Idea Transcript
Does Print Media Influence Public Opinion? An Analysis of Marijuana Legalization Opinion Coverage in Newspapers Julia Verbrugge December 2014
Abstract This paper studies the influence of mass media on public opinion of marijuana legalization. Using a measure of media framing in the most widely circulated newspapers in 11 states over the period from 2010 to 2014, I seek to explain the changes in public opinion on the issue of marijuana legalization. To identify the position of the print media, I rely on editorials and op-ed columns, which I then use to evaluate the changes in newspaper support over time. Controlling for social demographic factors of each state, I find that there is an 8.24 percent—statistically significant and positive—relationship between media framing and public opinion over time. I also find insignificant relationships between changes in media support and changes in public support over the period of interest. Analysis of public opinion trends on framing shifts over time also suggests that direction of causality is such that the media influences public opinion. In sum, the paper uncovers preliminary relationships between the media’s coverage of the topic and the public’s support of marijuana legalization.
I. Introduction
Multiple studies have evaluated the role that mass media plays in shaping public opinion and driving public policy innovation, but none have done so with the highly topical subject of recreational marijuana legalization. This study seeks to answer the following question: does the framing of marijuana legalization in newspaper opinion content influence public opinion on a state-by-state basis? To be clear from the outset, I define framing as the way in which newspapers define and package the issue (Colvin, 2006) in order to position the issue so that it conveys certain meaning (Menashe and Siegel, 1998). Additionally, I define newspaper opinion content as inclusive of both editorial articles as well as op-ed columns. Before diving into analysis, it is necessary to set up a basic context for the study. The path towards decriminalization, medical marijuana legalization, and now recreational marijuana
1
legalization, has been a winding one, as reflected on the graph of support over time shown in Figure 1, which provides an overview of support for marijuana legalization from 1969 through 2014. As apparent in the graph, support has been steadily increasing since the 1990s. Before that, however, there exist a few interesting trends worth pointing out. For instance, in June 1971, President Nixon declared the national “War on Drugs.” However, between 1973 and 1977, eleven states decriminalized marijuana possession, during which time support levels spiked (“History of Marijuana”). Within just a few years, though, the tide shifted, leading up to the drug war hysteria of the late 1980s, directly following Nancy Reagan’s “Just Say No” campaign. Disapproval became so extreme that in 1989, 64 percent of Americans polled saw drug abuse as the nation's "number one problem,” up from 5 percent in 1985 (“History of Marijuana”). This is reflected in the lower support levels in the late 1980s. Since then, support for legalization has steadily increased, and my interest is in the last 4 years in particular, during which it has became a hotly debated topic in the media and politics. The growth in support is also apparent in statelevel public policy innovation, the details of which are shown in Figure 2. According to the Marijuana Policy Project (2014), as of December 2014, recreational marijuana is legal in 4 states: Alaska, Colorado, Washington, and Oregon (Washington D.C. voted to legalize but this decision needs to undergo Congressional review). Given its relevance in the news and public policy, understanding the way in which public opinion on marijuana legalization develops is important. Beyond its topical nature, evaluating the relationship between media framing and public opinion also proves important for theoretical and public policy reasons. Within economic theory, the topic is relevant because it explores the idea that demographic factors are not adequate for a full explanation of public opinion. Numerous studies (Wald et al. 1996, Haider-Markel and Meier 1996) note that demographics are important
2
explanations for public opinion. One study by Erikson, Wright and McIver (1993) argues that public opinion in states reflects demographic characteristics of residents, but also acknowledges that a mysterious factor labeled ‘state culture’ substantially influences public opinion. Thus, this study provides relevant theoretical information in that it evaluates the role the media plays in influencing public opinion while also taking demographics into account. In doing so, it also expands upon recent Economics literature that addresses how media bias can affect attitude and voting behaviors (Chiang and Knight 2011, DellaVigna and Kaplan 2007). The topic is also relevant in the realm of public policy, because if mass media shapes public policy preferences, it could influence public policy innovation regarding the legalization of marijuana. Numerous researchers, like Erikson, Wright and McIver (1993), have argued that public opinion in states affects state policies. This finding has implications for the political process, because it means that those who shape media messages could have influence over public opinion and even political outcomes. This study also advances over relevant existing literature. Though a large body of literature exists on mass media’s impact on public opinion, there is not yet literature specifically on the media’s coverage of marijuana legalization. That being said, there are 3 areas of literature that I looked into: literature that connects media to marijuana legalization, literature that influences my methodology, and literature on the topic in a broader sense. Within the context of marijuana, Golan (2010) looked into differences in framing strategies between newspapers and between op-eds versus editorials. For the development of my methodology, literature on gay marriage and framing proved most useful. For example, Chomsky and Barclay (2006) looked into how public opinion on gay marriage shifts due to media. They find that the media’s position, as indicated by editorials and op-ed columns, “has a significant effect on the level of public
3
support for gay rights, even after taking into account social demographic factors.” Colvin (2006), on the other hand, looks into how framing of gay rights issues in the media influences public policy innovation, concluding that media coverage does not have a significant impact. More generally, much literature exists on the broader topic of how media influences public opinion. Page, Shapiro, and Dempsey (1987) did a comprehensive analysis entitled “What Moves Public Opinion,” which looks into how news content moves with the public’s policy preferences. I also looked to McCombs and Shaw (1972), who evaluated the power of mass media in setting political agendas given that media chooses what to publish and how to frame it. With all of the aforementioned literature and context in mind, I set out to evaluate the relationship between newspaper’s opinion coverage on marijuana legalization and public opinion on the topic on a state-by-state basis over time. To do so, I collect data on demographics and public opinion for multiple years and construct a media framing measure using a manually created dataset. To measure media’s framing of the topic over time, I identify the largest newspapers in the states in my sample and find all relevant opinion content from 2010 to 2014, which I then manually code based on positive/negative framing. In doing this, I assume that the polling datasets provide representative samples and that the opinions content as a whole accurately represents the policy stance of newspapers. I then calculate changes in framing over time for each state. With all of my data, I regress public opinion on media framing, taking into account all relevant social demographic factors. This leads to a positive and statistically significant result. I also compare changes in support levels in states where newspapers become more supportive over time versus less supportive over time, finding insignificant differences. Lastly, I complete a trends analysis on public opinion, finding that the trends begin after framing shifts take place in newspapers, suggesting that the media is influencing public opinion, and not
4
the other way around. All in all, relationships exist between the newspaper’s coverage and public opinion, and my analysis suggests that the relationship is causal such that the newspaper’s opinion content moves public opinion on the topic of marijuana legalization. The paper is organized as follows. Section II explains my data, section III discusses my methodology, and section IV presents my results. Section V provides a discussion of these results, all of which are summarized in the conclusion, section VI.
