PARTISAN MEDIA, ELECTORAL PREDICTIONS, AND WISHFUL THINKING

Transcription

1 Public Opinion Quarterly, Vol. 82, Special Issue 2018, pp PARTISAN MEDIA, ELECTORAL PREDICTIONS, AND WISHFUL THINKING KATHLEEN SEARLES* GLEN SMITH MINGXIAO SUI Abstract Public predictions about the results of forthcoming elections have important consequences for political campaigns and support for governmental institutions. This research examines whether partisan media facilitate wishful thinking, which occurs when candidate preference biases electoral predictions. We use two experiments to test whether partisan news exposure increases wishful thinking about election outcomes. Our results suggest that when partisan media cover the in-party as winning the horse race, likeminded viewers are more likely to predict the in-party will win. On the other hand, when partisan media cover the in-party as losing the horse race, like-minded viewers adjust their predictions downward. The effects of unfavorable horse-race coverage on electoral expectations are magnified when viewers watch the partisan news outlet they prefer. Meanwhile, watching counter-attitudinal media appears to affect electoral predictions only when partisans do not also watch like-minded news. Overall, we show that where people get their news, and how partisan media cover the horse race, has important effects on electoral expectations. Kathleen Searles is an assistant professor of political communication in the Department of Political Science and Manship School of Mass Communication at Louisiana State University, Baton Rouge, LA, USA. Glen Smith is an associate professor of political science at the University of North Georgia, Oakwood, GA, USA. Mingxiao Sui is an assistant professor of media and communication in the School of Arts and Humanities at Ferrum College, Ferrum, VA, USA. The authors would like to thank the Pew Research Center for sharing their data, and Christopher Mann, without whom they would not have access to CCES data. They would also like to thank Martin Johnson, Johanna Dunaway, Anthony Fowler, Yphtach Lelkes, participants in the LSU Political Science Research Brownbag, and participants in the Manship School of Mass Communication Lunch and Learn for their helpful feedback. Research assistance was provided by Jonathan Nickens, Brian Watson, Charlotte Bellotte, Simeon Burns, and Cana Kim. This work was supported by the National Science Foundation [ to Stephen Ansolabehere, Samantha Luks, and Brian Schaffner]; the Belo Corporation Professorship, and the Darlene & Thomas O. Ryder Professorship (held by Searles), both in the Manship School of Mass Communication and Louisiana State University. *Address correspondence to Kathleen Searles, Department of Political Science and Manship School of Mass Communication, Louisiana State University, 240 Stubbs Avenue, Baton Rouge, LA USA; doi: /poq/nfy006 Advance Access publication March 29, 2018 The Author(s) Published by Oxford University Press on behalf of the American Association for Public Opinion Research. All rights reserved. For permissions, please

2 Partisan Media Effects on Electoral Predictions 303 When making predictions about which candidate is likely to win an election, voters may behave strategically (Downs 1957; Cox 1997), forming accurate electoral expectations by drawing on media coverage and polls. Alternatively, they may ignore such information and instead form electoral expectations based on their preferences. How people form their electoral predictions is an important question because they can affect a person s decision to vote, donate money, or volunteer in a campaign. Inaccurate predictions can cause people to spend their time and money on a candidate that has little chance of winning the election. In a world without wishful thinking, partisans could devote more resources to candidates that stood a better chance of success (Fisher 2004). For example, Mutz (1995) found that poll coverage affected campaign donations, such that those leading in the polls received more donations. In this way, donors electoral expectations shaped their decision to allocate money to candidates. Perhaps most importantly, false electoral predictions can undermine the perceived legitimacy of institutions and officials by affecting whether electoral losers perceive their defeat as legitimate. Though an ignominious honor, a long tradition of scholarship elevates the role of the loser in a democracy: acceptance of a losing outcome is essential to legitimate governance (Easton 1965; Riker 1982; Anderson et al. 2005). If people truly believe that their preferred candidate will win the election, they may be more likely to attribute losses to voter fraud. In a political system of checks and balances, the losing side can create legislative gridlock if they perceive the opposition came to power illegitimately. How do people form their electoral predictions? Predictions about future electoral outcomes are often biased by cognitive processes. For example, when people lack information about how the electorate is likely to vote, they tend to assume that others will support their preferred candidate also called the false consensus effect (Ross, Greene, and House 1977). Collectively, these cognitive biases result in wishful thinking, a cognitive and motivational bias in which an individual s desire for a preferred outcome increases their judgment of the likelihood of that outcome occurring. Although wishful thinking is well documented in psychology and political science (Abramson et al. 1992; Meffert and Gschwend 2011), the role that partisan media play in facilitating this process has garnered little scholarly attention. Partisan media could potentially enable wishful thinking by exposing people to a biased selection of opinion polls that confirm viewers desired outcome. Research examining television news found that all outlets prefer to cover polls that suggest a close race; however, the authors also found that both MSNBC and Fox News s poll coverage demonstrated an aggregate preference for the in-party candidate (Searles, Ginn, and Nickens 2016). Thus, partisan media may make it less likely that partisans will be exposed to polls suggesting unfavorable outcomes. For example, a recent study by Hollander (2014) found that exposure to Fox News was significantly correlated with belief in a

