Misinformation or Expressive Responding? What an inauguration crowd can tell us about the source of political misinformation in surveys Brian F. Schaffner (Corresponding Author) University of Massachusetts Amherst 200 Hicks Way Amherst, MA 01003 (413) 545-0416 schaffne@polsci.umass.edu Samantha Luks YouGov 805 Veterans Blvd. Suite 202 Redwood City, CA 94063 (650) 462-8009 sam.luks@yougov.com FORTHCOMING AT PUBLIC OPINION QUARTERLY Running header: Misinformation or Expressive Responding Word count: 2,745 1
Brian F. Schaffner is professor in the Department of Political Science at the University of Massachusetts Amherst, Amherst, MA. Samantha Luks is Managing Director, Scientific Research at YouGov, Redwood City, CA. The authors thank John Bullock, Scott Blinder, Tatishe Nteta, and Meredith Rolfe for feedback on an earlier version of this paper. The authors are aware of no conflicts of interest involved in the publication of this article.* Address correspondence to Brian Schaffner, University of Massachusetts Amherst, Department of Political Science, 200 Hicks Way, Amherst, MA 01003, USA; email: schaffne@polsci.umass.edu. 2
The public s party-driven misinformation and misperceptions about politics has drawn a great deal of attention from scholars over the past decade. While much of this research assumes that the misinformation documented by survey researchers is an accurate reflection of what individuals truly believe, other scholars have suggested that individuals intentionally and knowingly provide misinformation to survey researchers as a way of showing support for their political side. To date, it has been difficult to adjudicate between these two contrasting explanations for misperceptions. However, in this note, we provide such a test. We take advantage of a controversy regarding the relative sizes of crowds at the presidential inaugurations of Donald Trump in 2017 and Barack Obama in 2009 to ask a question where the answer is so clear and obvious to the respondents that nobody providing an honest response should answer incorrectly. Yet, at the same time, the question taps into a salient political controversy that provides incentives for Trump supporters to engage in expressive responding. We find that there is clear evidence of expressive responding and that this behavior is especially prevalent among partisans with higher levels of political interest. Our findings provide support for the notion that at least some of the misinformation reported in surveys is the result of partisan cheerleading rather than genuinely held misperceptions. 3
As high levels of political misinformation have attracted increasing attention from scholars and the public alike, we still lack a clear understanding of why so many Americans appear to be misinformed about politically-relevant facts. The most prominent explanation for misinformation is derived from the theory of directionally motivated reasoning the tendency of individuals to engage new information with a motivation to reach a particular conclusion (Kunda 1990). That is, partisans tend to seek out information that reinforces their political beliefs, and reject or counter-argue information that challenges those beliefs (Taber and Lodge 2006). This theory assumes that the misinformation documented by survey researchers is an accurate reflection of what individuals truly believe. For example, when a large proportion of Republicans say that they think Obama is a Muslim, they truly believe that to be the case. A recent challenge to the belief that misperceptions are truly held by individuals comes from Bullock et al. (2015) and Prior et al. (2015), who argue that at least some of the misperceptions that respondents provide in response to survey questions are not genuinely held beliefs. This body of work proposes an alternative explanation, called expressive responding, whereby individuals intentionally provide misinformation to survey researchers as a way of showing support for their political viewpoint. By this reasoning, at least some of the Republicans who say that Obama is a Muslim may not actually believe that to be true, but they are using the question about his citizenship status to express their disapproval of him or his presidency. To date, it has been difficult to adjudicate between these two contrasting explanations for misperceptions. Bullock et al. (2015) and Prior et al. (2015) conducted experiments where some subjects were offered material rewards for either answering questions correctly or admitting that they don t know the answer. The logic behind this approach is to increase the importance of accuracy motivations for respondents. In both studies, the partisan polarization in misperceptions 4
declined under the incentive conditions, suggesting that some of the misinformation reported in surveys may indeed arise from expressive responding. However, Flynn et al. (2017) cite two reasons for why this evidence is not entirely conclusive. First, the incentive conditions do not consistently generate more accurate responses to factual questions, a pattern we might expect to see under conditions of expressive responding. In other words, while polarization in knowledge decreases, the material incentives do not always reduce the overall levels of misinformation. A second related critique is that the patterns from those experiments may simply arise from the fact that respondents tend to assemble survey responses off the top of their heads from a biased set of considerations (Zaller 1992). Therefore, the incentive condition