the previous conclusion.

Systematics and Abduction Some Implications for Phylogenetic Inference Once we acknowledge the formal structure of the why-questions we ask in systematics, and the abductive form of inference to hypotheses, there are distinct implications for the following issues:

1.

‘Parsimony’ vs ‘maximum likelihood’ methods.

2.

‘Bayesian’ inference.

The third common method of phylogenetic inference to be considered is the attempted use of Bayesian inference.

Systematics and Abduction Bayesian inference

P(H) • P(E | H) Bayes’ Theorem

P(H | E) = P(E)

• The goal of Bayesian inference is to quantify changes in belief in a hypothesis as a result of testing. • This is completely separate from the initial abductive inference of a hypothesis.

initial belief in hypothesis H + recent test evidence E = revised belief in H

First, we need to understand Baye's Theorem and why/how it is used in science in general. As to why the theorem is used, Bayesian inference attempts to quantify changes in belief in hypotheses as a result of the introduction of test evidence. As we have already seen, hypothesis testing is a matter of inductive inference. In order to engage in testing, one first must have a hypothesis. Obtaining that hypothesis falls within the realm of abductive inference. The act of inferring a hypothesis subsequent to the observations of effects is completely separate from the act of then engaging in the testing of that hypothesis. To consider the use of Baye's Theorem, we would be restricted to the realm of testing (induction), not abduction. Herein begins the problems of trying to apply Bayesian inference to phylogenetic inference.

Systematics and Abduction Bayesian inference P(H) • P(E | H) P(H | E) = P(E) initial belief in hypothesis H; P(H)

test evidence E; P(E | H)

Bayes’ Theorem

revised belief in hypothesis H; P(H | E)

This diagram summarizes the applications of the different components of Baye's Theorem. We have our hypothesis (H) that would have been inferred via abductive inference, plus the observed test evidence (E). Baye's Theorem is applied to the hypothesis and test evidence to conclude a revised belief in the hypothesis.

Systematics and Abduction Bayesian inference

P(H) • P(E | H) P(H | E) = P(E) H

The hypothesis to be tested.

E

All test evidence directly relevant to the truth of H.

P

The probability that something stated is true.

Let's now identify the symbols referred to in Baye's Theorem.

Systematics and Abduction Bayesian inference

P(H) • P(E | H) P(H | E) = P(E) P(H)

The prior probability of hypothesis H being true, independent of test evidence E. Derived from a previous abductive inference.

Hypothesis H would have been previously inferred to explain some set of observations. In the case of phylogenetic systematics, those observations would be character data. But, the approach in ‘Bayesian phylogenetic inference’ is to consider H to be all possible cladogram topologies, without any consideration of the observations being explained by those topologies. This is an incorrect interpretation of P(H)!

We can now identify the components in the theorem.

Systematics and Abduction Bayesian inference

P(H) • P(E | H) P(H | E) = P(E) P(E | H)

The conditional probability of finding test evidence E assuming that hypothesis H is true. This addresses the relevance of the evidence to the hypothesis being tested.

Often referred to as the likelihood, or predictive power of hypothesis H. In terms of testing, E should have the highest probability given H, and lowest probability given ~H.

We can now identify the components in the theorem.

Systematics and Abduction Bayesian inference

P(H) • P(E | H) P(H | E) = P(E) P(E)

The marginal probability of evidence E being observed independent of hypothesis H.

We can now identify the components in the theorem.

Systematics and Abduction Bayesian inference

P(H) • P(E | H) P(H | E) = P(E) P(H | E)

The posterior probability of hypothesis H being true given the existence of test evidence E.

This would be the revised belief in hypothesis H after considering test evidence E.

We can now identify the components in the theorem.

Systematics and Abduction ‘Bayesian’ inference The strange case of trying to apply Bayes’ Theorem to phylogenetic inference.

P(H) • P(E | H) P(H | E) = P(E)

P(Tree | data) =

P(Tree) • P(data | Tree) P(data)

The attempt to apply Bayesian inference to phylogenetic inference is based on the claim that the components of the theorem can be interpreted in terms of character data and cladograms ('trees'). From what we have covered in this course regarding the nature of abductive inference, the use of induction (sensu stricto) to test hypotheses, and the fact that Baye's Theorem is intended to be used relative to testing, it should be apparent that the modification of the Theorem shown here is incorrect. P(Tree) is meaningless because it does not refer to *the* hypothesis previously inferred by way of abduction and that is currently to be tested. Rather, systematists claim that P(Tree) comprises all possible 'tree topologies.' A 'tree topology' is a meaningless construct since it makes no reference to characters to be explained by the variety of hypotheses implied by a cladogram. P(data | Tree) is not equivalent to P(E | H) since (1) the 'data' are not valid test evidence, and (2) the 'Tree' cannot be equated with H since the former was never abductively inferred such that it can be considered relative to test evidence. P(data) is meaningless since this does not refer to test evidence. P(Tree | data) can not be equated with the posterior probability, P(H | E). As no testing has been performed, from which test evidence E can be used to determine changes in belief in a hypothesis, P(Tree | data) cannot be validly determined.

‘Bayesian’ Inference a-us b-us c-us

b-us a-us c-us

c-us a-us b-us

1.0

‘prior distribution’ P(tree)

0.5 0.0

P(data | tree): data + ‘tree model’ + ‘substitution model’

‘posterior distribution’ P(tree | data)

b-us a-us c-us

1.0

c-us a-us b-us a-us b-us c-us

48%

0.5

32%

0.0

20% topology / branch lengths

This illustration presents a schematic outline of how 'Bayesian' inference is performed in phylogenetics. Notice that 'prior probabilities,' P(tree), refers to all possible 'tree topologies,' not any legitimate prior probability. P(data | tree) refers to the actual process of inferring phylogenetic hypotheses. Of course, such an action would never occur in proper Bayesian inference. The inferred hypotheses are then assigned 'probabilities' based on topology and branch lengths. This could not be a posterior probability since no valid prior probability was available, plus, this is not the product of hypothesis testing.

‘Bayesian’ Inference – Some Problems a-us b-us c-us

b-us a-us c-us

c-us a-us b-us

1.0 0.5

‘prior distribution’ P(tree)

0.0

• These ‘topologies’ are not explanatory hypotheses – they refer to no effects or causal conditions. And these are not hypotheses to be considered for testing. There can be no prior probabilities, P(H), required for Bayesian inference.

Summary of problems with the attempt to present 'prior probabilities.'

‘Bayesian’ Inference – Some Problems • The ‘data’ do not constitute test evidence as required for Bayesian inference. • The ‘posterior probability’ is not equivalent to P(H | E) required for Bayesian inference. P(data | tree): data + ‘tree model’ + ‘substitution model’

‘posterior distribution’ P(tree | data)

b-us a-us c-us

1.0

c-us a-us b-us a-us b-us c-us

48%

0.5

32%

0.0

20% topology / branch lengths

Summary of problems with the attempt to present 'likelihoods' and 'posterior probabilities.'

‘Bayesian’ Inference – Some Problems

“Bayesian methods are closely related to likelihood methods, differing only in the use of a prior distribution of the quantity being inferred, which would typically be the tree [sic].” J. Felsenstein (2004: 288) Inferring Phylogenies

It is apparent from what we have just seen, that so-called 'Bayesian inference' used in phylogenetics is really nothing more than a derivation of maximum likelihood methods. This is not surprising, given that Baye's Theorem refers to the act of testing hypotheses, which is strictly inductive,, not their (abductive) inference.

‘Bayesian’ Inference • Is ‘Bayesian’ phylogenetic inference actually Bayesian? No (1) There are no valid hypotheses presented from which prior probabilities, P(H), can be given. (2) Character data are not valid test evidence E, from which P(E | H) can be derived. (3) The ‘posterior probabilities’ are meaningless since there can be no empirical comparison between original hypotheses, P(H) (since they aren't presented), and hypotheses subsequent to testing, P(H|E), due to the fact that no valid testing even occurred. Thus, no net change in belief in H has occurred.

(4) Bayesian inference is not intended to initially infer hypotheses. That task must first occur via abductive inference, after which testing using Bayesian techniques can proceed (but will not happen with cladograms).

Herein are some conclusions.

Systematics and Abduction ‘Bayesian’ Inference versus ‘Maximum Likelihood’

P(H) • P(E | H) P(H | E) = P(E) “Likelihood asks ‘what is the probability of the data given the model [p(data | model)]?’ The alternative question is: ‘what is the probability that the model is correct given the data [p(model | data)]?’ ” Wiley & Lieberman (2011: 219), Phylogenetics: Theory and Practice of Phylogenetic Systematics

It is instructive to look at the way likelihood and Bayesianism are contrasted in a recent book on phylogenetics. The difficulty is that the two questions shown here are irrelevant to one another, and as such cannot be regarded as alternatives. As we have seen, likelihood can be interpreted in two different ways, depending on how one interprets evidence, E. There is abductive evidence, in the form of observed effects we wish to explain; and there is test evidence. In the context of the two questions shown here, clearly the only way to interpret evidence/data is relative to hypothesis testing, not the abductive inference of a hypothesis. Similarly, asking the second question refers to posterior probabilities, which are determined as consequences of testing, not abductive inference.

Systematics and Abduction ‘Bayesian’ Inference versus ‘Maximum Likelihood’ P(H) • P(E | H) P(H | E) = P(E) Likelihood asks “what is the probability of the data given the model” [p(data | model)]? The alternative question is: “what is the probability that the model is correct given the data” [p(model | data)]? Wiley & Lieberman (2011: 219), Phylogenetics: Theory and Practice of Phylogenetic Systematics

Given that phylogenetic inference is abductive, these are irrelevant questions. So what we find is that the questions asked by Wiley & Lieberman are irrelevant to phylogenetic inference. It is only when we have a clear understanding of the abductive nature of hypothesis formation that we begin to see that methods intended for hypothesis inference cannot be compared in terms of likelihood and Bayesianism.

Systematics and Abduction ‘Bayesian’ inference

The basis for Bayesianism is not the inference of explanatory hypotheses, but rather the confirmation of those hypotheses. As a result, the inference of hypotheses is confused with the testing of hypotheses. There is nothing in the inferences of explanatory hypotheses that is ‘Bayesian.’

P(H) • P(E | H) P(H | E) = P(E) Bayes Theorem

The Philosophy of Biological Systematics Course Outline – Part 3

1.

The requirement of total evidence.

2.

Homology & homogeny & homoplasy.

3.

Character coding.

4.

The mechanics of hypothesis testing in biological systematics.

THE REQUIREMENT OF TOTAL EVIDENCE IN BIOLOGICAL SYSTEMATICS

Kirk Fitzhugh Natural History Museum of Los Angeles County

In this section of the course, we will examine a principle of reasoning that has frequently been discussed in the systematics literature for over 20 years: the requirement of total evidence (RTE). Oddly, for a subject that has received so much attention among systematists, the RTE is rarely correctly described, either by proponents or those who claim it can be ignored. We will make sense of the RTE with regard to inference in general, and from there we can readily see why it must be used in the inferences of systematics hypotheses. We also will examine problems with many of the standard arguments in systematics for endorsing the RTE.

‘Morphological’ observations Systematics hypotheses are routinely inferred from what are called ‘morphological’ characters...

It is common practice in biological systematics that hypotheses are inferred from what are often referred to as 'morphological' characters.

‘Morphological’ observations

‘Molecular’ observations

CCAGAGGCCCAACUGGUAAACGGGC CCG-AAGCUCAACGGGAUAAUGAGC CCG-AAGCCGAACGGGAAAACCGGC CC-CAAGCGC-AGGGGAGAA-GCGC CCG-ACGCCA-ACGGGAGAA-UGGC CCGUUUUCAG-UCGGGAAAAACUGA CCGUUACUCC-UCGGGAUAAAGGAG CCGUAAGAGG-ACGGGAUAAACCUC CCG-UAGGAG-GCGGGAUAU-CUCC CCG--UGCCG-GCGGGAUAU-CGGC CCG-AACUCG-ACGGGAUAA-CGAG CCG--ACUCG--CGGGAUAA-CGAG

...but it also is common to see hypotheses only inferred from what are called ‘molecular’ characters...

As well, we commonly see hypotheses inferred from what are called 'molecular' observations.

‘jaws’

‘Morphological’

A C B D

A B C D

sperm

Also, one might only rely on a particular subset of features, such as jaws or sperm, from which systematics hypotheses are inferred for each of these classes of characters.

We also encounter instances in which a particular worker is interested in a specific organ or tissue system, and will draw inferences from just those observations. For instance, among the marine worms, or polychaetes, it has not been uncommon for workers to place emphasis on aspects of jaw structures, from which they infer phylogenetic relationships. Alternatively, others focus on histological features, such as sperm morphology, and make statements about phylogeny based only on characters associated with those structures.

‘jaws’

‘Morphological’

A C B D

A B C D

sperm

‘Molecular’ CCGUUACUCC-UCGGGAUAAAGGAG CCGUAAGAGG-ACGGGAUAAACCUC CCG-UAGGAG-GCGGGAUAU-CUCC CCG--UGCCG-GCGGGAUAU-CGGC CCG-AACUCG-ACGGGAUAA-CGAG CCG--ACUCG--CGGGAUAA-CGAG

A B C D

CCAGAGGCCCAACUGGUAAACGGGC CCG-AAGCUCAACGGGAUAAUGAGC CCG-AAGCCGAACGGGAAAACCGGC CC-CAAGCGC-AGGGGAGAA-GCGC CCG-ACGCCA-ACGGGAGAA-UGGC CCGUUUUCAG-UCGGGAAAAACUGA

A B C D

And, it is common to see studies where separate sets of molecular data are used to infer different hypotheses.

We commonly see a similar situation with so-called 'molecular' data, where separate phylogenetic hypotheses are inferred from different sets of sequences for the same groups of organisms.

But.... There is another point of view, where it has been claimed that all available data should be treated together in the inference of systematics hypotheses. The question of whether or not to combine data sets has become known as the ‘requirement of total evidence debate.’

A D C B

All data: o1, o2, ... on + ox, oy, ... on CCGUUACUCC-UCGGGAUAAAGGAG CCGUAAGAGG-ACGGGAUAAACCUC CCG-UAGGAG-GCGGGAUAU-CUCC CCG--UGCCG-GCGGGAUAU-CGGC CCG-AACUCG-ACGGGAUAA-CGAG CCG--ACUCG--CGGGAUAA-CGAG CCAGAGGCCCAACUGGUAAACGGGC CCG-AAGCUCAACGGGAUAAUGAGC CCG-AAGCCGAACGGGAAAACCGGC CC-CAAGCGC-AGGGGAGAA-GCGC CCG-ACGCCA-ACGGGAGAA-UGGC CCGUUUUCAG-UCGGGAAAAACUGA

There is, however, another common point of view, where it is suggested that one should take all types of 'morphological' and 'molecular' observations, and treat them together in the inference of phylogenetic hypotheses.

‘Requirement of Total Evidence Debate’ (e.g. Felsenstein, 2004: Inferring Phylogenies)

Should one infer separate hypotheses from partitioned data sets, or make inferences from all available data combined?

These different opinions on the inferences of systematics hypotheses might best be referred to as the 'requirement of total evidence debate.' In other words, should one infer separate hypotheses from partitioned data sets, or make inferences from all available data combined?

This issue has especially been a concern since the publication of Kluge's (1989) paper on the subject, where he explicitly refers to the philosophical requirement of total evidence as providing the justification for all available data being treated together in the inference of phylogenetic hypotheses. Unfortunately, Kluge did not provide a detailed review of the requirement of total evidence (hereafter, RTE) or its relations to systematics inference. But, what is especially remarkable is that the subsequent discussions on the subject in the systematics literature have never included any extensive account of the requirement. While much has been written either defending the RTE or claiming that it can be ignored, neither advocates nor opponents have ever presented the philosophical basis of the requirement with sufficient accuracy to clearly defend their positions.

“Incongruence between different data sets remains one of the central issues in systematics.” M.S.Y. Lee, 2001, Mol. Biol. Evol. 18: 676.

It is not surprising that resolution of this issue has not been achieved since 1989. Many of the disagreements regarding systematics inference are due to the lack of clear philosophical understanding of what is required to justify the methods claimed to be appropriate for such inferences. The different opinions regarding the RTE is a good example. The 'debate' about the RTE is actually not a debate at all, but rather an ongoing and very basic misunderstanding of the requirement, which has often led to systematists engaging in philosophically, as well as scientifically unacceptable practices. For instance, the quote by Lee shown here is quite incorrect. 'Incongruence' is not an issue at all when one clearly understands the RTE.

GOALS 1. Relation of total evidence to inference. 2. Relation of total evidence to systematics. 3. Implications for systematics.

This presentation will address three issues regarding the RTE: (1) The first will be to show the relation of the RTE to inference in general. In other words, why are we supposed to apply the requirement and when are we supposed to apply it. (2) The second goal will be to show the actual relation between the RTE and the inference of phylogenetic hypotheses. (3) The third goal will be to show some of the implications for systematics once we correctly apply the requirement.

The basis of the requirement of total evidence – a simple example:

• 95% of Texans are millionaires • Kirk is from Texas • Kirk is a millionaire (with high probability)

Let's use a simple example to illustrate the importance of the RTE. We have the generalization that most people from Texas are millionaires, and we have evidence that I am from Texas. From these premises we could infer that I probably am a millionaire. We might even say that the premises support our conclusion to the extent that we are 95% certain that I am a millionaire. Note that the inference is not deductive, as indicated by the double line separating the premises from the conclusion.

The basis of the requirement of total evidence – a simple example:

• 95% of Texans are millionaires

(true)

• Kirk is from Texas

(true)

• Kirk probably is a millionaire (with high probability)

(true or false)

It is then important to recognize one of the fundamental rules that must be applied to any inference in order for it to be considered acceptable -- one must assume the premises are true. But, since this is a non-deductive inference, even if the premises are true, the conclusion is not necessarily true. While we might have very strong support for our belief that I am a millionaire, there remains the possibility that I am not.

The basis of the requirement of total evidence – a simple example:

• 95% of Texans are millionaires

(true)

>99% of museum curators are not millionaires

• Kirk is from Texas

(true)

Kirk is a museum curator

• Kirk probably is a millionaire

(true or false) Kirk probably is not a millionaire

Let's now consider a second, separate inference. We have one premise that states that very few museum curators are millionaires, and another premise that I am a curator. It would seem reasonable to conclude that there is a high probability that I am not a millionaire. Once again, while we assume the truth of the premises, they do not necessarily guarantee the truth of the conclusion.

The basis of the requirement of total evidence – a simple example:

• 95% of Texans are millionaires

(true)

• >99% of museum curators are not millionaires

• Kirk is from Texas

(true)

• Kirk is a museum curator

• Kirk probably is a millionaire

(true or false) • Kirk probably is not a millionaire

conclusions are contradictory What is most apparent in this example is that the two inferences lead to contradictory conclusions. Obviously, it would be irrational to ignore that I am a curator or that I am from Texas, since both types of evidence are relevant to the matter of inferring my financial status. The RTE has not been satisfied in these instances. In order to conclude that I am a millionaire, or not a millionaire, requires that we consider all evidence that can affect support for either of these conclusions.