II. Data This section presents the data on polling information and newspaper coverage. The tables included provide summary statistics from the data.
II.A. Polling Information To create a measure of public opinion, polling information on a state-by-state basis was needed. To collect this data, I utilized datasets made publicly available by the Pew Center for the People and the Press. Each year since 2010, Pew has conducted a Political Survey using a random sample of people contacted via landline and cell phone. The question of interest for this paper reads as follows: “Do you think the use of marijuana should be made legal, or not?” The wording of this question has not varied over time, making it a strong indicator for opinion on marijuana legalization. Unfortunately, Pew did not include this question in publicly available datasets for 2012. Furthermore, the full dataset has not yet been made available for the February 2014 Political Survey, which did include questions regarding marijuana legalization, since it is the largest survey Pew has conducted to date. Despite these data gaps, full datasets were available for 2010, 2011, and 2013, so this study focuses on those three years.
5
After downloading the datasets, I removed all unnecessary information and created a new spreadsheet of all the relevant information: state name, numerical state ID, race, income, age, education, urban density, sex, political party, and ideology. These demographic factors in particular are used because these same indicators have been used successfully in similar studies to account for state differences and political cultures (Erikson, Wright, and McIver 1993, Chomsky and Barclay 2006, Barclay and Fisher 2003). An advantage of the edited dataset is that all needed demographic information is available alongside the public opinion information, allowing for group-level analysis. In Tables 1 and 2, for instance, I provide summary statistics on support for marijuana legalization over time by state and by demographic group. When compiling data, I removed data for states that had less than 15 respondents for any given year because these sample sizes were too small for use in the analysis. The relatively small sample sizes for each state constitute one weakness of the polling data.
II.B. News Coverage The second phase of data collection required manual data creation, given that I compiled the dataset myself rather than using existing work. Inspired by the methodologies of previous studies (Chomsky and Barclay 2006, Colvin 2006), I used the most widely circulated newspaper as the primary news source for each state analyzed in this study. To find the needed circulation information, I used the Editor and Publisher’s International Yearbook, which provided state-level daily circulation numbers. Unfortunately, the Yearbook does not provide data on state-level distribution of national newspapers, like the New York Times and the Washington Post. For these, I made the assumption that these were the most widely circulated newspaper in their states of publication.
6
After identifying the newspapers of interest, I moved into data creation for the media framing measure. To construct this measure, I chose to analyze opinion content in newspapers in which enough relevant opinion content on marijuana legalization was published between 2010 and 2014. Opinion content includes both op-ed articles and editorials, in this case. I focused only on opinion content rather than all related newspaper content because it creates a more manageable data set, provides content that is easier to evaluate for framing, and because, as Golan (2010) notes, “highly subjective editorial and opinion pages provide readers with important benchmarks regarding salient issues against which readers can evaluate their own opinion.” Opinion content, as a whole, should provide relatively straightforward expressions of the newspapers’ policy preferences. To find all relevant articles, I used two electronic databases: LexisNexis and NewsBank’s Access World News. LexisNexis provides a comprehensive list of content from leading national newspapers included in this study, but covered few of the state-level newspapers. For the statelevel content, Access World News proved more useful. However, both had gaps in opinion content coverage of marijuana legalization, so I also used newspaper website archives to ensure my dataset was comprehensive. Because less content on the subject existed than originally anticipated, I focused on the top newspapers in 11 states: New York, New Jersey, Colorado, Oregon, Washington, Virginia, Maryland, Nevada, Texas, California, and Illinois. For states with two widely circulated newspapers, like Texas (Houston Chronicle and Dallas Morning News) and California (San Francisco Chronicle and the Los Angeles Times), both were used for the state’s media framing measure. To compute the media framing measure for each state for 2010, 2011, and 2013, I read each of the 160 opinion articles and manually coded them. If the article emphasized negative
7
ramifications of full legalization, highlighted the negative health consequences, or expressed concern about abuse by minors, I coded the article as a “0”. If the article pushed for full legalization and elucidated the positive effects of doing so, I coded the article as a “1”. Table A in the Appendix provides a snapshot of the dataset compiled, and Table 3 provides clippings of articles that I coded to illustrate my coding methodology. Also, it is important to note that articles that did not take any identifiable position were excluded from the sample. After doing this for every article in the selected newspapers, I calculated the average annual frame for each of the 11 states for 2010, 2011, 2012, 2013, and 2014. Figure 3 provides a visual for this information, and Table B in the Appendix shows the summary statistics of this data. Because public opinion information is not available for 2012 or 2014, I focus on the media framing measure for 2010, 2011, and 2013. As aforementioned, one weakness is that there are not a substantial number of opinion pieces on marijuana legalization in the chosen time frame. Gaps exist for the media framing measure in certain years as a result. Beyond that, very few newspapers have taken official editorial stances on the topic other than the New York Times, the New Jersey Star Ledger, and the Las Vegas Review Journal.
III. Methodology
My methodological approach to the question is twofold. First, I look to quantify the relationship between public opinion and frame using a regression while taking social demographic factors into account. Second, I compare the changes in public support over time between states with newspapers that become more supportive and states with newspapers that become less supportive over time. To do this, I rely on a t-test, Mann-Whitney test, and a followup regression.
8
III.A. Regression Analysis To model the source for shifts in public opinion, I use methods similar to those employed in relevant recent studies (Chomsky and Barclay 2006, Colvin 2006). I use a time-series regression model, shown below, and introduce a number of measures that shape public opinion in the 11 states in the sample. The unit of analysis is the individual in the regression given that the Pew datasets present public opinion on marijuana on the individual level. Therefore, I analyze how media framing impacts the way in which individuals in the 11 states think about marijuana legalization. Public Opinion = β0 + β1(Media Framing) + αX + ε All in all, I seek to regress public opinion on media framing while controlling for a variety of demographic factors in order to evaluate if and how the position of the print media has an effect on public support levels regarding recreational marijuana legalization. To identify the demographics necessary to include, I look to previous papers, and choose to focus on the following: race, population density of area, income level, education level, age, political party, and ideology (Wald et al., 1996). To include them in the regression, I create dummy variable indicators for all demographic factors of interest, allowing me to evaluate how differences in demographics influence the relationship between public opinion and media framing. Furthermore, as stated earlier, demographics prove highly important in determining policy preferences, so I control for them with the objective of isolating the influence of mass media on public opinion. After running this regression, I do it again and include state fixed effects to control for time-invariant differences between the states that the primary regression did not capture.
