The Effect of Ethnic Neighborhoods on Campaign Appeals: How Racial Context Interacts with Campaigns Messaging to Affect Public Political Behavior

Transcription

1 The Effect of Ethnic Neighborhoods on Campaign Appeals: How Racial Context Interacts with Campaigns Messaging to Affect Public Political Behavior Hans Hassell Cornell College Department of Politics 600 First St. SW Mount Vernon, IA Abstract Current trends suggest that in less than 30 years the United States will have a majority non-white population. These changes in racial demography have been shown to have real consequences on political attitudes and private voting behaviors. This paper argues that race also affects public political behaviors of white Americans and their responsiveness to campaign appeals. Using data from a congressional primary campaign, I examine the responses of white Republicans asked to display a yard sign of a white Republican running against a Latino Republican. Consistent with theories of racial threat, whites in Latino neighborhoods were more likely to request a sign. Using a field experiment run by the campaign, I show that the effectiveness of different campaign appeals are influenced by racial neighborhood context. Racial diversity plays a significant role in the public political behaviors of white Americans and how racial environments affect underlying behavioral motivations. Abstract Word Count: 147 Word Count (including references): 11,044

2 1 Official governmental estimates project that by as early as 2044 the United States will be a majority-minority country, with white Americans making up less than 50 percent of the US population (Frey 2014). Much of the discussion about the political implications of this demographic transformation has focused on how this shift will change the balance of power in the US. While minority populations, especially recent immigrants, are not necessarily locked into the Democratic party (Hajnal and Lee 2011), they are traditionally considered (with some small exceptions) to be more supportive of progressive policies and Democratic party priorities. 1 However, these changes in racial demographics not only affect the partisan balance of power by changing the composition of the electorate, they also affect the balance of power by influencing the political identities of white Americans. Racial demographics and increased salience of immigration influence white partisanship both on the individual (Bowler, Nicholson, and Segura 2006; Craig and Richeson 2014b; Hajnal and Rivera 2014) and macro level (Abrajano, Hajnal, and Hassell 2017) by shifting white support away from Democrats and toward the Republican party. The racial environment in which an individual lives has strong effects on the public opinions of white Americans by eliciting the perception of a racial threat (Abrajano and Hajnal 2015; Key 1949). The presence of large minority populations cause whites to place a higher priority on issues important to racial minorities (Abrajano and Hajnal 2015; Bobo and Johnson 2002; Fossett and Kiecolt 1989; Hopkins 2010; Key 1949), to take more conservative stances on those policies (Craig and Richeson 2014b; Hopkins 2009b, 2010), and to express more implicit 1 Given recent Republican Party politics and President Trump s use of inflammatory rhetoric directed at minorities, this trend of minorities leaning toward the Democratic Party may increase substantially. Similar previous rhetoric has significantly affected a political party s ability to appeal to minority voters (e.g. the rhetoric surrounding Prop 187 in California) (Segura, Falcon, and Pachon 1997).

3 2 and explicit racial bias (Craig and Richeson 2014a). Likewise, the larger the population of African-Americans or Latinos in a community, the more likely whites are to turn out to vote and to vote more conservatively (Abrajano and Hajnal 2015; Enos 2016; Giles and Buckner 1993; Hill and Leighley 1999; Key 1949; Leighley and Vedlitz 1999, but see Taylor 1998; Voss 1996). While we know that the racial environment has a strong effect on the political identities, attitudes, and voting behaviors of white Americans (all of which may not be obviously visible to neighbors and acquaintances), we know less about the interaction between the racial environment and public political behaviors. While race is a key component of political identity and shapes behavior, individuals may alter behaviors in to conform with social expectations especially when these behaviors are publicly visible (Berinsky 2004; Hopkins 2009a; LaPiere 1934). This paper seeks to understand how the racial environment also affects public political behaviors and the influence of campaign appeals in politics. Specifically, does the environment in which an individual lives affect the ability of campaigns and other political entities to mobilize white Americans to engage in public political behavior and does it change the effectiveness of certain types of campaign appeals? In this paper, I examine how the racial makeup of the neighborhood in which an individual lives affects the willingness of white respondents to publicly broadcast their support of a white candidate running against Latino candidates in a Republican primary election in Using a field experiment carried out in collaboration with the campaign, I examine the willingness of individuals to agree to place a yard sign in front of their home in support of the white candidate. Moreover, I also examine how campaign messages interact with the racial environment to condition the responses of whites and their public political behaviors. The campaign varied the