The basis of the requirement of total evidence – a simple example: RTE must be specifically considered in all non-deductive inferences. • 95% of Texans are millionaires

(true)

• >99% of museum curators are not millionaires

• Kirk is from Texas

(true)

• Kirk is a museum curator

• Kirk probably is a millionaire

(true or false) • Kirk probably is not a millionaire

One of the fundamental qualities of all non-deductive inferences is that the RTE must be specifically taken into consideration. This is because non-deductive conclusions are probabilistic in form; a conclusion has the potential to change in the event evidence is included or excluded from the premises.

Requirement of Total Evidence (RTE)

“[T]he credence which it is rational to give to a statement at a given time must be determined by the degree of confirmation, or the logical probability, which the statement possesses on the total evidence available at the time.” Hempel (1965: 64), Aspects of Scientific Explanation

First, it is essential to know what the RTE actually stipulates. A good, basic description was provided by Carl Hempel, quoted here. What Hempel is saying is that one's rational belief in a statement is a function of the evidence providing support for that statement. The rational acceptance of a conclusion must therefore be based on the total evidence appropriate for such support.

Requirement of Total Evidence (RTE) “The general consideration underlying the requirement of total evidence is obviously this: If an investigator wishes to decide what credence to give to an empirical hypothesis or to what extent to rely on it in planning his actions, then rationality demands that he take into account all the relevant evidence available to him; if he were to consider only part of that evidence, he might arrive at a much more favorable, or a much less favorable, appraisal, but it would surely not be rational for him to base his decision on evidence he knew to be selectively biased.” Hempel (1962), Deductive nomological vs. statistical explanation. In: Minnesota Studies in the Philosophy of Science , pp. 98-169.

Requirement of Total Evidence (RTE) “When two sound inductive arguments thus conflict, which conclusion, if any, is it reasonable to accept, and perhaps act on? If the available evidence includes the premises of [two different] arguments, it is irrational to base our expectations concerning the conclusions exclusively on the premises of one or the other of the arguments; the credence given to any contemplated hypothesis should always be determined by the support it receives from the total evidence available at the time... What the requirement of total evidence demands, then, is that the credence given to a hypothesis h in a given knowledge situation should be determined by the inductive support, or confirmation, which h receives from the total evidence e available in that situation.” Hempel (1966), Recent problems of induction. In: Mind and Cosmos, pp. 112-134.

Requirement of Total Evidence (RTE)

Carnap (1951: 211), Logical Foundations of Probability A more technical description of the RTE was provided by Rudolf Carnap, shown here.

Requirement of Total Evidence (RTE)

Carnap (1951: 211), Logical Foundations of Probability This is the characterizaton of the RTE provided by Carnap which appears on the page in the previous slide. An important item to notice is that Carnap (cf. earlier quote from Hempel) refers to the RTE relative to confirmation. Carnap did not specifically recognize abductive inference as a type of reasoning separate from induction, and his objective was the relation of evidence to induction-as-testing. So we might ask: Does the RTE apply to abduction? The key to answering that question lies in the paragraph above...

Requirement of Total Evidence (RTE)

Carnap (1951: 211), Logical Foundations of Probability Carnap notes that this rule is a rule "of the methodology of induction." In other words, a rule regarding the procedure of inductive reasoning. As a rule of induction overall, this would then apply to abduction. Certainly even when one engages in an abductive inference to a particular explanatory hypothesis, it is the premises of that inference that provide some level of justification for believing that hypothesis to be more credible than other hypotheses, even if only as a matter of conjoining particular theories with effects.

RTE must be specifically considered in all non-deductive inferences.

• 95% of Texans are millionaires

(true)

• >99% of museum curators are not millionaires

• Kirk is from Texas

(true)

• Kirk is a museum curator

• Kirk probably is a millionaire

(true or false) • Kirk probably is not a millionaire

Total Relevant Evidence: Evidence is relevant if it has an effect, either positive or negative, on the support for a hypothesis.

While it is commonly stated that all 'available' evidence must be considered in applying the RTE, it is more accurate to say all available *relevant* evidence must be considered. Evidence has relevance if its inclusion in, or exclusion from an inference has a positive or negative effect on support for a conclusion. In the examples just presented, the fact that I am a museum curator and from Texas are pieces of evidence relevant to belief in a given conclusion about my wealth. Other information, however, such as the size of the planet Jupiter or the number of segments of a worm, are clearly irrelevant to the conclusion. It is the matter of evidential relevance that is one of the most important features to take into consideration in biological systematics inference.

The RTE and Evidential Relevance

Carnap (1951: 347), Logical Foundations of Probability

A technical description of evidential relevance was provided by Rudolf Carnap, shown here.

RTE is always satisfied in valid deductions

Animals with hair are mammals (True) Kirk has hair

(True)

Kirk is a mammal

(True)

As noted already, the RTE is only referred to in the context of non-deductive reasoning, i.e. abduction and induction. A basic property of any valid deductive inference is that if the premises are true, then the conclusion is guaranteed to be true. There are no degrees of support as in non-deductive inference, so a conclusion is either completely supported by the premises or not supported at all. With regard to deduction, the RTE is automatically satisfied. The consequence is that the requirement never needs consideration since the conclusion must be necessarily correct.

RTE is always satisfied in valid deductions

Animals with hair are mammals (True) Kirk has hair

(True)

Kirk hates raw tomatoes

(True!)

Kirk is a mammal

(True)

(Notice that the added premise, ‘Kirk hates raw tomatoes,’ is irrelevant to the conclusion.)

For instance, we can place additional information in the premises, and the conclusion will not be changed. Additional information is *irrelevant* to the conclusion.

Some Conclusions

Ž

The RTE is one of the basic rules for rational reasoning.

Ž

The RTE applies to all non-deductive inferences.

Ž

The only basis for not considering evidence is to show it to be irrelevant to a given conclusion.

We can now identify several conclusions. The first is that the RTE is one of the basic rules for logical reasoning. By denying relevant evidence to play a part in our inferences, we commit ourselves to accepting conclusions that have a lower logical probability. The second point is that the requirement applies to all non-deductive inferences. The RTE is automatically satisfied in deduction. And third, the only basis for excluding evidence from an inference is if one can show that such evidence is irrelevant to a conclusion -- that the evidence has no effect on support, either positive or negative. Each of these issues has profound implications for biological systematics inference.

GOALS 1. Relation of total evidence to inference.

2. Relation of total evidence to systematics. 3. Implications for systematics.

We can now examine the actual relation of the RTE to biological systematics inference.

The requirement of total evidence and phylogenetic inference:

To characterize the RTE in relation to the inferences of biological systematics hypotheses, we once again have to acknowledge the nature abduction. In this example, we have observed that indivduals to which species hypotheses c-us and d-us refer have character 1, whereas we have observed members of other species with 0.

The requirement of total evidence and phylogenetic inference:

“Why do individuals to which species hypotheses c-us and d-us refer have character 1 in contrast to 0 as observed among other individuals?” Observations (effects):

members of species c-us and d-us have character 1

We would then ask the why-question shown here.

a

b

c

d

0

0

1

1

The requirement of total evidence and phylogenetic inference: Phylogenetic Theory: If character x(0) exists among individuals of a reproductively isolated, gonochoristic or cross-fertilizing hermaphroditic population and character x(1) originates by mechanisms a, b, cK n, and becomes fixed within the population by mechanisms d, e, fK n (=ancestral species hypothesis), followed by event(s) g, h, iK n, wherein the population is divided into two or more reproductively isolated populations, then individuals to which descendant species hypotheses refer would exhibit x(1). Observations (effects):

members of species c-us and d-us have character 1

a

b

c

d

0

0

1

1

To answer this why-question by way of a causal explanation, we would invoke some aspect of evolutionary theory. Because the subject of our question is shared similarities distributed among members of two or more species, the general concept of 'descent with modification' provides the necessary causal component to answer such questions. A formal representation of the theory is shown here.

The requirement of total evidence and phylogenetic inference: Phylogenetic Theory: If character x(0) exists among individuals of a reproductively isolated, gonochoristic or cross-fertilizing hermaphroditic population and character x(1) originates by mechanisms a, b, cK n, and becomes fixed within the population by mechanisms d, e, fK n (=ancestral species hypothesis), followed by event(s) g, h, iK n, wherein the population is divided into two or more reproductively isolated populations, then individuals to which descendant species hypotheses refer would exhibit x(1). Observations (effects):

members of species c-us and d-us have character 1

Causal Conditions (phylogenetic hypothesis): Character 1 originated by some unspecified mechanism(s) within a reproductively isolated population with character 0, and character 1 became fixed in the population by some unspecified mechanism(s) (= ancestral species hypothesis), followed by an unspecified event(s) that resulted in two reproductively isolated populations.

a

b

c

d

0

0

1

1

0

0

1

1

From these premises, we infer vague accounts of past causal conditions that at least initially enable us to answer the why-question. In this instance, we might conclude that there existed a reproductively isolated ancestral population with character 0, and in this population property 1 arose and became fixed, and subsequently this population split, resulting in populations to which species hypotheses c-us and d-us refer, of which we observe members in the present.

A simple example of evidential relevance and the RTE.

data set α

Consider the following example. Let's say we have made observations from among these individuals, represented in data matrix α.

data set α

From this data matrix, which also implies our why-questions, this is the cladogram inferred.

S1

S2

S3

data set α

Recall, however, that cladograms imply at least two classes of explanatory hypotheses: (1) the origins/fixation of novel characters among individuals in reproductively isolated ancestral populations (arrows), and (2) population splitting events (S).

S1

S2

S3

data set α

data set β

In addition to the observations summarized in data matrix α, let's say that another set of observations are available, in data matrix β.

S1

S5

S2

S4

S6

S3

data set α

data set β

A separate inference is made using this second data matrix, giving the second cladogram shown on the right. In contrast to the first cladogram, this second cladogram implies an entirely different set of causal events.

S5

S1

S2

S4

S6

• each inference is non-deductive S3

• conclusions are contradictory • total evidence requirement is violated

data set α

data set β • beware the reification of cladograms!

Each of these abductive inferences leads to sets of explanatory hypotheses accounting for observed features, and therefore each provides answers to two different sets of why-questions. The respective conclusions are, however, contradictory. They hypothesize the past existence of entirely different sets of causal conditions and events. In this example, the RTE has been violated. Clearly, the explanation of one set of features has relevance to the explanation of the other, especially when we consider that the explanatory context of each hypothesis is past individual organisms exhibiting properties in need of explanation. As a result, the plausibility of either hypothesis is compromised. Another way to characterize this problem is that in order to justify maintaining these separate data sets, one must postulate separate evolutionary histories to have occurred in entirely distinct common ancestral individuals, but, somehow these evolutionary pathways have miraculously converged to the fully integrated, individual organisms we now observe. Obviously, this is not a realistic proposition. An alternative view might be to consider ancestral entities other than whole organisms, but this requires completely altering the nature of our causal questions from ones which refer to the properties of individual organisms to the properties of individual parts of organisms. This is a potentially acceptable alternative, as in the case of 'gene trees,' but requires very careful consideration of the types of why-questions we are asking, what theories are to be used, and ensuring that evidential relevance does not apply to different sets of observations. It is when we clearly recognize that the goal of phylogenetic inference is to causally account for observed regularities among individual organisms in different species that we usually find the analysis of separate data sets is incorrect.

The Importance of Considering Evidential Relevance – An Example

Why-Questions

Observations a b c d e 1: 2: 3: 4: 5:

0 0 0 0 0

0 0 0 1 1

0 0 1 1 0

1 1 1 1 0

1 1 1 1 1

‘Why do members of species d-us & e-us have property 1(1) in contrast to 1(0)?’ ‘Why do members of species d-us & e-us have property 2(1) in contrast to 2(0)?’ ‘Why do members of species c-us, d-us, & e-us have property 3(1) in contrast to 3(0)?’ ‘Why do members of species b-us, c-us, d-us, & e-us have property 4(1) in contrast to 4(0)?’ ‘Why do members of species b-us & e-us have property 5(1) in contrast to 5(0)?’

We can use a simple example to summarize the magnitude of the problem of ignoring the RTE. Given the data matrix on the left, it not only summarizes our observation statements but also the why-questions shown here.

The Importance of Considering Evidential Relevance – An Example SEPARATE Inferences for Each Why-Question (assumes observations are not relevant to each other) (1) ‘Descent with modification’ theory

(4) ‘Descent with modification’ theory

abcde 1: 0 0 0 1 1

abcde 4: 0 1 1 1 1

H: abc(de)

H: a(bcde)

(2) ‘Descent with modification’ theory

(5) “Descent with modification” theory

abcde 2: 0 0 0 1 1

abcde 5: 0 1 0 0 1

H: abc(de)

H: acd(be)

(3) ‘Descent with modification’ theory abcde 4: 0 0 1 1 1 H: ab(cde)

One could argue that each why-question should be answered separately, giving five respective answers. These separate inferences would assume that none of these observations are relevant to each other with respect to providing causal understanding. But for this to be the case, one would have to provide an empirical basis for claiming such irrelevance.

The Importance of Considering Evidential Relevance – An Example SINGLE Inference for Why-Questions “Descent with modification” theory 1: 2: 3: 4: 5:

abcde 00011 00011 00111 01111 01001

Since the observations are relevant to each other, they must be explained together.

H: a(b(c(d e))) (d e):

1(1) and 2(1) became fixed in ancestral population, with subsequent splitting event resulting in two reprpductively isolated populations (d-us & e-us);

(c(d e)):

3(1) became fixed in ancestral population, with subsequent splitting event resulting in reprpductively isolated populations (c-us, d-us, e-us);

(b(c(d e))):

4(1) became fixed in ancestral population, with subsequent splitting event resulting in reprpductively isolated populations (b-us, c-us, d-us, e-us);

(b) & (e):

Ad hoc hypotheses – 5(1) became separately fixed among individuals to which hypotheses b-us and e-us refer.

Clearly, the observations are relevant to one another for the fact that the whyquestions are answered by way of the same theory.

More Conclusions

Ž

Phylogenetic inference is non-deductive, i.e. abductive.

Ž

The explanation of shared similarities is by way of inferring the past existence of individuals - thus, the relevance of causally accounting for a set of shared characters must be considered in conjunction with other shared characters.

Ž

All systematics inferences are subject to the requirement of total evidence – there are no exceptions.

There are several additional conclusions. The first is that inferences from shared similarities to phylogenetic hypotheses are not deductive. Such inferences are instead abductive. The goal of biological systematics is to causally account for the differentially shared features we observe within and among organisms. Second, given the abductive nature of systematics, it becomes important to recognize that causal explanations of any set of shared similarities must be in the context of past individuals. Thus, explaining one set of shared similarities can have relevance to the explanations of other shared similarities, since all of the properties observed are accounted for by past events involving individual organisms. The end result is that the RTE cannot be avoided in systematics inference, as has often been the case.

GOALS 1. Relation of total evidence to inference. 2. Relation of total evidence to systematics.

3. Implications for systematics.

We can now examine some of the implications of the RTE for systematics.

IMPLICATIONS: RTE Justification Philosophical Justification for Total Evidence: Carnap (1950), Hempel (1965), et al.:

- rationality - logical reasoning

Cladistic Justification for Total Evidence: Nixon & Carpenter (1996): ‘parsimony’ – all available evidence should be used because the most parsimonious cladogram has maximum explanatory power.

Rather than correctly focusing on evidential relevance and rationality as the basis for the RTE, there has been the suggestion among some systematists that the criterion of parimony provides justification. Unfortunately, this association of the RTE with parsimony is incorrect, and is common in the secondary literature.

Rational Reasoning Inference Rules • assume truth of premises • satisfy requirement of total evidence Inference (abductive) ‘Parsimony’ Explanatory Power Examining the relation of the RTE to parsimony in abductive inference makes it clear that parsimony has no relation to the RTE. Rational reasoning is determined by inference rules that dictate the content of the premises we use in an inference. The use of 'parsimony,' or what is more properly regarded the application of theory to effects, is an action that occurs *within* a given inference. It is this application of theory to effects in an abductive inference that ensures our hypotheses have maximum explanatory power. And of course, by explaining observed effects by way of a common cause theory then ensures that the parsimony criterion is fulfilled. We want to maintain the integrity of our perceptions as they exist in the premises as much as possible in the hypothesized explanation. Thus, the RTE is a criterion applied *prior to* the act of making an inference, whereas invoking what is said to be parsimony describes the relations between premises and conclusions *in* an inference. Parsimony, as it has often been construed in phylogenetic inference, has nothing to do with the RTE.

“The justification for total evidence is essentially the argument for parsimony as an integral part of method in science.” Schuh (2000: 152)

Unfortunately, this incorrect association of the RTE with parsimony has become common in the secondary literature, pointing again to the lack of a cogent philosophical foundation for systematics inference.

“The theoretical rationale for total evidence is the same argument as that for parsimony as an integral part of method in science....” Schuh & Brower (2009: 158)

Unfortunately, this incorrect association of the RTE with parsimony has become common in the secondary literature, pointing again to the lack of a cogent philosophical foundation for systematics inference.

IMPLICATIONS • ‘Simultaneous analysis’ • ‘Partitioned analysis’ (including ‘supertrees’)

• Cladogram comparisons

We can now examine some implications of the RTE for some of the standard approaches taken in phylogenetic systematics.

IMPLICATIONS ‘Simultaneous Analysis’ – Kluge (1989)

Observations

Cladogram

α: o1, o2, ... on + β: oa, ob, ... on

A B C D

Simultaneous analysis, frequently though incorrectly called 'total evidence analysis,' has been a common approach, which advocates claiming it applies the RTE. Given a set of all available observations, these are to be considered together in the inference of phylogenetic hypotheses.

IMPLICATIONS ‘Simultaneous Analysis’ – Kluge (1989) Observations

Cladogram

α: o1, o2, ... on + β: oa, ob, ... on

A B C D

• Not justified by ‘parsimony’ or ‘explanatory power.’

• Justified by the RTE (but evidential relevance is not usually considered).

While simultaneous analysis has the potential to satisfy the RTE, the requirement is not justified by the parsimony criterion or the need to maximize explanatory power. While the RTE can justify what is called simultaneous analysis, proponents have not recognized that evidential relevance is the key component. In fact, it is useless to refer to such inferences as 'simultaneous' or 'total evidence' analyses. These are nothing more than unnecessary phrases for what can only be regarded as proper, rational reasoning.