simply changes what types of considerations respondents sample; it does not prove that respondents are intentionally answering incorrectly. Thus, as Flynn et al. (2017) note in a recent review, Assessing the extent to which reported misperceptions are sincere (reflecting confidence in an incorrect answer) versus expressive (intentionally and knowingly reporting an incorrect answer to engage in partisan cheerleading) remains an important topic for future research (2017, p.139). The difficulty thus far has been in determining whether respondents intentionally and knowingly provide incorrect answers to questions in order to engage in partisan cheerleading. What would be useful then is an extreme case one that would provide a more definitive test of whether there is at least some expressive responding to factual questions on surveys. In this research note, we provide such a test. As we discuss below, we take advantage of a controversy regarding the relative sizes of crowds at the presidential inaugurations of Donald Trump in 2017 and Barack Obama in 2009 to ask a question where the answer is so clear and obvious to most respondents that almost nobody providing an honest response should answer 5
incorrectly. At the same time, the question taps into a salient political controversy that provides incentives for people especially supporters of Trump to engage in expressive responding. Thus, by design, our test allows us to essentially rule out the possibility that incorrect responses are the result of truly held beliefs or a biased sampling of information. Taking advantage of this design, we find that there is clear evidence of expressive responding and that this behavior is especially prevalent among partisans with higher levels of political engagement precisely those respondents who have both the contextual knowledge and motivation to engage in the behavior. Design On the day following Donald Trump s inauguration a debate erupted over the factually clear point that fewer people attended Trump s inauguration than had come to either of Barack Obama s inaugurations in 2009 and 2013. Even the White House Press Secretary addressed the news media to promote the notion that more people attended Trump s inaugural, stating that it was the largest audience ever to witness an inauguration period both in person and around the globe. What made the controversy particularly noteworthy was the existence of aerial photographs (taken from the Washington Monument) showing that many more people were on the mall in 2009 at Obama s inaugural than in 2017 at Trump s (see Figure 1). Despite this photographic evidence, as well as other empirical indicators of crowd size such as subway ridership, the Trump administration persisted in their insistence that more people attended his inauguration. [INSERT FIGURE 1 ABOUT HERE] 6
This remarkable episode provided us with a unique opportunity to gauge the extent to which people were willing to provide incorrect answers to survey questions and, especially, a way of determining whether some people engage in expressive responding. On the two days immediately following the controversy (January 22nd - 23rd), we fielded a brief survey of a nationally representative sample of 1,388 U.S. adults, administered online by YouGov. Respondents were randomized into two conditions. In both conditions, we showed respondents the pair of images comparing the Obama inauguration crowd with that for Trump s (see Figure 1). However, we did not identify these images for respondents; we merely labeled them as image A and image B. To respondents who were randomly assigned to the first condition, we asked a simple question: Which photo has more people? Respondents in our second condition received the same set of photographs, but were asked to identify which photo was from Trump s inauguration and which was from Obama s. The first condition essentially levels the playing field for all respondents; to anyone with reasonable vision, the correct answer to this question is obvious. Thus, there are two possible reasons that a respondent would choose image A as having more people: (1) they failed to read the question or accidentally clicked on the wrong box (measurement error), or (2) their incorrect answer was intentional. If we find differential levels of selecting image A based on one s political allegiances, then the latter explanation expressive responding is almost certainly the culprit. After all, there is no reason to expect that Trump supporters are more prone to misreading or accidental clicking in surveys (see the Appendix for confirmation of this). Likewise, incorrect responses to this question could not result from a biased sampling of considerations by respondents since the evidence is clear and accessible to nearly everyone (Zaller 1992). 7