4 3 treatments in order to determine how the effect of appeals to values and appeals that remind voters to consider their neighbors in their decision to act varies by neighborhood racial demographics. The effect of these two frames, I contend, is conditional on the racial makeup of the subjects precinct. First, examining the overall response of whites to the campaign s request, I find that the demographics of respondents neighborhoods are correlated with the willingness of individuals to post a yard sign in their yard in support of the white candidate. Consistent with theories of racial threat, as the percentage of non-white residents in the neighborhood increases, whites are more likely to be willing to publicly express their views in support of the white candidate running against Latino opponents. Second, I also find that subjects living in homogenous neighborhoods, whether largely white or largely non-white, were more likely to respond to campaign appeals designed to prompt respondents to consider their neighbors in their motivations to place a sign. Whites living in heterogeneous neighborhoods were less likely to respond to appeals that advocated respondent action in order to communicate to their neighbors. Prompting whites to think of their neighbors in communities where they are in the minority prompts considerations of racial threat. Prompting whites to think of their neighbors in largely homogeneous white areas prompts considerations of solidarity (Campbell 2006). For whites in heterogeneous communities, the effect of both these considerations is smaller or possibly non-existent. Reminding people about their neighbors is most effective at increasing participation when the community is easily identifiable as an ingroup or an out-group.

5 4 Lastly, I find that the racial environment has a significant effect on the effectiveness appeals to values in mobilizing white Americans. I show that residents in different racial neighborhoods respond differently to appeals to values compared to a standard control appeal. The results of the campaign conducted experiment show that appeals to values are most effective at recruiting white respondents when respondents live in neighborhoods that are largely nonwhite. Compared to the baseline treatment, white voters in high density Latino neighborhoods were more likely to agree to post a yard sign when the campaign included references to the candidate s and the respondent s shared values. 2 All together, these findings indicate that racial and ethnic environments have a strong effect on public political behaviors and the ability of campaigns to utilize certain types of appeals. These findings show that the effect of race and racial demographics is not limited to voting and other actions taken privately that are not publicly visible, but rather that race and the racial environment have a substantial effect on how white Americans participate publicly in politics. Racial Threat and Political Participation Theories of racial threat date back to V.O. Key s (1949) observation that white turnout and support for conservative politicians in the South was correlated with the population of African-Americans in the community. Key argued that this racial threat motivated the political engagement of white Americans and that whites became more politically involved and ideologically conservative because they were threatened by the presence of African-Americans. 2 While these appeals were intentionally designed by the campaign to subliminally remind voters of the racial differences between the primary candidates, it is not entirely clear that they had the effect of being implicit racial appeals (Banks and Bell 2013; Mendelberg 2001). Thus, in this paper I refer to them as value appeals.

6 5 More recent explorations of race and political behavior have likewise found similar effects for the presence of Latinos in the community (Abrajano and Hajnal 2015; Hopkins 2010) The perception of threat has a significant effect on an individual s policy preferences. In the aftermath of terrorist attacks, or in situations where mortality is made salient, individuals are more likely to endorse conservative policies and support conservative leaders (Nail et al. 2009). These effects, however, are not limited to physical threats (Cottrell and Neuberg 2005). Race and ethnicity are key constructs in the identification of in-group and out-group perceptions which contribute to perceptions of threat (Kam and Kinder 2012). Indeed, since Key s finding over a half century ago, scholars have continued to find evidence that the presence of minority populations causes whites to support conservative policy positions (Abrajano and Hajnal 2015; Gilens 1995; Hopkins 2010; Newman 2013). In addition to holding more conservative political policy positions, whites also respond to an increased racial threat by becoming more conservative in their self-reported ideology and partisan identification (Abrajano et al. 2017; Craig and Richeson 2014a, 2014b). As the perception of a threat from individuals unlike themselves increases, white citizens are also more likely to become politically active in ways that allow them to privately express their political opinions. The presence of large populations of African-Americans increases white voter turnout in the area (Enos 2016; Giles and Buckner 1993; Hill and Leighley 1999; Key 1949; Leighley and Vedlitz 1999; Matthews and Prothro 1963), although other studies have questioned this relationship (Carsey 1995; Taylor 1998; Voss 1996). Yet, while the connection between the racial homogeneity and private political behavior is well established, the connections between the racial environment and public political behaviors have not been investigated.

7 6 The Link between Public Political Behaviors and Private Attitudes While race and racial cues can have significant effects on opinions and private behaviors such as voting, how people act publicly may be different from how they respond when asked their opinions or even how they vote in the privacy of the ballot booth. Voting and identifying with a party are not the only way that citizens participate (Brady, Verba, and Schlozman 1995) and influence political outcomes (Bergan 2009; Bergan and Cole 2014). Recent work has even shown that the placement of yard signs can have a small but significant effect on election outcomes especially in low profile races such as primaries (Green et al. 2016; Kam and Zechmeister 2013). Thus, it is important to know whether opinions and unobservable behaviors follow the same pattern as publicly observable behaviors. Some argue that individuals responses to questions about opinions or preferences are different than their actions. Richard LaPiere s (1934) seminal study on the attitudes and actions of hotel and restaurant purveyors towards Chinese-Americans found individuals expressed opinions fundamentally different from their actions. In a like manner, other scholars have shown that individuals often lie or refuse to respond publicly when they hold socially unacceptable views (Berinsky 1999, 2004; Hopkins 2009a). Thus while individuals may espouse one set of opinions, their public behaviors may be entirely different. On the other hand, more recent work has found that considerations of race affect political behaviors in the same way that they affect political opinions and that injecting race into consideration of public policy affects both opinions about the issue and actions in support or opposition to the issue (Hassell and Visalvanich 2015). Indeed, in many ways public political expression is similar to other more private forms of political expression. Private political