IMPLICATIONS ‘Partitioned Analysis’ – Miyamoto & Fitch (1995)

Observations

Cladograms

Consensus cladogram

A B C D

α: o1, o2, ... on A B C D A B C D

β: oa, ob, ... on

A common claim is that observations can be partitioned into discreet data sets, from which separate phylogenetic hypotheses are inferred, and relationships are then presented in the form of consensus cladograms.

IMPLICATIONS ‘Partitioned Analysis’ – Miyamoto & Fitch (1995) Observations

Cladograms

Consensus cladogram

A B C D

α: o1, o2, ... on A B C D A B C D

β: oa, ob, ... on

•

Violates the RTE.

•

Denies the explanatory goal of systematics.

Partioned analysis is, however, in violation of the RTE. Not surprisingly, advocates of partitioned analysis have never correctly described the RTE such that justification for partitioning data could be regarded as acceptable. Just as important is the recognition that consensus techniques have no part to play in the explanations of characters since consensus trees lack the necessary empirical basis to be regarded as causal constructs. Consensus trees are merely graphical representations of topological similarities among tree shapes. Since the phylogenetic hypotheses are independent of one another, their comparison is uninformative. Consensus techniques are too often used under the mistaken assumption that cladograms are objects, and, as objects, can be compared. This is not only incorrect, but also denies the explanatory role of cladograms.

The Myth of Applying Total Evidence... Sometimes “Combining all the data into a single analysis assumes that all the data reflect the same evolutionary history.” “Combining data would be a good strategy if the phylogenetic method used was consistent, such that by adding more data the method would always converge on the correct [sic] tree.” Page & Holmes (1998: 282)

Misconceptions regarding the RTE and the defense of partitioned analyses can be found in the secondary literature. For instance, Page & Holmes (1998) conflat the explanation of effects per the RTE with the peculiar point of view that one must first assume completely identical causal events. Obviously, there is no foundation for such a position from either what the RTE stipulates or from the abductive nature of biological systematics inference. The idea that statistical consistency can be a defense for partitioned analysis fails because, as we have already discussed, consistency is irrelevant to abduction.

Observations

Hypotheses A B C D

α: o1, o2, ... on

3 2 1

A C E F

β: oa, ob, ... on

5 4

6

Supertree ‘phylogeny’

Supertree matrix tree components 123456 A 100110

B 1 t C 1 a x D 1 a E ?

1 1 1 ? F ??

0 1 1 ? ?

? 1 ? 1 1

? 1 ? 0 0

E F A B C D

? 0 ? 1 1

A recent approach, called 'supertrees,' has gained some popularity, especially because advocates have claimed that the method is a legitimate substitute for considering the RTE. The basis for inferring supertrees is that a large amount of character data are being produced in different studies, but often from groups of taxa which do not completely overlap, making it difficult to combine these data into a more comprehensive phylogenetic analysis. It is claimed that from separate data sets, which have overlapping taxa, each of the separate hypotheses can be characterized by their individual components, and that these components, for each associated taxon, can then be coded into a new data matrix. From this matrix is then inferred a 'supertree phylogeny' which includes all taxa from among the originally separate phylogenetic analyses.

Hypotheses

Observations

A B C D

α: o1, o2, ... on

3 2 1

A C E F

β: oa, ob, ... on

5

6

4

Supertree matrix tree components 123456 A 100110

B 1 t a C 1 x D 1 a E ?

Supertree ‘phylogeny’ E F A B C D

10??? 11110

11??? ??101 F ???101

•

Violates the RTE.

•

Denies the explanatory goal of systematics.

•

Supertrees lack empirical interpretation.

The supertree approach is, however, defective because it violates the RTE, thereby denying the ability to even interpret a 'supertree phylogeny' as any kind of explanation for our observations. This also leaves supertrees as nothing more than graphic diagrams lacking any empirical interpretation. The advocates of supertrees are mistaken in their view that this procedure can be a substitute for applying the RTE.

IMPLICATIONS The Error of Cladogram Comparison – I A B C D

Data set 1: o1, o2, ... on

= Explanation, E1, 2...n

A C B D

Data set 2: ox, oy, ... on

= Explanation, Ex, y...n

Let's now look at two types of common errors that are encountered when systematists attempt to compare cladograms. In this example, we have two sets of observations, data sets 1 and 2, from which we infer respective phylogenetic hypotheses. Each of these cladograms represent distinct sets of explanatory hypotheses intended to explain respective sets of data. Recall that character data comprise part of the premises used to infer a cladogram, which is intended to account for those data. Proponents claim that these respective cladograms can be compared relative to the same group of organisms.

IMPLICATIONS The Error of Cladogram Comparison – I A B C D

Data set 1: o1, o2, ... on

= Explanation, E1, 2...n Cladograms inferred from separate data sets cannot be meaningfully compared. Each cladogram represents a separate explanatory hypothesis - each of which is irrelevant to the other.

A C B D

Data set 2: ox, oy, ... on

= Explanation, Ex, y...n

Such comparison is, in fact, meaningless since the two explanatory hypotheses have no relevance to one another, causal or otherwise. Asserting such relevance is impossible since the only relevance a phylogenetic hypothesis has is to the shared properties it is intended to explain among a group of individuals.

IMPLICATIONS The Error of Cladogram Comparison – II A B C D ‘theory’ α

= Explanation, E1, 2...n

Data: o1, o2, ... on A C B D ‘theory’ β

= Explanation, Ex, y...n

There is the additional approach to comparison, shown here. It is common practice to explore the similarities and differences of cladograms inferred from different causal theories. For instance 'parsimony' versus 'maximum likelihood' versus 'Bayesianism.'

IMPLICATIONS The Error of Cladogram Comparison – II A B C D ‘theory’ α

= Explanation, E1, 2...n

Data: o1, o2, ... on A C B D ‘theory’ β

= Explanation, Ex, y...n

Cladograms inferred from different causal theories cannot be meaningfully compared. What requires comparison are the theories themselves.

Unfortunately, comparing these cladograms is uninformative for the same reasons pointed out earlier. If there is to be any reasonable and productive comparison, it would be at the level of the theories and causal hypotheses being considered for providing explanations. Comparisons of cladograms inferred from different theories is only a process of comparing 'groups' of taxa, whereas the salient matters of interest are the actual causal accounts implied by cladograms. As a consequence, comparisons of cladograms inferred from different theories is without merit.

IMPLICATIONS The Error of Character Mapping

A B C D

Data: o1, o2, ... on

= Explanation, E1, 2...n

In addition to the error of comparing cladogram topologies either inferred from separate data sets or different theories, there is the other common mistake of mapping characters onto a cladogram that was inferred from another set of data. As we have seen throughout much of this course, this is part of the fallacy of reification with regard to cladograms. We start with the inferences of phylogenetic hypotheses for some set of data, presented as a cladogram.

IMPLICATIONS The Error of Character Mapping

A B C D

= Explanation, E1, 2...n

Data: o1, o2, ... on

Data: ox, oy, ... on

map ox, oy, ... on onto cladogram

“Explanation, Ex, y...n”

Mapping characters onto a cladogram inferred from other data is meaningless. No explanations are provided for the mapped characters since those characters were not used to infer the cladogram.

Subsequent to the inference of that cladogram, additional observations are considered for explanation. But rather than combining those new data with the earlier observations, per the RTE, one simply maps the new data onto the preexisting cladogram. These new data are then explained in the context of the cladogram. The problem, however, is that the cladogram only has explanatory relevance to the character data used to abductively infer the cladogram. Any act of mapping other characters onto the cladogram will be meaningless from the perspective of the cladogram offering any sort of causal understanding of those mapped characters.

IMPLICATIONS The Error of Character Mapping – an example “How can we make statements about the evolution of organ systems?... If we have a tree,... we can map the characters from organ systems onto this tree and then try to ‘read’ its history. I regard this as an enormously important thing.” A. Schmidt-Rhaesa (2007: 1), The Evolution of Organ Systems

Prominent examples of character mapping can be found in the recent book, 'The Evolution of Organ Systems.' Indeed, Schmidt-Rhaesa suggests the mapping characters onto previously inferred phylogenetic hypotheses is entirely acceptable.

IMPLICATIONS The Error of Character Mapping – an example

“Summary of the distribution of striation patterns, HMS pattern, and some autapomorphies concerning musculature.” A. Schmidt-Rhaesa (2007: fig. 5.16), The Evolution of Organ Systems

Throughout the book are cladograms onto which various organ systems are mapped.

IMPLICATIONS The Error of Character Mapping – an example

K. Fitzhugh (2009), BioScience 59: 85-86

In my review of Schmidt-Rhaesa's book, I point out the significant problems associated with character mapping.

A recent systematics publication further illustrates the problem. The example shown here is an excellent descriptive paper on a new member of the polychaete clade ('family'), Syllidae. Members of the new genus and species, Ramisyllis multicaudata, are unusual because of the highly branched bodies; a condition otherwise known only among members of another syllid species in the genus Syllis, S. ramosa.

“The phylogenetic position of R. multicaudata gen. et sp. nov. indicates that branching has evolved independently in Ramisyllis gen. nov. and Syllis. This is supported by differences in the branching process between the two taxa: in S. ramosa branching is initiated by segment addition at the parapodium, whereas in R. multicaudata gen. et sp. nov. segments are added from a region between parapodia.” – Glasby et al. 2012: 481

The authors inferred phylogenetic relationships among representative members of syllid species and genera using DNA sequences. Based on these relationships, they note that members of Ramisyllis form a clade separate from Syllis. And since the only other syllids with branching bodies are members of Syllis, the authors suggest that the cladogram shown here indicates that branching has been independently derived in Ramisyllis multicaudata and Syllis ramosa (but note that S. ramosa was not included in their cladogram). The authors also note that the processes of branching are different between the two groups. Since the cladogram only provides a series of explanatory hypotheses accounting for sequence data, it is not possible to rely upon that topology to make evolutionary claims about other characters, such as body branching or chaetal forms. These are features that can only be explained in conjunction with, not separate from, the DNA sequence data. Mapping these morphological characters onto this cladogram has no empirical meaning.

Final Conclusions Ž

The RTE prevents inferences of cladograms from separate data sets, unless the criterion of irrelevance is satisfied. Thus, ... • to violate the RTE is to reason irrationally, • comparing cladograms inferred from different data sets is empirically meaningless, and • any methods that violate the RTE, such as partitioned analysis and supertrees, are scientifically unacceptable.

In conclusion, the principle consequence of correctly applying the RTE is that the common practice of inferrnig phylogenetic hypotheses from separate data sets is unacceptable unless one can clearly establish that explaining sets of observations are in fact irrelevant to one another. For this to be accomplished, one must assume that common ancestry in the form of past individual organisms comprising reproductively isolated populations, and population splitting events apply to one set of observations but not to other observations. More generally, we may also conclude that to violate the RTE is to reason irrationally and as a result, unscientifically. The comparison of hypotheses inferred from different data sets is empistemically meaningless. And lastly, the RTE clearly establishes as unacceptable methods such as partitioned analysis, including supertree methods, or comparisons of cladograms inferred from different theories.

The Philosophy of Biological Systematics Course Outline – Part 3

1.

The requirement of total evidence.

2.

Homology & homogeny & homoplasy.

3.

Character coding.

4.

The mechanics of hypothesis testing in biological systematics.

Homology vs. Homogeny

Richard Owen vs. E. Ray Lankester

J. Kirk Fitzhugh Natural History Museum of Los Angeles County

In this part of the course, we will examine introduction of the concepts of 'homogeny' and 'homoplasy' by E. Ray Lankester in 1870, and his suggestion that these terms replace 'homology.' The basis for this examination is derived from what we have already seen with regard to the inferential basis of biological systematics and the correct application of the requirement of total evidence. We will investigate the inferential bases of homology, homogeny, and homoplasy, and their relations within systematics.

Homology is indeed morphology’s central conception.... Hall (1994: 2)

homologue

homologous

While a great amount of attention has been given to the subject of homology, the definition of the term has had a convoluted past, made more so by the fact that a host of other terms and concepts have also been introduced and applied, sometimes without careful consideration of meanings or theoretical foundations. The varied definitions of homology, homologue, and homologous are good examples of these problems. As a result, there is no consensus as to what one means when they use the term homology, and the matter has been raised to an even higher level of confusion in recent years by some practitioners of cladistics. For instance, the variety of opinions to be found in the volume shown here makes this problem all too clear.

Homology: “Attributes of two organisms are homologous when they are derived from an equivalent characteristic of the common ancestor.” Mayr (1982: 465), The Growth of Biological Thought

definition of homology? or, description of the application of the concept?

Consider this definition of homology given by Mayr (1982), which is commonly seen in biology. Is this statement a definition of the term homology, or simply a rather imprecise description of the application of the concept we refer to as homology? A proper definition should give the meaning of a word in terms of a specific concept or subject which the word is intended to represent.

“In general, homology means inferred common ancestry, although it is commonly misused to mean similarity.” Moritz & Hillis (1996: 7), Molecular Systematics

In the purported definition of homology as 'Shared similarity due to common ancestry,' there is no specified subject for which meaning of the word homology can be applied. Homology does not merely refer to shared similarities or to just common ancestry. The actual subject which gives homology it's meaning, and thus allows for a proper definition, will be presented later in this talk, and will form the basis for some suggested modifications in our terminology.

Homologues? Or, homology?

Take for instance this block of 1999 postage stamps from South Korea. Is it the concept of homology or homologue that is being depicted among these vertebrate forelimbs? One could easily argue that only structural correspondence is being identified between past and present individuals. Or, one might take a more causal approach and interpret what is shown in the context of evolutionary theory. The ambiguity associated with the use of terms such as homologue, homologous, and homology, and whether or not these terms refer to different concepts, continues to be unresolved. Yet, if homology is such a fundamental concept to all of biology, we should be concerned that one of the central tenets of this field of science remains poorly articulated.

The Common Terms Used Today: homologue homologous homology homoplasy

In addition to the terms homology, homologue, and homologous, matters become more complicated when we take into consideration the concept of homoplasy, which has especially been in common use in systematics for the past 40 years. Loosely defined, and as will be discussed later, homoplasy is defined as an explanation of shared similarities by processes other than common descent or common ancestry. Unfortunately, the term has acquired some peculiar meanings in recent years, which makes its conceptual foundation almost as tenuous as that of homology.

What about Lankester (1870)? homologue homologous homology homoplasy homogeny

Edwin Ray Lankester 1847-1929

The term homoplasy was originally coined in a paper by the British evolutionary biologist E. Ray Lankester in 1870. A matter not often considered is that in the same publication, Lankester also introduced the term homogeny. In fact, the basis of Lankester's paper was to replace the term homology, and the associated terms homologue and homologous, with the two new terms homogeny and homoplasy. Lankester's ideas gained only partial acceptance, as is obvious by the common, though irregular reference to homoplasy in 20th century evolutionary biology. The term homogeny, however, has received virtually no consideration. What will be pointed out in this part of the course is that Lankester's replacement of the term of homology with the dual terms homogeny and homoplasy were in fact correct, contrary to past commentaries against the use of homogeny in lieu of the continued use of the homology in evolutionary biology.

Richard Owen 1804-1892

homologue ---homology

In order to properly evaluate and defend Lankester's position, we must clear away some misconceptions that have surrounded the concepts of homologue and homology for over a century. To do this, it is necessary to examine what was first said regarding these concepts. In this case, we need to be aware of the meanings of these terms as they were most notably presented by the 19th century British anatomist, Richard Owen.

Owen, 1846

The first formal treatments of the concepts of homologue and homology were provided by Owen in his 1846 paper, entitled, "Report on the Archetype and Homologies of the Vertebrate Skeleton." While often cited and discussed in the evolutonary literature, the interpretations of Owen's statements regarding homologue and homology are varied. But, as we will see, a careful examination of his writings do not support many of these views.

Homologue : ‘same word’ The corresponding parts in different animals being thus made namesakes, are called technically ‘homologues’. Owen (1846: 173)

‘radius’

“Homologue: the same organ in different animals under every variety of form and function.” Owen (1843), Lectures on... Invertebrate Animals

Owen's (1846) definition of homologue is essentially the same as what he gave in his 1843 book on invertebrates. The 1843 definition only appears, however, in the glossary of the book, without additional comment or analysis. It is interesting to note that it has been this definition of homologue which evolutionary biologists have often referred to as the definition of homology. As we will see, Owen never intended such an equivalence, and, this has been one of the misapplications of terms and definitions that has contributed to the longstanding problems associated with both words in evolutionary biology.

“Relations of homology are of three kinds:” Owen (1846: 175)

Special Homology General Homology Serial Homology

Homology: ‘the study of similarity’

In speaking of homology, Owen (1946) was very clear that the term refers to three conditions. Two of these conditions, special and general homology, are important to examine in order to understand how Owen was using the term homology. In contrast to the Latin derivation of homologue, the term homology means 'the study of similarity.' Owen clearly understood the derivations of the terms he was using, and, as will be evident in reviewing his ideas on homology, he intended a very different meaning for that term in comparison to homologue. This is in contrast to the common practice of equating homologue with homology in the evolutionary literature. But, as will be shown next, Owen never provided particularly clear definitions of special and general homology, which has allowed for a variety of interpretations of his concepts among evolutionary biologists in their attempts to adopt homology for their own use. This absence of clear definitions on Owen's part would appear to be the principle cause of the confusion regarding the definition of homology in evolutionary biology.

Special Homology ...the correspondency of a part or organ, determined by its relative position and connections, with a part or organ in a different animal; the determination of which homology indicates that such animals are constructed on a common type....

‘radius’

Owen (1846: 175)

‘Special homology’ refers to the causal relations of homologues, or homologous parts, to an unspecified “common type” or archetypal form.

‘special homology’

Shown here is Owen's (1846) characterization of 'special homology.' What Owen is saying is that while homologues indicate shared similarities, special homology denotes the causal relations of homologues to a "common type" or archetype. In Owen's conception, an archetype was a reflection of the general plan of the creator, from which all species had been separately formed.

Special Homology ...when, for example, the correspondence of the basilar process of the human occipital bone with the distinct bone called ‘basioccipital’ in a fish or crocodile is shown, the special homology of that process is determined. Owen (1846: 175)

Owen provides the following example of special homology when he says, "...when the correspondence of the basilar process of the human occipital bone with the distinct bone called basioccipital in a fish or crocodile is shown, the special homology of that process is determined." The occiptial bone is seen here at the base of this human skull.

Cod This example of special homology can be seen in a series of illustrations presented by Owen of the homologous bones among vertebrate skulls, where the basioccipital or basilar process is shown here in a fish...

Crocodile ... in a crocodile

Ostrich .... a bird

Hog ...a hog

Human ...and a human. It is after recognizing this correspondence of parts, as homologues, that special homology is inferred to provide initial causal understanding by regarding shared similarities as derived from the archetypal plan of the Creator. Note, however, that in making reference to special homology, there is no direct comparison with the archetype; simply reference to the archetype as accounting for the regularities one has observed.

