The Impact of Media Endorsements in Legislative Elections

Transcription

1 The Impact of Media Endorsements in Legislative Elections Kyle A. Dropp Department of Government Dartmouth College Christopher Warshaw Department of Political Science Massachusetts Institute of Technology First draft: 2012/10/15 This draft: 2014/06/25 Assistant Professor, Department of Government, Dartmouth College, Assistant Professor, Department of Political Science, Massachusetts Institute of Technology, 1

2 In this study, we investigate how newspaper endorsements influence election outcomes. Scholars have puzzled over the fact that newspaper endorsements only seem to have a small impact on election outcomes. Building upon a literature in political behavior and psychology, we argue that newspaper endorsements function as credible source cues that voters use as an information shortcut to evaluate candidates. This implies that the impact of newspapers should depend on the level of ideological congruence between the voter and the source of the endorsement. We examine this hypothesis using separate conjoint-based survey experiments nationally and in a single media market. We show that newspaper endorsements have a small average treatment effect on candidate support. But the effect varies across individuals. As expected, voters reward congressional candidates when they perceive that an endorsing newspaper is ideologically similar to their preferences but punish endorsed candidates when they perceive that the source is ideologically distal. These findings have implications for the burgeoning literatures on elite cues, voting behavior, and representation. 2

3 Newspapers are one of the most important sources of information about politics for many voters. Despite the graduate decline in newspaper readership in recent years, a quarter of Americans still read newspapers everyday and nearly 40% read a newspaper regularly. 1 Moreover, most newspapers make political endorsements. In 2012, 76 of the 100 most widely circulated newspapers endorsed a presidential candidate. 2 Most newspapers also make endorsements in Congressional elections, as well as in state and local elections. Furthermore, newspapers regularly endorse candidates in both primary and general elections. For instance, the Austin American-Statesman endorsed six candidates from both parties in the 2012 congressional primary elections 3 and the Detroit News endorsed 13 candidates from both parties in 10 congressional districts. 4 It is important to understand how these endorsements influence politics. Previous observational research has found suggestive evidence that exposure to newspaper endorsements influences voters decisions. Studies have found that endorsements may boost candidate support in city council elections (Krebs, 1998), state house races (Mason, 1973), U.S. Senate contests (Druckman and Parkin, 2005; Kahn and Kenney, 2002), gubernatorial races (Hollander, 1979), presidential elections (Chiang and Knight, 2011; Erikson, 1976; Robinson, 1974) and parliamentary elections (Ladd and Lenz, 2009). However, these observational studies may over-estimate the impact of endorsements for a variety of reasons. First, voters are more likely to read newspapers that confirm their pre-existing views (Arceneaux and Johnson, 2013; Hollander, 2008; Stroud, 2011). As a result, they may be predisposed to support 1 Pew Research Center, 2 American Presidency Project php

4 the same candidate as the newspaper. Second, newspapers could choose to support highquality candidates that are likely to win with or without an endorsement, thereby creating the false impression that media endorsements affect candidate support. Another weakness of previous studies is that few studies have compared the impact of endorsements relative to candidate attributes such as occupation or gender or relative to candidate policy positions, and none have examined whether the impact of newspaper endorsements is heterogeneous across individuals. This is striking because there are an abundance of theoretical reasons to expect heterogeneous treatment effects in the impact of endorsements. Most importantly, endorsements function as credible source cues that citizens can use as an efficient information shortcut (Arceneaux and Kolodny, 2009; Druckman, 2001; Lupia, 1994). We hypothesize that when voters see an endorsing group as aligned with their values, they can trust that the group would make the same decisions they would have made with complete information. Conversely, if people see an endorsing group in opposition to their values and interests, then they can assume that the group would make the opposite decision than they would (Brady and Sniderman, 1985). This suggests that voters are more likely to support a candidate that receives an endorsement from a like-minded group, while political endorsements from groups individuals dislike makes them less likely to support a candidate. Our study addresses both of these weaknesses in past studies. We use a choice-based conjoint design among a national sample of more than 2,500 adults and 550 Washington, D.C. area residents to identify the impact of newspaper endorsements on citizens votes in legislative elections. Most importantly, we randomize which candidate receives a newspaper endorsement, along with other attributes of candidates. Conjoint analyses have been widely 4

5 used in marketing research (Gustafsson, Herrmann, and Huber, 2010) and they have recently been applied to political science (Hainmueller, Hopkins, and Yamamoto, 2014; Hainmueller and Hopkins, 2012). This design enables us to identify the causal effects of newspaper endorsements on voting decisions. It also enables us examine whether the ideological congruence between voters and newspapers moderates the impact of newspaper endorsements in legislative elections. We find strong evidence that newspaper endorsements matter. As expected, their effect varies depending on the ideological congruence between newspapers and voters. Citizens use their perception of the media outlet s ideological stance as a heuristic: citizens are more likely to support endorsed candidates when they perceive that the newspaper is ideologically similar to their preferences, and less likely to support the endorsed candidate when they perceive that the newspaper is ideologically distinct from their preferences. These results indicate that newspaper endorsements function as credible source cues that citizens can use as an efficient information shortcut to evaluate the spatial position and valence of a candidate. This study advances our understanding of the impact of the media on elections. Most importantly, our findings show that media endorsements can have a large effect on the type of congressional candidates that are elected to Congress. Voters reward or punish endorsed candidates based on the relative spatial locations of media outlets, candidates and voters. But our results also have implications for strategic candidates and interest groups. Our findings suggest that candidates should seek to obtain endorsements from newspapers that are ideologically similar to the median voter in their district. In contrast, it is not in the best interest of candidates to receive an endorsement from a newspaper that is ideologically distinct from the median voter. 5