8 7 participation is a means of self-expression in society, and as such, social constructs and social cues affect turnout (Garcia Bedolla and Michelson 2012; Popkin 1994; Rogers, Fox, and Gerber 2012), party identification (Berelson, Lazarsfeld, and McPhee 1954; Campbell et al. 1960; Klar 2014; Popkin 1994; Sinclair 2012), campaign giving (Hassell 2016; Sinclair 2012), and public opinion (Klar 2014). Social constructs also play a key role in public activist behaviors. Decisions to become publicly involved in a campaign or movement are often driven by intrinsic motivations that define participants and provide a sense of self-worth (Han 2014). Campaign Actions and their Effects on Public Behavior The social political environment affects individual motivations to participate. In environments where they are in the majority, people tend to participate for social solidarity reasons (Campbell 2006) or to conform to others that are like them (Cialdini et al. 2006; Gerber and Rogers 2009). Campaigns that act to increase social pressure in these environments where the individual is similar to others in the neighborhood has large effects on political views (Sinclair 2012), political action (Gerber, Green, and Larimer 2008; Panagopoulos 2010; Sinclair 2012), and even partisan identification (Klar 2014; Sinclair 2012). In contrast, in environments where the individual is in the minority and noticeably different from those in the community, participation is motivated by political considerations. In these environments participation is not an act meant to generate solidarity but rather a politically motivated act (Campbell 2006). Specifically looking at the act of placing a yard sign, Makse and Sokhey (2014) build off this previous work and argue that the relative heterogeneity of the political context changes the underlying motivations for the placement of yard signs. For individuals in homogeneous environments where they are in the majority, the public display of support for a candidate is

9 8 motivated by a desire to relate to others similar to themselves. In contrast, for individuals living in neighborhoods that do not reflect their own political preferences the motivation to publicly state political preferences is driven by a desire to disseminate information and persuade others. While not specifically testing individual motivations for placing or not placing a sign, Makse and Sokhey examine the explanations given for posting a yard sign by individuals who did place signs. Their findings however, are mixed. While individuals living in more politically homogeneous neighborhoods, as measured by presidential vote share, were less likely to indicate that they placed signs to let their neighbors know, a phrase they argue reflects communicative motivations. Yet, at the same time an increased volume of same party signs in the immediate vicinity increased the likelihood of respondents indicating that they placed signs to let their neighbors know where they stood. In short, these contradictory findings indicate that it is not entirely clear how the neighborhood political environment shapes public displays of political advocacy. 3 Another alternative explanation is that the phrase let your neighbors know measures communicative motivations when individuals reside in locales where they are in the political minority and at the same time reflects an act of solidarity with similarly minded neighbors in homogenous neighborhoods. In communities where their preferences align with their neighbors, respondents could want their neighbors to know about their political preferences in order to better fit in with their community. In contrast, in communities where their preferences are not in the majority, respondents could want their neighbors to know their political preferences to disseminate information to their neighbors and persuade them. Thus, focusing respondents on the 3 It is fair to note as well, however, that these tests could be seen as testing the effect of different levels of context as well (personal networks as opposed to neighborhood communities). Regardless, these findings do not provide a consistent message about the effect of the political environment on public displays of political advocacy.

10 9 effect of communicating their preferences to their neighbors could motivate action among both those in homogenous like-minded neighborhoods and also those in neighborhoods where they are in the political minority. Lastly, we know little about how the rhetoric of campaigns interacts with the racial demography to influence political behavior and participation. Specifically, the use of values as a part of a call to action is not uncommon among campaign messaging. Indeed, using data from the s sent out by campaigns from a random selection of congressional districts in the 2012 congressional elections, I find that candidates appeal to shared values in just under 10 percent of all s sent. Moreover, we should expect that the use of values as an appeal to become involved in a campaign might become a more prominent aspect of political campaigns as these sorts of appeals have often been used in previous campaigns in opposition to out group threats. In short, appeals to values have become emblematic of the culture wars that have been a major cleavage in American politics over the past few decades (Layman 2001). Expectations and Hypotheses As outlined above, theories of racial threat posit that whites who reside in communities with larger minority populations are more likely to become engaged with politics and to hold more conservative positions. Although these findings have focused entirely on private political behaviors and have not been tested with public political behaviors, theory leads us to believe that these findings will hold in the same way for public political behaviors as they do for private public behaviors. This leads us to the first hypothesis: Hypothesis 1: As the percentage of the minority population increases, whites will be more likely to engage in public political behaviors.