9
As is clear from the summary statistics provided, most states display increases in public support levels no matter what the demographic characteristics of the state are. Even more conservative states like Alabama and Kansas display upward trends in support, indicating that there exist trends towards increased support outside of politics and demographics. Because of this, I expect that even when a newspaper is unsupportive of marijuana legalization, the state’s population will still become increasingly supportive of legalization, but less so than in states where newspapers advocate strongly for legalization, like in Nevada. To control for these time trends, I include dummy variables for each year in the analysis. To further strengthen the regression, I separate the sample into groups that are more or less likely to read the newspaper, according to the Pew Research Center’s State of the Media (2012). The report concludes that people who are above 45 and have high levels of education and high income are most likely to read the newspaper. That said, the relationship between public opinion and framing should be most pronounced within these demographic groups, so I run regressions on these groups individually. Because there is no question on the Pew poll that I used in my dataset about readership of newspapers, I use these demographic factors as proxies.
III.B. Framing and Support Shifts Analysis The second piece of analysis splits newspapers into 3 categories: newspapers that become more supportive of marijuana legalization over time, newspapers that become less supportive, and newspapers that display no change in position. The average media framing measure for each newspaper for each year indicates these changes in support over time. For clarification, I define an “up frame” as the instance when a newspaper becomes more supportive of full marijuana legalization between 2010 and 2013. I define a “down frame” as the instance when a newspaper
10
becomes more opposed to legalization policies between 2010 and 2013. From my sample size of 11 newspapers, 2 exhibit down frames and 3 exhibit up frames. The remaining 6 newspapers have framing approaches that do not change significantly. With this division in place, I evaluate how changes in framing in newspaper opinion content impacts changes in public opinion. To do so, I run two separate t-tests. The first compares the change in public support levels from 2010 to 2013 between states where there are up frames and states where there are not up frames. The second compares the change in public support levels from 2010 to 2013 between states where there are down frames and states where there are not down frames. I then supplement the t-test analysis with a Mann-Whitney test analysis to measure sensitivity. The Mann-Whitney test, or Wilcoxon rank sum test, is a nonparametric test that compares two unpaired groups. To complete it, I rank all values in change in public opinion for the 11 states without paying attention to whether the value is for an up frame, down frame, or other frame state. I assign the smallest value for change in public opinion a rank of 1, and the largest a rank of 11. I then average the ranks by group in the same way that I separate the frame groups for the t-tests. In other words, I compare the average rank between up frames and other frames, and also compare the average rank between down frames and other frames. I add this additional test as a supplement to the t-tests because the MannWhitney test doesn’t require the underlying assumption of normality when comparing samples, and is thus a generalization of the t-test. Lastly, I aggregate the data to the state-year level and regress the change in opinion from 2010 to 2013 on indicators for up frame and down frame, as shown in the regression below. Change in Public Opinion = β0 + β1(Indicator for Up Frame) + β2(Indicator for Down Frame) + ε
11
To address direction of causality, I complete my analysis work by evaluating public opinion trends before the changes in framing and comparing them to public opinion trends during the framing shifts.
In completing these regressions and t-tests I make numerous necessary assumptions. The first few relate directly to my datasets in that I assume that both the polling datasets and the newspaper framing datasets are representative. In the Pew datasets, samples are relatively small, vary in size over the years included, and also change in terms of people chosen as respondents. Despite these potential challenges, it proves necessary to assume that the people surveyed in the 11 states analyzed provide an accurate representation of the populations in those states. A similar assumption is necessary for the collected opinion content, which I assume, on an aggregate level, is representative of the newspapers’ policy preferences. Op-ed content can diverge significantly from a newspaper’s official editorial stance, but for the purposes of this study, I assume that the opinion content as a whole represents the newspapers’ views. Because of how few newspapers have released official editorial stances on the topic of recreational marijuana legalization, including op-ed content is a necessity. Beyond these assumptions of representativeness, I also assume that people read the most widely circulated newspaper in their state of residency, and more specifically, that they read the opinion content. To strengthen this assumption, as aforementioned, I identify the demographic groups that are most likely to read the newspaper and separate them in part of the analysis. Finally, I assume that nothing, other than the variables I control for, is correlated with both newspaper framing and public attitudes towards marijuana. I assume that I have not omitted any variables, and that the following controls prove sufficient: time trends, political party,
12
ideology, income, education, race, and urban density. I make this assumption given my controls because these control variables have been successfully employed in similar studies and because it is necessary to assume the absence of confounding variables left unaccounted for in order to successfully estimate the relationship between media framing and public opinion. Because I recognize that there might be differences across states that I am not yet controlling for, I use state fixed effects as a part of my regression analysis. In doing so, I find that the regression loses its statistically significant result.