6 This paper proceeds as follows. First, we discuss previous work on the impact of newspaper endorsements. Next, we discuss our theory and hypotheses. Third, we discuss our research design. Then, we discuss our findings. Finally, we briefly conclude. The Impact of Media Endorsements Scholars have long found that most voters pay only fleeting attention to politics (Berelson, Lazarsfeld, and McPhee, 1954; Delli Carpini and Keeter, 1993). They are not performing deep, independent research on political candidates. Instead, voters look for shortcuts and credible cues such as a candidate s party affiliation, ideology, viability, endorsements, or appearance to save time, reduce information costs, and ultimately inform their voting decisions (Lau and Redlawsk, 2001). Newspapers function as one of these cues. Previous research has found that exposure to newspapers influences voters decisions (e.g., Dilliplane, 2014; Gerber, Karlan, and Bergan, 2009). Moreover, there is observational evidence that newspaper endorsements have buoyed candidate support between two to five points for candidates in city council elections (Krebs, 1998), U.S. Senate contests (Druckman and Parkin, 2005; Kahn and Kenney, 2002), gubernatorial races (Hollander, 1979), presidential elections (Chiang and Knight, 2011; Erikson, 1976; Robinson, 1974) and parliamentary elections (Ladd and Lenz, 2009). Robinson (1974) finds that voters were more likely to support Nixon in 1972 if they read a newspaper endorsing Nixon. Krebs (1998) demonstrates that city council candidates in Chicago received more votes if they were endorsed by the Chicago Tribune or the Sun Times. Erikson (1976) finds that newspaper endorsements in the 1964 presidential elections increased 6

7 the vote share of the endorsed candidate by five percentage points. Kahn and Kenney (2002) found significant positive effects of endorsements in U.S. Senate races on the comparative feeling thermometer score in American National Election Survey data. Similarly, Druckman and Parkin (2005) find that the slant of newspapers had an impact in the 2000 Minnesota Senate election. Hollander (1979) finds evidence that support for a candidate in Maryland s 1976 gubernatorial primary increased after they received a newspaper endorsement. 5 Ladd and Lenz (2009) show that newspaper endorsements in the U.K. s 1997 parliamentary elections substantially increased support for the party that was endorsed. Finally, Chiang and Knight (2011) show that credible newspaper endorsements increased candidate support in the 2000 and 2004 presidential elections. Cumulatively, these studies lead to the following hypothesis: H1: Newspaper endorsements increase the average level of support for candidates. Scholars have argued, however, that the impact of endorsements is heterogeneous across different individuals. Most importantly, partisan bias shapes the way individuals translate information into their political attitudes and opinions (Bartels, 2002; Taber and Lodge, 2006; Gaines et al., 2007) and serves as a perceptual screen through which the individual tends to see what is favorable to his partisan orientation (Campbell et al., 1980, p. 133). Previous work has focused on the relationship between the ideology of the endorser and individuals. If people see an endorsing group as aligned with their values, they can trust that the group would make the same decisions they would have made with complete information. Conversely, if people see it in opposition to their values and interests, then they 5 Note, however, that this study is limited to a single newspaper endorsement and also does not provide tests for the statistical significance of this increase (Chiang and Knight, 2011). 7

8 can assume that the group would make the opposite decision than they would (Brady and Sniderman, 1985). Thus, endorsements function as credible source cues that citizens can use as an efficient information shortcut. These cues help individuals to make correct voting decision (Arceneaux and Kolodny, 2009; Druckman, 2001; Lupia, 1994, but see Kuklinski and Hurley, 1994). As a result, voters are more likely to support a candidate that receives an endorsement from a like-minded group, whereas political endorsements from groups individuals dislike makes them less likely to support a candidate. This suggests newspaper endorsements should increase the average level of support for candidates when individuals perceive that the ideology of the newspaper is congruent with their own ideology, and decrease the average level of support for candidates when individuals perceive that the ideology of the newspaper is incongruent with their own ideology. No previous study has examined whether newspaper endorsements function as credible source cues that citizens can use as an efficient information shortcut to support or oppose candidates. But several recent studies have examined whether other types of endorsements are used as a heuristic by voters. Arceneaux and Kolodny (2009) find that Republicans that were contacted by a liberal political group in a Pennsylvania state house election were less likely to support a Democratic candidate. They argue that Republicans used the endorsement as a negative voting cue, which helped some Republicans compensate for their lack of political awareness. Similarly, Hopkins and Ladd (2014) examine the impact of Fox News Channel s national expansion on candidate preferences for 22,592 respondents to the 2000 National Annenberg Election Survey. This can be seen as a study on endorsements since Fox News Channel clearly supported George W. Bush rather than Al Gore. While they find an average treatment effect indistinguishable from zero, Hopkins and Ladd (2014) find a sizable effect 8