11 10 Second, I also examine the effect of priming individuals to think explicitly about the neighborhood in which they live in their decision to engage in public political behaviors. Because the kinds of people that individuals think about when prompted to think about their neighbors varies, again I hypothesize that the effect of this prompt will vary by neighborhood. Since individuals in white homogenous neighborhoods are more likely to express solidarity with their neighbors, prompting individuals in homogenous white neighborhoods to think about their neighbors should increase participation; this expectation leads to the third hypothesis: Hypothesis 2: Whites in homogeneous white neighborhoods will be more likely to engage in public political behaviors when primed to consider their neighbors. We next turn to whites living in homogeneous minority neighborhoods. Individuals living in areas where they are in the minority are more likely to participate for political motivations (Campbell 2006). These individuals are motivated to participate to prevent what they perceive as negative political outcomes that are the result of their neighbors preferences. Whites living in neighborhoods where they are the racial minority who are prompted to think about their neighbors should also be more likely to engage in public political behaviors because they have a greater perception of political and racial threat. This motivates the fourth hypothesis: Hypothesis 3: Whites in homogeneous minority neighborhoods will be more likely to engage in public political behaviors when primed to consider their neighbors. This leaves us with those individuals residing in heterogeneous neighborhoods where there is a relative balance of whites and minorities. In these instances, when prompted to post a yard sign to tell their neighbors about their political preferences, it is unclear whether these residents are motivated by solidarity or political threat. With neither a large threat nor a large community with which to identify, it is possible that neither motivation is as strong as it is in the homogeneous white or homogenous minority neighborhoods. As such, appeals to engage in

12 11 public political behaviors in heterogeneous neighborhoods should not be as effective because the encouragement of political motivations or solidarity motivations is not as powerful as it is in homogenous environments. For those respondents living in precincts that are largely homogeneous and white, this should prompt them to consider their white neighbors. For respondents living in precincts with a low percentage of white voters, this should raise considerations of race and how the respondent is different racially from the rest of the neighborhood. For racially mixed precincts, the neighbors cue prompts neither threat nor solidarity. Hypothesis 4 Individuals in heterogeneous neighborhoods will be less likely to engage in public political behaviors when primed to consider their neighbors. Lastly, I also examine the interaction between appeals to values and the neighborhood in which the respondent lives. The use of political threats is generally most effective when those threats are more easily perceive. The effective priming of these cues will be most effective in contexts where the perceived differences in values are greatest. Thus, these appeals to values appeals should be most effective in areas with a higher presence of minorities, where individuals will be more likely to encounter the context necessary to prime those cues. Hypothesis 5: Appeals to values will be more effective at mobilizing whites in areas with higher percentages of minority residents Data and Research Design To test these hypotheses and examine how racial neighborhoods affect public political behaviors and interact with campaign appeals to influence public political participation, I took advantage of a campaign of a white candidate for Congress running in a competitive Republican primary in 2014 against other Latino Republican candidates to challenge an incumbent Latino

13 12 Democrat in a relatively competitive district. The median income in the congressional district is slightly above the US median income, but most importantly, the district in which the primary took place is a majority Latino district. Within the congressional district there was also wide variation in the relative racial diversity of precincts. The mean percentage of registered voters who were white in each precinct ranged from just below 10 percent to almost 98 percent. 4 Just like African-Americans, white Americans feel threatened by Latinos (Abrajano and Hajnal 2015; Chavez 2008). Studies of immigration coverage provide a growing evidence that the media overwhelming focus on an immigrant threat narrative that links immigration to economic costs, social dysfunction, illegality, and cultural decline (Abrajano et al. 2017; Dunaway, Branton, and Abrajano 2010). Whites feel the same racial threat and react to Latino populations and react with increasing conservatism and backlash against those communities (Abrajano and Hajnal 2015). The candidate was a quality challenger who had previous political experience as an elected city official, was the only non-hispanic candidate in the district, and had a clearly identifiable white name. Because of his status as the lone non-hispanic white candidate, race implicitly became an issue that infused the campaign and the campaign s message. The campaign also made strategic decisions to highlight the ancestry of the candidate s opponents and their stances on various policies related to the Latino community and immigration in the literature it distributed. When questioned about the overt focus of the campaign s negative message on his opponents heritage, the candidate explained that his feeling was that the election seemed to be too focused on Latino heritage and values and should instead be about America. 4 The average precinct was 40 percent white with a standard deviation of roughly 25 percentage points.