O. Haas & G.G. Simpson. 1946. Analysis of some phylogentic terms, with attempts at redefintion. Proc. Amer. Phil. Soc. 90: 319-349. Special homology: “It will be noticed... that ‘constructed on a common type’ may imply a slight foreboding of common origin.” (pp. 320-321)

Not surprisingly, Owen's concept of special homology has been readily adopted by evolutionists by replacing the archetype theory with that of common ancestry, as indicated in this statement by Otto Haas and G.G. Simpson (1946), which deals extensively with homology and homoplasy.

General Homology “A higher relation of homology is that in which a part or series of parts stands to the fundamental or general type, and its enunciation involves and implies a knowledge of the type on which a natural group of animals...is constructed.” Owen (1846: 175)

‘General homology’ denotes the causal relations of homologues, or homologous parts, to the archetype.

Next we need to consider Owen's 'general homology.' What Owen is saying in the quote shown here is that general homology denotes the causal relations of homologues to the actual archetype. Since, for example, the archetype for all vertebrates must be a generalized concept, the level of structural generality referred to between homologues and archetype is far greater than that considered in special homology. Hence, Owen's distinction between special and general forms of the concept.

General Homology

Thus when the basilar process of the human occipital bone is determined to be the ‘centrum’ or ‘body of the last cranial vertebra,’ its general homology is enunciated. Owen (1846: 175)

Using once again the example of the basilar process of the human occipital bone, Owen illustrates the concept of general homology when he points out that this structure is actually a very highly modified vertebral centrum.

General Homology “The relation between a part or series of parts to the fundamental or general type.”

centrum

‘Ideal Typical Vertebra’

In other words, the basilar process is considered equivalent to the centrum of a vertebra in the archetype. The archetype of all vertebrates as concieved by Owen contains a skeleton composed of a series of vertebrae. Owen was following the point of view introduced in the late eighteenth century that the vertebrate skull is a series of very highly modified vertebrae. He regarded the skull as being composed of four cranial vertebrae, colored here in the archetype, with the basilar process in the human skull being equivalent to the centrum of the last of these vertebrae. General homology provides more complete causal understanding of homologues by relating the features of observed organisms directly to the general archetypal form from which all variations would have been derived by the Creator for each species.

“If the special homology of each part... [is] recognisable..., shall we close the mind’s eye to the evidences of that higher law of archetypal conformity on which... the lower and more special correspondences depend?” Owen (1846: 301)

Since Owen intended the concept of homology to be causal, it is not surprising that there would be a consistent relation between special and general homology insofar as both rely on the same theory of archetypal form. The difference between them is only the degree to which an archetype is invoked as a matter of providing causal understanding. I have included this quote from Owen's (1846) paper to point out this relation, which has been claimed not to exist by some evolutionary biologists.

O. Haas & G.G. Simpson. 1946. Analysis of some phylogentic terms, with attempts at redefintion. Proc. Amer. Phil. Soc. 90: 319-349.

General homology: “...phenomena which do not presuppose two or more organisms to be compared with each other, therefore [general homology lies] outside the original scope of homology, as conceived by Owen in 1843” (p. 321)

“...the term homology is... equivalent to [Owen’s] special homology of 1847.” (p. 321)

In their discussion of homology, Haas & Simpson (1946) interpreted general homology to be "phenomena which do not presuppose two or more organisms to be compared with each other, therefore general homology lies outside the original scope of homology, as conceived by Owen in 1843." Clearly, Haas & Simpson have not correctly interpreted Owen's concept of general homology. Owen was clear that the same causal theory applies in both general and special homology, and as a result, is applied to homologues. Therefore, comparisons between two or more organisms would certainly apply in general homology. The only difference between general and special homology being that the actual archetype is recognized in the former and only alluded to in the latter. Note as well that, contrary to what Haas & Simpson claim, Owen never mentions homology in his 1843 book on invertebrates. His only mention is of homologue. As a result of their misinterpretation, Haas & Simpson claim that the modern evolutionary concept of homology is only equivalent to Owen's special homology. There is nothing in what Owen says that would lead to this conclusion.

Definitions sensu Owen homologue:

a named feature observed among two or more individuals.

homologous:

features which are homologues.

homology:

a causal hypothesis accounting for shared similarities, i.e. homologues, among members of two or more species by some common cause.

We can now recognize formal definitions of homologue and homologous that are consistent with the useage intended by Owen and still applicable in evolutionary biology. This also provides part of the proper context for evaluating Lankester's (1870) work. For homologue, the definition provided by Owen (1843, 1846) still stands as named features observed among two or more individuals. The term homologous is defined as features that are homologues. While this definition is consistent with the same Latin derivation as homologue, it does differ from the common view that the term homologous is only applied to features which have been accounted for by way of homology. As I will show later in relation to Lankester's (1870) work, this misapplication is easily corrected. With regard to homology we can now more readily deal with the issue of properly defining this term.

homology:

an explanatory hypothesis accounting for shared similarities, i.e., homologues, among members of two or more species by some common cause .

(Darwinian):

an explanatory hypothesis accounting for shared similarities, i.e., homologues, among members of two or more species by descent with modification.

(Owenian):

an explanatory hypothesis accounting for shared similarities, i.e., homologues, among members of two or more species by reference to a ‘common type.’

But, it might be suggested that we restrict the definition of homology to only an evolutionary context, in which only descent with modification or common ancestry is applied. But, this alternative would deny the older Owenian definition, which has the same form, but makes reference to an archetype theory. The problem of attempting a wholesale replacement of the definition of homology by the adoption of evolutionary theory denies the possibility of reasonably discussing the original, pre-Darwinian definition. A proper definition of homology should allow for the application of both theories in their respective historical contexts. The more general definition of homology suggested in the previous slide subsumes both types of application, therefore is more appropriate than the simple transfer of meaning from the Owenian to the Darwinian form.

homology: sensu Owen

?

an explanatory hypothesis accounting for shared similarities, i.e., homologues, among members of two or more species by reference to a ‘common type.’

– an explanatory hypothesis accounting for shared similarities, i.e., homologues, among members of two or more species by descent with modification. HOMOGENY & HOMOPLASY

The second possibility, which I believe to be the only viable choice, would be to restrict the definition of homology to only its original, pre-Darwinian application of referring to causal hypotheses based on the archetype concept of Owen. We should then apply a different term to hypotheses using descent with modification or common ancestry. This is where the work of E. Ray Lankester (1870) comes into consideration. There are several benefits to this approach. First, we accurately recognize the fact that homology, as it was originally intended, refers to causal hypotheses that are different from those hypotheses based on evolutionary biology. In this way, the original conceptual integrity of the term homology is not diminished and the explanatory distinctions between the two classes of hypotheses are established. What has confounded evolutionary biology has been the use of the pre-Darwinian term in a context for which it was not intended. The result has been a consistently poor and often times misleading definition.

1870. Annals and Magazine of Natural History, Series 4, 6: 34-43.

Lankester's (1870) paper is "On the use of the term homology in modern zoology, and the distinction between homogenetic and homoplastic agreements." It is regrettable that his views have not received more attention, especially in the field of phylogenetic systematics, where Lankester's term homoplasy is so often applied.

Lankester (1870: 42): “...under the term ‘homology,’ belonging to another philosophy, evolutionists have described and do describe two kinds of agreement – the one, now proposed to be called ‘homogeny,’ depending simply on the inheritance of a common part, the other, proposed to be called ‘homoplasy,’ depending on a common action of evoking causes or moulding environment on such homogenous parts, or on parts which for other reasons offer a likeness of material to begin with.” “In distinguishing these two factors of a common result we are only recognizing the principle of a plurality of causes tending to a common end....” Lankester (1870: 41): “Homoplasy includes all cases of close resemblance of form which are not traceable to homogeny....”

Lankester notes that "...under the term homology, belonging to another philosophy..." meaning the Platonic philosophy of which Owen was a follower, "...evolutionists have described and do describe two kinds of agreement -- the one, now proposed to be called homogeny, depending simply on the inheritance of a common part, the other, proposed to be called homoplasy, depending on a common action of evoking causes or moulding environment on such homogenous parts, or on parts which for other reasons offer a likeness of material to begin with." Lankester is claiming that homology cannot effectively accommodate the two classes of causal explanations biologists had begun using with the recent introduction of natural selection theory: common ancestry and the need to invoke separate origins in other instances. He offers the term homogeny for the former type of explanation and homoplasy for the latter.

Lankester (1870: 36): “It will be found... necessary to have two terms in place of the one ‘homologue,’ and to broadly distinguish the nature of the resemblances to which they are applied. Structures which are genetically related, in so far as they have a single representative in a common ancestor, may be called homogenous.” “We may trace an homogeny between them, and speak of one as the homogen of the other.”

Lankester also thought it necessary to replace the term homologue. This was apparently based on the view that homologues are recognized subsequent to the inference of homology. So if, as Lankester maintained, homology must be replaced by terms indicating hypotheses of common ancestry and separate origins, then the term homologue must also be replaced. Thus, in an instance of a hypothesis of homogeny, one would say shared similarities are homogens, and in the case of homoplasy, shared similarities are homoplasts.

Owen / Darwinians, according to Lankester observed shared similarities homology: explanation via archetype or descent with modification

homologues: effects explained

We can summarize Lankester's basis for replacing homology with homogeny and homoplasy. Based on the observation of shared similarities, Owen and early evolutionists would have inferred a hypothesis of homology, which causally accounts for these similarities by way of either an archetype in the case of Owen, or by descent with modification in the case of evolutionists. The effects which have been explained are regarded as homologues.

Owen / Darwinians, according to Lankester observed shared similarities homology: explanation via archetype or descent with modification common ancestry

separate origins

homologues: effects explained

Lankester claimed that with the advent of evolutionary theory, homology refers to two classes of causal explanations, in the form of common ancestry and separate origins.

Owen / Darwinians

Lankester

observed shared similarities

observed shared similarities

homology: explanation via archetype or descent with modification common ancestry

homogeny: common ancestry

homoplasy: separate origins

homogens

homoplasts

separate origins

homologues: effects explained

(replaces ‘homologue’)

A more recent exposition, which perpetuates some of these misinterpretations of homology, is the magnum opus of Stephen Jay Gould (2002), who presents some peculiar views regarding Lankester's (1870) paper on homogeny and homoplasy. As I have tried to show in this talk, it is only when we carefully read what Owen wrote that it becomes apparent that much of the history of the evolutionary concept of homology has been frought with errors, due in large part to Owen's obscure writing style. But once we recognize that Owen intended a clear distinction between the concept of homologue as referring to effects, and the concept of homology as representing causal explanations for those effects, these topics become open to more cogent treatment.

Owen / Darwinians

Lankester

observed shared similarities

observed shared similarities

homology

homogeny

homoplasy

homologues

homogens

homoplasts

Because of his interpretation of the relation of homologue to homology, this then was the basis for Lankester's replacing the term homologue with the terms homogen and homoplast. Lankester's interpretation of the relations between homology and homologue is, however, incorrect.

“The attempt [by Lankester] was an interesting one, but most morphologists wisely adhered to the old concept of homology....” Russell (1916: 267)

Lankester's (1870) suggested replacement of homology in evolutionary biology has been almost completely ignored, as indicated by Russell's (1916) review of the situation. However, there have never been any rigorous arguments presented that effectively counter Lankester's position.

“But morphology is a much more complex subject than it at first appears, as has lately been well shown in a remarkable paper by Mr. E. Ray Lankester, who has drawn an important distinction between certain classes of cases which have all been equally ranked by naturalists as homologous. He proposes to call the structures which resemble each other in distinct animals, owing to their descent from a common progenitor with subsequent modification, homogenous; and the resemblances which cannot thus be accounted for, he proposes to call homoplastic.” Darwin (1872: 385)

Interestingly, in the last edition of "The Origin of Species" (6th edition, 1872), Darwin appears to approve of Lankester's change of terms.

St. George Mivart. 1870. On the use of the term “homology.” Ann. Mag. Nat. Hist. 6 (ser. 4): 113-121. “...it is desirable to retain the word ‘homology’... in the very sense Professor Owen gave to it – namely, a close resemblance of parts as regards their relation to surrounding parts, to whatever cause that resemblance may be due, whether genetic or otherwise.” (p. 115)

An extensive criticism of Lankester's (1870) paper was published by St. George Mivart (1870). Mivart was willing to acknowledge Lankester's terminology, but only as components within a broader concept of homology. As an anti-Darwinian, Mivart sought to maintain Owen's archetype concept. In so doing, he seemed to at least be willing to expand the meaning of homology to the point that it can accommodate both archetype and evolutionary points of view, when he says, shown here, that resemblance may be due to factors which are genetic or otherwise. In fact, Mivart's statement effectively leaves us with the first definition of homology I proposed earlier, which simply stipulates that one has applied a common cause.

St. George Mivart. 1870. On the use of the term “homology.” Ann. Mag. Nat. Hist. 6 (ser. 4): 113-121. 1. non-homologous analogues

14. serial homotrophes

2. homologous analogues

15. vertical homotrophes

3. homogenetic homologues

16. lateral homotrophes

4. developmental homogens

17. antero-posterior homotrophes

5. ancestral homogens

18. actinologous homologues

6. homoplastic homologues

19. serial actinologues

7. homogenetic serial homologues

20. secondary actinologues

8. homoplastic serial homologues

21. serial secondary actinologues

9. vertical homologues

22. correlated secondary actinologues

10. lateral homologues

23. correlated serial secondary actinologues

11. antero-posterior homologues

24. special homologues

12. radial homologues

25. general homologues

13. homotrophic homologues

As an apparent result of such a compromise, Mivart (1870) recognized no fewer than 25 different types of relations within homology. I am not aware of anyone ever following his suggestions. But, this is a good example of the confusion surrounding homology which began early on in the development of evolutionary biology.

O. Haas & G.G. Simpson. 1946. Analysis of some phylogentic terms, with attempts at redefintion. Proc. Amer. Phil. Soc. 90: 319-349. “... homogeny (Lankester) [is]... a synonym of homology.” (p. 344)

The more typical reaction to Lankester's paper has been to simply regard homogeny as equivalent to the evolutionary connotation of homology.

Gould’s (2002: 1069-1089) misunderstanding of Lankester

Homology Homogeny

Homoplasy

A more recent exposition, which perpetuates some of these misinterpretations of homology, is the magnum opus of Stephen Jay Gould (2002), who presents some peculiar views regarding Lankester's (1870) paper on homogeny and homoplasy. As I have tried to show in this talk, it is only when we carefully read what Owen wrote that it becomes apparent that much of the history of the evolutionary concept of homology has been frought with errors, due in large part to Owen's obscure writing style. But once we recognize that Owen intended a clear distinction between the concept of homologue as referring to effects, and the concept of homology as representing causal explanations for those effects, these topics become open to more cogent treatment.

“Homology is resemblance due to inheritance from a common ancestry.” “Homoplasy is resemblance not due to inheritance from a common ancestry.” Simpson (1961: 78)

But while homology has been adopted by evolutionists, the concept of homoplasy has been recognized to some extent since 1870 in essentially the form suggested by Lankester. In fact, one of the best expositions on the subject was in the Haas & Simpson paper of 1946, cited earlier.

homology

Was Lankester (1870) correct in replacing ‘homology’ with ‘homogeny’ and ‘homoplasy?’

homogeny

homoplasy

Given the confusion surrounding the definitions of homologue and homology, it's not surprising that only Lankester's (1870) term homoplasy has survived. But, in light of what has been presented here regarding the correct definition of homology, a pertinent question is whether or not Lankester was correct in removing homology from evolutionary biology in favor of homogeny and homoplasy. I believe the answer is 'yes,' and the final part of this lecture will be a defense of Lankester's position.

Owen / Darwinians homologues: shared similarities in need of explanation

homology: explanation via archetype or descent with modification

common ancestry

separate origins

As noted earlier, Owen defined homologue as characters to which the same name is applied. Homologues, or homologous features, are then recognized as in need of explanation. This leads to inferences of homology as causal accountings for those homologues. This relation between homologue and homology differs from Lankester's (1870) view that the term homologue refers to effects already explained. Lankester was correct to point out that with the advent of evolutionary theory, explanatory hypotheses can be in the form of common ancestry or separate causes.

Lankester, slightly modified homologues

homogeny: common ancestry

homoplasy: separate origins

homogens

homoplasts

Upon applying the correct interpretation of homologue, if we accept Lankester's (1870) replacement of homology with homogeny and homoplasy, then effects can be explained as homogens and homoplasts, respectively. And the term homologue can still be maintained in its proper, original context.

Can ‘Homology’ Survive? – The Current Situation A.

B.

C.

D.

theoryx + homologues 1

theoryx + homologues 2

theoryx + homologues n

abduction

abduction

abduction

homology1

homology2

homology n

requirement of total evidence: discard homology hypotheses

theoryx + homologues 1, 2, ...n abduction

cladogram(s): ‘homology’ + homoplasy hypotheses

As a result of the interpretation of the inference of homology and our inferences of systematics hypotheses discussed in this course, we can identify the following relations. For separate sets of homologues, one applies the theory of common ancestry, and abductively infers respective homology hypotheses. But since we must abide by the requirement of total evidence, which states that all relevant evidence must be considered in inferences, we must effectively discard our original homology hypotheses, and instead apply the theory of common ancestry to all homologues together in a single inference, which concludes a phylogenetic hypothesis or set of mutually exclusive hypotheses. A cladogram might then be regarded as implying 'new' homology hypotheses, which account for some shared similarities by way of common ancestry, and homoplasy hypotheses, which are ad hoc explanations of other shared similarities accounted for by causal conditions other than strict common ancestry.

Can ‘Homology’ Survive? – The Current Situation A.

theoryx + homologues 1

theoryx + homologues 2

abduction

abduction

theoryx + homologues n abduction

Steps A-C are unnecessary. B.

C.

D.

homology1

homology2

homology n

requirement of total evidence: discard homology hypotheses

theoryx + homologues 1, 2, ...n abduction

cladogram(s): ‘homology’ + homoplasy hypotheses

But, such a relation between homology and phylogenetic hypotheses is incorrect because there is no basis for including steps A through C, as one would have to knowingly violate the requirement of total evidence, which is unwarranted. As a result, there is no basis for treating homology as a set of independent inferential acts separate from the more inclusive inference of what are commonly called phylogenetic hypotheses, or any of the other standard biological systematics hypotheses known as 'taxa.'

Solving the Problem – I theoryx + homologues 1, 2, ...n

abduction (per requirement of total evidence)

cladogram(s) (composite hypothesis) = ‘homology’ + homoplasy hypotheses hypotheses

shared similarities = ??? shared similarities = homoplasts

While shared similarities explained by way of homoplasy would be referred to as homoplasts, there is no equivalent term to be applied in "homology" hypotheses. These latter features cannot be termed homologues, since this concept is already applied to all named shared similarities in need of explanation. It is by the proper recognition of the nature of our causal hypotheses in systematics that Lankester's suggestion of completely removing the term homology from phylogenetic inference has its strongest justification, especially when coupled with the fact that, as I mentioned earlier, the proper definition of homology should be restricted to the sense originally intended by Owen.