3 304 Searles, Smith, and Sui Republican victory during the 2012 presidential election. Fox News viewers were more likely to be both surprised by electoral losses and hold negative perceptions of democratic institutions following the election. Although previous results are suggestive, they are based on cross-sectional surveys that do not test for causation. Are Fox News s viewers simply more likely to engage in wishful thinking, or does exposure to partisan media enhance the effect of candidate preference on electoral predictions? To better isolate the effects of partisan media on electoral predictions, we conducted two survey experiments during the months prior to the 2014 midterm elections. In study 1, participants were shown segments from Fox News or MSNBC that suggested the in-party would gain majority status during the 2014 US Senate elections. The purpose of the first experiment is to examine whether favorable partisan media horse-race coverage facilitates wishful thinking. The second experiment uses a more representative sample to pose a conservative test of wishful thinking by examining whether wishful thinking persists in the face of unfavorable partisan media horse race coverage. Additionally, study 2 builds on previous work that emphasizes viewer choice (Arceneaux and Johnson 2013) by allowing participants to choose which news outlet they watch. Partisan Media and Wishful Thinking The predictions people make are often influenced by cognitive biases, resulting in wishful thinking. Evidence for wishful thinking spans decision-making contexts (Babad 1997), countries (Babad 1995), and disciplines including political science and social psychology (Hayes 1936; Cantril 1938; Olsen 1997; Krizan and Windschitl 2007), but is particularly pronounced in the case of elections (Granberg and Brent 1983; Abramson et al. 1992; Krizan, Miller, and Johar 2010; Meffert and Gschwend 2011; Meffert et al. 2011). In a political context, wishful thinking is conceptualized as a positive relationship between candidate preference and electoral expectations. Wishful thinking is not limited to inaccurate predictions: even people who predict an electoral outcome accurately may have reached their conclusion through biased processes. For example, a Trump voter may have accurately predicted the 2016 election outcome, but that does not mean her prediction was untethered to preference. What causes wishful thinking during election campaigns? One possible explanation is the false consensus effect, a type of attribution bias in which people see their personal choice as dominant and alternative choices as uncommon (Ross, Greene, and House 1977). In political contexts, this effect manifests in voters assumptions that the majority will vote for the same candidate (Mullen et al. 1985; Krueger and Clement 1994). Projection effect is another possibility, in which people project their preferred electoral outcome on their election predictions (Lazarsfeld, Berelson, and Gaudet 1944; McAllister and Studlar 1991).

4 Partisan Media Effects on Electoral Predictions 305 Both effects are reinforced by peer groups (Fischer and Budescu 1995; Mutz 1998) and information deficiency: in the absence of information, people make assumptions regarding majority opinions. However, information deficiency is not a necessary condition for wishful thinking. Indeed, some types of information such as horse-race coverage on partisan media outlets may facilitate wishful thinking by providing viewers with the necessary ingredients to maintain belief that their party is winning. It is well documented that people go to great lengths to buttress their preferred conclusions (Kunda 1990; Taber and Lodge 2006). To this end, people seek out news sources they expect will confirm their desired opinion (Stroud 2011). As these same people believe like-minded sources to be more credible than counter-attitudinal sources, the result is a feedback loop in which congenial information presented by a preferred source is perceived as persuasive, anchoring existing beliefs (Levendusky 2013). Exposure to horse-race coverage, which has the potential to inform electoral predictions (Iyengar, Norpoth, and Hahn 2004; Blais, Gidengil, and Nevitte 2006; Dunaway 2008), may provide the impetus voters need to adjust electoral expectations, assuming the information is accurate. But if the sampling of opinions by media favors the in-party, rather than correcting electoral expectations, partisan media might facilitate wishful thinking by exposing viewers to congenial election coverage. Indeed, other studies find differences in programming content for MSNBC and Fox News (Feldman et al. 2011; Smith and Searles 2014), including poll coverage specifically (Searles, Ginn, and Nickens 2016). Before we generate expectations for how partisan media coverage might affect wishful thinking, we test our assumption that partisan news coverage of the horse race significantly differs from mainstream news. We look to content data from the 2012 presidential election collected for a special report titled Winning the Campaign 2012 by the Project for Excellence in Journalism (PEJ). The PEJ coded campaign related stories for tone toward the candidates from August 27 through October 21, 2012 (see Online Appendix 1 for more details). The PEJ defines poll coverage as stories that report polls, while horserace coverage discusses electoral standings broadly. The results support our assumption that partisan news coverage of the horse race significantly differs from mainstream news, as horse-race and poll coverage on partisan media were more positive toward the in-party candidate and more negative toward the out-party candidate. Table 1 shows the total number of minutes devoted to the horse race and polls on Fox News, MSNBC, and broadcast news by candidate and tone. Numbers in the top part of table 1 show the coverage of Obama, while the bottom portion shows coverage of Romney. On Fox News, horse-race coverage of Obama was three times more negative (170 minutes) than positive (51 minutes), which was in similar proportion to poll-specific coverage. In stark contrast to Obama s coverage, Romney received twice as much positive (128 minutes) as negative (59 minutes)