In analyzing responses to this question, we focus on two moderators the respondent s support for Trump and how politically engaged the respondent is. To measure support for Trump, we use two measures the respondent s 2016 vote choice and the extent to which the respondent approves or disapproves of Trump. We add the latter measure because it provides a more nuanced measure of support for Trump since it allows us to distinguish between those who strongly approve and those who only somewhat approve of him. For political engagement, we use the respondent s level of education. While education has been used extensively as a proxy for political engagement in a wide array of studies (Zaller 1992), the Appendix re-creates our main results using an alternative measure the respondent s self-reported interest in politics. The measures of presidential vote choice and educational attainment come from questions that are part of YouGov s profile battery, and thus for nearly all of the respondents the answers to these questions were collected before those respondents encountered our survey. Presidential approval was measured during the administration of the survey. In our weighted sample, 38% of respondents voted for Trump, 41% said they had voted for Hillary Clinton, and 21% were non-voters (we excluded third party voters from our analysis). 28% strongly approved of Trump, 11% somewhat approved, 10% somewhat disapproved, and 35% strongly disapproved (with 15% not sure). Finally, in terms of education, 17% of respondents in our sample reported having at least a college degree, with the remaining 83% reporting lower levels of education. For respondents in our second condition, the question does not have a clear accessible answer. Accordingly, this question cannot provide us with the same strong test of expressive responding, but it does provide us with a baseline of misperceptions as an additional reference point for the results for the first condition. 8
Results Figure 2 shows the proportion of respondents who selected the incorrect photo (image A) as the one that had more people. Note that the rate of selecting that photograph among nonvoters (3%) and Clinton voters (2%) is negligible essentially in line with what we would expect simply from measurement error. However, a much higher percentage of Trump voters 15% selected image A as showing more people. A similar pattern is clear for Trump approval. The difference in selecting the incorrect photo based on one s vote choice or approval of Trump is statistically significant (p<.01) and substantively meaningful, especially considering the clarity of the question and the accessibility of the evidence. This could be considered a lower bound estimate for the amount of expressive responding in surveys since many Trump voters were likely unaware of the controversy when we conducted our poll. [INSERT FIGURE 2 ABOUT HERE] Figure 3 presents the proportion of respondents giving the incorrect answer to the crowd size question based on our two measures of Trump support. Starting with panel A, for respondents who were non-voters in 2016 or reported voting for Clinton, there were only modest (and statistically indistinguishable) differences based on levels of education. This is what we would expect for these groups as there is no incentive for either engaged or unengaged respondents who are not Trump supporters to claim that the photo associated with his inauguration has more people in it. However, among Trump supporters, we find a significant divide between those with college degrees and everyone else. While only 11% of less-educated Trump supporters chose the wrong image as having more people, 26% of Trump supporters with college degrees selected the incorrect photo (difference of proportions p =.054). 9
[INSERT FIGURE 3 ABOUT HERE] That politically engaged Trump supporters were more than twice as likely to select the wrong photo makes sense from the perspective of expressive responding. After all, Trump supporters would only be motivated to choose the wrong photograph as a way of engaging in partisan cheerleading if they understood the context in which the question was being asked. Since Trump supporters with more engagement in politics were more likely to have understood the controversy about the inauguration crowds, they were more likely to choose the wrong photo as a way of supporting Trump in the debate. Panel B in Figure 3 shows similar patterns, this time by approval of Trump. This plot provides even stronger evidence that expressive responding is mostly likely to occur among those who are highly motivated and highly aware. Specifically, only a small percent of those who somewhat approve of Trump chose the wrong photograph, and the incidence of choosing the wrong photo is not related to education among this group. Only those who strongly approve of Trump show noteworthy levels of expressive responding. This pattern also occurs at a higher rate among the more educated strong approvers than those without college degrees (p =.025). The other half of respondents received the same set of photographs, but were asked which photo was from Trump s inauguration and which was from Obama s. Figure 4 shows the proportion of respondents giving the wrong answer to this question based on their educational attainment and our two measures of Trump support. Not surprisingly, we see more incorrect responses across the board when it comes to this question; however, we also see similar partisan differences as we found for the crowd size question. About 40% of Trump voters mis-identify which picture went with which president s inauguration, compared to about 20% of non-voters and less than 10% of Clinton voters. Additionally, we find a striking reversal for the moderating 10