IV. Results This section contains my empirical results. I first discuss the results of the regression quantifying the relationship between public opinion and framing. Following that, I present the ttest and Mann-Whitney results that address how changes in framing impact public opinion over time. Finally, I address the challenge of causality. The first regression I run is of public opinion on opinion content frames over the time period. My hypothesis is that framing of opinion content in newspapers influences public opinion, as evident in β1, at a statistically significant level. Public Opinion = β0 + β1(Media Framing) + αX + ε I include race, sex, education, income, political party, ideology, and urban density as controls (αX) by creating dummy variables for all possible values of each of the factors. I also include the years as dummy controls in order to control for time trends. Results of this regression can be seen in Table 4. As is evident, a positive framing of marijuana legalization in newspaper opinion content is associated with an 8.24 percent increase in public support of legalization. The result is statistically significant at the 95 percent level with a p-value of .047 and a confidence interval ranging from .001 to .1636. The time trends and social demographic factors can also be
13
seen in Table 4. Some of the demographics are of more importance than others, such as sex, political party, race, and ideology. To briefly evaluate the relationship between frame and readers’ predispositions, I drop all demographic controls in the regression but leave in the year dummies. When I do so, the frame effect jumps from 8.24 percent to 12 percent and is statistically significant with a p-value of .001, as shown in Table 5. Given that this coefficient is higher when I drop the demographic controls and that I think the relationship between readers’ predispositions and the dependent variable is positive, the relationship between frame and readers’ predispositions is positive as well. Essentially, dropping the demographic controls allows me capture this positive relationship between frame and predispositions. Going back to the first regression with demographic controls, I expect the results to be stronger in demographic groups that are more likely to read the paper, including older people, more educated people, and wealthier people. These factors serve as proxies for whether the individuals polled read the paper. To test this hypothesis, I run the same regression on these specific demographic groups. All results are shown in Table 6. First, I run the regression on people with different income levels, once using a sample of people whose annual incomes are above $75,000 and again with people whose incomes fall below that threshold. The coefficient is 4 percent higher for people whose annual income is above $75,000; more specifically, a change to a positive frame is associated with a 10 percent increase in support for legalization among richer people. However, the results are not statistically significant. Then, I run the same regression for people who have at least an associate’s degree and compare the results to people who have lower levels of educational attainment. These two groups have almost identical coefficients (.083 and .085), and neither result is statistically significant. Lastly, I run the
14
regression for people who are fifty years old or above and compare the results to people younger than that. The coefficient for people aged 50 or above is only .6 percent higher, and again, the results are insignificant. In all of these cases, the 95 percent confidence intervals include zero. The statistical insignificance of these results could also be attributed to the lower sample sizes given that they are cut in half. Regardless, the fact that none of these lead to significant differences is not ideal, because I would expect that people who are more likely to read the newspaper are also more likely to be influenced by the framing of the opinion content. To complete the regression analysis, I return to the initial full regression and include state fixed effects to address the possibility of omitted variable bias. I recognize that there could be some time-invariant differences that have not already been controlled for between the 11 states that might be correlated with the independent and dependent variables. Adding state fixed effects allows me to control for these unobserved differences, and doing so leads to a negative 10 percent insignificant correlation between framing and public opinion. The full regression output including state fixed effects is shown in Table 7. When completing the regression, I omit the dummy indicator for California and find that only two states have statistically significant coefficients. Texas has a coefficient of -0.248 and Virginia has a coefficient of -0.1647, which makes sense given the relative conservativeness of these states. The 95 percent confidence interval of the regression of public opinion on frame goes from -.342 to .139, so I cannot reject the hypothesis that there is a positive effect of framing on public opinion. This range also includes the coefficient of 8.24 percent, as found in the first regression. However, the negative coefficient remains surprising, and suggests that if there is an effect, it is likely very small. The time-invariant differences between the states seem to account for more of the relationship. It is also important to note that adding more variables while keeping the same number of observations
15
lowers the statistical strength of the regression. With more observations, I would likely see more of a significant relationship between framing and public opinion.
All in all, it appears that there is a correlation between framing and public opinion, even after taking social demographic factors into account. This supports my hypothesis that print media framing influences public opinion, and the results are high in magnitude, because an 8 percent increase in public support for legalization is significant. However, it remains disconcerting that these effects are not higher for demographic groups more likely to read the opinion content. It is also predicted based on the results of adding state fixed effects that the relationship between framing and public opinion is very small.
The second analysis I conduct looks into another hypothesis: that shifts in opinion content framing should influence shifts in public opinion. To analyze this hypothesis, I now complete a full analysis using a t-test, Mann-Whitney test, and regression. Using the designated groups of “up frame”, “down frame”, and “other frame”, I compare movement in public opinion between 2010 and 2013. To do so, I run two separate t-tests. The first compares the change in public opinion between states with up frames and states with other frames. The second compares the change in public opinion between states with down frames and states with other frames. The results of these t-tests are displayed in Figure 4. Neither t-test provides statistically significant results, which can again partially be attributed to the sample size. However, the relationships do match my hypothesis. States with up frames experience an average 14 percent increase in public support, while states without up frames experience lower increases of 10 percent. States with down frames experience increases in support of 3 percent, while other states experience average
16
increases of 13 percent. As aforementioned, it is not surprising that down frame states also experience increases in support levels, because time trends suggest that support is increasing in all states regardless. Though insignificant at the 95 percent level, the magnitudes of difference are large, because a 14 percent increase in public support could influence passage of legislation on a state level. To supplement the t-test analysis, I also run a Mann-Whitney two-sample test. This tests the hypothesis that two independent samples are from populations with the same distribution. To put it another way, I answer the following question: if the groups are sampled from populations that have identical distributions, what is the chance that random sampling would lead to the mean ranks being as far apart as I observe in this study? As in the t-tests explained previously, I compare up frame states to other frame states, and also run a test comparing down frame states to other frame states. Thus, my null hypothesis is that the samples were drawn from populations with the same distribution. Results of my Mann-Whitney tests are shown in Table 8. In my comparison between up frames and other frames, I find insufficient evidence to reject the null hypothesis, given a p-value of 0.8383. In my comparison between down frames and other frames, I again find insufficient evidence to reject the null hypothesis due to a p-value of 0.1573. These results do not necessarily means that the two populations are the same in both tests; however, as with the t-tests, I simply lack compelling evidence to prove that they differ. Lastly, I incorporate these findings into a regression, in which I aggregate the data to the state-year level and then regress the change in opinion from 2010 to 2013 on indicators for up frame and down frame. I show the regression equation below, the results of which are evident in Table 9. Change in Public Opinion = β0 + β1(Indicator for Up Frame) + β2(Indicator for Down Frame) + ε
17
The regression outputs a β0 of .1247, β1 of .02, and β2 of -.0923, and although neither of the coefficients for the frame indicators prove statistically insignificant, the relationships match expectations. Relative to people in states with newspapers that don’t exhibit a shift in framing, residents of states with up frames, on average, are 2 percent more likely to support marijuana legalization. Furthermore, residents of states with down frames, on average, are 9 percent less likely to support legalization, relative to residents in states with other frames. None of the aforementioned analyses provide convincing evidence for direction of causality to suggest that newspaper content influences public opinion. To address this, I did an analysis on public opinion trends before the changes in framing and compared them to public opinion trends during the framing shifts as a preliminary defense of causation. Results of this analysis are shown in Figure 6. The figure shows that the significant trends begin after newspapers begin to write increasingly framed opinion content, suggesting that the media drives public opinion. For example, in states where the newspapers become increasingly supportive of marijuana legalization between 2011 and 2013 (up frames), public opinion, on average, actually fell 3 percent from 2010 to 2011 before jumping an average 17% between 2011 and 2013. All states show upward trends from 2011 to 2013, including states where newspapers exhibit down frames, so it remains difficult to separate the time trends from the impact of framing. On the whole, however, the distinguishable trends from 2011 to 2013, when the newspapers began to shift their frames, provides introductory evidence to support the hypothesis that newspapers influence public policy preferences, and not the other way around. I return to the initial question: does the framing of marijuana legalization in newspaper opinion content influence public opinion on a state-by-state basis? Unfortunately, the results of all of the aforementioned analyses provide no conclusive answer to this question. The
18
correlations in the regression suggest that a relationship exists, and the final trends analysis suggests that direction of causality is such that newspapers influence public opinion. The t-tests and Mann-Whitney tests provide support as well in that they illustrate that support moves with shifts in framing, but they lack statistical strength. All in all, I can conclude that framing of marijuana legalization in print media opinion content relates to public opinion on the topic, but with this dataset, I cannot decipher the specifics of that relationship.