5 306 Searles, Smith, and Sui Table 1. Tone of horse-race and poll coverage of the 2012 election Horse-race coverage Poll coverage Broadcast Fox News MSNBC Broadcast Fox News MSNBC Obama Positive 65 (30%) 51 (15%) 237 (51%) 10 (43%) 11 (15%) 34 (83%) Negative 58 (27%) 170 (48%) 91 (20%) 1 (4%) 39 (51%) 2 (5%) Neutral 94 (43%) 131 (37%) 132 (29%) 12 (52%) 26 (34%) 5 (12%) Total Romney Positive 27 (13%) 128 (39%) 17 (3%) 2 (8%) 42 (60%) 2 (4%) Negative 104 (50%) 59 (18%) 522 (82%) 19 (80%) 6 (9%) 42 (86%) Neutral 76 (37%) 144 (44%) 95 (15%) 3 (12%) 22 (31%) 5 (10%) Total Note. Cells contain the total number of minutes devoted to stories about the candidates by tone and source, with column proportions in parentheses. Horse-race coverage includes poll specific coverage and other discussion of candidate strategy and momentum. horse-race coverage from Fox News, and polling coverage was even more lopsided. For MSNBC, roughly half of Obama s horse-race coverage was positive, compared to roughly 20 percent negative horse-race coverage; again, poll coverage was even more lopsided. Meanwhile, both horse-race and poll coverage of Romney was overwhelmingly negative on MSNBC. Broadcast news was more balanced than partisan media in their horse-race and poll coverage, but coverage was still more negative toward Romney relative to Obama. Altogether, the content analysis suggests there are empirical reasons to believe that partisan news coverage of the election is distinct. Previous work finds that such slanted coverage has an effect on vote choice and attitudes toward the candidate (Levendusky 2013; Smith and Searles 2014). Given that partisan media are more likely to present coverage that favors the in-party candidate s electoral standing, our first hypothesis simply posits that partisan media exert influence on expectations for all viewers. This influence is in part driven by design, as partisan media influence is more likely when exposure is forced, or when viewers are treated with partisan media they may not watch otherwise (Arceneaux and Johnson 2013). However, previous work does not examine electoral expectations specifically. Thus, building on this research, we test whether exposure to partisan news horse-race coverage affects all viewers perception of the in-party s chances. Hypothesis 1: Exposure to horse-race coverage on partisan news will affect electoral predictions.

6 Partisan Media Effects on Electoral Predictions 307 However, the relationship between partisan news and wishful thinking for like-minded viewers is more interesting. There are reasons to believe partisan media play a role in reinforcing like-minded viewers electoral expectations, given shared in-party affinities. Viewers tend to trust like-minded media more than neutral or opposing media, which is a requisite for persuasion (Miller and Krosnick 2000; Levendusky 2013). Given that partisan media commonly present favorable coverage of the in-party (as our analyses above suggest), partisan media should have an easier time convincing like-minded viewers that their preferred candidate will be victorious. In particular, research suggests that people tend to uncritically accept information that supports their preferred opinions (Kunda 1990; Ditto and Lopez 1992). In this way, partisan horse-race coverage that depicts in-party standings as favorable will facilitate wishful thinking among like-minded viewers. But what if coverage does not support their preferred electoral outcome? In other words, what happens when viewers hear from a trusted partisan news source: your party is losing? Indeed, wishful thinking describes the potential for motivated reasoning to shape likelihood judgments, whether faced with favorable or unfavorable information (Windschitl et al. 2013). In the case of partisan news, we have already seen that horse-race coverage tends to favor the in-party. And while it is easy to see why favorable horse-race coverage might promote wishful thinking, it is not immediately obvious how unfavorable coverage might affect wishful thinking. There are limits to people s willingness and ability to resist opposing information (Kunda 1990); for example, Redlawsk and Lau (2006) show that repeated exposure to counter-attitudinal information can change opinions. Moreover, like-minded viewers perceive partisan news as credible (Levendusky 2013), and are more likely to accept information from a source they trust (Miller and Krosnick 2000), perhaps even if that information is contrary to their belief. Still, while people expect congenial information from partisan media and these sources typically deliver it is unlikely that a single unfavorable story will be enough to unmoor electoral preferences. Exposure to one bad story on a like-minded outlet is likely not sufficient to motivate viewers with a preference for the in-party to change their mind. Attitude change as a result of exposure to unfavorable partisan news coverage is particularly unlikely given single-shot exposure during a low information election. Moreover, viewers likely perceive coverage from a like-minded source, whether unfavorable or favorable, through a different lens than coverage from a counter-attitudinal source, such that coverage on a congenial source will be perceived as more favorable relative to coverage on a counter-attitudinal outlet, regardless of the content (Dilliplane, Goldman, and Mutz 2013). Most compelling is evidence that shows wishful thinking is pervasive and difficult to overcome with countervailing information. In fact, previous studies found that even when people were provided with political knowledge or poll information, they still failed to overcome wishful thinking bias