14 13 However, it is important to note that the campaign, recognizing that whites made up a minority of the district s Republican voters, did not exclusively target white voters. The campaign produced several direct mail pieces and television ads in Spanish that were targeted to appeal to Latino voters. 5 Experimental Treatments To mobilize voters leading up to the Republican primary in August of 2014 the campaign used a small in-house paid phone bank as part of its efforts to contact voters, encourage support of the candidate in the upcoming primary, and place yard signs. 6 This phone bank targeted all non-hispanic registered Republican voters in the congressional district who had not already requested an absentee ballot and had voted in one of the last five elections including primary and statewide special elections. 7 Calls were made over a two-week period about three weeks before the primary. The contact rate for the study was 13 percent, generating a total of 760 respondents. It is important to recognize that the sample of respondents used here (white Republicans in one congressional district) perhaps limits the ability to draw wide conclusions about the 5 The mailers to Latino voters (and some of the advertisements) printed in Spanish were attacks on the other candidates. Given some perceptions that negative campaign tactics drive down turnout, it could be that the campaign was attempting to demobilize those individuals rather persuade Latinos to vote for the candidate. However, recent evidence from field experiments examining the effect of negative campaign mailers shows that these mailers actually increase the respondents intentions to vote, although these effects are not long-lasting (Doherty and Adler 2014). 6 The campaign phone bank employed three callers, all women in their late teens and early twenties, supervised directly by the campaign manager. 7 Although the campaign did not differentiate between white and other non-hispanics, the sample was almost entirely white. Less than two percent of the population in the congressional district is Asian. Likewise African-Americans made up a similarly small portion of the district and the scarcity of Blacks identifying with the Republican Party make it so that the sample of respondents was likely almost entirely white. Non-Hispanics were identified by the campaign through the voter file and additional proprietary information.

15 14 effects of racial demographics on the public political behaviors of all Americans. Republicans as a group are more likely to have higher levels of racial resentment and as such may be more susceptible to priming of racial threats (Feldman and Huddy 2005). However, there is also strong reason to believe that these effects would be similar or even greater elsewhere. The effects of racial cues on political behaviors beyond holding opinions are evident in experiments using more liberal and younger samples (Hassell and Visalvanich 2015). Evidence of the effect of racial threats on changing partisan identification are actually stronger among Democrats and Independents than they are for Republicans (Abrajano et al. 2017; Sniderman and Piazza 1993), and negative racial stereotypes are more effective when used against Democratic candidates than against Republican candidates (Stephens-Dougan 2016). Thus, if previous work suggests that Democrats and Independents are more responsive to the effects of racial neighborhoods and implicit racial appeals and that these appeals work better against Democratic candidates, the findings presented here offer a conservative test of the hypothesis. In addition, using a Republican primary also has the inherent advantage of eliminating partisan cues which are one of the major confounds of previous work on the effect of racial cues. Because of the belief that the treatment effect of the various appeals varies by the demographics of the precinct where the respondent lives, this analysis uses a randomized block design by precinct to ensure adequate representation of treatments in precincts with high, medium, and low white populations. Precincts were divided into three groups by the percentage of white registered voters who made up the precinct. 8 Because our hypotheses suggest different outcomes for homogeneous white precincts, heterogeneous precincts, and homogeneous Latino precincts, treatments were randomized by block to ensure balance at each level. The first block 8 The campaign provided the information on the percentage of Latinos in each precinct from their voter file.

16 15 contained precincts where the white registered voters made up zero percent to 33 percent of the population of registered voters. The second and third blocks contained precincts with 34 to 67 percent white voters and 68 to 100 percent white voters, respectively. Of the 760 respondents, 185 were in the high-density Latino precincts, 210 were in the heterogeneous precincts, and 365 were in the homogenous white precincts. 9 Within each of these blocks all of the voters within each precinct were randomly assigned to receive one of the three scripts. Hotelling balance tests using age and gender indicated that there were no significant differences between the groups. Although there may be a disconnect between formal jurisdictions such as a precinct or other official government administrative unit and the neighborhood in which an individual lives (Wong et al. 2012), precinct racial demographics provide a reasonable measurement strategy for measuring racial heterogeneity. Precincts have been used extensively as a standard unit in other studies that are interested in the racial demographics and their effect on behavior (Carsey 1995). Some scholars have argued that some findings detailing the effect of local contexts have cherrypicked geographical units to support their own hypotheses. These scholars have argued for the need to run multiple checks using more than one geographical unit (Tam Cho and Baer 2011). Indeed, there is some evidence that across smaller geographic units there is no relationship between minority population and white racial attitudes (Oliver and Mendelberg 2000). To guard against the possibility that the results presented here are an artifact of the arbitrary geographic unit, I also matched respondents addresses to their census tract. The models shown below were then re-run using the percent of the census tract that was white as an alternative to precinct 9 Because we analyze neighborhood racial demographics on a continuous level and were not analyzing the responses by blocked group, we split the groups by percentage rather than dividing the groups equally. The purpose of the blocking was to ensure proper representation of the full range of racial precincts in each treatment group.