Solving the Problem – II Reviving Lankester (1870) theoryx + homologues 1, 2, ...n

abduction (per requirement of total evidence)

cladogram(s) (composite hypothesis) = homogeny + homoplasy hypotheses hypotheses

homogens homoplasts

In fully applying Lankester's suggestions, a phylogenetic hypothesis consists of homogeny hypotheses, for which those shared similarities accounted for by way of common ancestry are termed homogens. A phylogenetic hypothesis can also contain homoplasy hypotheses, referring to homoplasts.

The Etymology of Terms Does Matter theoryx + homologues 1, 2, ...n

abduction (per requirement of total evidence)

cladogram(s) (composite hypothesis) = homogeny + homoplasy hypotheses hypotheses

homogens homoplasts

And especially when we look at the Latin derivations of these terms, we recognize that the application of Lankester's views are more precise than trying to adapt homology to our purpose.

The Etymology of Terms Does Matter theoryx + homologues 1, 2, ...n

‘same word’

abduction (per requirement of total evidence)

cladogram(s) (composite hypothesis) ‘same origin’

‘same kind’

= homogeny + homoplasy hypotheses hypotheses

‘to mold into the same’

homogens homoplasts

‘molded into the same’

Consistent with Owen's use, the term homologue, meaning 'same word,' accurately describes our naming of similar properties among organisms. Homogeny means 'same origin,' and homogen means 'same kind.' Homoplasy means 'to mold into the same,' which encompasses such causal processes as convergence and parallelism, and homoplast means 'molded into the same.'

Homologues, Homogeny, & Homoplasy The Proper Relations to Phylogenetic Inference

Causal theory:

Observations: (homologues)

If character x(0) exists among individuals of a reproductively isolated, gonochoristic or cross-fertilizing hermaphroditic population and character x(1) originates by mechanisms a, b, cJ n, and becomes fixed within the population by mechanisms d, e, fJ n (=ancestral species hypothesis), followed by event(s) g, h, iJ n, wherein the population is divided into two or more reproductively isolated populations, then individuals to which descendant species hypotheses refer would exhibit x(1).

– Members of species b-us and c-us have character 1(1) in contrast to 1(0) as seen among members of other species; – Members of species b-us and c-us have character 2(1) in contrast to 2(0) as seen among members of other species; – Members of species a-us and c-us have character 3(1) in contrast to 3(0) as seen among members of other species.

data matrix a-us b-us c-us 1: 0 2: 0 3: 1

1 1 0

1 1 1

We can now fully realize the proper application of Lankester's (1870) terms in the context of the abductive inferences of systematics hypotheses. The example shown here, and in the next slide, is the inference of phylogenetic hypotheses. Note in this instance that our named shared characters are homologues.

Homologues, Homogeny, & Homoplasy The Proper Relations to Phylogenetic Inference Causal theory:

Observations: (homologues)

Causal conditions:

If character x(0)...

– Members of species...

data matrix a-us b-us c-us 1: 0 2: 0 3: 1

1 1 0

1 1 1

homogeny hypothesis 1. Character 1(1) originated in a population of individuals with 1(0) by unspecified mechanisms, subsequent to which 1(1) became fixed in the population by unspecified mechanisms, followed by a population splitting event by unspecified mechanisms, leading to individuals observed in the present with 1(1) and to which species hypotheses bus and c-us also apply. homogeny hypothesis 2. Character 2(1) originated in a population of individuals with 2(0) by unspecified mechanisms, subsequent to which 2(1) became fixed in the population by unspecified mechanisms, followed by a population splitting event by unspecified mechanisms, leading to individuals observed in the present with 1(1) and to which species hypotheses bus and c-us also apply.

a-us b-us c-us 3(0) 6 3(1)*

3(0) 6 3(1)* 1(0) 6 1(1) 2(0) 6 2(1)

homoplasy hypothesis, ad hoc hypothesis of parallelism. Character 3(1) originated by unspecified mechanisms in two separate reproductively isolated populations of individuals with 3(0), subsequent to which 3(1) became fixed in both populations by unspecified mechanisms. This hypothesis is also a subset of species hypotheses a-us and c-us.

The recognition of homogeny and homoplasy hypotheses are then parts of the conclusions of the inference.

Taxonomy of Terms

homologues / homologous features: hypotheses accounting for perceptions of similar properties among organisms. abduction

“These lizards have blue tails (in contrast to brown tails).”

homogeny: causal hypothesis accounting for homologous features, as homogens, by ‘descent with modification.’

homoplasy: ad hoc hypothesis accounting for homologous features, as homoplasts, by separate, intraspecific events of parallelism or convergence.

“These lizards have blue tails because they are descended from members of a common ancestral population that had this feature.”

“These lizards have blue tails because of causal events other than common ancestry.”

In properly applying Lankester's (1870) dual concepts of homogeny and homoplasy in evolutionary biology, we can provide definitions of terms which no longer suffer from the confusion that has befallen homology for over 100 years. This is especially significant in light of the fact that the definition of homology should be limited to hypotheses involving Owen's (1846) archetype theory and not evolutionary theories. Hypotheses involving the evolutionary theories are best accommodated by the terminology offered by Lankester (1870). I suggest the following taxonomy of terms: We can readily maintain the concept of homologue as it was originally intended by Owen (1843, 1846), defined as hypotheses accounting for our perceptions of similar properties among organisms. This is indicated by our activity of applying the same names to characters of individuals. As it is the case that explanatory hypotheses are inferred to account for observed effects, the recognition that phylogenetic inference is abductive makes the dual recognition of homogeny and homoplasy more relevant than has been presented in the past. Lankester's homogeny is an accurate place holder for hypotheses accounting for homologous features, as homogens, by way of descent with modification or common ancestry. By referring to homogeny in evolutionary biology, we avoid the historical baggage of misinterpretations associated with homology. Which, as Lankester noted, had its origin in an entirely different philosophy. Lankester's term homoplasy remains as he originally intended, as an ad hoc hypothesis accounting for homologous features, as homoplasts, by separate, intraspecific events of either parallelism or convergence. More generally, homoplasy simply recognizes the ad hoc nature of the hypothesis, as a statement that causal events other than common ancestry must be invoked to account for homologues.

Taxonomy of Terms homologues / homologous features: hypotheses accounting for perceptions of similar properties among organisms.

analogues / analogous features: different structures or characters with similar functions and/or superficial similarity of appearance.

abduction

homogeny: causal hypothesis accounting for homologous features, as homogens, by ‘descent with modification.’

analogy: the study of different structures or characters with similar functions and/or superficial similarity of appearance.

homoplasy: ad hoc hypothesis accounting for homologous features, as homoplasts, by separate, intraspecific events of parallelism or convergence.

And we also can continue to use Owen's (1843, 1846) analogue and analogy.

Taxonomy of Terms observed effects

homologues / homologous features

analogues

why-question(s)

theory applied

archetype theory

evolutionary theory

abductive inference explanatory hypotheses

homology

homogeny

homoplasy

analogy

From a broader perspective, we can summarize relations between homology, homogeny, and homplasy as follows. For shared similarities, homologues, differentially observed among individuals, we implicitely or explicitly formulate why-questions in relation to facts in need of explanation. The alternative theories most commonly applied have been that of the archetype, as in the case of Owen's work, or evolutionary theories, principally descent with modification or common ancestry. Regardless of the theory applied, the nature of the inference is abductive, leading to respective explanatory hypotheses. Hypotheses derived from the archetype theory should be referred to as homology, whereas hypotheses from evolutionary theory are in the form of homogeny and homoplasy. By fully understanding the original foundations of these concepts, we might surmize that had Lankester's (1870) terminology been adopted early on, much of the confusion surrounding the interpretation and use of homology might have been avoided. To rectify the situation in the present simply requires that we acknowledge several conditions: the first being correct distinctions between Owen's and Lankester's views; the second being the fact that the subjects referred to in the definitions of homology, homogeny, and homoplasy are explanatory hypotheses associated with different causal theories; and lastly, that we must overcome the all too common practice of disregarding the requirement of total evidence in systematics.

The Myth of ‘Tests’ of Homology Patterson (1982) suggested two principle ways to test homology hypotheses: (1) topographic similarities and similarities of structural components; and, (2) character congruence on a cladogram. But, it is important to recognize that Patterson’s interpretation of ‘homology’ is equivalent to what should actually be regarded as homologues, not homology (or homogeny). The correct framework for testing hypotheses of homologues is presented next. But what is referred to as cladistic ‘character congruence’ cannot function as a test of hypotheses of homologues or homology.

Earlier in the course we have discussed the mechanics of hypothesis testing, especially as it involves the inferences (usually deductive) of potential test consequences, followed by the process (inductive) of determining whether or not such consequences are the case. While we will specifically look at the process of testing phylogenetic hypotheses later in the course, we now need to address the common view that homology is tested. The work by Patterson (1982) has been influential with regard to testing homology, where it was suggested that homology is tested by way of (1) similarities of topographic relations and structural components, and (2) congruence of characters on cladograms. The problem, however, is that such testing refers not to homology, but rather homologues. As a consequence, we need to examine Patterson's 'tests' in the context of homologues.

The Myth of ‘Tests’ of Homology – Testing Homologue Hypotheses The Correct Structure of Determining Potential Test Evidence for Hypotheses of Homologues (formerly Patterson’s ‘test of similarity’)

Theory of Feature X:

Homologue Hypothesis to be tested:

Potential Test Evidence:

features recognized as X are composed of subsidiary features or components x1, x2, x3... xn, and X has topographic relations r1, r2, r3... xn to other features.

specimens A, B, and C have feature X.

if specimens A, B, and C do each have the homologue X, then this feature will also have subsidiary features, or components x1, x2, x3...xn, and topographic relations r1, r2, r3... xn to other features.

Since Patterson's (1982) tests of homology are actually tests of hypotheses of homologous characters, the inference of potential consequences that might validly test those hypotheses would have the form shown here. Notice that the potential consequences for testing homologues are similar to Patterson's suggested examination of structural similarities.

The Myth of ‘Tests’ of Homology – Testing Homologue Hypotheses The Correct Structure of a Test of Hypotheses of Homologues

‘Humans have forelimbs’

Theory of ‘Forelimbs’:

Homologue Hypothesis to be Tested:

‘Bats have forelimbs’

features recognized as ‘forelimbs’ are composed of internal bones x1, x2, x3... xn, and ‘forelimbs’ have topographic relations r1, r2, r3... xn to the head, vertebrae, etc. humans and bats have structures called ‘forelimbs.’

An example, derived from the previous slide, of the inference of consequences that could act as potential test evidence for the homologous character, 'forelimbs.' The act of performing the test would, however, be inductive.

The Myth of ‘Tests’ of Homology – Testing Homologue Hypotheses The Correct Structure of a Test of Hypotheses of Homologues

‘Humans have forelimbs’

Theory of ‘Forelimbs’:

Homologue Hypothesis to be Tested: Potential Test Evidence:

‘Bats have forelimbs’

features recognized as ‘forelimbs’ are composed of internal bones x1, x2, x3... xn, and ‘forelimbs’ have topographic relations r1, r2, r3... xn to the head, vertebrae, etc. humans and bats have structures called ‘forelimbs.’ if specimens of humans and bats each do have the same structures called ‘forelimbs,’ then this feature will also be composed of internal bones x1, x2, x3... xn, and ‘forelimbs’ have topographic relations r1, r2, r3... xn to the head, vertebrae, etc.

An example, derived from the previous slide, of the inference of consequences that could act as potential test evidence for the homologous character, 'forelimbs.' The act of performing the test would, however, be inductive.

The Myth of ‘Tests’ of Homology – Testing Homologue Hypotheses The Correct Structure of a Test of Hypotheses of Homologues

‘Humans have forelimbs’

Theory of ‘Forelimbs’:

Homologue Hypothesis to be Tested:

‘Bats have forelimbs’

features recognized as ‘forelimbs’ are composed of internal bones x1, x2, x3... xn, and ‘forelimbs’ have topographic relations r1, r2, r3... xn to the head, vertebrae, etc.

humans and bats have structures called ‘forelimbs.’

Test Observations:

specimens of humans and bats have internal bones x1, x2, x3... xn, and ‘forelimbs’ have topographic relations r1, r2, r3... xn to the head, vertebrae, etc.

Conclusion:

homologue hypothesis is confirmed.

An example, derived from the previous slide, of the inference of consequences that could act as potential test evidence for the homologous character, 'forelimbs.' The act of performing the test would, however, be inductive.

The Myth of ‘Tests’ of Homology The ‘Test of Congruence’ is Not a Valid Test

data matrix a-us b-us c-us 1: 0 2: 0 3: 1

1 1 0

1 1 1

a-us b-us c-us 3(0) 6 3(1)* 3(0) 6 3(1)* 1(0) 6 1(1) 2(0) 6 2(1)

The Problem: A phylogenetic hypothesis cannot function as a test of the effects that have been used to abductively infer that hypothesis. A phylogenetic hypothesis only provides a causal accounting of one’s observations, not any kind of ‘test.’ Thus,... • the congruence of observational hypotheses (homologues, not homology!) of 1(1) and 2(1) do not say anything about the integrity of those observations; • The ‘incongruence’ of 3(1) on the cladogram only indicates that the explanation of that similarity must be by way of causal events other than strict common ancestry. The fact that ad hoc hypotheses (homoplasy) are required to explain homologues is to only offer an explanation. But that explanation, ad hoc or otherwise, can not serve as a test of a homologue hypothesis.

As described here, the 'test' of congruence for homologue (not homology) hypotheses is invalid.

Concepts of Homology and Associated Terms: A Clue to the Confusion Topographical & Structural Similarity

Character Congruence

Character Incongruence

Owen

homology

--------------

analogy

Lankester

homology

homogeny (= apomorphy)

homoplasy

Patterson

homology synapomorphy

synapomorphy

homoplasy

de Pinna

primary homology

secondary homology

homoplasy

Lipscomb

homology

synapomorphy

homoplasy

Brower & Schawaroch

observation

homology synapomorphy

homoplasy

Hennig

homology

synapomorphy

convergence

Mayr

homology

homology

convergence/ parallelism

Ball

homology

synapotypy

convergence/ parallelism

Author

From: Schuh (2000: 71), Biological Systematics: Principles and Applications

The Philosophy of Biological Systematics Course Outline – Part 3

1.

The requirement of total evidence.

2.

Homology & homogeny & homoplasy.

3.

Character coding.

4.

The mechanics of hypothesis testing in biological systematics.

The Inferential Basis for Coding Characters Observations ‘Characters’ 12345 a-us 0 0 0 0 0 b-us 1 1 0 0 1 c-us 1 1 1 1 2 d-us 1 1 1 1 2

The process of coding our observations into a data matrix has long been recognized as of fundamental importance to phylogenetic inference. Yet, no distinct philosophical criteria have been provided for deciding how to code those observations, especially for the treatment of what are commonly referred to as 'multistate characters.'

CRITERIA FOR CORRECTLY CODING OBSERVATIONS

• explicitly recognize why character coding is necessary in biological systematics. • accurately represent perceptual beliefs. • accurately convey the form of why-questions.

There are three basic criteria that must be met in order for any coding of characters to be epistemically and scientifically acceptable. The first criterion is that one's approach must be consistent with the goal of inference in biological systematics.

How Should Observations Be Coded? Common (incorrect) answer - choose a ‘coding method’

“The main problem in choosing an appropriate coding method is to arrive at an accurate division of characters and character states so that they reflect the relationships of the organisms.” Kitching et al. (1998:29), Cladistics

Too often, the issue of how to code observations has focused on selecting from among prescribed 'coding methods.' The problem, however, is that simply selecting a particular method of coding does not address the real issue. Our objective is to accurately convey our observations.

How Should Observations Be Coded? The correct first step...

In order to answer this question, we first have to answer these questions: 1.

‘Why do we need to code observations?’ • in order to causally account for shared characters, as answers to our why-questions.

2.

‘Why do we code some, but not all, observations?’ • evidential relevance, via the requirement of total evidence.

To properly determine how observations of characters of organisms should be coded into a data matrix first requires that we answer several fundamental questions.

How Should Observations Be Coded? The next step...

By representing our observation statements and why-questions as accurately as possible.

Theories/Laws (causal basis)

Systematics: causal understanding

Observations (descriptions)

data matrix (why-questions)

Next we need to acknowledge that the goal of character coding is to not only represent as accurately as possible our observation statements regarding the differentially shared characters of organisms, but such coding must also accurately convey our why-questions. This makes sense given that the inferences we produce in biological systematics are intended to answer why-questions regarding observations in need of explanation. The reason we engage in these practices is because the goal of both science and systematics is to acquire causal understanding.

How Should Characters Be Coded? – 1. Common Methods

‘Character’ 1: appendages 0. absent 1. present ‘Character’ 2: distal ends of appendages 0. entire 1. bifid At least three distinct coding 'methods' are commonly used for what are referred to as 'multistate characters.' Using the set of hypothetical organisms shown here, some workers would recognize two 'binary characters' -- these 'characters' being 'appendages' and 'distal ends of appendages.' Each of these 'characters' then have two 'states.'

How Should Characters Be Coded? – 2. Common Methods

‘Character’: appendages 0. absent 1. distally entire 2. distally bifid

An alternative procedure is to treat all the observed conditions as a single 'multistate character,' with three 'states.'

How Should Characters Be Coded? – 3. Common Methods

Absence/Presence Coding (Pleijel, 1995) ‘Character’ 1: appendages 0. absent 1. present ‘Character’ 2: distally entire appendages 0. absent 1. present ‘Character’ 3: distally bifid appendages 0. absent 1. present

A third method, known as 'absence/presence' or 'a/p coding,' was advocated by Pleijel (1995), and has been used in recent phylogenetic analyses, especially regarding polychaetes. In a/p coding, each condition observed of an organism is considered as a separate 'character,' while absence or presence of the condition comprise the 'states.'

How Should Observations Be Coded?

Do any of these coding methods accurately represent our observations and why-questions? No.

There are several fundamental reasons why these 'coding methods' cannot accurately represent our observations and why-questions. These issues will be examined next.

CRITERIA FOR CORRECTLY CODING OBSERVATIONS

• explicitly recognize why character coding is necessary in biological systematics. • accurately represent perceptual beliefs. • accurately convey the form of why-questions.

We first addressed the fact that character coding is of importance in biological systematics because we must be able to accurately represent both our observation statements as well as the why-questions that serve as the basis for our abductive inferences to particular explanatory hypotheses. We now need to clarify several long-standing misconceptions associated with character coding.

Observation Statements are Abductively Inferred I. SENSE DATA, i.e., non-verbal, mental images.

abduction: we apply various theories of objects to our sense data to infer hypotheses that explain what we perceive.