7 308 Searles, Smith, and Sui (Babad, Hills, and O Driscoll 1992; Krizan, Miller, and Johar 2010; but see Dolan and Holbrook 2001). Such wishful thinking failures reflect patterns well known in motivated reasoning literature: people are motivated to accept information that suggests their preferred candidate will win, and discount information that suggests they might lose (Irwin 1953; Ditto and Lopez 1992). When people encounter unfavorable information, they may act as motivated skeptics, counter-arguing and becoming more entrenched in their existing attitudes (Taber and Lodge 2006). This process is analogous to rooting for a sports team, in that fans continue to believe their team will win despite overwhelming evidence to the contrary. Thus, given the strength of wishful thinking, and people s motivation to maintain their existing electoral expectations, we put wishful thinking to a strict test. Simply put, regardless of whether horse-race coverage is favorable or unfavorable, we argue that watching horse-race coverage on partisan news outlets will facilitate wishful thinking among like-minded viewers. Hypothesis 2: Exposure to horse-race coverage on partisan news will make like-minded viewers more likely to predict their party will win the election. In contrast to like-minded viewers, partisan media will have a difficult time convincing opposing partisans that their preferred outcome is unlikely. Although partisans often access both liberal and conservative media (Garrett 2009; Holbert, Hmielowski, and Weeks 2012), most do not find opposing sources to be credible (Stroud 2011; Smith and Searles 2014); and even if people watch opposing sources, they are unlikely to believe what they hear (Turner 2007). Previous research suggests that partisan media are largely unpersuasive when viewers believe they are biased (Levendusky 2013), and this is true whether information confirms viewers beliefs or not. Consequently, exposure to opposing media is unlikely to affect viewers electoral expectations. In a situation where partisans watch both like-minded and opposing media, they are likely to accept messages regarding the status of the race from like-minded media and ignore messages from opposing media, regardless of whether these messages are favorable or unfavorable. Given these expectations, we propose the following hypotheses regarding exposure to ideologically diverse partisan media. Hypothesis 3: Exposure to horse-race coverage on partisan news will not significantly affect electoral predictions among out-party viewers. For a realistic test of these hypotheses, what is needed is a low-information electoral context with actual stakes, to ensure that exposure to horse-race information is meaningful. We conducted two survey experiments testing the

8 Partisan Media Effects on Electoral Predictions 309 effects of conservative and liberal media on electoral expectations during the 2014 midterm elections. Midterms garner less attention than presidential elections, and therefore provide a low-information environment in which to test for partisan news effects. Of course, the downside of a low-information context is that it is not generalizable to high-information environments like presidential elections. However, non-presidential elections are far more common and thus, we believe, of interest. Study 1: Experimental Evidence for Wishful Thinking To test our hypotheses, we utilize an experiment to establish the effects of partisan news exposure on electoral expectations given candidate preference (wishful thinking). A survey was administered via Qualtrics from November 1 to November 6, We designed a 2 (Fox, MSNBC) x 2 (no alternative, alternative) in which participants are randomly assigned to one of four conditions: 1) watch clip from Fox News (only, with no alternative), 2) watch clip from MSNBC (only, with no alternative),3) watch clip from Fox News followed by MSNBC, and 4) watch clip from MSNBC followed by Fox News. We use between-group comparisons to examine whether partisan media affect electoral predictions. All participants are exposed to some content predicting the election outcome to ensure a baseline level of knowledge, while participants in the other two treatment groups are also exposed to clips from the partisan news alternative. This allows us to examine the effect of adding likeminded and opposing media to participants information environment. For example, if Group 1 watches a clip from Fox News (only), while Group 3 watches the same Fox News clip followed by a MSNBC clip, the posttest difference between the two groups measures the effect of adding MSNBC coverage. Although this approach does not allow us to measure the individual effects of each source, it does provide a measure of how much partisan media affect electoral predictions. Given the random assignment of experimental groups, any difference between the electoral predictions of people in the Fox News group and MSNBC group can be attributed to partisan media exposure. For study 1, we are not interested in comparing the effects of Fox News to MSNBC. Instead, our focus is on whether partisan media, in general, influence electoral expectations. This approach also improves external validity by better replicating a real-world environment where partisan media viewers frequently watch both like-minded and opposing media (Garrett 2009; Garrett, Carnahan, and Lynch 2013). However, to ensure the robustness of our findings, we replicate these effects with a more conservative test in study 2.

9 310 Searles, Smith, and Sui STIMULUS Participants watched a five-minute television clip, which discussed in-party chances of winning the majority in the Senate following the 2014 election. Akin to the PEJ s operationalization of electoral coverage, this clip is best characterized as horse-race coverage, in which electoral standings are discussed in terms of strategies and tactics. We chose to focus on the midterm election broadly rather than candidates specifically to make the stimulus applicable to the broadest number of respondents. Also, this was a timely topic of discussion on the partisan networks given pundit discussion of a GOP wave election. The stimuli were selected to ensure that they are as alike as possible in terms of format and structure. Stimuli were also similar in content: the clips from each source featured pundits discussing the likelihood that the in-party would win the Senate majority, without reference to polls. We include additional details on our systematic collection and selection of clips using SnapStream for stimulus creation in Online Appendix 2. While there are some downsides to using actual television clips as stimuli, we adopted this approach for several reasons: first, as we are interested in the effects on viewers who watch these outlets, it is not feasible for us to create stimuli from scratch. Also, since we posit effects for television, we believe using actual clips advantages us by increasing the intensity of the stimulus. In addition, these actual television clips better resemble what individuals will be exposed to in the electoral environment (as opposed to a manipulated script, for example) while avoiding issues related to confounds (Dafoe, Zhang, and Caughey 2016). EXPERIMENTAL PROCEDURE Participants filled out a pretest, which included questions on their demographics, news consumption, and political attitudes. We then randomly assign participants to a treatment. Posttest questions queried participants about their electoral expectations and preferences. Participants were then debriefed. The order of questions within the pre- and posttest are randomized, as are response options. The stimulus was first pilot-tested on 148 undergraduate students (see Online Appendix 2). To ensure randomization of key covariates among experimental conditions, we estimated logistic regression analyses and no relevant predictors were significant (see Online Appendix 3). PARTICIPANTS A total of 1,034 participants were recruited from Mechanical Turk (MTurk). A growing body of work demonstrates that MTurk offers a more diverse set of subjects than student samples (Huff and Tingley 2015), and results using the platform replicate causal relationships found in other convenience and representative samples (Mullinix et al. 2015). There were about 260 participants per condition. The sample included about percent men and 49.17