9 of Fox News access on the vote intentions of Republicans and pure independents. Thus, it appears that the Fox News endorsements act as a cue for like-minded voters. Cumulatively, this leads to the second hypothesis: H2: Newspaper endorsements increase the average level of support for candidates when individuals perceive that the ideology of the newspaper is congruent with their own ideology, and decrease the average level of support for candidates when individuals perceive that the ideology of the newspaper is incongruent with their own ideology. Research Design - Conjoint Analysis As we have previously noted, the observational, cross-sectional design of most previous studies on media endorsements are particularly susceptible to selection bias. 6 First, voters are more likely to read newspapers that confirms their pre-existing views (Arceneaux and Johnson, 2013; Hollander, 2008; Stroud, 2011). Therefore, they may be predisposed to support the same candidate as the newspaper. Second, newspapers could choose to support highquality candidates that are likely to win with or without an endorsement, thereby creating the false impression that media endorsements affect candidate support. For instance, the New York Times endorsed three Democrats and two Republicans in general elections for State Senate races in 2012, an indication that they have supported candidates out of line with their editorial board s liberal ideological orientation. We address these problems by administering two choice-based conjoint experiments among a national sample of more than 2,500 adults and 550 Washington, D.C. area residents (for more on conjoint designs see Hainmueller, Hopkins, and Yamamoto (2014) and Hainmueller and Hopkins (2012)). This conjoint design enables us to assess the causal impact of news- 6 Exceptions are Ladd and Lenz (2009) which exploits panel-data, and Chiang and Knight (2011), which uses a regression-discontinuity design. 9

10 paper endorsements on citizens votes in legislative elections. It also enables us to assess whether the treatment effect of endorsements varies due to the level of ideological congruence between the newspaper and voter. National Study This study, conducted in January and March 2014 using Survey Sampling International, Inc., an online survey research firm, places respondents in the position of a voter making decisions in a hypothetical Congressional election. The introduction to the survey reads: For the next few minutes, we are going to ask you to act as if you were about to cast a vote for the candidate that will represent you in the U.S. Senate. We will describe to you several pairs of candidates running for election to the U.S. Senate. For each pair of people, please indicate your attitudes towards the two candidates and which one you would prefer to represent you. Even if you aren t entirely sure, please indicate which of the two you prefer. [Figure 1 about here] We then show respondents a screen with profiles of two candidates, as displayed in Figure 1, and require a choice between each pair of candidates. We vary the profiles of the two candidates on five different attributes: each candidate s party identification, occupation, and issue positions on an assault weapons ban and abortion. 7 We randomly assigned candidates a first name and last name from a list of 20 common first names and surnames to make 7 Marketing scholars are highly concerned with the potential for information overload in conjoint analysis studies (Acito, 1979). Various scholars have recommended using at most 5, 6 or 10 attributes in conjoint studies (Green and Srinivasan, 1978; Malhotra, 1982). We chose these issues since they are each salient policy areas, but it is plausible that members of each party could take a liberal or conservative position on a given issue. Here is the full list of traits for each attribute: Party Identification - Republican, Democrat, Independent; Position on Abortion - Pro-Choice, Pro-Life; Position on Assault Weapons Ban - Supports Ban, Opposes Ban; Newspaper Endorsements - Washington Post, Washington Times, None; Occupation - Business Owner, Car Dealer, Carpenter, Doctor, High School Teacher, Lawyer, Member of Congress, Military Veteran 10

11 the decision task more realistic and provide a signal of candidate gender and race/ethnicity. Finally, we vary whether the respondent received an endorsement from the largest newspaper by circulation in the respondent s ZIP Code. 8 A respondent living in Milwaukee, Wisconsin, might see an endorsement from the Milwaukee Journal Sentinel and a respondent living in San Francisco, California, might see an endorsement from the San Francisco Chronicle. Presenting respondents with an endorsement from a local source with which nearly all respondents should be familiar enhances the study s external validity and introduces important variation to respondents evaluations of the source s credibility. Each respondent evaluates three binary comparisons in succession. On separate screens, respondents select their preferred candidate, then evaluate the first candidate in the comparison, and finally evaluate the second candidate. 9 For each respondent, we randomly assign the order of the attributes to rule out primacy and recency effects. 10 To measure outcomes, we ask respondents to report a preference for one of the two profiles. Our analyses use the binary responses as the primary outcome variable with a 1 indicating the selected candidate profile. This design has the advantage that it forces respondents to 8 We obtained the newspaper circulation data from the Audit Bureau of Circulations. accessabc.com/products/newsgeo.htm 9 We presented the candidate information on each of the three screens. We believe that using three separate screens rather than one makes the decision task easier for respondents because they do not have to scroll, they see the information three times instead of once, and there are fewer unique questions per screen. Respondents provide their overall level of support, perceived ideology, and perceived qualifications for each candidate. Each item is asked on a seven point scale. Overall Level of Support: On a scale from 1 to 7, where 1 indicates that you would never support this candidate, and 7 indicates you would always support this candidate, where would you place CANDIDATE? Perceived Ideology: Thinking about politics these days, how would you describe the political viewpoint of CANDIDATE? Very liberal, Liberal, Slightly liberal, Moderate, Slightly conservative, Conservative, Very conservative Perceived Quality: On a scale from 1 to 7, do you believe CANDIDATE has the qualifications, skills, and experience to be an effective U.S. Senator? 10 However, the order of the attributes does not change for each respondent across the three binary comparisons to reduce the task s complexity. Moreover, we restrict the randomization of the attribute order such that the issue positions always appear together in a block in a randomized order (Hainmueller and Hopkins, 2012). 11