role of educational attainment. In this condition, where respondents were asked to match the photographs to the correct president s inauguration, contextual knowledge of the crowd controversy is no longer needed to know which answer would favor Trump. Accordingly, the effects conditioned by education are flipped. Specifically, low-education Trump voters were about twice as likely to select the wrong image as college-educated Trump voters were. [INSERT FIGURE 4 ABOUT HERE] Another way of looking at these findings is to note that high education Trump voters were about equally likely to select the wrong photograph in both conditions (26% in Figure 3, panel b and 23% in Figure 4, panel b), despite the first question being much easier than the other. At the same time, nearly half of less educated Trump voters provided the wrong answer to the question about which photograph went with which president s inauguration while only one-inten provided the wrong answer to the easy question about which photograph had more people in it. In other words, college-educated Trump supporters provide incorrect information at roughly the same rate across the two conditions, but less-educated Trump supporters are highly affected by whether the answer to the question was clear or not. Panel B in Figure 4 shows these patterns based on approval of Trump. A similar pattern emerges college educated strong approvers are about equally likely to pick the wrong photo in this condition as they were in the previous one, but those with less education now do so at a much higher rate. Furthermore, this pattern of picking the wrong photograph extends to less educated respondents who somewhat approve or are not sure. This suggests that for less politically engaged Trump supporters, misinformation may stem from the tendency to guess the option that is more favorable to Trump when they are not sure of the answer. 11
Conclusion Our study provides strong evidence that the most politically engaged Trump supporters provided expressive responses to a straightforward question related to a controversy over inauguration crowd sizes. In fact, the best way to think about this result is that it helps to establish that much of the misinformation documented among the most engaged Trump supporters appears to be the result of expressive responding. Why? Because high education Trump supporters were just as likely to provide the wrong answer in the condition where the answer was in plain view as they were when the answer was not as obvious. If some other process accounted for misinformation among politically engaged Trump supporters, then we should have seen higher levels of misinformation in the second condition when the answer was less clear. In such a scenario, there is more of an opportunity for motivated reasoning to come into play; yet, there was no statistically significant increase in misinformation. This is an important result for contextualizing the scholarship on misinformation. Scholars have consistently shown that misperceptions are most common among the most politically engaged; yet, our results suggest that much of those misperceptions are not genuine. It is worth noting that the nature of our study only allows us to establish the existence of expressive responding on one side of the political spectrum. Future research may attempt to design similar experiments to test whether Democrats also engage in this behavior. Nevertheless, this research provides an important contribution by showing that expressive responding can account for at least some of the substantial levels of political misinformation commonly reported in surveys, especially when that misinformation comes from the most politically engaged partisans. The consequences of this finding are important for considering the implications of 12
partisan misperceptions. As Bullock et al. note, If Democrats and Republicans perceive different realities, then the incentives for incumbent politicians to pursue policies that generate objectively good policies may be reduced (2015). These concerns may be alleviated, to some extent, if at least some of the misperceptions reported in surveys are the result of partisan cheerleading rather than truly held misinformation. 13
Appendix Using Self-Reported Political Interest Rather than Education In the results presented in the paper, we use educational attainment as a proxy for political engagement (Zaller 1992). Here, we demonstrate that we find similar patterns of results when we use self-reported political attention instead. In our sample, 51% of respondents reported that they pay attention to politics most of the time, with the remaining 49% selecting either some of the time, only now and then, or hardly at all. We compare the former high interest group to the latter low interest respondents. Figures A1 and A2 reproduce the results from Figures 3 and 4 in the main paper. The patterns in these figures largely match what is shown in the paper. To the extent that there are differences, it is that high and low interest respondents who support Trump are about equally likely to provide the wrong response to the question about which photograph goes with which president s inauguration whereas low education respondents were more likely to provide the wrong response compared to those with college degrees. This difference likely occurs because people over-report their political interest in the self-report question and thus that group includes more people who are less engaged with politics than the group of college educated respondents does. Nonetheless, the key patterns presented in the paper (particularly those from Figure 3 in the paper and Figure A1 here) are quite consistent regardless of which measure of political engagement we use. Ruling Out Alternative Explanations 14