10 Partisan Media Effects on Electoral Predictions 311 percent women, with percent identifying as Republicans, 43 percent as Democrats, and an average age of 37 years old (SD = 11.73). More details on the participants can be found in Online Appendix 2. MEASURES For the outcomes of interest, respondents were asked, If you had to guess, which party do you think will win the Senate majority in the upcoming election? To assess electoral preference, respondents were also asked, If you had to pick, which party would you like to win the Senate majority in the upcoming election? Response options for both questions include the Democratic Party or the Republican Party. Examining the relationship between electoral preference and electoral expectation is the traditional approach to measuring wishful thinking in the literature (Granberg and Brent 1983; Abramson et al. 1992; Krizan, Miller, and Johar 2010; Meffert et al. 2011). Finally, we create a categorical treatment variable indicating exposure to MSNBC (only), Fox then MSNBC, MSNBC then Fox, with Fox News (only) as the baseline. Study 1: Results How does watching partisan media affect electoral expectations? To test hypothesis 1, we first look to differences between groups to flesh out whether watching partisan media affected electoral expectations. To do so, we use between-group analyses to test whether partisan media affect electoral predictions ( Fox (only) MSNBC (only) ). Figure 1 graphs the treatment effects for participants predicting a Republican victory in the 2014 midterm election by preference for electoral outcome; logistic regression models are available in Appendix A. Consistent with hypothesis 1, the results suggest that watching partisan media had a substantial effect on electoral predictions. Looking to the effects of partisan media exposure for all participants (see first panel of figure 1), roughly two-thirds (66%) of those watching just Fox News (only) believed Republicans would win the Senate, compared to 42 percent among those watching MSNBC (only). Thus, watching Fox News instead of MSNBC produced a 24-percentage-point difference (χ 2 = 29.93, p <.001) in electoral predictions. Additionally, only 48 percent of those also watching Fox then MSNBC thought Republicans would win the Senate. In other words, simply adding the MSNBC clip to Fox News exposure was enough to change viewers electoral predictions by 18 percentage points (χ 2 = 18.06, p <.001). To test hypothesis 2, which posits that exposure to horse-race coverage on like-minded media will make people more likely to predict their party will win the election, we use group differences ( Fox then MSNBC Fox (only) & MSNBC then Fox MSNBC (only) ) to measure the effects of

11 312 Searles, Smith, and Sui Figure 1. Electoral predictions by treatment and electoral preference (study 1). Error bars represent 95 percent confidence intervals. Electoral prediction is a binary measure indicating belief in a Republican (versus Democratic) victory. adding a like-minded source. In this way, we draw comparisons between groups that are exposed to a baseline of electoral coverage. The middle panel of figure 1 shows participants preferring a Democratic victory: 55 percent exposed only to Fox News believed the opposing party (Republicans) would win the Senate. But that number dropped to 34 percent among the Fox then MSNBC group. In other words, the addition of MSNBC coverage (Fox then MSNBC Fox only) made those preferring a Democratic majority 21 percentage points (χ 2 = 15.38, p <.001) more likely to predict Democrats would win. The third panel in figure 1 shows the effects of partisan media on participants that preferred a Republican victory: 67 percent of those watching MSNBC (only) thought the Republican Party would win the Senate, compared to 83 percent of those watching MSNBC then Fox. This suggests that watching Fox News after MSNBC made Republicans 16 percentage points more likely (χ 2 = 5.02, p =.025) to predict their preferred party would win the Senate. In short, these results support hypothesis 2: exposure to likeminded partisan media made viewers more confident that the party they prefer would win the election. Shifting our attention to hypothesis 3, which posits that partisan media will not significantly affect electoral predictions among viewers in the opposing