II. OBSERVATION STATEMENTS: “There are sabellid and serpulid polychaetes in this dish.” Observation statements are explanatory hypotheses, accounting for our sense data by the fact that objects and events exist.

WE PERCEIVE OBJECTS BY WAY OF THE VARIOUS PROPERTIES, OR CHARACTERS OF THOSE OBJECTS. WE DO NOT PERCEIVE OBJECTS BY WAY OF THE DISTINCTION OF ‘CHARACTERS / STATES.’

As we have seen thus far in this course, abductive inference is the most common and important mode of reasoning we use every day. Indeed, the very act of making observations and conveying them as observation statements can be characterized as the product of abduction. From our sense data, in the form of, for example, mental images, it is by way of abductive inference that we develop explanatory hypotheses to answer the implicit (or explicit) question of why one observes these objects. Such explanatory hypotheses are in the form of observation statements about our surroundings. In other words, we account for our sense data by inferring that objects and events exist independent of an observer. It is from our observation statements that we then ask why-questions regarding the properties of the objects perceived, which in the case of biological systematics, leads to additional abductive inferences that are ontogenetic, tokogenetic, specific, phylogenetic, etc., in form.

Mistake of the ‘Character’ / ‘State’ Distinction ‘Character’:

leaf margins

‘States’:

(0) smooth

‘Characters’ and ‘States’ ‘Characters’

(1) serrate

a-us Taxa b-us c-us d-us (2) undulate

12345 00000 11001 11112 11112 ‘State’

The most common approach in biological systematics has been to make distinctions between what are referred to as 'characters' and 'states.' In the example shown here, one would claim that 'leaf margins' is the 'character,' while there are distinct 'states' exhibited by individual leaves with respect to margins. This character-state distinction is extended to the data matrix, where each cell is said to contain a 'state' and each column represents a 'character.'

Properties of leaf margins: (0) leaf margins are smooth Properties of Things are always indicated by Subject - Predicate Relations One does not observe ‘characters’ with ‘states.’ Rather, what we observe are objects by way of their characters (= properties), and those observations are communicated by subject-predicate relations.

(1) leaf margins are serrate

(2) leaf margins are undulate subject

predicate

The distinction between 'characters' and 'states' is not epistemically correct, and does not accurately convey the nature of our observation statements. To correctly communicate what we perceive, we must express observations as subject-predicate relations. We perceive objects based on the properties they manifest. Statements as to the nature of all things can only be in the form of statements that relate a property, using a predicate-based language, to a subject. In the example shown here, there are different observed properties of leaves. The condition of each leaf must be stated as a complete proposition that identifies the relation between the subject, i.e. leaf margin, and a predicate that identifies the property of the subject. Once we correctly communicate our observations as subject-predicate relations, the problems of coding 'methods' become obvious.

The correct interpretation of the components of a phylogenetic data matrix.

The consequence of properly recognizing subject-predicate relations is that each cell in a data matrix is not a 'state,' but rather is a proposition about a property of an individual, which is expressed in subject-predicate form. Each column of the matrix is not a 'character,' but instead is the subject that is associated with individuals we have observed. To be acceptable, a coding procedure must correctly convey our observation statements in the form of subject-predicate relations.

Beware The Misuse of Absence

• the ‘absence’ of a property is not itself a property. • properties of things only exist and can be perceived by positive instances. • ‘absence’ is a relational concept rather than a descriptive concept - the ‘absence’ of a property in one individual is to presume the presence of the property in another individual. -------------------------------------------------------------------------Example: ‘Appendages are absent in species A,’ should be reworded as, ‘Lateral body wall margins are convex.’

Another problem with coding has been the misuse of the concept of absence. While we might say, for example, 'appendages are absent in this individual,' this is not an accurate description of the properties we actually observe of that individual. The three issues shown here indicate why this is so. These consequences are especially detrimental to a/p coding, which applies a concept of 'absence' that is both inappropriate for reflecting observations, as well as not correctly representing the subjects which we use in observational propositions.

The Myth of ‘Multistate’ Characters

(1) We observe objects on the basis of their characters or properties; (2) We communicate our observations by way of subject-predicate relations (the ‘character-state’ distinction is epistemically incorrect); (3) Thus, it is incorrect to refer to organisms as having ‘multistate’ characters. A character cannot exhibit ‘states.’ The correct phrase is multi-predicate subjects.

There will be instances in which a subject can be characterized by multiple predicates, similar to what is commonly, though incorrectly, referred to as a 'multistate' condition. Consider the situation shown here, where distal ends of appendages exhibit three properties: entire, bifid, or trifid.

A common error in most instances of ‘Multistate’ Coding Another reason it is incorrect to refer to the ‘character-state’ distinction

correct (1)

(0)

0: lateral body margins are convex

incorrect

1: lateral body margins are appendages (0)

(1)

(3)

subject-predicate relations are not correctly identified

(0)

(1)

0: distal ends of appendages are entire 1: distal ends of appendages are bifid subject-predicate relations correctly identified

The example shown here addresses how to correct the error of 'multistate' coding. Among the hypothetical organisms shown on the left, we must recognize there are two distinct subjects: 'lateral body margins' and 'distal ends of appendages' (right). In the case of the lateral body margins, there are two sets of subject-predicate relations: body wall is convex (not 'appendages are absent') or margins developed as appendages. Distinct subject-predicate relations are also recognized for distal ends of appendages: entire or bidif. Notice that in correctly conveying the properties of these individuals it is not possible to treat these as 'multistate' since this would not allow us to identify the distinctly different properties of the two different subjects which require explanation.

Some Conclusions (1) Most ‘multistate’ and all ‘absence/presence’ codings are not acceptable for the following reasons: • Observations cannot be accurately summarized as subject-predicate relations. • The concept of ‘absence’ is not correctly applied. (2) When recoded correctly, most ‘multistate characters’ are reduced to subjects with only two possible predicates.

There are several conclusions we can draw from what has been presented thus far.

Correct Coding of Multi-Predicate Subjects An Example

There will be instances in which a subject can be characterized by multiple predicates, similar to what is commonly, though incorrectly, referred to as a 'multistate' condition. In the example shown here, the distal ends of appendages exhibit three possible properties: entire, bifid, or trifid.

Correct Coding of Multi-Predicate Subjects • There are two subjects that first must be recognized. • Predicates applied to those subjects are then identified. • The coding of observations, as subject-predicate relations can proceed.

1. Lateral Body Margins (subject)

2. Appendages (subject)

(0) (0)

(1)

(2)

(1) 0: appendages are entire

0: lateral body margins are convex

1: appendages are bifid

1: lateral body margins are appendages

2: appendages are trifid

1. subject-predicate relations

2. subject-predicate relations

We first must acknowledge that there are two subjects: lateral body margins and distal ends of appendages. The multiple predicates can then be associated with relevant subjects, to accurately convey all observation statements.

CRITERIA FOR CORRECTLY CODING OBSERVATIONS

• explicitly recognize why character coding is necessary in biological systematics. • accurately represent perceptual beliefs. • accurately convey the form of why-questions.

We initially addressed the fact that character coding is of importance in biological systematics because we must be able to accurately represent both our observation statements as well as the why-questions that serve as the basis for our abductive inferences to particular explanatory hypotheses. We then clarified several long-standing misconceptions associated with character coding. Given that abductive inferences in biological systematics serve as answers to specific why-questions, then our process of coding characters into a data matrix must also be able to convey our why-questions. We now need to examine the nature of those matrices to understand how those why-questions are conveyed in a data matrix.

Character Coding: Observations and Why-Questions I. SENSE DATA, i.e., non-verbal, mental images.

abduction: we apply various theories of objects to our sense data to infer hypotheses that explain what we perceive.

II. OBSERVATION STATEMENTS: “There are sabellid and serpulid polychaetes in this dish.” Observation statements are explanatory hypotheses, accounting for our sense data by the fact that objects and events exist. III. WHY-QUESTIONS: Observation statements lead to our asking why-questions to acquire causal understanding of phenomena. In the case of biological systematics, the subjects of our observation statements to which why-questions are applied are the properties, or characters of organisms.

As we have saw earlier, it is the abductive inferences of our unexpected or surprising observation statements that lead to why-questions. Since why-questions are intimately connected to observation statements, those questions must appear in some capacity in data matrices.

A Phylogenetic ‘Data’ Matrix Represents Observations and Why-Questions subjects

Subject-Predicate relations among individuals to which species hypothesis d-us applies

Under the assumption that the goal of biological syestmatics inference is to causally account for differentially shared characters among individuals, then in addition to a data matrix containing our observation statements as subject-predicate relations, our why-questions also must be accurately and explicitly represented.

A Phylogenetic ‘Data’ Matrix Represents Observations and Why-Questions subjects

“Why do individuals to which species hypotheses c-us and d-us refer have character 4(1), in contrast to 4(0) as observed among members of other species?” Note that ‘4(1)’ codes the subject-predicate relation from our observations. Subject-Predicate relations among individuals to which species hypothesis d-us applies

Our why-questions are represented, in that a column in a data matrix, which indicates a particular subject, also represents a why-question as to the existence of different subject-predicate relations among individuals. But, as we have already seen, why-questions have a contrastive form. A data matrix must not just represent why-questions, but it must do so in a way that accurately represents the required form of such questions.

Why-Questions How we usually ask them

“Why P?” Example: “Why do these specimens have lateral body wall extensions called ‘appendages’?”

It is essential to know the formal structure of the why-questions we ask. We usually think of why-questions as simply having the form, "Why P?", or "Why is it the case that x is P?" This form is, however, incomplete and thus does not fully represent the basis for such questions.

Why-Questions The proper form: Contrastive questions

“Why P in contrast to X?” Example: “Why do these specimens have lateral body wall extensions (= appendages) in contrast to other specimens with convex body walls?”

P

X

The correct form of why-questions is that they are 'contrastive.' In other words, we ask questions that contrast the surprising condition in need of being explained with the expected condition that has already been explained. In the case of systematics-based observations, our contrastive why-questions are of the form shown here.

Why-Questions & Character Coding The basis for outgroup comparison

“Why P in contrast to X?” Example: “Why do these specimens have lateral body wall extensions (= appendages) in contrast to other specimens with convex body walls?”

P

X

In order to fully appreciate that data matrices do represent our why-questions, we should recall the formal structure of such questions. While we usually ask whyquestions as, "Why P?" or "Why is it the case that x is P?", this form is incomplete and thus does not fully represent such questions.

Why-Questions & Character Coding The basis for outgroup comparison fact / ingroup

“Why P in contrast to X?”

foil / outgroup

Example: “Why do these specimens have lateral body wall extensions (= appendages) in contrast to other specimens with convex body walls?”

P

X

Since why-questions are contrastive, instead of "Why P?", the actual question is "Why P in contrast to X?" What is unexpected, and in need of explanation, stands in contrast to what we already know. Hence, it is the unexpected nature of an experience relative to what was expected as a matter of routine that forms the basis of the why-question. What are presented here are the formal components of any contrastive question. Notice that the traditional use of 'outgroups' in systematics data matrices is actually intended to designate the contrast class, the sets of observations that already have been explained, relative to the 'ingroup' which consists of those observations in need of explanation.

A Phylogenetic ‘Data’ Matrix Represents Observations and Why-Questions subjects

contrast class

what requires explanation

“Why do individuals to which species hypotheses c-us and d-us refer have character 4(1), in contrast to 4(0) as observed among members of other species?” Note that ‘4(1)’ codes the subject-predicate relation from our observations.

Subject-Predicate relations among individuals to which species hypothesis d-us applies

It is with the inclusion of an 'outgroup' that each column in the matrix conveys each why-question in the correct contrastive form.

Character Coding Summarizes Our Why-Questions An example with a multi-predicate subject

1. Lateral Body Margins (subject)

2. Appendages (subject)

(0) (0)

(1)

(2)

(1) 0: appendages are entire

0: lateral body margins are convex

1: appendages are bifid

1: lateral body margins are appendages

2: appendages are trifid

There will be instances in which a subject can be characterized by multiple predicates, similar to what is commonly, though incorrectly, referred to as a 'multistate' condition. Consider the example shown here, where why-questions regarding two sets of subject-predicate relations need to be represented in a data matrix.

Character Coding Summarizes Our Why-Questions An example with a multi-predicate subject

1. Properties of Lateral Body Margins

q1: why do individuals to which species hypotheses a-us, b-us, and d-us refer have lateral body wall margins as appendages in contrast to convex body walls?

1 (0)

(1)

0: lateral body margins are convex 1: lateral body margins are appendages

outgroup: a-us b-us c- us d-us

0 1 1 1

The coding of the two alternate properties of the body wall, either convex or developed as appendages, not only summarizes our observations among individuals in each species, but must also represent the specific why-question. Notice that the structure of the why-question recognizes the convex body wall condition among members of species a-us as plesiomorphic - a condition that would have been previously explained. As noted earlier, the basis for outgroup comparison is that it designates the contrast class within our why-questions.

Character Coding Summarizes Our Why-Questions An example with a multi-predicate subject

2. Properties of Appendages

(0)

(1)

q2a:

why do individuals to which species hypothesis b-us refers have distally entire appendages in contrast to bifid or trifid appendages?

q2b:

why do individuals to which species hypothesis c-us refers have bifid appendages in contrast to entire or trifid appendages?

q2c:

why do individuals to which species hypothesis d-us refers have trifid appendages in contrast to entire or bifid appendages?

1 2

(2)

0: appendages are entire 1: appendages are bifid 2: appendages are trifid

outgroup: a-us b-us c-us d-us

0 1 1 1

? 0 1 2

In the case of the distal ends of appendages, since there is no recognized outgroup and thus no designated plesiomorphic condition, there are three why-questions. Within each question, all alternate properties are treated as 'outgroups,' or contrast classes. In this situation, it is during the inference of a phylogenetic hypothesis that each of these questions is answered by way of being considered relative to answers to other contrastive questions. This is somewhat reminiscent of the old concept of 'functional outgroups' and 'functional ingroups.'

More Conclusions (1) A character matrix represents two fundamental qualities needed for inferring explanatory hypotheses: • observations of shared characters are represented as subject-predicate relations. • the why-questions that refer to observed shared characters. (2) ‘Absence/presence coding’ cannot meet the criteria in (1) because: • observations are not correctly represented as subject-predicate relations. • ‘absence’ is not treated in the correct observational context.

There are two fundamental conclusions we can draw from this analysis. The most basic conclusion is that there is no one particular coding 'method' that can be advocated for all situations. Rather, the epistemic and philosophical criteria that have been presented constrain all coding practices. The other issue is that 'absence,' as a descriptive term, must be used with caution.

The Philosophy of Biological Systematics Course Outline – Part 3

1.

The requirement of total evidence.

2.

Homology & homogeny & homoplasy.

3.

Character coding.

4.

The mechanics of hypothesis testing in biological systematics.

Ernst Mayr and the Limits of Biological Systematics

In order to examine the mechanics of hypothesis testing in biological systematics, we need to place testing within the context of our desire to acquire understanding, since it is the acquisition of such understanding that is the goal of scientific inquiry. For this purpose, we will briefly examine the aspects of understanding in biology that were developed by evolutionary biologist, Ernst Mayr. This topic was first addressed at the start of the course, but it will be important to review this topic here since it is the process of hypothesis testing that leads to our revising or replacing the initial understanding we first acquire by way of abductive inference. The next six slides are repeated here from the first lecture (notes for these slides can be found with that lecture).

“...proximate causes govern the responses of the individual (and his organs) to immediate factors of the environment while ultimate causes are responsible for the evolution of the particular DNA code of information with which every individual of every species is endowed.” Mayr (1961: 1503)

1904-2005

Mayr suggested that biological inquiry seeks to acquire understanding that is causal, and that such causal understanding can be separated into 'proximate' and 'ultimate' causes. Mayr originally published his idea of proximate and ultimate causes in biology in 1961. What might be noticed is that proximate causes refer to those causes that only occur within an organism during its lifetime. Ultimate causes, on the other hand, transcend lifetimes.

“...biology can be divided into the study of proximate causes, the subject of the physiological sciences (broadly conceived), and into the study of ultimate (evolutionary) causes, the subject matter of natural history....” Mayr (1982: 67)

1904-2005

Mayr's 1982 monograph includes a good section discussing the proximate-ultimate distinction.

“The proximate causes of an organism’s traits occur within the lifetime of the organism.... The ultimate causes occur prior to the lifetime of the organism, within the evolutionary history of the organism’s species.” Beatty (1994: 334)

In his analysis of Mayr's proximate/ultimate distinction, Beatty (1994) offers a very good characterization, shown here.

Biological Understanding sensu Mayr proximate

ultimate

ontogenetic / functional

evolutionary

We can now begin to summarize Mayr's view of causal understanding in biology with the more general goal of science we examined earlier. We can see that proximate understanding refers to ontogenetic and functional aspects during the lifetime of an individual organism. Ultimate understanding refers to evolutionary causes that can apply to groups of organisms over time.

Biological Understanding sensu Mayr descriptive biology

ultimate

proximate

(observation statements) “It is sometimes overlooked how essential a component in the methodology of evolutionary biology the underlying descriptive work is. ”

Mayr (1982: 70)

ontogenetic / functional

evolutionary

Goal of Science – acquire ever-increasing understanding: • descriptive • causal - proximate / ultimate • predictive

But in addition to proximate and ultimate understanding, Mayr was very clear in his writings on the subject that there is a third dimension to understanding, what he referred to as 'descriptive biology.' Mayr was correct that in order to pursue either proximate or ultimate understanding, one must already have observations of effects that are in need of explanation. These effects are in the form of the properties, features, characters, etc., of organisms, that we communicate by way of our observation statements. Notice that Mayr's descriptive, proximate, and ultimate understanding are consistent with the goal of science presented earlier. To acquire ever-increasing understanding we see that it must be descriptive as well as causal, and also predictive. We seek descriptive understanding of what we perceive, as well as offering possible past causes that explain what we observe in the present. And we attempt to apply our understanding into the future with predictions of effects due to causal conditions that exist in the present.

Biological Understanding sensu Mayr descriptive biology

proximate

ultimate

ontogenetic / functional

evolutionary

(observation statements)

To what extent is biological systematics successful at acquiring ever-increasing understanding that is descriptive, proximate, and especially ultimate? An important part of this course will be to examine the extent to which descriptive, proximate, and ultimate understanding is acquired in biological systematics. Since we have already examined the abductive nature of reasoning to explanatory hypotheses for initial understanding, we are now prepared to examine how the act of testing accomplishes the task of increasing our causal understanding, which is the most fundamental goal of scientific inquiry.

(1913-1976)

Hennig, W. 1966. Phylogenetic Systematics

Recall that we saw earlier that one of the best examinations of the nature of causal relationships in systematics can be found in Willi Hennig's (1966) book, Phylogenetic Systematics.