12 make tradeoffs between candidates. As a robustness check, we use a seven-point variable assessing support for each candidate separately. 11 Our conjoint experimental design has several advantages over prior observational and experimental approaches to understanding citizens vote choices. The fully randomized design allows us to estimate the unique impact of each candidate attribute on the probability of being preferred (e.g. how much does the probability of supporting a candidate increase when the candidate in question is pro-choice on abortion?). Moreover, since we randomize all the attributes and measure their effects on the same scale, the design allows us to examine the relative importance of each attribute (e.g., endorsements versus issue positions). Data We utilized Survey Sampling International, Inc. (SSI) to recruit a total of 2,608 U.S. residents to two identical online surveys. The first survey was in the field from Jan. 28, Jan. 31, 2014 and the second survey was in the field from March 9-11, SSI samples are not as representative as the best national probability samples but significantly outperform convenience samples. Furthermore, many political scientists have used SSI samples for their survey research projects (Berinsky, Margolis, and Sances, 2013; Malhotra, Margalit, and Mo, 2013). We use post-stratification weights to adjust the final respondent data for common sources of survey error (non-response, coverage error, etc.). 12 However, the results are substantively similar with or without weights. 11 This outcome has the advantage that it allows us to investigate how certain attributes affect absolute levels of support for a particular candidate. 12 The weights adjust the sample to the demographic and geographic distributions for U.S. registered voters from the November Supplement of the 2012 Current Population Survey (CPS). The results are unchanged with or without using these weights. 12

13 Regression Models We study whether an endorsement from a local newspaper influences candidate support using a series of OLS regressions. Our primary specification is (1) Support i = α 0 +β 1 Endorsement i +β 2 Occupation i +β 3 P arty i +β 4 Abortion i + β 5 W eapons i + u i where the dependent variable is support for the U.S. Senate candidate and the predictors are the five randomized attribute dimensions 13 This design provides a non-parametric, unbiased estimate of the average effect of each candidate attribute, including newspaper endorsements, since the values of all attributes for both profiles are randomly assigned and, therefore, orthogonal to one another (Hainmueller, Hopkins, and Yamamoto, 2014). 14 Next, we examine the interaction between endorsement treatment group assignment and the level of ideological congruence between newspapers and voters. We construct a measure of ideological congruence by asking individuals to evaluate their own political ideology and the ideology of the largest newspaper in their ZIP Code on a seven-point scale, 15 placing these metrics on a common scale between 0 and 1, and then calculating the absolute value of the difference between the voter s ideology and their perceived ideology of the newspaper The models also include fixed effects at the Metropolitan Statistical Area (MSA) level and for the sequence (i.e., 1, 2, or 3) of the binary comparison. We have run OLS and logistic regressions when the main dependent variable is a binary support measure, along with OLS and proportional odds logistic regressions when the main dependent variable is the 7-point support measure for individual candidates. In all cases, the results are qualitatively similar for both the linear and nonlinear models. We use the OLS models for ease of interpretation. When making any predictions, we will use coefficients from the logistic or ordered logistics models. 14 The standard errors are clustered by respondent to account for the fact that the choices made by a single respondent may not be be independent across the three comparisons (Hainmueller, Hopkins, and Yamamoto, 2014). 15 We ask both questions on the same seven-point scale. Response options were very liberal, liberal, somewhat liberal, moderate, somewhat conservative, conservative, and very conservative. 16 This variable ranges from 0 to 1, where 0 is complete ideological congruence and 1 is complete ideological dissonance. Republicans mean score on this measure was.42, compared with.25 for Democrats. It is also worth noting that we asked respondents to rate the ideology of their local newspaper prior to the endorsement modules. 13