12 Partisan Media Effects on Electoral Predictions 313 party, we again use group differences ( MSNBC then Fox Fox (only) & Fox then MSNBC MSNBC (only) ), this time to measure the effects of adding opposing sources. While like-minded media increased predictions that the preferred party will be victorious, watching counter-attitudinal media after likeminded media had no discernable effect on electoral predictions. Of those preferring a Democratic victory, watching MSNBC (only) made them three percentage points less likely (χ 2 = 0.41, p =.519) to predict a Republican victory compared to those who watched MSNBC then Fox. Meanwhile, of those preferring a Republican victory, watching Fox (only) made them 11 percentage points more likely (χ 2 = 3.74, p =.053) to predict a Republican victory than those in the Fox then MSNBC group. It appears that watching opposing media after like-minded media has little effect on electoral predictions. These results provide some support for hypothesis 3, as exposure to opposing media had little effect on electoral expectations when viewers also watched like-minded media. These results offer evidence that partisan news facilitates wishful thinking by providing information that supports like-minded viewers preferred electoral outcome. Watching partisan media from either side of the aisle has an effect on electoral expectations, but perhaps most interestingly, watching both conservative and liberal media buoyed rather than tempered like-minded viewers expectations. In other words, exposure to favorable coverage from a likeminded source was enough for viewers to overcome unfavorable information from the opposing source. Consequently, partisan media are likely to facilitate wishful thinking among like-minded viewers when horse-race coverage is favorable toward the in-party. While our stimuli in study 1 expose participants to the sort of coverage that typifies these partisan news sources, as the PEJ data demonstrate, it is unclear whether we would see the same effects given unfavorable coverage. The true test of partisan media influence pits unfavorable horse-race coverage against viewers wishful thinking. People s desire to believe their party will win (wishful thinking) may motivate them to counter-argue the message, which could result in even more confidence in an electoral victory. Conversely, viewers may accept the negative information because they trust the source, resulting in less confidence that one s party will win the election. To apply this more conservative test of our hypotheses, we conduct a second survey experiment also during the 2014 midterm elections on a more representative sample. The purpose of study 2 is to examine whether our hypotheses hold up when messages from partisan news suggest the in-party s electoral standing are unfavorable. These tests yield compelling evidence for the relationship between partisan news and wishful thinking, and specifically, for the mitigating effect of unfavorable partisan news coverage on wishful thinking.

13 314 Searles, Smith, and Sui Study 2: Additional Experimental Evidence for Wishful Thinking The aim of this experiment is to replicate and extend the effects uncovered in study 1 by using a generalizable sample and dissonant partisan news stimuli. We use data from an embedded survey experiment in the 2014 pre-election wave of the Cooperative Congressional Election Study (CCES), fielded in October 2014 on 1,000 respondents (see Online Appendix 4 for more details). For the purposes of this study, we focus on a subset of the sample that received the treatment of interest (n = 506). 1 The CCES was conducted on the Internet and utilized an opt-in nonprobability sample-matching method to approximate a sample representative of the US adult population (Ansolabehere and Schaffner 2012). To measure the effects of partisan news on actual partisan news viewers, we utilize a modified patient preference trial (PPT) design that builds a patient s preference for treatment into the experimental design (Arceneaux and Johnson 2013). Specifically, we ask participants to state their news preference before being randomly assigned to either receive their preferred news source ( choice ) or not ( no choice ). The advantage of this approach is that it captures expressed source preference without those preferences being influenced by the stimuli, and allows us to test whether the results would generalize to a more natural environment where people can choose their preferred news source. STIMULUS Participants watch a 30-second television clip depicting unfavorable horse-race coverage. Each partisan news clip discusses which party will win the Senate majority in the 2014 election. Unlike study 1, the Fox News clip suggested that Democrats would gain Senate majority, while the MSNBC clip suggested Republicans would win. We again collected clips using SnapStream, and selected stimuli that are as alike as possible in terms of format and structure. For study 2, we also included a broadcast television treatment as a control, in which the reporter talked about both parties as competitive in the race for Senate majority. Additional details appear in Online Appendix 4. EXPERIMENTAL PROCEDURES Participants were informed that: Next you will watch a TV news clip on which political party will win the majority in the US Senate in the Not utilized in this study is a second manipulation in which participants were randomly assigned to receive coverage that specifically discussed poll results. We focus on the horse-race coverage results here for substantive and methodological reasons, but primarily to maintain continuity between study 1 and study 2.

14 Partisan Media Effects on Electoral Predictions 315 election. Please pick the news network you most prefer. If you don t like any of these networks, or do not watch television news, you may select the, I feel lucky option. We asked participants to reveal their true preferences by selecting their news source from the following items: MSNBC, Fox, NBC/CBS/ ABC, or I feel lucky. Given our modified PPT approach, participants were randomly assigned to choice (50 percent) or no choice (50 percent): those assigned to choice watched their preferred news source, while those assigned to no choice were randomly assigned to watch one of the other two networks (50/50). Thus, all participants watch one of three horse-race clips: MSNBC, Fox News, or broadcast. After viewing the clip, participants answered questions regarding their attitudes toward each party, and each party s chances of winning the Senate majority. Both the order of questions and response options are randomized. PARTICIPANTS A total of 1,000 participants were recruited for the 2014 CCES pre-election wave module. Participants were primarily women (53.5 percent), with an average age of 49 years (SD = 16.81). Most participants attained education beyond high school (71.7 percent). Participants identified primarily as Caucasian (73.6 percent), with 13 percent identifying as African American, 6.8 percent Hispanic, and 2.9 percent Asian. In terms of party identification, 49.7 percent identified as Democrats (including lean Democrat), 15.4 percent identified as Independents, and 34.1 percent identified as Republicans (including lean Republican). The participants scored an average of 4.15 (SD = 1.74) on a seven-point ideology scale where 1 = Very Liberal and 7 = Very Conservative. MEASURES For the dependent variable, respondents were asked to indicate how certain they are: The Republican/Democratic Party will win the Senate in the upcoming election. Respondents could either select a number on a 100- point ruler, anchored by Very Uncertain (0) and Very Certain (100), or Don t Know. Participants that selected Don t Know were dropped from the analysis (n = 154). We reverse-coded evaluations of the Democratic Party to create a scale in which low numbers indicate uncertainty in a Republican victory (or high certainty in a Democratic victory), and high numbers indicate high certainty in a Republican victory (or low certainty in a Democratic victory). To assess electoral preference, we categorized respondents as preferring a Republican or Democratic victory based on their expected Senate vote in 2014 (for more details, see Online Appendix 4). We then created a binary variable for Republican candidate preference in which 1 indicates the participant prefers a Republican candidate, and 0 indicates the participant prefers a Democratic candidate. Finally, we created a categorical treatment variable