17 16 demographics. Using census tract rather than precinct does not change the results reported here. These alternative specifications using census tract are available in the online appendix. One other alternative explanation is that it is not racial demography that impacts the decision making of individuals to display yard signs, but rather the partisan environment. This is not the case. The online appendix also includes models that show the effect of the different appeals interacted with the 2012 presidential vote. There is no significant effect of partisan neighborhoods on the willingness to display a yard sign in this Republican primary. 10 The campaign used three different scripts to place yard signs. The full phone script is available in the online appendix. The scripts were all the same with the exception of the request at the end of the script encouraging individuals to put a yard sign in their yard supporting the candidate. The campaign s standard phone script prompted respondents to consider the American values (in contrast to unspoken Latino values) and asked respondents if they would be willing to post a sign in their yard to express your support of our candidate [Candidate Title and Surname] and our values? This treatment intentionally attempted to play off implicit racial associations between the shared values of white voters and the candidate. Although the names of the candidate s Latino primary opponents were not mentioned in the script, the script did state explicitly that the candidate was the best candidate to defeat the incumbent, who also had a distinctly Latino surname. The name of the candidate sponsoring the phone call was also unequivocally white. The campaign chose the message specifically to highlight American values 10 Nor is there any effect on the substantive results shown here if both partisan demographics and racial demographics are included.

18 17 to cue the racial aspects of the campaign and the ethnicity of the candidate s opponents, both in the primary and in the general election. 11 The second version of the text asked respondents if they would be willing to post a sign in their yard to let your neighbors know you support [Candidate Title and Surname]? This alternative treatment directly measures the effect of considerations of the racial neighborhood on political action by making neighborhood salient in the minds of the respondents when they are considering the request. For those respondents living in precincts that are largely homogeneous and white, this should prompt them to consider their white neighbors. For respondents living in precincts with a low percentage of white voters, this should raise considerations of race and how the respondent is different racially from the rest of the neighborhood. For racially mixed precincts, the neighbors cue does not prompt considerations of homogeneity, but rather that of a diverse group of neighbors. The last version of the text acted as a control and only asked respondents if they would be willing to post a sign in their yard to express your support of [Candidate Title and Surname]? Without the appeal to values or references to neighbors, this version acts as a control and a way to gauge the baseline response when individuals are asked by the campaign to post a sign in their yard. Yard signs were delivered by the campaign to individuals who indicated they were willing to post a yard sign in their yard within a week of the phone calls. 11 While the campaign intended the message to be an implicit racial cue, it is not entirely clear whether the cue raised the same subliminal messages about race that other implicit racial primes that political science has studied in the past. Thus for this purpose, this article refers to it as a values appeal.

19 18 Differentiating between Treatments One concern is that the differences between the values treatment and the control are slight enough so that they are indistinguishable between each other and might not be sufficient to work effectively. It is important that there be substantive differences between the treatment and the control so that the appeals can prompt different considerations when responding. While it would be nice to have a manipulation check built into the process, this was not feasible to do in conjunction with the campaign s desire to maximize votes and the placement of yard signs. Thus, to test whether the treatments were distinguishable from each other, I tested these scripts with 204 respondents (145 white respondents) recruited on Amazon Mechanical Turk. Respondents were randomly shown either the control script or the implicit script. To gauge whether individuals could differentiate between the two scripts respondents were also asked to write a response to the question If you were to support this candidate what reasons would you give? Responses to this question were then coded to identify whether the respondent had indicated shared values in the text response. A significantly higher percentage of respondents who received the implicit script treatment indicated that if they were to support the candidate, it would be because of values. Less than 1% of respondents (less than 1% of white respondents) in the control group used language referring to values compared to 8.8% of respondents (8.0% of white respondents) who received the implicit treatment (a difference significant at p<.01 (p<.05 for white respondents), on a one-tailed test). 12 This is not a large percentage overall, but it is important to remember that there were a number of other 12 Because the hypothesis is a directional hypothesis (we would not expect more respondents in the control group to mention values more often than those in the values treatment) I use a onetailed test here. All others are two-tailed tests.