14 We examine the impact of endorsements separately for respondents with High Ideological Congruence, Medium Ideological Congruence, and Low Ideological Congruence. 17 On a 0 to 1 scale, High Ideological Congruence respondents are 0 or 0.17 scale points from the newspaper, Medium Ideological Congruence respondents are 0.33 to 0.5 scale points from the newspaper, and Low Ideological Congruence respondents are more than 0.5 scale points away from the newspaper. Results Figure 2 displays the average treatment effects among registered voters who read their local paper at least weekly. The results indicate that newspaper endorsements have a small, positive effect on candidate support. Endorsements increase candidate support by about 4 points compared to no endorsement, a statistically significant effect. 18 The magnitude of these effects is similar to the impact of candidates Party ID or issue positions. Democratic candidates receive about 4 points more support than Independent candidates, while Republican candidates receive about 2 points less support than Independent candidates. Pro-choice candidates receive about 5 points more support than pro-life candidates. Candidates that favor an assault weapons ban are no more likely to garner support than candidates that oppose it. Respondents are 10 to 12 points more likely to support a candidate whose previous occupation is a military veteran compared with a car dealer. 17 This approach is similar to the one in Bartels and Johnston (2013), which constructed a measure of Subjective Ideological Disagreement with the Supreme Court of the United States. They ask for respondents ideological preferences along with their perception of the Supreme Court s ideological tenor (e.g., liberal, conservative, moderate). The authors create a four item measure with strong agreement, tacit agreement, moderate disagreement and strong disagreement rather than using a distance metric. 18 The average treatment effect among registered voters who report reading a newspaper in the previous 24 hours is also 4 percentage points (p <.05). The average treatment effect among all registered voters is 2 percentage points (p <.16). 14

15 [Figure 2 about here] Figure 3 displays how the impact of newspaper endorsements varies as a function of ideological congruence between the newspaper and the voter for all voters (Panel A), for weekly newspaper readers (Panel B) and for respondents who have read a newspaper in the past 24 hours (Panel C), respectively. Among all voters, endorsements have a large, positive and statistically significantly effect of approximately 5 percentage points when there is high ideological congruence (low distance) between the newspaper and the voter. When there is medium ideological congruence, endorsements have a positive, borderline statistically significantly effect of approximately 4 percentage points. However, when there is a low level of congruence between the newspaper and the voter, an endorsement reduces candidate support by a large, negative, and statistically significantly 9 percentage points. 19 The punishment effect among adults with low ideological congruence with their local newspaper is sizable. The treatment effects are similar among adults who read a newspaper at least weekly and among adults who reported reading a newspaper in the past 24 hours. For example, among adults who read a newspaper at least weekly, an endorsement when there is high ideological congruence boosts candidate support by 6 percentage points and an endorsement when there is low congruence reduces candidate support by 12 percentage points. [Figure 3 about here] 19 Tables 1-3 in the Appendix display how the impact of newspaper endorsements varies as a function of ideological congruence for all registered voters, for respondents who read their local paper at least weekly, and for respondents who have reported reading any paper in the previous 24 hours. 15

16 Washington, D.C. Area Study Results from the national study indicate that endorsements exert a positive overall effect on candidate support and that the magnitude of this effect depends on the relative spatial location of voters, newspapers, and candidates. This study varied whether respondents received an endorsement from the largest newspaper by circulation within the respondent s ZIP code. It did not, however, assess the impact of varying a newspaper s ideology within a given designated market area because most markets have a single, dominant media outlet. We address this point with a Spring 2013 survey of residents in the Washington, DC metro area, where there are clearly differentiated liberal and conservative newspapers. Respondents viewed an introduction similar to the national study 20 and then evaluated five binary comparisons, as displayed in Table 4, with each comparison displayed on a new screen. We vary the profiles of the two candidates on six different attributes: each candidate s race/ethnicity, party identification, issue positions on an assault weapons ban and abortion, and whether they received a Washington Times endorsement, a Washington Post endorsement, or no endorsement. 21 As in the national study, we randomly assign the order of the attributes. After viewing the binary comparisons, respondents report a preference for one of the two profiles and assess each candidate on a seven-point continuous support variable. 20 For the next few minutes, we are going to ask you to act as if you were about to cast a vote for the candidate that will represent you in the U.S. Congress. We will describe to you several pairs of candidates running for election to Congress. For each pair of people, please indicate your attitudes towards the two candidates and which one you would prefer to represent you. Even if you aren t entirely sure, please indicate which of the two you prefer. 21 Here is the full list of traits for each attribute: Race/Ethnicity - White, Black, Hispanic, Asian-American; Party Identification - Republican, Democrat, Independent; Position on Abortion - Pro-Choice, Pro-Life; Position on Assault Weapons Ban - Supports Ban, Opposes Ban; Newspaper Endorsements - Washington Post, Washington Times, None; Occupation - Business Owner, Car Dealer, Carpenter, Doctor, High School Teacher, Lawyer, Member of Congress, Military Veteran 16

17 [Figure 4 about here] Data We utilized Survey Sampling International (SSI) to recruit a total of 550 Washington, D.C. metropolitan area residents to an online survey. 22 The results reported below have been weighted to match geographic distributions from the Washington DC metropolitan area in the March Supplement of the 2010 Current Population Survey (CPS). 23 Models We study whether an endorsement from a local newspaper influences candidate support using a series of OLS regressions. Our primary specification is (1) Support i = α 0 +β 1 Endorsement i +β 2 Occupation i +β 3 P arty i +β 4 Abortion i + β 5 W eapons i + β 6 Race i + u i where the dependent variable is support for the legislative candidate and the predictors are the five randomized attribute dimensions. 24 Next, we examine whether endorsements have heterogeneous impacts based the ideological congruence between voters and newspapers. We measured respondents ideological positions with eight binary questions on policy issues such as abortion and the minimum wage. 25 Based on the newspapers previous endorsement behavior and individual-level survey data, we characterize the Washington Post as a liberal newspaper and the Washington 22 We restricted our sample to include only adult citizens who resided within 35 miles of Washington, DC. Slightly modifying the distance metric does not change the substantive findings. 23 The weighted analyses use post-stratification weights to adjust the final respondent data for common sources of survey error (e.g., non-response, coverage error, etc.) The results are similar with or without survey weights. 24 The models also include fixed effects for the sequence (i.e., 1, 2, 3, 4, or 5) 25 We asked about for the following issues: raising the minimum wage, allowing same-sex marriage, providing universal health care, eliminating the estate tax, reinstating the assault weapons ban, limiting greenhouse gas emissions and abortion. We use this definition of ideology in the foregoing results section rather than 17