15 316 Searles, Smith, and Sui indicating exposure to Fox News or MSNBC, with broadcast television as the baseline. To measure news preference, we rely on participants revealed preferences in selecting the source they would most prefer to watch. Participants are identified as having a preference for MSNBC if they select MSNBC, likewise for Fox News and broadcast. To ensure randomization of key covariates among experimental conditions, we estimated logistic regression analyses, and no relevant predictors are significant (see Online Appendix 5). Study 2: Results Figure 2 graphs the treatment effects for participants average certainty of a Republican Senate victory in the 2014 election by preference for electoral outcome. The full regression models are available in Appendix A. Similar to study 1, and consistent with hypothesis 1, the results suggest that partisan media affect electoral predictions for all viewers. However, this time, given exposure to clips discussing an in-party loss, we see significant effects in the opposite direction. Looking to the results among all participants (see graph on the left), exposure to Fox News made viewers less certain of a Republican victory, while exposure to MSNBC made viewers more certain that Republicans Figure 2. Electoral predictions by treatment and electoral preference (study 2). Error bars represent 95 percent confidence intervals. Certainty of a Republican victory is measured on a scale, with higher numbers denoting increased certainty that Republicans will win the Senate.

16 Partisan Media Effects on Electoral Predictions 317 would capture the Senate. Specifically, those watching Fox News were roughly 15 points less certain (p <.05) of a Republican victory than those watching the MSNBC clip. In other words, partisan media effects persist (akin to study 1), but the nature of content changes the relationship. It appears that partisan media can move electoral expectations in a positive or negative direction depending on whether they are favorable or unfavorable toward the in-party. To test hypothesis 2, we can also look to figure 2; recall that hypothesis 2 posits that exposure to like-minded partisan media whether favorable or unfavorable increases the likelihood that people will predict their preferred party will win the election. Interestingly, both Fox News and MSNBC had nearly identical effects on their like-minded viewers, but in the opposite direction. Relative to people watching broadcast news, watching MSNBC made those preferring a Democratic candidate nearly 11 points more certain that the other party would win the Senate (p <.05). Meanwhile, Fox News made participants preferring a Republican candidate 11 points less certain that Republicans would win the Senate (p <.05), compared to those exposed to broadcast. It appears that, on the rare occasion partisan media call a race unfavorable to the in-party, such coverage can overcome wishful thinking for like-minded viewers. Contrary to hypothesis 2, wishful thinking does not persist for like-minded viewers regardless of the direction of horse-race coverage. In support of hypothesis 3, we find no evidence that partisan media affected electoral predictions among viewers in the opposing party. The results in the middle and left graphs in figure 2 suggest that Fox News and MSNBC had no effect on the electoral expectations of those preferring a Democratic and Republican victory, respectively. Indeed, people with a preference for a Democratic candidate had similar predictions whether watching Fox News or broadcast coverage. The same was true for those preferring a Republican victory and watching MSNBC. In addition to supporting hypothesis 3, these results demonstrate that receiving favorable news from opposing media does not augment electoral expectations. Given previous research on selective exposure and media choice (Druckman, Fein, and Leeper 2012; Arceneaux and Johnson 2013), we leverage our experimental design to examine whether partisan media have different effects on those that would normally watch the news source. The purpose of this analysis is to flesh out whether the experimental effects are simply an artifact of forced exposure to partisan media among respondents who would not watch it otherwise. Figure 3 shows treatment effects for participants that were exposed to the news source they prefer. When participants are exposed to the news source they prefer, both MSNBC and Fox News had a significant and large effect on electoral expectations. In fact, those choosing the MSNBC clip were nearly 20 points more certain that Republicans would win relative to those selecting the Fox News clip. In other words, partisan media can reduce wishful thinking even when we limit effects to those who prefer it.

17 318 Searles, Smith, and Sui These results seemingly conflict with previous research suggesting that allowing participants to choose their information source weakens opinion change (Druckman, Fein, and Leeper 2012; Arceneaux and Johnson 2013). The main difference in our study is that participants were shown counter-attitudinal information by their preferred information source. In a low-information environment, partisan media may be able to move opinions among like-minded viewers if they present counter-attitudinal information. Though seemingly counterintuitive, it is rational for viewers to update their electoral expectations in the face of bad news from an unexpected source. Calvert s (1985) model of political advice suggests that unexpected electoral predictions from a biased adviser here Fox News or MSNBC reports provides a strong signal with regard to information accuracy. Under such circumstances, viewers could seek alternative explanations, taking additional effort and resources, but other source predictions are likely to be noisy. In this way, bad electoral news from an in-party news source is valuable for boundedly rational decision-makers in a way that the same news from an out-party news source or good news from the likeminded source is not. Figure 3. Electoral predictions by treatment and electoral preference for preferred news source (study 2). This figure only includes participants exposed to their preferred news source. Error bars represent 95 percent confidence intervals. Certainty of a Republican victory is measured on a scale, with higher numbers denoting increased certainty that Republicans will win the Senate.