V. Discussion Numerous limitations exist in this study. First and foremost, the sample size of 11 newspapers is too small to make decisive conclusions with. Although I analyze 160 articles, the average framing measure method I use shrinks this to 1 frame number per year per newspaper per state. These small sample sizes could not be resolved due to the general lack of opinion content on the topic of marijuana legalization, however. Figure 5 illustrates the issue of limited print media coverage on the topic, displaying the fact that terms like “recreational marijuana” and “marijuana legalization” only entered the newspaper lexicon in the previous 3 years. Related studies rely on at least 14 years of data (Chomsky and Barclay, 2006), and this study draws from only 3 years. Adding more years of polling data and framing information would significantly strengthen the analysis, but this is not yet possible. Furthermore, very few newspapers have published official editorials asserting their position on recreational marijuana policies, leaving me to rely more heavily on op-ed content, which has more variation in bias. The missing years in polling data and gaps in media framing data also become a primary limitation, as does the reality of the difficulties in ensuring a comprehensive opinion content dataset. It proved difficult to find all relevant articles in the
19
databases and newspaper archives, but I remain confident that for the newspapers in my sample set, the articles coded make up a sufficiently inclusive sample. In addition to the limitations related to sample size shortfalls, there exist a few issues in approach to the analysis. For instance, a bias could exist in the way in which I coded the opinion content as positive or negative. To check for this bias, I asked another student to read 10 random pieces of opinion content from the study and code them as positive or negatively framed. The student’s framing results matched mine for all of the articles, which validates my framing analysis and supports the robustness of the measure. A final limitation is the difficulty in fully controlling for social demographic factors in the regression. For example, I did not include religion in my analysis because of gaps in the data, but this could be an important driver of public opinion. To put it another way, it is difficult to isolate the influence of media framing on public opinion due to the plethora of other factors at play, as shown with the introduction of state fixed effects. The discussion of this difficulty leads right into the main challenge someone could raise to my conclusions: how can I assert direction of causality? The regression suggesting a positive and statistically significant relationship between framing and public opinion shows correlation rather than causation. A challenge is deciphering whether this relationship exists because newspapers influence public opinion, or because public opinion influences newspaper content. As Chomsky and Barclay (2006) posit, the media may alter their messages in response to existing public policy preferences on the local level. Newspapers are profit-driven commercial entities, which provides an incentive to please readers by aligning their content with existing opinions. This reflects the point of Hamilton (2004), who argues that the media takes stances in reflection of their audience’s viewpoints, as evident in the demographic characteristics of
20
audience members. In response to this challenge of causality, I admit it is difficult, using my data, to provide proof of causality. However, the analysis previously outlined on public opinion trends before and during newspaper framing shifts serves as a preliminary defense. Another, more minor, challenge that could be brought up is the presence of strong time trends in support levels. Regardless of demographic composition, all states appear to display higher support levels for legalization over time. To address this challenge, I controlled for time trends in the regression. With more time and resources, I would attempt to get the full datasets for Pew’s 2014 Political Survey. This would allow me to increase my sample size in public opinion as well as include more available opinion content. For statistical strength, the analysis requires more years of information, so I would also try to get more detailed 2009 data. This would also allow me to strengthen my argument of direction of causality. Additionally, I would increase my state sample size above 11 by searching for more news coverage on LexisNexis, Newsbank’s Access World News, and newspaper archives. Essentially, additional time and resources would allow for increased sample sizes, though doing so could require waiting for continued increases in relevant articles on marijuana legalization.
VI. Conclusion This study provides numerous findings regarding the relationship between print media and public opinion. First, in the specific case of marijuana legalization, it presents the finding that a relationship exists between the way in which newspapers frame the topical issue and the public’s support for recreational marijuana becoming fully legalized. It also addresses the commonly addressed question of whether the media drives public opinion, or whether the media adapts to the preferences of their audiences for profit-based reasons. Preliminary evidence in this
21
paper suggests the media is leading public opinion. Lastly, the findings suggest that frame shifts over time move with public opinion on a state-by-state basis. It is also interesting to note the variety of perspectives that newspapers take regarding marijuana legalization. In similar studies, like Chomsky and Barclay (2006), the newspaper coverage was relatively homogeneous on the topic of gay marriage, a detail that the authors note as a possible explanation for why the newspaper had such a significant influence. Within marijuana legalization, far less content has been published in state-level newspapers, and the opinion coverage to date varies quite a bit within each newspaper. As the topic continues to come to the forefront of social and political discussions, I posit that articles will converge towards more positive framing. Proponents of legalization have achieved significant victories in legislation to date, and with growing support by the public, I anticipate more support for these proponents moving into the 2016 election year and beyond. Results of this paper provide evidence for a relationship between print media and public support for legalization, and also suggest that the direction of influence is such that the public alters its opinion based on the media. Thus, this remains an important topic to study moving forward as more content and relevant polls emerge, because if public opinion shifts as a result of the position of print media sources, then the media messages may hold political power.