18 Times is a conservative newspaper. 26 We then can categorize liberals viewing a Washington Post endorsement or conservatives viewing a Washington Times endorsement as high on an ideological congruence scale; alternatively, liberals viewing a Washington Times endorsement or conservatives viewing a Washington Post endorsement are categorized as low on an ideological congruence scale. Results Figure 5 displays the average treatment effects across our entire sample and indicates, consistent with the national study, newspapers have a small, positive effect on candidate support. The Washington Post endorsement increases candidate support by approximately five percentage points, a statistically significant result, and a Washington Times endorsement increases support by two percentage points, though this coefficient is not statistically significant. 27 The magnitude of these effects is slightly smaller than the impact of candidates party affiliation or issue positions. Democratic candidates receive about the same level of support as Independent candidates, while Republican candidates receive about 9 points less support than Independent candidates, a result of the Democratic-leaning Washington, D.C. the standard 7-point self-reported political ideology question. We believe that responses to this eight question battery better capture an individual s ideological position than a single question. Nevertheless, figures and tables using this standard political ideology question as an independent variable produce similar results. 26 The Washington Post endorsed the Democratic candidate for President in the past three elections, whereas the Washington Times endorsed the Republican contender in each contest. The vast majority of respondents perceived the correct ideological placement of these newspapers. Fifty-eight percent of our respondents indicate that the Washington Post is liberal on a 7-point scale, while 56 percent indicate that the Washington Times is conservative on a 7-point scale. Namely, the average ideological placement of the Washington Post is 3.1, where 1 is very liberal and 7 is very conservative, and the average placement of the Washington Times is 4.9. In fact, fewer than one in 10 respondents incorrectly place the Washington Post to the right of the Washington Times on the 7-point ideological scale. In a separate question, eight times as many respondents say the Washington Post has a liberal, pro-democratic bias than a conservative, pro-republican bias, whereas three times as many respondents say the Washington Times has a conservative, pro-republican bias as say the paper has a liberal, pro-democratic bias. 27 The Post endorsement is significant at p <.05 level 18

19 metropolitan area. Pro-choice candidates receive about 7 points more support than pro-life candidates. Candidates that favor an assault weapons ban receive about 10 points more support than candidates that oppose it. While a candidate s race does not appear to affect voters decisions overall, respondents exhibit a noted aversion toward candidates with certain occupational backgrounds. They are 15 to 20 points more likely to support a candidate whose previous occupation is a doctor or military veteran compared with a carpenter or car dealer. [Figure 5 about here] Next, we examine whether the impact of Post and Times endorsements varies based on the level of ideological congruence between the newspaper and voters. Figure 6 shows the impact of newspaper endorsements by ideological subgroup. Conservatives are 3 percentage points more likely to support a candidate endorsed by the Washington Times (although this difference falls below conventional levels of statistical significance) and 15 points less likely to support candidates endorsed by the Washington Post, a highly statistically significant difference. The magnitude of this negative effect among conservatives is similar to the effect of a candidate affiliating with the Democratic Party, a candidate favoring the assault weapons ban, or a candidate having a pro-choice position on abortion, and it is considerably larger than the impact of a candidate s race or occupation. Among liberals, a Washington Times endorsement has no effect on candidate support. However, a Washington Post endorsement increases the probability that a liberal will support a candidate by 10 percentage points, a highly significant finding. The magnitude of this positive endorsement effect is about the half the size of the effect of a candidate favoring the assault weapons ban and approaches 19

20 the size of the effect of a candidate taking a pro-choice position on abortion. [Figure 6 about here] Conclusion Newspapers have broad and loyal readerships and commonly endorse candidates in both general and primary elections. This study has examined the circumstances when these political endorsements influence candidate support and election outcomes. Overall, in separate studies among a national sample and in a single large media market, we find that newspapers increase the average level of support for candidates and that endorsements have heterogeneous impacts across groups. Most importantly, we find that the level of ideological congruence between citizens and newspapers affects the impact of the endorsement. Citizens are more likely to support candidates that receive an endorsement when they perceive that the newspaper is ideologically similar to their preferences. However, they are less likely to support the candidate that receives an endorsement when they perceive that the newspaper is ideologically distinct from their preferences. This suggests that newspaper endorsements can function as credible source cues that citizens use as an efficient information shortcut to evaluate the spatial position and valence of a candidate. Future work should examine the mechanisms that underlie the impact of newspaper endorsements. For instance, do newspaper endorsements influence voters perceptions of candidates ideology? Alternatively, perhaps they influence voters perception of candidates underlying valence or quality. Most importantly, our findings show that media endorsements can have a large effect on 20