20 19 characteristics including past political and military experience that were more salient in the script and might be perceived as more legitimate reasons for supporting a candidate. 13 In addition, to show that respondents recognized the use of values as an attempt by the candidate to appeal to voters in the values script, respondents were asked explicitly whether the script equated supporting our values with supporting the candidate. To not contaminate the openended responses, this question was asked on a separate screen after respondents were asked about their reasons for supporting the candidate. Almost 80% of those individuals in the implicit treatment group (81% of whites in that group) compared to less than 60% in the control group (63% of whites in that group) indicated that the script did equate supporting values with the support of the candidate, a difference significant at p<.01 (p<.01 for whites only), on a one-tailed test. While there are obviously differences in the sample of respondents and differences between reading the script on Mechanical Turk and hearing the script in a phone call in the field experiment, these results do indicate that individuals can find substantive and meaningful differences between the two scripts. 14 The Effects of Racial Context on Public Political Behaviors I begin by examining the overall effect of the racial demographics of the precinct on the likelihood of a respondent being willing to put up a yard sign. Of those reached, 16 percent of 13 In addition, it is likely that values (especially the clearly conservative values identified in the script) are a more salient issue and motivating factor for conservatives than they would be for liberals thus making any differences here smaller than what we would expect in the field experiment. 14 Hearing the script may make the values appeal even less subtle because the values cues were at the end of the script as part of the campaign s request to put a yard sign in the respondent s yard. Without the ability to review the script, respondents might be more likely to rely on what they last heard and is on the top of their head.

21 20 individuals indicated that they would put up a yard sign that would be delivered to them by the campaign. 15 There were, however, big differences in the willingness of individuals to place a yard sign by precinct demographics. The first two models in Table 1 predict the likelihood of a respondent being willing to place a yard sign. The first model contains only those individuals in the control group using a linear model. This allows the examination of the overall effect of neighborhood effects on the likelihood of posting a yard sign without the treatments. Because predictions for the interaction of some of the other treatments with the racial demographics are non-linear, the second model examines a model with a quadratic term for the racial demographics. Consistent with Hypothesis 1, as the percentage of the white voters as a percentage of all registered voters in the precinct increases, respondents are less likely to be willing to place a yard sign in their yard supporting the white candidate in the race While we did not specifically measure compliance with the request, the campaign personally delivered each sign the week following the phone contact. 16 The quadratic model suggests that the likelihood of placing a yard sign remains relatively stable until whites made up a sizeable percentage of the population. Similarly, a non-parametric examination of the results using a LOWESS with a bandwidth of.2 suggests that the decline of respondents willingness to post yard signs begins at the point where 25% of the population is white. Standard LOWESS bandwidths of.8 suggest a more linear decline consistent with model 1 in Table 1.

22 21 Table 1: Probability of Posting a Yard Sign by White Percentage of Population in Precinct Control Group Only Full Sample Past Non-Primary Voters Past Primary Voters (1) (2) (3) (4) (5) Post Sign Post Sign Post Sign Post Sign Post Sign Percent White ** (0.630) (4.205) (4.176) (4.349) (6.480) Percent White Squared * * * * (3.682) (3.657) (3.930) (5.478) Values Cue * (1.152) (1.337) (2.081) Neighbors Cue 3.029* 3.509* (1.232) (1.710) (1.989) Implicit Race Cue * Percent White (5.000) (5.408) (8.180) Implicit Race Cue * % White Squared (4.460) (4.895) (7.053) Neighbors Cue * % White ** ** (5.835) (7.034) (8.750) Neighbors Cue * % White Squared ** ** (5.178) (6.138) (7.540) Constant * ** ** ** (0.373) (0.966) (0.959) (1.112) (1.614) Observations Pseudo R-Squared Log-Likelihood Logit Coefficients. Standard errors clustered by precinct in parentheses. ** p<0.01, * p<0.05, two-tailed tests. Figure 1 graphs the predicted probabilities of taking a sign as the population becomes more homogeneous using the linear model. White Republicans living in minority precincts are much more likely to post a sign in their yard in support of the white candidate. A ten percent increase in the white percentage of all registered voters in the precinct corresponds with roughly

23 22 a 2 percent decrease in the likelihood of a respondent indicating a willingness to post a sign. Stated differently, if 100 percent of the registered voters in a precinct were white, then the percentage of white registered Republicans being willing to post a sign would be 21 points lower (30 percent compared to 9 percent) than in precincts where the percentage of white registered voters approached zero. Figure 1: Probability of Posting a Yard Sign Note: Shaded areas represent 95% confidence intervals The higher response rate of white Republicans living in precincts where there is a higher concentration of minorities strongly supports Hypothesis 1 indicating a strong effect of the perception of a minority threat on the willingness to post a yard sign. White voters living in neighborhoods with a larger percentage of minority voters are more perceptive of a threat from this group and are more likely to take public political action to counter the perceived threat.

