Viewers like you: the effect of elite co-identity reinforcement on U.S. immigration attitudes

ABSTRACTAs the political salience of immigrant-origin minorities continues to increase in the United States and Europe, researchers are increasingly focused on understanding what interventions reduce exclusionary attitudes. While several recent studies have examined the effect of different narrative and interpersonal communication techniques, few have focused on the role of the “messenger” that delivers these techniques. Drawing from psychological research on persuasion, we hypothesize that anti-exclusionary messages are more persuasive when delivered by elite messengers who reinforce shared identities. To test this, we conduct a large, pre-registered survey experiment exposing a sample of American adults to audio messages on immigration from persuasive elites performed by professional voice actors. We find that a persuasive message only shifts attitudes about immigration when elites include co-identity reinforcement primes. These findings offer additional nuance to the literature on immigration attitudes, persuasion, and elite-led public opinion, and have important implications for immigration advocacy work.KEYWORDS: Immigration attitudespersuasionelitespublic opinionidentity Disclosure statementNo potential conflict of interest was reported by the author(s).Notes1 While this study focuses on explicit prejudice, researchers have studied interventions for reducing implicit prejudice as well. For a thorough comparative investigation of these techniques, see Lai (Citation2014).2 This is consistent with recent research on elites and attitudes related to race, ethnicity, and immigration.3 https://immigrationforum.org/landing_page/bibles-badges-business/4 There is evidence that generic elite/party cues – those from “Republicans,” “Democrats,” or fictional candidates (see Arcenueax 2008; Bullock 2011; Ciuk and Yose 2016; Druckman et al 2013) – are often not powerful enough to persuade co-partisans to update their attitudes. Instead, evidence suggests that cues from known elites like the president of the United States, are the most persuasive (Agadjanian 2020; Lenz 2012; Nicholson 2012; Barber and Pope Citation2019). We suspect that generic party cues are often weak because of issues of informational equivalence where the respondent might project onto the elite their perception of the elite’s strength of partisan identity or core political values, for example.5 We pre-registered our design and analyses with OSF.io on August 7, 2020, before full data collection commenced on August 10, 2020. Full pre-analysis plan is included in Appendix B. IRB approval was acquired for all pilot tests and the full survey experiment at each of the authors’ institutions.6 We discuss our decision to measure our moderator before treatment in Appendix A.7 In Appendix A, we also display tests for non-random attrition and show balance for pre-registered pre-treatment covariates across treatment conditions.8 We ensured compliance with the treatment in a few different ways. First, we ensured that respondents had working speakers and could watch video on their device by having them pass a video and audio check (picture of a cow and audio of a dog barking) before they could proceed to the video. The video itself was hosted on YouTube to maximize compatibility across mobile devices and browsers and minimize streaming issues. We set a timer for the duration of the video time so that respondents could not skip the video without waiting for 20 s, removed the scroll bar from the video so they had to watch the entire thing, and embedded a code at the end of the video that respondents had to enter into a text box correctly before proceeding.All respondents so exposed were informed of the deception immediately at the conclusion of the survey. As part of the experimental design, this deception was necessary so that all the messengers’ statements were standardized. The statements, however, were crafted after real messages made by real political elites. To isolate the effect of the speaker in our experiment we need to use the same exact message for each one; otherwise, we would have confounded different messages with different speakers. The deception employed involved minimal risk to subjects. While the deception may alter respondents’ impressions of real people, this alteration only extended until the end of the survey administration. No respondent completed participation without being informed about the deception, and so it will not adversely affect any participant's rights or welfare.9 This message was derived from speeches that the late Republican Senator from Arizona, John McCain, made on immigration. Though a decorated ideological conservative who was once his party’s presidential nominee, McCain held very centrist views on immigration policy and persuaded many Republican voters and legislators to take more liberal positions in his attempts to pass comprehensive immigration reform. For more on this message construction, see Appendix A.10 For full wording for all questions see Appendix C. We measured but did not include an item on English-language only policy because a double negative was clearly confusing for respondents and does not positively correlate with other immigration policy attitudes. While the inclusion of the item, a departure from our pre-analysis plan, does not change substantive finding of the study, we have omitted it from all analyses.11 Pre-registration is posted but currently embargoed on OSF. For review purposes, an anonymized pre-registration has been included in the appendix of this manuscript. We report non-covariate adjusted means in the body of the manuscript and leave adjusted means, which are substantively identical, in the appendix.12 We further probe this relationship by breaking Republicans into strong versus weak partisanship (as measured by branching partisan identity question), those with stronger versus weaker partisan identity (Huddy, Mason, and Aarøe Citation2015), and those high or low in self-monitoring (Connors Citation2020). Full wording of all items in Appendix C.13 By examining partisan subgroups separately, we are theoretically assuming, of course, that respondent partisanship is moderating the treatment effects we observe in each, though we acknowledge that making a causal moderation claim here requires randomly assigning respondent partisanship which is infeasible (Kam and Trussler Citation2017). Our covariate-adjusted regression estimates in Appendix D help assuage concerns by assessing our assumed model’s sensitivity to the most theoretically relevant confounders.14 For context, this change in attitude is about two-thirds the size of the gap in immigration policy attitudes between college and non-college educated Americans, about one-half the size of the gap in immigration policy attitudes between white and non-white Americans, and about three-quarters the size of the gap in immigration policy attitudes between Independents and Republicans. A one-fifth of a standard deviation shift in attitudes is similar to the Paluck et al. (Citation2021) meta-analytic estimates of the effects of prejudice reduction experiments in large samples.15 Results for individual attitude items are included in Appendix Table D3. We find that the message is most likely to move respondents’ attitudes about a pathway to citizenship, job training programs, welfare benefits, and establishing a Hispanic heritage month, all policies aimed at accommodation. It did not, however, change attitudes about deportation.16 Hannity is notably anti-immigrant in his current programming in 2020 but in 2012 famously pivoted to an accommodationist messaging urging Republicans to embrace a pathway to citizenship.17 These divergent responses are likely due to the fact that our treatment message is predominantly focused on moving attitudes about policy rather than about who immigrants are.18 We did not pre-register looking at effects by pre-existing levels of immigration attitudes, so this analysis can be treated as exploratory; however, we feel that this analysis is very important for organizations hoping to shift immigration attitudes to see whether the effects are concentrated, maybe, among those Republicans who already felt warmer toward immigrants, for example.19 We did not actually give respondents the opportunity to sign a petition, but respondents were not aware of this when they indicated their intentions. They were debriefed on the purpose of this measure after they completed the survey.20 Research suggests that asking about attitude change exhibits poor measurement properties (Graham and Coppock Citation2020). We are limited, however, by our use a fictional elite speaker with little auxiliary information with which to otherwise form attitudes (or counterfactual assessments).