22
References Barclay, Scott and Shauna Fisher. 2003. The states and the differing impetuses for divergent paths on same-sex marriage, 1990—2001. Policy Studies Journal 31(3): 331-352. Chomsky, Daniel and Scott Barclay. 2006. Shaping the same sex marriage discussion: Mass media, government action, social demographic factors, and public opinion in the 50 States. Paper presented at the annual meeting of the American Political Science Association, Marriott, Loews Philadelphia, and the Pennsylvania Convention Center, Philadelphia, PA. Chiang, Chun-Fang and Brian Knight. 2011. Media bias and influence: Evidence from newspaper endorsements. Review of Economic Studies 78(3):795-820. Colvin, Roddrick. 2006. Understanding policy adoption and gay rights: The role of the media and other factors. The Innovation Journal 11(2): 1-16. DellaVigna, Stefano and Ethan Kaplan. 2007. The Fox News effect: Media bias and voting. The Quarterly Journal of Economics 122(3): 1187-1234. Erikson, Robert S., Gerald C. Wright, and John P. McIver. 1993, Statehouse democracy: Public opinion and policy in the American states. New York: Cambridge University Press. Golan, Guy J. 2010. Editorials, op-ed columns frame medical marijuana debate. Newspaper Research Journal 31(3): 50-60. Haider-Markel, Donald P. and Kenneth J. Meier. 1996. The politics of gay and lesbian rights: Expanding the scope of the conflict. Journal of Politics 58: 332-349. Hamilton, James T. 2004. All the news that’s fit to sell: How the market transforms information into news. Princeton: Princeton University Press. History of marijuana in the United States. (2014, June 1). Retrieved December 10, 2014, from http://www.pbs.org/wgbh/pages/frontline/shows/dope/etc/cron.html Marijuana state policies. (2014, November 10). Retrieved December 10, 2014, from http://www.mpp.org/ McCombs, Maxwell E., and Donald L. Shaw. 1972. The agenda-setting function of the press. Public Opinion Quarterly, 36(2): 176-187. Menashe, Claudia L. and Michael Siegel. 1998. The power of a frame: An analysis of newspaper coverage of tobacco issues--United States, 1985-1996. Journal of Health Communication 3: 307 325.
23
Page, Benjamin I., Robert Y. Shapiro, and Glenn R. Dempsey. 1987. What moves public opinion?. American Political Science Review 83(1): 23-43. Pew state of the news media 2012. (2012, February 11). Retrieved December 10th, 2014, from http://www.stateofthemedia.org/2012/newspapers-building-digital-revenues-provespainfully-slow/ Wald, Kenneth, James W. Button and Barbara A. Rienzo. 1996. The politics of gay rights in American communities: Explaining antidiscrimination ordinances and policies," American Journal of Political Science 40(1): 1152-1178.
24
Table 1: Support for Marijuana Legalization by Demographic Groups Year 2010
2011
2013
2014
By age: 18-29 30-49 50-64 65+
57.3% 43.7% 44.4% 24.8%
54.0% 48.0% 47.1% 33.9%
66.2% 56.4% 53.6% 37.0%
70.0% 56.0% 55.0% 32.0%
Sex Male Female
47.0% 36.5%
48.3% 42.4%
46.8% 57.6%
. .
Race White Black Hispanic
41.4% 38.7% 37.0%
44.7% 49.0% 41.6%
51.0% 55.0% 56.5%
55.0% 60.0% 43.0%
Urban Density Rural Suburban Urban
36.9% 37.2% 44.4%
38.9% 37.9% 51.1%
48.1% 45.8% 53.3%
. . .
Political Party: Democrat Republican Independent
48.9% 22.3% 49.1%
55.0% 28.3% 50.0%
59.8% 37.1% 57.8%
63.0% 39.0% 58.0%
Income |t|
[95% Conf. Interval]
0.041 0.041 0.038 0.043 0.047 0.052 0.045 0.043
1.99 0.39 2.07 -0.58 -1.2 -0.34 -0.78 3.08
0.047 0.7 0.039 0.564 0.232 0.735 0.435 0.002
0.001 -0.065 0.004 -0.110 -0.147 -0.120 -0.124 0.048
0.164 0.097 0.153 0.060 0.036 0.085 0.053 0.216
0.039 0.051 0.071 0.065 0.092 0.331 0.083 0.029 0.132 0.121 0.126 0.121 0.122 0.127
3.12 -1.9 -2.4 0.15 -0.67 -1.74 0.05 3.76 -0.11 1.52 1.88 1.7 1.36 1.34
0.002 0.058 0.017 0.878 0.501 0.083 0.96 0 0.915 0.129 0.06 0.09 0.174 0.179
0.046 -0.196 -0.311 -0.117 -0.241 -1.223 -0.159 0.052 -0.273 -0.053 -0.010 -0.032 -0.073 -0.078
0.200 0.003 -0.031 0.137 0.118 0.075 0.168 0.167 0.245 0.420 0.485 0.444 0.406 0.419
0.136 0.048 0.049 0.048 0.047
0.65 -0.37 -0.39 0.38 1.22
0.515 0.708 0.694 0.704 0.224
-0.178 -0.112 -0.115 -0.075 -0.035
0.356 0.076 0.076 0.111 0.148
28
ideology _cons
(2) conservative 3 4 (5) liberal
-0.014 0.226 0.291 0.456 -0.005
0.057 0.058 0.065 0.078 0.134
-0.25 3.9 4.49 5.85 -0.04
0.804 0 0 0 0.969
-0.125 0.112 0.164 0.303 -0.267
0.097 0.339 0.418 0.609 0.257
Notes: This table provides the detailed results of the regression of public opinion on media framing. It includes dummy variable indicators for 7 demographic factors (income, party, race, sex, education, density, and ideology) as well as for time trends. The coefficient between public opinion and framing comes out to 8.24 percent, and is statistically significant.
29
Table 5: Regression without Demographic Controls
Notes: I run the same regression as shown in Table 4 but remove all demographic controls. As is evident, the coefficient jumps to 12.06 percent from 8.24 percent. This analysis allows me to pinpoint the correlation between frame and readers’ predispositions. Given that this coefficient is higher without the demographic controls and that I think the relationship between readers’ predispositions and the dependent variable is positive, the relationship between frame and readers’ predispositions is positive as well.
