Attribute Affinity: U.S. Natives’ Attitudes Toward Immigrants

In fact, irrespective of other abilities, most natives prefer potential immigrants with high skill levels (using level of education as a proxy for skill). These studies find that highly skilled immigrants are preferred to low-skilled immigrants, regardless of a native’s own skill level, suggesting that this preference is not driven by different levels of threat perception about one’s own job.

Political scientists also incorporate social identity into studies of public opinion. Most notably, Kinder and Kam (2009) focus on what they argue is a tendency to “divide the world into in groups and out groups” or ethnocentrism (76). For instance, the authors argue that high levels of ethnocentrism among U.S. natives drive negative attitudes toward immigrants.

Although drawing from common in-group identity, our argument is not about any aspect of identity becoming superordinate or recategorized – instead we suggest that simply the recognition of shared characteristics on one dimension may override aversion along another.

In this study, we use frequency of attendance at religious services as a proxy for religiosity. To test its validity, we correlated frequency of attendance with other measures of religiosity. See footnote 19 for details.

Though, as we note below, such a pattern does not hold for individuals from “outsider” religions, such as Muslims in the present day.

While attitudes toward Muslims tend to differ by political affiliation, Americans on both sides of the aisle support keeping Muslims out of the country or subjecting them to additional surveillance (Sides and Mogahed 2018). The same survey evidence suggests that Americans view Muslims as religious and believe they hold outdated views of women, gays, and lesbians.

Attitudes toward Muslims have likely been influenced by the September 11, 2001 events and the resulting portrayal of Muslims in the media, where they are frequently associated with threatening images of terrorists (Cimino 2005; Oswald 2005; Putnam 2002; Schildkraut 2002; Shaheen 2003). It should be noted that Creighton and Jamal (2015) find that U.S. natives are equally opposed to Muslim and Christian immigrants. While U.S. Christians are explicitly biased against Muslim immigrants, they also hold implicit biases against Christian immigrants of specific nationalities and ethnicities.

YouGov interviewed 1089 respondents from May 2 to May 6, 2013. Then, respondents were matched down to a sample of 1000 to produce the final dataset. The respondents were matched on gender, age, race, education, party identification, ideology, and political interest. Using the 2010 American Community Survey, YouGov then weighted the matched set of survey respondents to known characteristics of the general population of the U.S.

We emphasized that we were solely concerned with legal immigrants to rule out natives’ attitudes toward illegal immigrants, which may be driven by other socially undesirable characteristics (Wright et al. 2015).

We oversampled religious respondents on MTurk using contacts from previous studies on religion because, in general, respondents on MTurk are less likely to be affiliated with a religion than the overall United States population (Berinsky et al. 2012).

SSI recruits participants through various online communities, social networks, and website advertisements. SSI makes efforts to recruit hard-to-reach groups, such as ethnic minorities and seniors. These potential participants are then screened and invited into the panel. When deploying a particular survey, SSI randomly selects panel participants for survey invitations. We did not employ quotas but asked SSI to recruit a target population that matched the (18 and over) 2010 census population on education, gender, age, geography, and income (based on the premeasured profile characteristics of the respondents). The resulting sample is not a probability sample but a diverse national sample. SSI samples have been used in a number of recent publications in political science.

Hainmueller et al. (2015) test several vignette experimental designs and find advantages to both measures, both in terms of their accuracy in calculating effects and predicting real life outcomes.

We selected major world religions: Christianity (Catholic and Protestant denominations), Islam, and Hinduism. We also included atheists and Jews because we wanted enough variation in our variable of interest and, particularly, enough variation in religious affiliation for each country of origin in our experiment.

Although we assigned the attribute-levels at random, we included one restriction over the possible profiles that the respondents saw. “Atheist” is one of the levels of the religion attribute and it is unrealistic to present respondents with atheist immigrants who attended religious services. Therefore, we restricted the possible combinations of immigrants such that atheist immigrants would always “Never” attend religious services. Hainmueller et al. (2014) state that the randomization of profiles (an identification assumption in conjoint analysis), requires that the randomization scheme assign a non-zero probability to all the possible attribute combinations for which the potential outcomes are defined. Because the control observation for each attribute-level is all other attribute-levels, removing theoretically problematic combinations makes it impossible to analyze causal quantities for these attribute-levels. We are not particularly concerned, as these combinations are not crucial for our results.