7 Classes of Causal Relationships

1. ontogenetic (semaphoront)

6

2. cyclomorphic 3. sexual dimorphic

7 4

2

4. tokogenetic 5. polymorphic 6. specific (species)

1

3 5

7. phylogenetic

TAXA = Explanatory hypotheses

Hennig, W. 1966. Phylogenetic Systematics

Shown here is Hennig's (1966) well known figure 6, which we often see reproduced in other works on the principles of biological systematics. It is in this figure that Hennig identifies the fundamental classes of relationships used in systematics. But too often, what is not recognized is that Hennig pointed out that all of these relationships deal with individual organisms. He discussed in great detail seven classes of relationships involving organisms, all of which are shown in his diagram. Ontogenetic relationships. Where we speak of an individual at a particular point in it's life history. Cyclomorphic relationships. Where there are seasonal phenotypic differences among individuals of different generations. Sexual dimorphic relationships. The phenotypic differences between males and females. Tokogenetic relationships. Parents producing offsrping as a result of reproductive events (tokogeny). Polymorphic relationships. Different phenotypes expressed among individuals in a population. Specific relationships. Refers to species hypotheses, accounting for features among a group of organisms that are reproductively isolated from other groups. Phylogenetic relationships. The most general type of relationship in systematics, accounting for shared features among organisms to which different species hypotheses refer, as well as strictly asexual or strictly self-fertilizing hermphroditic organisms.

7 Classes of Causal Relationships

1. ontogenetic

6

Proximate

2. cyclomorphic 3. sexual dimorphic

7 4

2

4. tokogenetic Ultimate 5. polymorphic 6. specific (species)

1

3 5

7. phylogenetic Descriptive Biology (observation statements)

Using Hennig's (1966) figure 6, we can clearly identify the three broad classes of causal understanding recognized by Ersnt Mayr, that were referred to earlier.

“Science depends on judgments of the bearing of evidence on theory.... One of the central aims of the philosophy of science is to give a principled account of those judgments and inferences connecting evidence to theory.” Peter Lipton (2001: 184, Inference to the best explanation). In: A Companion to the Philosophy of Science.

Recall that we saw this quote near the start of the course, noting that a key part of our emphasis will be on recognizing the relations between evidence and biological systematics hypotheses. We have already seen that these relations occur in different ways, depending on what we mean by 'evidence,' as well as our objectives in maintaining particular relations. While most of our interest thus far has been in 'evidence' as it relates to abductive inference, we now need to examine the nature of the 'evidence' required to properly test hypotheses. As with abduction, Liption's statement also applies to any examination of test evidence relative to inferences.

Understanding: Initial versus Continued Level of Understanding

initial sketch --full

Mode of abduction Inference (inferences of explanatory hypotheses)

induction (hypothesis testing)

continued (goal of science)

To speak of testing as a fundamental part of our process of acquiring understanding, we need to identify the relations between initial and continued understanding. These relations are best presented according to the types of inference used. In most of this course, our attention has been focused on the acquisition of initial understanding by way of abductive inference. But as we have already seen, the abductive inferences of systematics hypotheses usually only allow for very vague causal understanding. Such limited hypotheses are known as 'explanation sketches.' In order to pursue continued understanding, by way of testing, we need explanatory accounts that provide more complete causal conditions, from which we can engage in testing by way of induction, and thus pursue continued understanding.

The Myths of... • hypothesis testing, and • support measures

...in Phylogenetic Systematics

A prominent topic for many years in biological systematics has been the claim that the views of philosopher of science, Karl Popper, are faithfully followed. This association of Popper to systematics is interesting since Popper was primarily concerned with testing, and denied there can be a logic of theory or hypothesis formation. As a consequence, systematists routinely confuse the process of hypothesis inference with the testing of hypotheses. In lieu of testing, there has been the alternative view that techniques other than testing can be used to evaluate 'support' for hypotheses. In this part of the course, we will deal with these issues by correctly identifying the nature of testing and the consequences for systematics. Since the emphasis on testing in biological systematics has mainly focused on phylogenetic hypotheses, especially in the form of cladograms, this will be the main focus here.

Why Do We Test Hypotheses? C The goal of science is to acquire causal understanding of what we observe around us. C The start to acquiring that understanding is the abductive inferences of hypotheses that provide tentative explanations of the objects and events we observe. C To evaluate if a hypothesis is acceptable as an explanation, the hypothesis must be tested. C Since a hypothesis makes specific causal claims, then there are consequences regarding those claims that can be sought. C Testing is the process of determining whether or not those consequences actually occur.

How Does One Evaluate a Phylogenetic Hypothesis? Hypothetico-Deductive (‘Popperian’) Testing

Statistical Testing

‘Other’ Approach

New Shared Similarities as Tests

Bootstrap

Bremer Support Analysis

Jackknife Permutation Tests

Regarding the evaluation of phylogenetic hypotheses, there have been three common approaches. The first of these, which also is the oldest, is the hypothetico-deductive method of testing, which most often has been claimed by systematists to follow the views of the philosopher of science, Karl Popper. The standard claim in systematics has been that the introduction of new shared characters have the capacity to test a previously inferred phylogenetic hypothesis. The second view is founded on the assumption that phylogenetic hypotheses are statistical constructs, such that several popular methods of statistical hypothesis testing can be employed. And finally, there is a popular approach among many systematists that is neither hypothetico-deductive nor statistical. This is called Bremer support analysis, where it is claimed that support for a hypothesis can be assessed based on the number of extra steps required by cladograms to result in the collapse of a particular clade.

How Does One Evaluate a Phylogenetic Hypothesis? Hypothetico-Deductive (‘Popperian’) Testing

Statistical Testing

Can new shared similarities serve as tests; i.e. are they deducible consequences?

Are phylogenetic hypotheses statistical in form? Are they amenable to standard approaches to testing in statistics?

‘Other’ Approach

Can ‘cladograms’ of different length be rationally compared?

In order to critically evaluate each of these approaches to testing, or determining evidential support for phylogenetic hypotheses, there are specific questions that need to be asked regarding each point of view.

PART I The Hypothetico-Deductive Myth in Biological Systematics

In this part of our examination of testing, we will examine the long-standing claim that systematics hypotheses are routinely tested in the Popperian or hypotheticodeductive spirit.

A Common Myth:

The use of derived character distributions as articulated by Hennig (1966) appears to fit the hypothetico-deductive model best. Gaffney (1979: 84)

The hypothetico-deductive or Popperian view in systematics has focused almost entirely on the testing of phylogenetic hypotheses, and especially the view that newly observed shared characters serve as the 'evidence' for testing those hypotheses. This has, however, led to the unfortunate consequence that the acts of inferring phylogenetic hypotheses and the subsequent testing of those hypotheses have become confused with one another.

A Common Myth: ...phylogenetic relationships of the species is tested by means of other series of characters : by trying to bring the relationships indicated by the several series of characters into congruence . Hennig (1966: 122) A B C D

A B C D new characters added: H is “tested.” phylogenetic hypothesis, H

“H is corroborated.”

Indeed, we can go all the way back to Hennig (1966), who suggested that character data have the capacity to test phylogenetic hypotheses. As new characters are added to a data matrix, those characters are supposed to function as tests of a previous hypothesis. For instance, we have a hypothesis that has been inferred from some set of characters, and additional characters are then 'predicted.' It is said that these new characters have the capacity to test this hypothesis. In the event these 'predicted' characters are then observed, and found to have distributions 'congruent' with the original hypothesis, it is said this hypothesis has been 'corroborated' (sensu Popper).

A Common Myth: ...phylogenetic relationships of the species is tested by means of other series of characters : by trying to bring the relationships indicated by the several series of characters into congruence . Hennig (1966: 122) A D

A B C D

C B

new characters added: H is “tested.” phylogenetic hypothesis, H

“H is falsified.”

Alternatively, if the new characters are not distributed as 'predicted,' such that the hypothesis required to accommodate observations has a different topology, then it is claimed that the earlier hypothesis has been 'falsified.' There are, however, two fundamental errors in reasoning here. One error is that these hypotheses can not be compared - it is meaningless to compare phylogenetic hypotheses that have been derived from different data sets. This issue was addressed in the earlier lecture on the requirement of total evidence. The second error has to do with recognizing the proper testing of hypotheses, and what qualifies as deductive consequences or predictions that can serve as potential test evidence. These are issues to be covered in this part of the course.

A Common Myth:

The distribution of characters provides potential tests of the relationships of taxa showing the characters. Wiley (1981: 110)

We will examine the mechanics of testing explanatory hypotheses, especially to point out that, contrary to the accepted view regarding the use of shared similarities as tests of phylogenetic hypotheses, shared similarities have no part to play in testing. In fact, the more fundamental conclusion that will be presented is that testing is a difficult task in the historical sciences, and one that is to be rarely accomplished with any degree of severity in biological systematics. Unfortunately, the hypothetico-deductive approach fails to accurately represent scientific activities, since it denies consideration of the abductive inferences of theories and hypotheses. It is in this latter realm that systematics primarily resides, rather than the limited perspectives of Popperian testing.

Examples of the Testing Myth

In order to test [sic] current hypotheses of arthropod evolution, we have analysed three independent [sic] lines of evidence: a phylogenomic dataset of 198 genes, a new miRNA dataset and a large morphological dataset. Rota-Stabelli et al. (2010: 7) Errors: (1) violation of requirement of total evidence; (2) meaningless cladogram comparisons; (3) incorrect testing.

The quotes shown in the next four slides exemplify the fundamental misunderstanding of the nature of test evidence and testing.

Examples of the Testing Myth For the first time, our data and analyses resolve the broad-scale relationships within Mollusca with strong support [sic]. This allows us to gain an understanding [sic] not only of the relationships of modern molluscs but also of the numerous Palaeozoic forms of molluscs. Smith et al. (2011: 3) Errors: (1) violation of requirement of total evidence; (2) confuses abductive support with test support; (3) incorrect claim of increased understanding.

Examples of the Testing Myth Our phylogenetic results indicate that a comparison of existing data with lampreys will provide an adequate test [sic] of the hypothesis [of vertebrate monophyly], because together, these taxa circumscribe the clade of all living vertebrates.... Heimberg et al. 2010: 19381

Errors: (1) violation of requirement of total evidence; (2) meaningless cladogram comparisons; (3) incorrect testing.

Examples of the Testing Myth

Errors: (1) violation of requirement of total evidence; (2) meaningless cladogram comparisons; (3) incorrect testing.

‘Science’ according to Popper Historical

Generalizing & Experimental

specific events (effects)

laws/theories

Goal:

causal explanation

unification

What are tested:

hypotheses of causal conditions

laws/theories

predictions of specific effects

predictions of specific effects

history, phylogenetics, paleontology (partim)

physics, biology, sociology (partim)

Units of interest:

Tests: Relevant to:

NB: It is unrealistic to strictly interpret all fields of science as either ‘historical’ or ‘generalizing/experimental.’

The characteristics of the 'generalizing/experimental' and 'historical' sciences especially point out that the focus in each is different, except with regard to the nature of testing theories or hypotheses in both. But, there are some fundamental differences in the ways testing is actually accomplished.

Testing: Experimental vs. Historical Sciences Present (=Observation)

Known Cause (experiment) Known Effect

While the focus of this part of the course is on the testing of explanatory hypotheses, most discussions about testing use examples from the experimental sciences. There are some important differences between these fields regarding the nature of testing, which need to be briefly mentioned. What is of principle interest in the experimental sciences is testing, usually theories, by way of controlled experiments. A theory or hypothesis is tested by providing controlled causal conditions in the present. In other words, the causal conditions are known to us. It is then a matter of observing whether or not a predicted effect occurs. What you will notice is that both cause and effect can be observed. We have the opportunity to know both.

Testing: Experimental vs. Historical Sciences Present (=Observation)

Past Known Cause (experiment) Known Effect

‘HISTORICAL’

In the case of the historical sciences, what we know in the present are observed effects that are in need of being explained. The difficulty is that the cause(s) that explains those effects occurred in the past, so no longer exist. As a result, the cause is often unobserved and unknown.

Testing: Experimental vs. Historical Sciences Present (=Observation)

Past Known Cause (experiment) Unknown Cause (not observable)

explanation

Known Effect

We seek explanatory hypotheses

‘HISTORICAL’

Thus, we infer an explanatory hypothesis to account for the observed effects. It is this hypothesis that we then want to test. But in comparison to the experimental sciences, where the relations between cause and effect can be known, the fact that a past causal event is usually not known can make it very difficult to test explanatory hypotheses since the relevant effects needed for a test might not be available. As we will see later, this is a distinct limitation that applies to the testing of phylogenetic, as well as many other classes of hypotheses in biological systematics.

Testing: Experimental vs. Historical Sciences Present (=Observation)

Past

Future Known Cause (experiment)

Unknown Cause (not observable) explanation We seek an explanatory hypothesis

‘HISTORICAL’

Known Effect

‘EXPERIMENTAL’

In the case of the experimental sciences, we know to some extent a cause or set of causes (causal or experimental conditions) in the present.

Testing: Experimental vs. Historical Sciences Present (=Observation)

Past

Future Known Cause (experiment)

Unknown Cause (not observable) explanation We seek an explanatory hypothesis

‘HISTORICAL’

prediction

Effect (potentially observable)

Known Effect

‘EXPERIMENTAL’

The concern then is to infer predictions of what effects should be observed in the future, given the known causal conditions.

“Predictions require a knowledge of the total cause; postdictions [= explanations], or statements about past events, can be based on partial effects, on records.” Reichenbach (1956: 23), The Direction of Time

Reichenbach (1956) provided a particularly clear description of the difference between prediction and explanation which points to the distinction between experimental and historical approaches. The experimental realm requires knowledge as complete as possible of causal conditions, from which can then be predicted specific future effects. The basis for explanation, however, can come from minimally complete knowledge of observed effects. And, as will be shown later, it is the stipulation of past causal, or initial conditions as fully as possible that are also required for testing explanatory hypotheses.

The Origins of Tests: The Deductive Paradigm • the inference of a hypothesis is not deductive, but instead is abductive. • if hypothesis H is true, then (deductively) observation e is predicted to be true. • the testing of a hypothesis is not deductive, but instead is inductive.

In order to discuss the testing of biological systematics hypotheses, we first need to recognize that the inferential structure for deriving potential tests is usually deductive. In its simplest form, the act of inferring a possible test of a hypothesis is to deduce from that hypothesis observable conditions that should be expected if the hypothesis is true. But the act of actually testing a hypothesis is not deductive, but rather inductive. Similarly, the inference of an explanatory hypothesis is abductive, not deductive. It is important that we recognize the distinctions between the inferential structures used to formulate a test, as opposed to the inferential structures used for deriving a hypothesis or subsequently testing it.

The Basic Relations Between Abduction, Deduction and Induction

• abduction: inference of a hypothesis. • deduction: inference of potential test evidence. • induction:

the act of testing the hypothesis by way of the presence or absence of test evidence.

Summary of the relations between the different classes of reasoning commonly used in science.

Testing Explanatory Hypotheses: A hypothesis must first be inferred Present (=Observation)

Past

Explanatory Hypothesis

abduction

Known Effect

In order to discuss the mechanics of testing phylogenetic hypotheses, we first need to recognize the logic required to infer those hypotheses. This is critical to later examining problems with the way phylogenetic hypotheses are said to be tested. As we have already seen in the course, explanatory hypotheses are inferred by that type of non-deductive reasoning known as abduction, or abductive inference.

Abductive Inference: The Inference of Explanatory Hypotheses Present (=Observation)

Past Explanatory Hypothesis

abduction

Known Effect

• T1, T2, ..., Tn:

causal theory(ies), background knowledge

• e:

known effect(s)

• H, or, c1, c2, ..., cn:

hypothesized past causal conditions (explanatory hypothesis)

The formal structure of an abductive inference is that a causal theory (or theories) is applied to observed effects in need of being explained. From these premises we infer an explanatory hypothesis that states at least some of the causal events or initial conditions that might have occurred in the past, and that account for effects observed in the present.

Formal Explanatory Structure

• T1, T2, ..., Tn:

causal theory(ies), background knowledge

• H:

hypothesized causal conditions [past]

• e:

known effect(s) [present]

For the purpose of considering the testing of an explanatory hypothesis, we can arrange the hypothesis into a deductive framework that characterizes our explanation of the observed effects. In other words, if we apply a causal theory (or theories) to our explanatory hypothesis, then the effects we observe would be expected as instances of the theory.

Testing Explanatory Hypotheses Inference of potential test evidence

• T1, T2, ..., Tn:

causal theory(ies), background knowledge

• H:

hypothesized causal conditions [past]

• e:

known effect(s) [present] • e!!:

predicted effect(s) [future]

nb: These deduced effects cannot be character data. It is from this deductive form that we could then infer observations we would expect to find if it is the case that the explanatory hypothesis accurately states past causal events that led to the effects we observe now. It is especially important at this point to note that we are attempting to judge whether or not this explanatory hypothesis best accounts for effects we have observed. This means that the effects we predict as potential test evidence should be of a form that could only occur as a result of the causal conditions we think existed in the past. Therefore, the effects we seek to find for the process of testing are related as specifically as possible to the hypothesized causal events. As we will see later, this requirement has distinct implications for correctly testing phylogenetic hypotheses.

Testing Explanatory Hypotheses Present (=Observation)

Past Explanatory Hypothesis

Specific Causal Condition(s)

Future abduction

deduction

Known Effect

(evidence1) test that should be performed

induction

testing of hypothesis by seeking occurrences of test evidence

(evidence2)

We can now summarize the relations between abductive inference of an explanatory hypothesis and the testing of that hypothesis. It is from effects observed in the present that we infer by way of abduction an explanatory hypothesis. The effects, plus theory, comprise our initial evidence upon which the hypothesis is concluded. From the specific causal conditions stated in that hypothesis, we deduce effects that should be observed that are only possible because the specific causal conditions that occurred in the past would allow for those effects. The deduction of such effects provides the basis for the tests that need to be performed. The act of testing the hypothesis is, however, inductive, where the hypothesis is either accepted or rejected, based on what we find during the tests. Notice that the test results are the evidence that either confirms or disconfirms the hypothesis. Since no test can guarantee the truth of a hypothesis, and a disconfirmed hypothesis simply leaves us with alternative hypotheses to consider, testing is always inductive.

Testing Explanatory Hypotheses: a simple example (from Popper, Objective Knowledge)

Present (=Observation)

“There is a dead rat in the room.”

We can use a simple of example to illustrate what has just been discussed. This example is derived from one provided by Karl Popper, and is one of the few where he specifically addressed the testing of an explanatory hypothesis as opposed to theories. Suppose that upon walking into a room, I observe a dead rat on the floor.