30
Table 6: Regression Results for more Likely Newspaper Readers Coefficient
P-value
95% Confidence Interval
Education High (at least Associates) Low
0.0832452 0.085513
0.115 0.204
-0.02042 -0.046553
0.18691 0.2175811
Income High (75,000+) Low
0.105789 0.0648352
0.145 0.212
-0.03678 -0.03697
0.2483601 0.16664
0.09156 0.084824
0.121 0.149
-0.02425 -0.030445
0.20738 0.2000947
Age 50 or above Below 50
Notes: This table repeats the regression from Table 4, but separates the sample into different demographic groups based on education, income, and age. These factors serve as proxies for reading the newspaper. I expect that the coefficients would be higher for groups that are more likely to read the newspapers. However, the results do not show significant differences.
31
Table 7: Regression with State Fixed Effects
Public Opinion
Coefficient
Std Error
t
P>|t|
[95% Confidence Interval]
Frame
-0.1007645
0.1226957
-0.82
0.412
-0.3415277
0.1399987
Density 2 (sparsest) 3 4 5 (densest)
-0.0153675 -0.0158228 0.0229966 0.0601378
0.0482359 0.0489777 0.0493515 0.0482842
-0.32 -0.32 0.47 1.25
0.75 0.747 0.641 0.213
-0.1100197 -0.1119306 -0.0738448 -0.0346093
0.0792848 0.0802851 0.119838 0.1548849
Education 2 3 4 5 6 7 8 (most education)
0.0013948 0.1906241 0.2843841 0.1982282 0.1714614 0.1415526 0.1424413
0.1326167 0.1208788 0.1251385 0.1216458 0.1224056 0.1268823 0.1347122
0.01 1.58 2.27 1.63 1.4 1.12 1.06
0.992 0.115 0.023 0.104 0.162 0.265 0.291
-0.2588361 -0.0465738 0.0388275 -0.0404747 -0.0687326 -0.1074258 -0.1219015
0.2616257 0.4278221 0.5299407 0.4369311 0.4116554 0.390531 0.4067841
Income 2 3 4 5 (richest)
-0.0166416 -0.0453843 -0.0014751 -0.0270958
0.0437528 0.0468112 0.0524752 0.0455932
-0.38 -0.97 -0.03 -0.59
0.704 0.333 0.978 0.552
-0.1024967 -0.137241 -0.1044461 -0.1165625
0.0692135 0.0464724 0.1014958 0.0623709
Party (2) Dem (3) Ind
0.1247417 0.1186708
0.0429599 0.0395167
2.9 3
0.004 0.003
0.0404424 0.041128
0.2090409 0.1962136
Race 2 (black) 3 (asian) 4 5 6 7 (hispanic)
-0.0655839 -0.170823 -0.0262438 -0.0726592 -0.5777333 0.0423235
0.0513466 0.0716978 0.0647347 0.0897484 0.3309759 0.0827481
-1.28 -2.38 -0.41 -0.81 -1.75 0.51
0.202 0.017 0.685 0.418 0.081 0.609
-0.1663402 -0.311514 -0.1532714 -0.2487707 -1.2272 -0.1200513
0.0351723 -0.030132 0.1007838 0.1034523 0.0717337 0.2046982
32
Ideology 2 (conservative) 3 4 5 (liberal)
-0.0097939 0.2266995 0.2891781 0.4669464
0.0570368 0.0583729 0.0649914 0.0785504
-0.17 3.88 4.45 5.94
0.864 0 0 0
-0.121716 0.1121555 0.1616469 0.3128087
0.1021282 0.3412435 0.4167094 0.6210842
Sex (1) Male
0.1089243
0.0293508
3.71
0
0.0513298
0.1665188
State ID 8 - CO 17 - IL 24 - MD 32 - NV 34 - NJ 36 - NY 41 OR 48 - TX 51 - VA 53 - WA
0.0511468 0.0086456 -0.0487377 -0.2220461 0.0162138 -0.0069191 -0.0387779 -0.1957071 -0.1353096 0.0189383
0.0897186 0.0644903 0.0811003 0.1731962 0.0935968 0.063897 0.0964513 0.0940124 0.0638964 0.0600709
0.57 0.13 -0.6 -1.28 0.17 -0.11 -0.4 -2.08 -2.12 0.32
0.569 0.893 0.548 0.2 0.863 0.914 0.688 0.038 0.034 0.753
-0.124906 -0.1179024 -0.2078792 -0.5619053 -0.1674493 -0.1323028 -0.2280423 -0.3801856 -0.2606922 -0.0989376
0.2271996 0.1351935 0.1104037 0.1178131 0.1998769 0.1184645 0.1504865 -0.0112286 -0.009927 0.1368143
_cons
0.1726547
0.1552699
1.11
0.266
-0.1320282
0.4773376
Notes: To check for omitted variables, I run the same regression as shown in Table 4, but also include dummy indicators for State ID, thus adding state fixed effects. With state fixed effects added in, the coefficient between public opinion and framing becomes negative (-10 percent) and insignificant. Two states added have statistically significant coefficients: Texas and Virginia. Overall, the negative relationship is surprising, but the 95 percent confidence interval is very large and crosses zero. Therefore, I cannot reject my hypothesis that there is a positive effect, but think that if there is an effect, it is very small.
33
Table 8: Mann-Whitney Two-Sample Test Results
Notes: This table provides the results to the Mann-Whitney two-sample test conducted as the non-parametric supplement the t-test previously done. It is used to answer the following question: if the groups are sampled from populations that have identical distributions, what is the chance that random sampling would lead to the mean ranks being as far apart as I observe in this study? As in the t-tests, I compare up frame states to other frame states, and also run a test comparing down frame states to other frame states. The results above provide insufficient evidence to reject the null hypothesis that the distributions are the same because both p-values are statistically insignificant.
34
Table 9: Regression for Change in Opinion on Framing Shifts
Notes: This table builds on the t-test comparing change in public opinion from 2010 to 2013 with the type of newspaper frame shift. I aggregate the data to the state-year level and regress the change in opinion from 2010 to 2013 on indicators for up frame and down frame. Results are not statistically significant but match expectations, as an “up frame” is associated with a jump in public support while a “down frame” is associated with a drop in support.
35
Figure 1: Support for Marijuana Legalization over time
Should Marijuana be Legal or Illegal in the United States?
Percent of those Surveyed
60 50 40 30 20 10 0
Year
Notes: Figure shows support levels for marijuana legalization from 1969 to 2014 in the United States. The data points are from the Pew Research Center for the People and the Press.