21 the type of congressional candidates that are elected to Congress. They also suggest that newspapers practice of endorsing both Democratic and Republican candidates in primary and general elections could reduce candidate support. In elections where voters have a low degree of ideological congruence with newspapers, such as Republican primaries in a media market with a left-leaning newspaper, we demonstrate that an endorsement can actually reduce candidate support. As an example, many have argued that the left-leaning Boston Globe s endorsement of former Governor Jon Huntsman in the New Hampshire presidential primary actually hurt Huntsman among voters. Mitt Romney s campaign specifically cited the ideological orientation of the newspaper in its reaction: The Globe has a liberal editorial page, and it s not surprising they would endorse Jon Huntsman. Mitt Romney was pleased to get the endorsement of the more conservative Boston Herald. 28 Our findings also have implications for voting behavior and representation. They suggest that ideological sorting among newspapers may contribute to the growing partisan polarization in the electorate. Indeed, newspaper endorsements may have contributed to the increase in partisan voting in U.S. elections in the past few decades (Bartels, 2000). They may also have contributed to increasingly polarized issue opinions if voters use editorial endorsements to inform their policy preferences. 28 HuffingtonPost, MittRomneyCampaignNotSurprisedBostonGlobeEndorsedJonHuntsman", MichaelCalderone,January6,2012, mitt-romney-boston-globe-boston-herald_n_ html 21

22 Figure 1: This graph shows an example of the conjoint table with various candidate attributes for the national study. Each attribute was fully randomized in our experiment, and each respondent in a treatment group viewed an endorsement from the largest newspaper by circulation in their ZIP Code. 22

23 Figure 2: National Study - Effect of Candidate Attributes on Candidate Support Endorsement: None Newspaper Effects of Candidate Attributes on Candidate Support Among Those Who Read Paper Weekly+ Party: Democrat Republican Independent Issue positions: Pro life Favor Assault Weapon Ban Occupation: Business Owner Car Dealer Carpenter Doctor High School Teacher Lawyer Member of Congress Military Veteran Change in Candidate Support These plots show estimates of the effects of the randomly assigned candidate attributes on the probability of being preferred by respondents. Estimates are based on an OLS model with clustered standard errors calculated at the respondent level; bars represent 95 percent confidence intervals. Results are from a national survey administered in January and March 2014 among a national sample of 2,608 adults conducted via Survey Sampling International, Inc. 23

24 Figure 3: National Study - Effect of Newspaper Endorsements on Candidate Support Endorsement: None (High Congruence) Newspaper (High Congruence) None (Medium Congruence) Newspaper (Medium Congruence) None (Low Congruence) Newspaper (Low Congruence) Panel A: Effect of Endorsement on Candidate Support By Low, Medium, and High Ideological Congruence Among All Voters Change in Pr(Vote for Candidate) Endorsement: None (High Congruence) Newspaper (High Congruence) None (Medium Congruence) Newspaper (Medium Congruence) None (Low Congruence) Newspaper (Low Congruence) Panel B: Effect of Endorsement on Candidate Support By Low, Medium, and High Ideological Congruence Among Weekly Newspaper Readers Change in Pr(Vote for Candidate) Endorsement: None (High Congruence) Newspaper (High Congruence) None (Medium Congruence) Newspaper (Medium Congruence) None (Low Congruence) Newspaper (Low Congruence) Panel C: Effect of Endorsement on Candidate Support By Low, Medium, and High Ideological Congruence Among Those Who Have Read Paper in Past 24 Hours Change in Pr(Vote for Candidate) These plots show estimates of the effects of a newspaper endorsement on the probability of being preferred by respondents. Estimates are based on an OLS model with clustered standard errors calculated at the respondent level; bars represent 95 percent confidence intervals. Results are from a national survey administered in January and March 2014 among a national sample of 2,608 adults conducted via Survey Sampling International, Inc. On a scale from 0 to 1, High Ideological Congruence respondents are 0 or 0.17 points from the newspaper, Medium Ideological Congruence respondents are 0.33 to 0.5 points from the newspaper, and Low Ideological Congruence respondents are more than 0.5 points away from the newspaper. 24

25 Figure 4: This graph shows an example of the conjoint table with various candidate attributes for the Washington, D.C. area study. Each attribute was fully randomized in our experiment, and each respondent viewed either a Washington Post endorsement, a Washington Times endorsement, or no endorsement. 25

26 Figure 5: D.C. study - Effect of Newspaper Endorsements on Preferred Candidate Endorsement: None Washington Post Washington Times Party: Republican Democrat Issue positions: Pro choice Favor Assault Weapon Ban Race: White Black Hispanic Asian Occupation: Business Owner Car Dealer Carpenter Doctor High School Teacher Lawyer Member of Congress Military Veteran Effects of Candidate Attributes on Candidate Support Change in Candidate Support These plots show estimates of the effects of the randomly assigned candidate attributes on the probability of being preferred by respondents. Estimates are based on an OLS model with clustered standard errors estimated by respondent; bars represent 95 percent confidence intervals. Results are from a survey of 550 Washington, D.C. metropolitan area residents conducted via Survey Sampling International, Inc. in Spring