Immigrants’ religions were: Protestant, Catholic, Mormon, Jewish, Muslim, atheist/agnostic, Hindu or Buddhist, and Eastern or Greek Orthodox.

Psychologists have long known that individuals often fail to accurately report their underlying mental processes (Nisbett and Wilson 1977). When individuals are asked to report their cognitive processes, they often do so on the basis of a priori judgments. Our results demonstrate that irrespective of the stated importance of religion, respondents’ own religion greatly influenced their preferences.

Because our treatment and outcome variables are binary (either an immigrant has or does not have an attribute and a respondent either accepts one or the other immigrant), the linear regression estimator is fully nonparametric.

The relatively small sample size of United States respondents identifying with the smaller religions randomly assigned to the immigrant profiles (Hinduism, Islam) prevent us from drawing inferences about these groups.

We additionally tested whether our measure of religiosity (attendance at religious services) corresponds to the salience of religiosity as an identity. In our YouGov—religion survey, we asked respondents how important religion is in their lives (“Extremely important”,“Very important”,“Somewhat important”,“Not very important”,“Not at all important”). Figure 1 in the Supporting Information shows that our measure of religiosity is highly correlated with self-reported levels of importance of religion, suggesting that our measure is capturing both the strength and salience of religiosity.

We exclude atheists from both the respondent sample and immigrant profiles since they are not considered religious people. However, the results do not change if atheists are included.

We ran another similar experiment in 2015 where we included an additional treatment arm that presented respondents with a Muslim immigrant without any information about religiosity. The experiment presented respondents with two immigrant profiles sequentially. We managed to replicate our results as we did with the conjoint analysis only in the first period, not the second. We find that, in line with our theory of attribute affinity, religious respondents prefer religious over non-religious Muslims. However, we are careful in interpreting these results because of their instability (see Supporting Information for details). Although we cannot explain this instability, we suspect that events relating to Islamic extremism that occurred between our experiments, such as the Charlie Hebdo attacks and the increased media coverage of the Islamic State, may have influenced our respondents’ answers in ways we did not anticipate. However, given the strong findings from our first two experiments—conducted using different methods on different samples—we are confident that the initial results we report in this paper are not simply statistical flukes.

The findings in this sample cannot rule out that this pattern is particular to religious Catholic respondents because in the full sample of religious respondents there is a slight preference for religious Muslims over non-religious Muslim immigrants (difference in means of 0.06, p-value = 0.86), but the difference is not statistically distinguishable from zero.

Similar to the MTurk–religion/religiosity experiment, we asked respondents to rate each immigrant on a 7-point scale (see Fig. 1). Including atheists that report attending religious services does not change the results.

The difference in means between religious respondents’ evaluations of religious and non-religious Muslims is practically the same size as when measured using a 7 point scale rating dependent variable and is significant at \(p < 0.08\).

Instead of seven pairs, we showed 10 pairs of profiles to respondents of minority religions such as Mormons, Greek Orthodox, Jews, Muslims, Buddhists, Hindus, and atheists. Therefore, when we pool all religious respondents, we are in fact over-representing these religions relative to their population size. We ran the analyses both with all profiles and as well as with just the first seven that each respondent saw. The results do not change with the modification in the sample size.

We designed the SSI–conjoint experiment in 2014 to specifically test our theory on religion and religiosity, but we also use the data to test the relevance of attribute affinity with respect to other interactive aspects of identity-specifically, between race on the one hand and language skills on the other. We find that although the statistical significance fails to reach standard levels, attribute affinity mitigates negative reactions in these cases. Although only suggestive, this evidence further strengthens our theory and emphasizes the importance of studying the effect of the interaction of different identity attributes on attitudes toward out-groups.