18 Partisan Media Effects on Electoral Predictions 319 Discussion The news media play a key role in helping Americans understand the electoral environment. Under normal circumstances, partisan media are likely to facilitate wishful thinking by presenting horse-race coverage that favors the in-party. According to our content analysis data, and the results of previous research (Searles, Ginn, and Nickens 2016), partisan media present favorable horse-race coverage, such that Fox News coverage is more likely to depict Republicans winning, while MSNBC suggests Democrats will be victorious. Our experimental results in study 1 demonstrate that this sort of typical coverage is influential on like-minded viewers, as exposure to favorable horse-race coverage increased wishful thinking among likeminded viewers. The reverse was observed when people watched coverage from the other side; however, watching coverage from both sides of the aisle buoyed wishful thinking. In other words, people may be swayed by coverage from the other side when watching it alone, but when it is accompanied by exposure to likeminded media, people respond as if they ve heard only good news. These results likely simulate what happens in an actual media environment. Research suggests that the audience for partisan media accesses a fairly diverse diet of news media (Holbert, Hmielowski, and Weeks 2012; Garrett, Carnahan, and Lynch 2013). This research suggests that cross-exposure is unlikely to blunt the effects of likeminded media on electoral expectations. Whether relying solely on their outlet of choice, or consuming more diverse sources, the result is the same: like-minded partisan media offer viewers the necessary fodder to bulwark their prediction that their party will be victorious. In short, our results do not suggest that exposure to ideologically diverse media is the panacea for wishful thinking. Study 2 puts wishful thinking to a stricter test, in which partisan media suggest that the in-party will lose the election. Contrary to our expectations, partisan media reduced wishful thinking among like-minded viewers. Fox News made people with a preference for Republican victory less confident that Republicans would win, and MSNBC made people with a preference for Democratic victory less confident that Democrats would win, compared to those watching broadcast news. It appears that partisans are more likely to accept counter-attitudinal horse-race information when it comes from a likeminded source. These findings come of some surprise given that wishful thinking has previously been found to be resistant to countervailing information in a variety of contexts (Babad, Hills, and O Driscoll 1992; Krizan, Miller, and Johar 2010). But given the strong signal of costly talk from a biased adviser, such a response can be characterized as rational given the resources required to assess alternatives. Altogether, our results suggest that an addendum may be needed: wishful thinking is persistent, but during a midterm election it may be mitigated by contrary evidence from a congenial source. Even more interesting is the implication that our study has for the news industry, as our results suggest that partisan media could potentially facilitate more accurate electoral predictions by simply accurately reporting the horse race, or at least occasionally reporting unfavorable electoral standings for the in-party.

19 320 Searles, Smith, and Sui It is important to point out that we conducted these experiments just prior to the 2014 midterm elections, a context that may allow for larger wishful thinking effects than presidential elections. The 2014 midterm battle for the Senate involved a number of different races in multiple states. This more complex environment likely renders partisan media horse-race coverage more meaningful. Partisan media effects are likely less pronounced in a presidential election that garners more media attention and involves only two candidates. Still, our results have important implications for the legitimacy of democratic institutions. On the one hand, by presenting biased coverage of a race, partisan media are potentially undermining the legitimacy of electoral outcomes. If partisans are surprised by their eventual defeat, they might attribute it to voting manipulation or fraud. For example, some Republicans believed that ACORN helped Obama win in 2008, while some Democrats believed that manipulation of polling machines helped elect Bush in If people only see coverage discussing the inevitable victory of their preferred candidate, and that candidate loses the election, they may attribute blame inaccurately. In efforts to maintain cognitive consistency, people may pin the loss on malfeasance, or worse, question the integrity of institutions. In sum, perceptions of who is likely to win or lose elections affect how people vote, whether they donate money, or whether they volunteer for a campaign. The very people that are active in political campaigns are also more likely to get their news from partisan media (Levendusky 2013). Taken together, our results suggest that partisan media are poised to affect the electoral expectations of the most politically active segment of the American public. Overall, we show that where people get their news, and how partisan media cover the political horse race, has important effects on electoral expectations. Appendix A. Regression analyses (study 1 and study 2) Table A1. Electoral predictions by treatment group and electoral preference (study 1) All participants Prefer Democrats Prefer Republicans MSNBC (.183)* 1.09 (.231)* 1.03 (.369)* Fox/MSNBC (.178)* (.222)* (.370)* MSNBC/Fox (.183)* (.225)* (.428) Constant (.131)* (.157) 1.76 (.289)* N 1, Note. Coefficients are from logistic regression models where the binary dependent variable is participants prediction that Republicans (1) or Democrats (0) would control the Senate after the 2014 midterm elections. Standard errors are in parentheses. The MSNBC, Fox/MSNBC, and MSNBC/Fox variables indicate the clip watched by the respondent, with Fox News as the baseline. The first column is the full model, and each subsequent column is restricted to either participants with a preference for a Democratic Senate majority or participants with a preference for a Republican Senate majority. These results were used to create figure 1. Tests are two-tailed; * p <.05.