27 Figure 6: D.C. study - Effect of Newspaper Endorsements on Preferred Candidate by Ideology Endorsement: None Washington Post Washington Times Party: Independent Republican Democrat Issue positions: Pro choice Favor Assault Weapon Ban Race: White Black Hispanic Asian Occupation: Business Owner Car Dealer Carpenter Doctor High School Teacher Lawyer Member of Congress Military Veteran Effects of Candidate Attributes on Candidate Support Among Liberals Change in Candidate Support Endorsement: None Washington Post Washington Times Party: Independent Republican Democrat Issue positions: Pro choice Favor Assault Weapon Ban Race: White Black Hispanic Asian Occupation: Business Owner Car Dealer Carpenter Doctor High School Teacher Lawyer Member of Congress Military Veteran Effects of Candidate Attributes on Candidate Support Among Conservatives Change in Candidate Support These plots show estimates of the effects of the randomly assigned candidate attributes on the probability of being preferred by respondents. Estimates are based on an OLS model with clustered standard errors; bars represent 95 percent confidence intervals. Results are from a survey of Washington, D.C. metropolitan area residents conducted via Survey Sampling International, Inc. in Spring 2013.

28 References Acito, Franklin An Investigation of the Reliability of Conjoint Measurement for Various Orthogonal Designs. In Proceedings, Southern Marketing Association 1979 Conference ( ). Lafayette: University of Southwestern Louisiana. Arceneaux, Kevin, and Martin Johnson Changing Minds or Changing Channels?: Partisan News in an Age of Choice. University of Chicago Press. Arceneaux, Kevin, and Robin Kolodny Educating the Least Informed: Group Endorsements in a Grassroots Campaign. American Journal of Political Science 53(4): Bartels, Brandon L., and Christopher D. Johnston On the Ideological Foundations of Supreme Court Legitimacy in the American Public. American Journal of Political Science 57(1): Bartels, L. M Beyond the Running Tally: Partisan Bias in Political Perceptions. Political Behavior 24(2): Bartels, Larry M Partisanship and Voting Behavior, American Journal of Political Science 44(1): Berelson, Bernard R., Paul F. Lazarsfeld, and William N. McPhee Voting: A Study of Opinion Formation in a Presidential Campaign. Chicago: University of Chicago Press. Berinsky, Adam, Michele Margolis, and Michael Sances Separating the Shirkers from 28

29 the Workers? Making Sure Respondents Pay Attention on Internet Surveys. American Journal of Political Science. Brady, H. E., and P. M. Sniderman Attitude Attribution: A Group Basis for Political Reasoning. The American Political Science Review pp Campbell, A., P. Converse, W. Miller, and D. Stokes The American Voter. University of Chicago Press. Chiang, Chun-Fang, and Brian Knight Media Bias and Influence: Evidence from Newspaper Endorsements. The Review of Economic Studies 78(3): Delli Carpini, Michael X., and Scott Keeter Measuring Political Knowledge: Putting First Things First. American Journal of Political Science pp Dilliplane, Susanna Activation, Conversion, or Reinforcement? The Impact of Partisan News Exposure on Vote Choice. American Journal of Political Science 58(1): Druckman, J. N On the Limits of Framing Effects: Who Can Frame? Journal of Politics 63(4): Druckman, J. N., and M. Parkin The Impact of Media Bias: How Editorial Slant Affects Voters. Journal of Politics 67(4): Erikson, Robert S The Influence of Newspaper Endorsements in Presidential Elections: The Case of American Journal of Political Science 20(2): Gaines, B. J., J. H. Kuklinski, P. J. Quirk, B. Peyton, and J. Verkuilen Same Facts, 29

30 Different Interpretations: Partisan Motivation and Opinion on Iraq. Journal of Politics 69(4): Gerber, Alan S, Dean Karlan, and Daniel Bergan Does the Media Matter? A Field Experiment Measuring the Effect of Newspapers on Voting Behavior and Political Opinions. American Economic Journal: Applied Economics 1(2): Green, Paul E., and Venkatachary Srinivasan Conjoint Analysis in Consumer Research: Issues and Outlook. Journal of Consumer Research pp Gustafsson, Anders, Andreas Herrmann, and Frank Huber Conjoint Measurement: Methods and Applications. Springer. Hainmueller, J., and D. Hopkins The Hidden American Immigration Consensus: A Conjoint Analysis of Attitudes Toward Immigrants.. Hainmueller, Jens, Daniel Hopkins, and Teppei Yamamoto Causal Inference in Conjoint Analysis: Understanding Multidimensional Choices via Stated Preference Experiments. Political Analysis 22(1): Hollander, Barry A Tuning Out or Tuning Elsewhere? Partisanship, Polarization, and Media Migration from 1998 to Journalism & Mass Communication Quarterly 85(1): Hollander, Sidney On the Strength of a Newspaper Endorsement. The Public Opinion Quarterly 43(3):