In this section, we address two potential alternative explanations for the results we find in Figure 2. The first possibility we address is the notion that the images may be more difficult for respondents to view on mobile devices. If that is true and if Trump supporters were more likely to take the survey on a mobile device than Clinton supporters, then that could account for the higher rate of choosing the wrong photograph for Trump versus Clinton voters. YouGov collects meta-data on the device used to take each survey. Approximately one-fourth of respondents took the survey on a mobile device. However, there were only small differences in mobile survey administration by vote choice and, in fact, Clinton voters were actually more likely to take the survey on a mobile device than Trump voters (26% and 21%, respectively). Thus, the usage of mobile devices cannot account for the results we present in Figure 2. A second possibility that could account for the results in Figure 2 is that Trump supporters are more likely to make response entry errors than Clinton supporters. If it was true that Trump supporters exhibit higher rates of measurement error in general, then this pattern could account for the fact that they were much more likely to choose the wrong image. There is no theoretical reason to think this is true, but to be sure we examined how respondents to the survey answered a question that is part of YouGov s demographic profile battery. Questions in the profile battery are periodically re-asked of respondents to ensure the validity of those measures. For this exercise, we used the question about a respondent s educational attainment since the question has six categories, making it more likely that an individual might accidentally select the wrong value. However, we did collapse the categories for some college and 2-year degree since moving between these categories is valid without making an error. Notably, 21% of Trump voters changed their response to the education question at some point during their time on the YouGov panel compared to 24% of Clinton voters did so. Of course, some of this 15
movement is likely to be actual change, as individuals achieve higher levels of education during their time on the panel (finishing college or post-graduate degrees). But a negative change in educational attainment cannot be explained by anything other than response entry error. Here, we find that 16% of both Clinton and Trump voters moved from more to less education at some point during their time on the panel. Thus, there is no evidence that Trump voters are more likely to commit response entry errors. Information on Survey The survey was conducted by YouGov. The survey was collected online on January 22-23, 2017. YouGov sampled 2,798 individuals from their panel and 1,417 completed the survey for a response rate (RR1) of 51.7%. YouGov panelists were invited to the survey with a generic invitation to prevent potential respondents from self-selecting into a topic of interest. The respondents who took this survey had participated in mostly non-political surveys during the preceding months. Specifically, only 22% of the surveys taken by our sample during the previous two months had any political content. The population under study was American adults. At the recruitment stage, respondents were invited based on their fit to interlocking demographic targets of gender x race x age x education, plus a marginal target for Census region. All sampling targets were loosely applied. That is, if a respondent started the survey after her target cell had been filled, she was permitted to complete the survey. The final sample of respondents to the survey was then weighted to a more complete population frame, selecting the closest matches to the population. The frame was constructed by stratified sampling from the full 2013 American Community Survey (ACS) sample with 16
selection within strata by weighted sampling with replacements. The sample was weighted based on age, education, gender, party identification, ideology, political interest and race. Wording of Questions Used in Analysis Figure 1 in the main text shows the images and wording for the question capturing the main dependent variable for a randomly selected half of the sample. The other half of the sample received the same preamble text and images, but the questio text instead read: From what event was each photo taken? <1> Photo A was from Donald Trump s 2017 Inauguration. Photo B was from Barack Obama s 2009 Inauguration. <2> Photo A was from Barack Obama s 2009 Inauguration. Photo B was from Donald Trump s 2017 Inauguration. Who did you vote for in the election for President? {response options 1 and 2 randomized} <1> Hillary Clinton <2> Donald Trump <3> Gary Johnson <4> Jill Stein <5> Evan McMullin <6> Other {please specify} <7>Did not vote for President Political interest: 17
Some people seem to follow what s going on in government and public affairs most of the time, whether there s an election going on or not. Others aren t that interested. Would you say you follow what s going on in government and public affairs...? Most of the time Some of the time Only now and then Hardly at all Don t know Wording of the variables used for weighting: Education: What is the highest level of education you have completed? Did not graduate from high school High school graduate Some college, but no degree (yet) 2-year college degree 4-year college degree Postgraduate degree (MA, MBA, MD, JD, PhD, etc.) Age: In what year were you born? [TEXT ENTRY] 18
Gender: Are you male or female? Male Female Race: What racial or ethnic group best describes you? White Black or African American Hispanic or Latino Asian or Asian American Native American Mixed Other Middle Eastern Party Identification: Generally speaking, do you think of yourself as a...? Democrat Republican Independent Other 19