Testing Explanatory Hypotheses: a simple example (from Popper, Objective Knowledge)

Present (=Observation)

All Why-Questions are ‘Contrastive’ “Why is there a dead rat in the room?” [in contrast to no rats in the room]

As with any observation that presents us with unexpected or surprising facts, or facts not readily placed within an already accepted theoretical framework, we ask why-questions. As we have already seen, a why-question has a contrastive form - the question being asked is in reference to observations that are in need of explanation in contrast to our usual expectations of phenomena of the type that have already been explained. It is because the dead rat in the room is outside of my normal expectations that I have a basis for asking the why-question.

Testing Explanatory Hypotheses: a simple example (from Popper, Objective Knowledge) Present (=Observation)

Q: “Why is there a dead rat in the room?” [in contrast to no rats in the room] A:

“It ate rat poison.”

Subsequent to my asking this why-question, I might infer that the rat is dead because it must have eaten rat poison that is present in the room.

Testing Explanatory Hypotheses: a simple example (from Popper, Objective Knowledge)

Present (=Observation)

Past

Abduction

ˆ H, Causal Condition: this rat ate at least 0.1 gram of poison.

• Theory: when a rat eats more than 0.1 gram of poison, it will die. • Effect: there is a dead rat in the room.

My answer to the why-question is the product of an abductive inference. I apply a theory regarding the eating of poison by rats to my observation of the dead rat, to conclude that the past event of the rat eating poison might have occurred.

Testing Explanatory Hypotheses: a simple example (from Popper, Objective Knowledge)

Present (=Observation)

Past

Explanation

• Theory: when a rat eats more than 0.1 gram of poison, it will die.

ˆ Effect: there is a dead rat in the room.

• H, Causal Condition: this rat ate at least 0.1 gram of poison.

From either a predictive or explanatory perspective, observing the dead rat would be expected given that it did in fact eat the poison.

Testing Explanatory Hypotheses: a simple example (from Popper, Objective Knowledge) Present (=Observation)

Past

Explanation

• Theory: when a rat eats more than 0.1 gram of poison, it will die. • H, Causal Condition: this rat ate at least 0.1 gram of poison.

ˆ Effect: there is a dead rat in the room.

Prediction: traces of poison will be present in the dead rat.

In order to test the claim that the hypothesis does account for what is observed, we need to deduce other, independent effects from the causal event. The goal of the test is to determine whether or not the hypothesized causal condition is true in explaining the presence of the dead rat. What therefore provides useful test evidence are predicted effects related as closely as possible to the causal event stated in the hypothesis. In other words, we want to find test evidence with the lowest probability of occurrence if the hypothesized causal event did not occur. In this case, we can predict that such test evidence would be an effect in the form of the presence of poison or poison residue in the rat's body. This effect is independent of the fact that the rat is dead, since there could be other causes for the rat dying. And, the presence of poison in the rat would be the most specific effect directly produced by the hypothesized past causal event.

Testing Explanatory Hypotheses: a simple example (from Popper, Objective Knowledge) Present (=Observation)

Past

• Theory: when a rat eats more than 0.1 gram of poison, it will die. • H, Causal Condition: this rat ate at least 0.1 gram of poison.

Prediction: traces of poison will be present in the dead rat.

Test of H: conduct an autopsy of the rat and look for traces of rat poison.

Performing the test of the hypothesis then involves putting ourselves in a position to observe whether or not poison is present in the rat. In this case, the test conditions would involve performing an autopsy on the rat and analyzing tissues for the presence of specific types of poison.

Testing Explanatory Hypotheses: a simple example (from Popper, Objective Knowledge) Present (=Observation)

Past

Observed effect • Theory: when a rat eats more than 0.1 gram of poison, it will die.

e (poison is present), ˆ H is confirmed

or

• H, Causal Condition: this rat ate at least 0.1 gram of poison.

-e (poison is absent), ˆ H is disconfirmed Prediction: traces of poison will be present in the dead rat.

Test of H: conduct an autopsy of the rat and look for traces of rat poison.

In the event poison is found in the rat, we might claim this is confirming test evidence supporting the hypothesis. Alternatively, if no poison is found, then the hypothesis has been disconfirmed. Such a result leaves us to consider other possible hypotheses, as well as having to decide which of these to test in the future. Notice that a statement of confirmation or disconfirmation is an inductive conclusion from the premises comprising the test conditions and test outcome.

Testing Explanatory Hypotheses: a simple example (from Popper, Objective Knowledge) Present (=Observation)

Past

Explanation

• Theory: when a rat eats more than 0.1 gram of poison, it will die. • H, Causal Condition: this rat ate at least 0.1 gram of poison.

ˆ Effect: there is a dead rat in the room.

Prediction: look for more dead mice.

There is an incorrect alternative to the example of testing just shown that we need to recognize. As we will see later, this example uses the popular approach seen in biological systematics. One might claim that a valid way to test the hypothesis that this rat died from eating poison would be to look for additional dead rats. The parallel to this in systematics is when it is claimed that additional observations of shared characters can serve to test a systematics hypothesis, such as a cladogram.

Testing Explanatory Hypotheses: a simple example (from Popper, Objective Knowledge) Present (=Observation)

Past

Explanation

• Theory: when a rat eats more than 0.1 gram of poison, it will die. • H, Causal Condition: this rat ate at least 0.1 gram of poison.

ˆ Effect: there is a dead rat in the room.

not confirming evidence

Prediction: look for more dead mice.

not valid predictions

The problem, however, is that the occurrences of more dead rats are not phenomena that are consequences of the hypothesis. Recall that the hypothesis addresses possible past causal conditions for the rat I found dead on the floor. Simply finding additional dead rats cannot serve as test evidence to support the hypothesis because there is nothing in such observations that address the causal conditions presented in the hypothesis. As we will see later, this is exactly why the traditional approach in systematics, especially cladistics, of claiming character data can test phylogenetic hypotheses is incorrect.

Evidence Used to Infer a Hypothesis Cannot Also Be Used to Test That Hypothesis “Indeed, any reason which we may state in support of any hypothesis must be other than, and independent of, the [effects being explained]. If we can only adduce the [effects] as evidence, we feel that our explanation is circular, and therefore quite unsatisfactory.” Popper (1972: 351), Objective Knowledge

• T1, T2, ..., Tn:

causal theory(ies), background knowledge

• H:

hypothesized causal conditions [past]

• e:

known effect(s) [present]

• e!:

predicted effect(s) [future]

Popper (1972) recognized the inadequacy of simply relying on the effects in need of explanation as the only basis for claiming 'support' for an explanatory hypothesis. Similarly, additional effects of the same type being explained by the hypothesis cannot serve as test evidence.

Ad hoc / Circular Support (K. Popper, 1992: 132-133)

“Why is the sea so rough today?” – “Because Neptune is very angry.” “By what evidence can you support your statement that Neptune is very angry?” – “Oh, don’t you see how very rough the sea is? And is it not always rough when Neptune is angry?”

This now brings us to an issue that is imporant to understand, especially since the topic of 'evidence' is so often referred to in biological systematics, not only in terms of inferring hypotheses but also with regard to testing those hypotheses or when speaking of hypothesis 'support.' Recall from the lecture of the overview of the nature of science and reasoning that evidence comprises the premises used to infer a conclusion. Thus, while we often speak of evidence, it is important that we speak of that evidence in the context of the type of inference we are using. For example, the evidence (premises) used to abductively infer an explanatory hypothesis will be very different from the evidence (premises) used in the act of testing that hypothesis. As a result, we need to distinguish between two meanings of 'evidence.' There is evidence that provides the basis for suggesting a hypothesis in the first place. And there is evidence used to judge whether or not that hypothesis is true. As we will see, making this distinction is important, and it is a distinction that is rarely acknowledged in biological systematics.

Quasi-Ad hoc / Circular Support (K. Popper, 1992: 132-133) “Why is the sea so rough today?”

– “Because Neptune is very angry.” “By what evidence can you support your statement that Neptune is very angry?”

– “Oh, don’t you see how very rough the sea is? And is it not always rough when Neptune is angry?”

“This explanation is found unsatisfactory because... the only evidence for the [explanation] is the [effects being explained]. The feeling that this kind of almost circular or ad hoc explanation is highly unsatisfactory, and the corresponding requirement that explanations of this kind should be avoided are... among the main forces in the development of science....”

This now brings us to an issue that is imporant to understand, especially since the topic of 'evidence' is so often referred to in biological systematics, not only in terms of inferring hypotheses but also with regard to testing those hypotheses or when speaking of hypothesis 'support.' Recall from the lecture of the overview of the nature of science and reasoning that evidence comprises the premises used to infer a conclusion. Thus, while we often speak of evidence, it is important that we speak of that evidence in the context of the type of inference we are using. For example, the evidence (premises) used to abductively infer an explanatory hypothesis will be very different from the evidence (premises) used in the act of testing that hypothesis. As a result, we need to distinguish between two meanings of 'evidence.' There is evidence that provides the basis for suggesting a hypothesis in the first place. And there is evidence used to judge whether or not that hypothesis is true. As we will see, making this distinction is important, and it is a distinction that is rarely acknowledged in biological systematics.

Evidence Used to Infer a Hypothesis Cannot Also Be Used to Test That Hypothesis

“...for evidence e to support hypothesis H (or for e to be a good test of H), ...e itself must not have been used in H’s construction.” Mayo (1996: 258), Error and the Growth of Experimental Knowledge

C The class of effects used to infer an explanatory hypothesis cannot then be used to predict additional, similar effects as potential test evidence.

A good review of this subject is presented by Deborah Mayo (1996).

Evidence Used to Infer a Hypothesis Cannot Also Be Used to Test That Hypothesis

“...if some particular feature of [theory] T was in fact tied down on the basis of e... then checking e clearly constitutes no real test of T.... even though e follows from T and hence not-e is, in Popper’s terminology, a potential falsifier of T – it wasn’t really a potential falsifier of T, since T was, because of its method of construction, never at any risk from the facts described by e.” Worrall in Mayo (1996: 264), Error and the Growth of Experimental Knowledge

Relations Between the Types of ‘Evidence’ in Biological Systematics

Evidence 1: The basis for initially suggesting a hypothesis. Since a hypothesis is abductively inferred, this ‘evidence’ consists of character data plus the causal theory(ies). The relation of evidence1 to the hypothesis is one effects [and theory(ies)] providing the initial basis for considering the hypothesis as a possible explanation of the effects. Evidence 2: The basis for judging a hypothesis to be true. This is the ‘evidence’ derived as deductive consequences from the hypothesis [and theory(ies), and which is sought during testing of the hypothesis. The relation of evidence2 to the hypothesis is that of deduced consequences (‘test evidence’) following from the specific causal conditions presented in the hypothesis in conjunction with the theory(ies).

This now brings us to an issue that is imporant to understand, especially since the topic of 'evidence' is so often referred to in biological systematics, not only in terms of inferring hypotheses but also with regard to testing those hypotheses or when speaking of hypothesis 'support.' Recall from the lecture of the overview of the nature of science and reasoning that evidence comprises the premises used to infer a conclusion. Thus, while we often speak of evidence, it is important that we speak of that evidence in the context of the type of inference we are using. For example, the evidence (premises) used to abductively infer an explanatory hypothesis will be very different from the evidence (premises) used in the act of testing that hypothesis. As a result, we need to distinguish between two meanings of 'evidence.' There is evidence that provides the basis for suggesting a hypothesis in the first place. And there is evidence used to judge whether or not that hypothesis is true. As we will see, making this distinction is important, and it is a distinction that is rarely acknowledged in biological systematics.

Evidence1 vs. Evidence2 (N.R. Hanson, 1958: 200, note 2)

“[There] must be an appreciation of the logical distinction between (1) reasons for accepting an hypothesis H, and (2) reasons for suggesting H in the first place. (1) is pertinent to what makes us say H is true, (2) is pertinent to what makes us say H is plausible. Both are the province of logical inquiry, although H-D theorists discuss only (1) saying that (2) is a matter for psychology or sociology – not logic. This is just an errorM. We are discussing the rationale behind the proposal of hypotheses as possible [explanations]. H-D theorists never raise the problem at all.”

The distinction made in the previous slide, between the evidence that allows us to abductively infer hypotheses and the evidence needed to test those hypotheses, is clearly recognized by N.R. Hanson (1958). Notice that Hanson stresses that there has been a tendency to ignore this distinction. It has been this failure to acknowledge the fundamental difference between these two types of evidence that has created significant problems in biological systematics, especially in misinterpreting what is required to actually test hypotheses.

Relations Between the Types of ‘Evidence’ in Biological Systematics The realm of systematics Evidence 1: The basis for initially suggesting a hypothesis.

ABDUCTIVE SUPPORT

U

evidence = effects in need of explanation + theory

Evidence 2: The basis for judging a hypothesis to be true – empirically assessing understanding.

TEST SUPPORT

Y

evidence = results of actual testing

As we have seen, the evidential support that has typically been the focus in biological systematics has been evidence1, i.e. the character data that suggest a possible explanatory hypothesis for those data. The evidence2 needed to critically judge hypotheses is almost never obtained.

Testing Phylogenetic Hypotheses (Correctly) Character data are not test evidence

A B C D

A B C D new characters added:

H is “tested.” phylogenetic hypothesis, H

“H is corroborated.”

A B C D

A D C B new characters added:

H is “tested.” phylogenetic hypothesis, H

H corroborated.” is falsified.” “H“is

We are now at a point that we can examine what is required to correctly test biological systematics hypotheses. Since the emphasis on testing in systematics has mainly focused on phylogenetic hypotheses, we will use these as examples.

Abductive Inference: The Inference of Explanatory Hypotheses Present (=Observation)

Past Explanatory Hypothesis

abduction

!

Known Effect

• T:

causal theory(ies)

• e:

known effect(s)

• H, or, c1, c2, ..., cn:

hypothesized past causal conditions (explanatory hypothesis) ! phylogenetic hypothesis

!

‘descent with modification’ shared similarities

Just to review what we have covered thus far, hypotheses in biological systematics are inferred by way of abductive inference. We apply particular theories to observed effects in need of being explained. From these premises we infer an explanatory hypothesis, or hypotheses, that states at least some of the causal events or initial conditions that might have occurred in the past to account for the effects observed in the present.

Phylogenetic Hypotheses, in the Form of ‘Cladograms,’ are ‘Explanation Sketches’ “What the explanatory analyses of historical events offer is, then, in most cases not an explanation..., but something that might be called an explanation sketch. Such a sketch consists of a more or less vague indication of the laws and initial conditions considered as relevant, and it needs ‘filling out’ in order to turn into a full-fledged explanation.”

Hempel (1965: 238), Aspects of Scientific Explanation

It also was noted earlier that many of the hypotheses inferred in biological systematics are nothing more than 'explanation sketches.' As we will see in this section, the fact that systematics hypotheses have this quality immediately precludes them from being properly tested.

Phylogenetic Hypotheses, in the Form of ‘Cladograms,’ are ‘Explanation Sketches’ 0 a-us

1 b-us

1 c-us

Individuals observed in the present, to which species hypotheses a-us, b-us, c-us refer.

Past tokogenetic events referring to character origin and fixation, leading to individuals observed in the present, represented by species hypothesis c-us. Population splitting event. Past tokogenetic events among members of an ancestral population, during which character 1 originated and became fixed in the population.

Notice that a ‘cladogram’ neither implies nor states the specific causal events that explain observed shared similarities. A cladogram is a very vague and incomplete explanation.

To solve the problem of systematics hypotheses being explanation sketches, rather than full explanations that can be empirically tested, we first need to identify what these sketches actually have to offer. Using the cladogram shown here, we can identify three classes of hypotheses. One of these is said to be a species hypothesis, e.g. c-us. The other two classes of hypotheses comprise what would be more generally called a phylogenetic hypothesis. In the case of the example shown here, the phylogenetic hypothesis is summarized by the cladogram. But two hypotheses in this cladogram are indicated what are commonly known as the 'internode' and 'node.' What you should notice for the three hypotheses shown on the cladogram is that all of them are extremely vague. None of them provides full explanatory accounts. As such, there would be no way to proceed with testing any of these hypotheses. It first would be necessary to fill out the causal conditions within each. This is a fundamental condition that is usually ignored in systematics.

Testing Phylogenetic Hypotheses: a simple example

Present (=Observations)

a-us: 0

b-us: 1

Individuals to which species hypotheses b-us and c-us refer have character 1, while members of other species have 0.

c-us: 1

We can examine a simple example to show the proper mechanics of testing phylogenetic hypotheses. Let's say we observe that all individuals to which species b-us and c-us refer have character 1, in contrast to members of all other species that have character 0.

Testing Phylogenetic Hypotheses: a simple example Present (=Observations)

a-us: 0

b-us: 1

Q: Why do individuals to which species hypotheses b-us and c-us refer have character 1, while members of other species have 0?

c-us: 1 effects, i.e., shared similarities, e1...en

Since what we observe among members of b-us and c-us are new or unexpected effects, these observations lead to the why-question shown here. As we have already seen, such questions are contrastive in form. What we would normally expect would be to observe individuals with character 0, whereas our new observations show individuals with a new character that is in need of being explained.

Testing Phylogenetic Hypotheses: a simple example Present (=Observations)

Past Q: Why do individuals to which species hypotheses b-us and c- us refer have character 1, while members of other species have 0?

a-us: 0 b-us: 1 c-us: 1

ˆ causal conditions, H a-us b-us c-us

abduction: • theory, T • e1...en

effects, i.e., shared similarities, e1...en

As we have already discussed, our attempt to answer the why-question occurs by way of abductive inference to an explanatory hypothesis.

Testing Phylogenetic Hypotheses: a simple example Present (=Observations)

Past A: Character 1 originated within a reproductively isolated population with character 0 and eventually became fixed throughout the population during tokogeny, with a subsequent population splitting event leading to individuals to which species hypotheses b-us and c-us refer.

ˆ causal conditions, H a-us b-us c-us

Q: Why do individuals to which species hypotheses b-us and c- us refer have character 1, while members of other species have 0?

a-us: 0 b-us: 1 c-us: 1

abduction: • theory, T • e1...en

effects, i.e., shared similarities, e1...en

In order to actually address the matter of testing any of the hypotheses implied by cladograms, we have to carefully expand these hypotheses to a fuller explanatory form. While the written form of the hypotheses implied by the cladogram are shown here, they are still far too incomplete to lead to proper testing.

Testing Phylogenetic Hypotheses: a simple example Present (=Observations)

Past A: Character 1 originated within a reQ: productively isolated population with character 0 and eventually became fixed throughout the population during tokogeny by causal events u, v, w, with a subsequent population splitting event by causal events x, y, z, leading to individuals to which species hypotheses b-us and c-us refer.

ˆ causal conditions, H a-us b-us c-us

Why do individuals to which species hypotheses b-us and c- us refer have character 1, while members of other species have 0?

abduction: • theory, T • e1...en

a-us: 0 b-us: 1 c-us: 1

effects, i.e., shared similarities, e1...en

What would need to be specifically filled out are the causal events associated with character fixation and subsequent population splitting.

Testing Phylogenetic Hypotheses: a simple example Present (=Observations)

Past

a-us: 0 0