36
Figure 2: Current State Policies on Marijuana
Note: The Marijuana Policy Project created this map of current state policies. It illustrates the variety of policies currently in place across the country and the progress that has been made in drug policy recently (given the full legalization measures in Alaska, Oregon, Washington, Colorado, and Washington D.C.).
37
Figure 3: Average Frames Over Time in States
Notes: The visual shows the average newspaper frame for each of the 11 states over the 3 years analyzed in order to express the shifts in frames over the years. Average newspaper frame per year was calculated by averaging the frames of all articles published by the paper in the specific year. Average frames closer to 1 (highlighted in dark blue) means the largest newspaper in the state expressed strong favorability towards marijuana legalization, and average frames closer to 0 (highlighted in light blue) means the largest newspaper in the state expresses strong opposition. Full summary of average frames available in Appendix Table B.
38
Figure 4: Changes in Public Support by Framing Shifts
Notes: T-test results comparing the average change in support for marijuana legalization in states with up frames (newspapers that become more supportive over time) against states with other frames. Figure on right compares the average change in support in states with down frames (newspapers that become less supportive over time) against states with other frames. Results match expectations, but neither is statistically significant (as evident in the high p-values).
39
Figure 5: Lack of Newspaper Content on Marijuana Legalization
Notes: Figure created used the Chronicle Tool from the New York Times. Figure shows the percent of total articles that the terms “recreational marijuana”, “marijuana legalization”, “medical marijuana”, and “marijuana” are included in over time. It illustrates the limitation that marijuana legalization has only become a more spoken about topic in newspapers in recent years.
40
Figure 6: Trends in Public Opinion Before and After Framing Shifts Change in Support by Framing Change 0.2 0.15 0.1 2010-2011 2011-2013
0.05 0 Up Frame
Other Frame
Down Frame
-0.05
Notes: This figure provides preliminary evidence suggesting direction of causality. It is similar to Figure 4, and shows average public opinion changes over two time periods, 2010-2011, and 2011-2013, when framing shifts took place. The figure groups states by the type of framing shift that the newspaper in the state exhibits.
41
Appendix Table A: Framing Analysis Results Year
Paper
Frame
2011
Las Vegas Review Journal
0
2014
Las Vegas Review Journal
1
2014
Las Vegas Review Journal
1
2014
Las Vegas Review Journal
1
2009
Washington Post
1
2010
Washington Post
1
2010
Washington Post
0
2013
Washington Post
1
2013
Washington Post
0
2014
Washington Post
1
2014
Washington Post
0
2014
Washington Post
0
2014
Washington Post
0
2009
Chicago Sun Times
1
2010
Chicago Sun Times
1
2010
Chicago Sun Times
1
2012
Chicago Sun Times
1
2012
Chicago Sun Times
1
2012
Chicago Sun Times
1
2012
Chicago Sun Times
1
2013
Chicago Tribune
1
2013
Chicago Tribune
1
2013
Chicago Tribune
0
2013
Chicago Sun Times
1
2013
Chicago Sun Times
0
2014
Chicago Tribune
0
2010
Denver Post
0
2012
Denver Post
1
2012
Denver Post
0
2013
Denver Post
1
2013
Denver Post
0
2014
Denver Post
1
2014
Denver Post
1
2010
Los Angeles Times
1
2010
San Francisco Chronicle
0
2010
San Francisco Chronicle
0
2010
San Francisco Chronicle
0
2010
San Francisco Chronicle
1
42
2010
San Francisco Chronicle
1
2011
Los Angeles Times
0
2011
Los Angeles Times
1
2011
Los Angeles Times
1
2011
San Francisco Chronicle
0
2011
San Francisco Chronicle
1
2011
San Francisco Chronicle
1
2011
San Francisco Chronicle
1
2012
San Francisco Chronicle
1
2012
San Francisco Chronicle
1
2012
San Francisco Chronicle
1
2012
San Francisco Chronicle
1
2012
San Francisco Chronicle
1
2013
Los Angeles Times
0
2013
Los Angeles Times
1
2013
Los Angeles Times
1
2013
San Francisco Chronicle
1
2013
San Francisco Chronicle
1
2013
San Francisco Chronicle
1
2014
Los Angeles Times
0
2014
Los Angeles Times
0
2014
Los Angeles Times
1
2014
Los Angeles Times
1
2014
Los Angeles Times
0
2014
Los Angeles Times
1
2014
Los Angeles Times
1
2014
Los Angeles Times
0
2014
San Francisco Chronicle
1
2014
San Francisco Chronicle
1
2014
San Francisco Chronicle
1
2014
San Francisco Chronicle
1
2014
San Francisco Chronicle
1
2014
San Francisco Chronicle
0
2014
San Francisco Chronicle
0
2014
San Francisco Chronicle
1
2014
San Francisco Chronicle
1
2014
San Francisco Chronicle
1
2014
San Francisco Chronicle
1
Notes: This table provides a snapshot into the framing analysis I conducted. After reading each article I coded it as positive “1”, or negative “0”. I then noted the article title.
43
Table B: Average Frames in States Over Time Year 2010 2011 2012 2013 2010 2012 2013 2014 2009 2010 2012 2013 2014 2009 2010 2013 2014 2011 2013 2014 2011 2013 2014 2010 2012 2013 2014 2010 2012 2013 2014 2010 2011 2012 2013 2014 2009 2010 2013 2014 2011 2012
State California California California California Colorado Colorado Colorado Colorado Illinois Illinois Illinois Illinois Illinois Maryland Maryland Maryland Maryland Nevada Nevada Nevada New Jersey New Jersey New Jersey New York New York New York New York Oregon Oregon Oregon Oregon Texas Texas Texas Texas Texas Virginia Virginia Virginia Virginia Washington Washington
44
Average Frame 0.7 0.70833333 1 0.83333333 0 0.5 0.5 1 1 1 1 0.70833333 0 1 0.5 0.5 0.25 0 1 1 1 1 1 1 0.8 1 1 0.33333333 0.6 1 1 0 0 0.5 0 0.44444444 1 0.5 0.5 0.25 0.92857143 0.75
2013 Washington 2014 Washington
0.66666667 1
Notes: This table shows the average media frame measure for each of the 11 states in my sample. To compile this, I coded 160 opinion pieces as positive (1) or negative (0) and calculated the average for each year for each newspaper.
45