Not sure Strength of Democratic ID: Would you call yourself a strong Democrat or a not very strong Democrat? Strong Democrat Not very strong Democrat Strength of Republican ID: Would you call yourself a strong Republican or a not very strong Republican? Strong Republican Not very strong Republican Independent Leaners: Do you think of yourself as closer to the Democratic or the Republican Party? Lean Democrat Independent Lean Republican Not sure Ideology: Thinking about politics these days, how would you describe your own political viewpoint? Very liberal Liberal 20
Moderate Conservative Very Conservative Not sure 21
References Bullock, John G., Alan S. Gerber, Seth J. Hill, Gregory A. Huber. 2015. Partisan bias in factual beliefs about politics. Quarterly Journal of Political Science 10: 519 578. Flynn, D.J., Brendan Nyhan and Jason Reifler. 2017. The Nature and Origins of Misperceptions: Understanding False and Unsupported Beliefs About Politics. Advances in Political Psychology, 38: 127-150. Kunda, Ziva. 1990. The case for motivated reasoning. Psychological bulletin. 108: 480. Prior, Markus, Gaurav Sood and Kabir Khanna. 2015. You cannot be serious: The impact of accuracy incentives on partisan bias in reports of economic perceptions. Quarterly Journal of Political Science. 10: 489-518. Taber, Charles S. Milton Lodge. 2006. Motivated skepticism in the evaluation of political beliefs. American Journal of Political Science. 50: 755-769. Zaller, John. 1992. The nature and origins of mass opinion. Cambridge University Press. 22
Figure 1: Screen shot of question shown to respondents in first condition Note: A randomly assigned one-half of the sample received this question. 23
Figure 2: Percent choosing wrong photograph as having more people by 2016 vote choice and Trump approval 20 20 Percent choosing wrong photo 15 10 5 3% 2% 15% Percent choosing wrong photo 15 10 5 4% 2% 4% 5% 14% 0 0 Non voters Clinton Voters 2016 Presidential Vote Trump Voters Strongly Somewhat disapprove disapprove Not sure Trump Approval Somewhat approve Strongly approve Note: Figure shows percent choosing image A in response to the question shown in Figure 1 with sampling weights employed. N = 158 non-voters, 275 Clinton voters, and 218 Trump voters. 24
Figure 3: Percent choosing wrong photograph as having more people by 2016 vote choice, approval of Trump, and education 30 26% 30 28% Percent choosing wrong photo 20 10 0 3% Non voters 3% 2% 1% Clinton Voters 2016 Presidential Vote 11% Trump Voters Low education High education Note: Figure shows percent choosing image A in response to the question shown in Figure 1 with sampling weights employed. High education respondents are those with at least a college degree and low education respondents are those with some college or less. Percent choosing wrong photo 20 10 0 6% 1% Strongly disapprove 2% 0% Somewhat disapprove 4% 5% Not sure Trump approval 5% 6% Somewhat approve 10% Strongly approve Low education High education 25
Figure 4: Percent making an incorrect match of the inauguration photograph to the president by 2016 vote choice, Trump approval, and education Percent choosing wrong inauguration 50 40 30 20 10 0 23% Non voters 9% 10% 3% Clinton Voters 2016 Presidential Vote 46% Trump Voters 23% Low education High education Note: Figure shows percent choosing image A as belonging to Obama s inauguration and image B as belonging to Trump s with sampling weights employed. High education respondents are those with at least a college degree and low education respondents are those with some college or less. Percent choosing wrong inauguration 50 40 30 20 10 0 8% 2% Strongly disapprove 10% 3% Somewhat disapprove 38% 14% Not sure Trump approval 32% 33% 6% Somewhat approve 46% Strongly approve Low education High education 26
Figure A1: Percent choosing wrong photograph as having more people by 2016 vote choice, Trump approval, and political interest 30 30 Percent choosing wrong photo 20 10 0 3% Non voters 2% 3% 1% Clinton Voters 2016 Presidential Vote 8% Trump Voters 18% Low interest High interest Note: Figure shows percent choosing image A in response to the question shown in Figure 1 with sampling weights employed. High interest are respondents who say they pay attention to politics most of the time, and low interest respondents are those who say they pay attention some of the time, only now and then, or hardly at all. Percent choosing wrong photo 20 10 0 5% 4% Strongly disapprove 6% 4% 4% 3% 3% 0% Somewhat disapprove Not sure Trump approval Somewhat approve 7% 18% Strongly approve Low interest High interest 27
Figure A2: Percent making an incorrect match of the inauguration photograph to the president by 2016 vote choice, Trump approval, and political interest 50 50 Percent choosing wrong inauguration 40 30 20 10 20% 27% 9% 7% 43% 39% Low interest High interest Percent choosing wrong inauguration 40 30 20 10 9% 3% 4% 18% 32% 41% 24% 23% 44% 43% Low interest High interest 0 Non voters Clinton Voters 2016 Presidential Vote Trump Voters Note: Figure shows percent choosing image A as belonging to Obama s inauguration and image B as belonging to Trump s with sampling weights employed. High interest are respondents who say they pay attention to politics most of the time, and low interest respondents are those who say they pay attention some of the time, only now and then, or hardly at all. 0 Strongly disapprove Somewhat disapprove Not sure Trump approval Somewhat approve Strongly approve 28