24 23 Cuing Neighbors The third model in Table 1 examines the effect of the treatments in comparison to the baseline model. Figure 2 looks at the marginal effect of the neighbors prime by precinct demographics relative to the baseline prompt taken from Model 3 in Table 1. As predicted in Hypothesis 2, relative to the baseline prompt, when prompted to consider their neighbors as a motivation for posting a sign white respondents are more likely to agree to post a sign in their yard in support of the white candidate when they live in a precinct that is homogenously white. Likewise, consistent with Hypothesis 3 respondents living in homogenous minority precincts are also more likely to be willing to post a yard sign compared to the baseline prompt. At the same time, respondents living in heterogeneous precincts are marginally less likely to agree to post a sign in their yard when asked to do so to tell their neighbors they support the white candidate. Put together this evidence provides strong support that prompting individuals to consider their neighbors has the biggest effect when those neighbors have a consistent identity. Figure 2: Marginal Effect of Neighbors Prompt Relative to Baseline

25 24 One potential problem with the results from the model above could be that the model forces a quadratic fit onto what could be a linear outcome for the baseline appeal. As noted in Table 1, the model without the quadratic term shows significant effects for a simple linear model. To allay concerns that the findings noted above are not the result of overfitting, we can look at the results within each randomized block. Table 2 provides the differences between the treatment prompting respondents to consider their neighbors and the control in each of the blocked groups. Table 2: Likelihood of Posting a Yard Sign by Neighborhood Diversity Neighbors Treatment Effect of Neighbors Treatment Control (N) (N) Homogeneous Latino (0% to 33% White) 21.21% % % Heterogeneous (33% to 66% White) 21.95% % % Homogeneous White (67% to 100% White) 5.60% % %** Two-tailed t-test: ** p<0.01, * p<0.05 The findings in Table 2 support the same findings predicted by Figure 2. Although the findings regarding the effect of the neighbors cue in homogeneous Latino neighborhoods do not match, this is because the effect shown in Figure 2 is greatest in neighborhoods with very few non-hispanic whites (<20% white) whereas the blocks shown in Table 2 include all respondents living in precinct that were up to 33% white. 17 Table 2 also shows that prompting voters to think of their neighbors is most effective when their neighbors are homogenous, regardless of whether those neighbors are similar or different to the respondent. Likewise, there is also a strong negative effect for the neighbors appeal in heterogeneous neighborhoods, although the 17 The effect of prompting whites to consider their neighbors in when considering whether to accept a yard sign in precincts where whites make up less than 20 percent of the population is 12%, but does not reach standard levels of statistical significance due to the small sample size (p<.24 two-tailed test).

26 25 differences do not quite reach standard levels of statistical significance on a two-tailed test (p<.098). Cuing Values Turning to the effect of the values appeal, I find suggestive results to support Hypothesis 5. Table 3 compares the response rate of individuals in the baseline control group to the values appeal group. Overall across all respondents, the effect of the value appeal is positive, although the effects to not reach standard levels of significance (p<.1, one-tailed test). Table 3: Implicit Racial Cues and the Likelihood of Posting a Yard Sign Control (N) Values Treatment (N) Effect of Values Treatment All Voters 14.29% % % Habitual Primary Voters 18.55% % % Non-Habitual Primary Voters 10.74% % %* Two-tailed t-test: ** p<0.01, * p<0.05 Table 3 also looks at the differences in the effects of the values appeal on voters who had previously voted in the primary election in 2012 and 2010 and voters who had not voted in both of those elections. 18 Because of concerns regarding the information environment about the campaign and how that environment might affect the influence of values appeals, I intentionally differentiated the findings between these two types of respondents for two reasons. Primary elections are low information affairs where most voters have little information about the candidates (Maisel 1982) especially in districts where the district is only a small portion of the overall media market (Campbell, Alford, and Henry 1984) as was the case in this district. As such, most voters are likely not to obtain very much information about the campaign and the 18 These effects are consistent when we look at voters who only voted in the primary in 2010 or only in the primary in 2012.

27 26 candidates. This would generally make voters more susceptible to primes as they have low information and may not have a strong impression about the candidates. People who have well formulated opinions and follow political matters more closely are less likely to be affected by the messages they receive (Zaller 1992). However, one group of individuals who are attentive and more knowledgeable, however, is regular primary voters (Norrander 1989). As such, because the candidate emphasized the use of symbolic phrases throughout the campaign, it could be that because of their awareness of the campaign they had already internalized the implicit racial connections that the campaign attempted to use to mobilize potential supporters through its emphasis of values. Thus, voters who were more likely to have information about the primary campaign may not differentiate between a standard invitation to post a sign in their yard in support of the candidate and the invitation to post a sign in their yard in support of the candidate and our values. The second and third row of Table 3 shows that the effect of the values treatment is negative (and not statistically significant) for voters who regularly participate in the primary, and positive and significant for those who do not. For individuals who do not regularly participate in primary elections and who were less likely to have extensive information about the campaign, however, the campaign s implicit racial cue did have a significant effect on the likelihood of participation.

28 27 Figure 3: Marginal Effect of Implicit Racial Appeal Relative to Baseline Moreover, consistent with Hypothesis 5, these effects interact with the racial demographics of the neighborhood in which the respondent lives. Figure 3 uses the fourth and fifth models in Table 1 to examine how the effects of the implicit racial appeal varies relative to