![Sample our Politics & International Relations journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5947/TOC-Subject-Sample-2021-Politics-International-Relations.jpg"})
Abstract
Explanations of Americans' attitudes toward immigration emphasize threats to national identity and culture. However, we do not know the specific sources of cultural threat or whether they operate locally. Native-born residents commonly voice concerns about the prevalence of Spanish, suggesting that foreign languages might be one such source of threat. This article uses survey experiments to provide one of the first causal tests of the impact of written Spanish on Americans' immigration attitudes. One experiment (N = 351) was conducted online with a nationally representative sample, while a second was embedded in an exit poll (N = 902). The experiments show that Spanish has differential impacts depending on Americans' prior contact with it. Among those who hear Spanish frequently in day-to-day life, seeing written Spanish induces anti-immigration attitudes. These findings suggest that language can foster cultural threat, and they highlight a mechanism through which local encounters can be threatening.
Acknowledgements
Data were collected partially through Time-sharing Experiments in the Social Sciences. The project was reviewed by Institutional Review Boards at Yale (711003296) and Harvard (F15546-101 and F16792-101). The authors also acknowledge assistance or advice from Matt Barreto, Adam Berinsky, Irene Bloemraad, Colin Brown, Nicole Deterding, James N. Druckman, Shang Ha, Gregory Huber, Gabriel Lenz, Arthur Lupia, Diana Mutz, Gloria Park, Robert D. Putnam, Mary Waters, Ismail White, Cara Wong, and participants at the aforementioned seminars. We also thank the anonymous reviewers of our paper for their helpful comments.
Funding
The authors are grateful to Harvard University's Center for American Political Studies, Hauser Center for Non-Profit Organizations, and Saguaro Seminar for Civic Engagement; the MIT Department of Political Science; the Yale University Center for the Study of American Politics; the Rappaport Institute for Greater Boston; and the ZEIT-Stiftung for institutional and funding support.
Notes
†Previous versions of this research were presented at the Eastern Sociological Society, the Midwest Political Science Association, the Harvard University Migration and Immigrant Incorporation Workshop, the Georgetown American Politics seminar, the Harvard Political Psychology and Behavior Workshop, the University of Virginia American Politics Seminar, and the Yale University American Politics Seminar.
1. We also show that self-reported exposure to Spanish is not correlated with immigration-related attitudes ex ante, a fact which reduces concerns that an alternate moderator is really at work.
2. A separate question asked respondents, “How frequently do you see Hispanic/Latino immigrants in your community?” The two measures have a Pearson's correlation of .66. (We employ the second in robustness checks, but focus on hearing Spanish since it directly measures exposure to the language.) The strong correlation between the questions suggests that the question on hearing Spanish captures actual contact with Latinos rather than hearing Spanish-language television.
3. Certainly, the absence of a correlation between exposure to immigrants and immigration attitudes does not contradict the claim that encountering Spanish can be a negative experience for English-speaking Americans. Instead, it simply indicates that other factors might also be at work in shaping their immigration attitudes, such as selective exposure.
4. The panel recruitment rate (AAPOR RR3) was 23% and the panel profile rate was 57%.
5. While Freedman (Citation2008) raises concerns about the applicability of generalized linear models to experimental data, Green (Citation2009) convincingly dismisses such objections for all but the smallest sample sizes.
6. Unless otherwise noted, all other questions have four response categories.
7. The result is very similar when we use logistic regression conditional on party identification to estimate the treatment effect, a precautionary step given the imbalance on that covariate.
8. Tests for interaction effects with the anti-immigration index confirm that prior exposure to Spanish is not masking the effects of other variables, such as education, income, partisanship, or ideology. The anti-immigration impact is stronger still among those with the most day-to-day exposure to Spanish. We do not observe any notable non-linear impacts.
9. For the binary variable, we used logistic regression; for the others, we used ordinary least squares. There is very little missing data, with only 1.8% of observations removed through list-wise deletion.
10. Seeing Spanish is measured via a binary, 0–1 indicator variable. Prior exposure to Spanish is measured as an ordinal variable ranging from 1 (never exposed to Spanish) to 5 (exposed to Spanish daily).
11. As noted above, we control for partisan identification to address the covariate imbalance. The experimental design precludes the need to control for additional predictors of immigration attitudes, since the treatment and control groups are not systematically different with regard to these characteristics. Nonetheless, if we include other control variables, such as political ideology, race, ethnicity, foreign birthplace, or gender, the results do not change.
12. In addition to the 144 respondents who saw the monolingual welcoming note, 70 respondents saw no welcoming note. If we include these respondents in the control group and include an indicator variable for those who saw no welcoming message, we recover substantively similar results.
13. Using t-tests on the sample that does not frequently hear Spanish, we see that those exposed to Spanish are less likely to link immigrants to rising crime (p = .09, two-sided test) or to higher taxes (p = .03, two-sided test).
14. In interpreting , readers should note that prior exposure to Spanish was not randomly assigned, so comparisons within a treatment group have no clear causal interpretation.
15. Everett has a median household income of $51,333, which almost exactly matches the national figure of $50,007. It is 33% foreign born. According to the same 2005–2007 American Community Survey data, Somerville's median household income was $59,146, and it was 27% foreign born.
16. Since Somerville and Everett are home to large Portuguese and Brazilian immigrant populations, we included mention of Portuguese in our question.
17. In Spanish, the text read, “Por favor, fíjense que uds. pueden contestar en español al otro lado.” The sentence means “Please be aware that you can answer in Spanish on the other side.” To maintain realism for Spanish-speaking respondents, the treatment survey included a Spanish translation on the back, while the back of the control survey was blank.
18. The surveys were given to respondents affixed to the clipboard with the English always facing up. Very few respondents ever turned over the survey, and those who did did so only after completing it. The treatment effect is characterized as seeing a single line of Spanish. Because the treatment was visible at the top of the survey, all 19 questions should be considered “post-treatment,” although we suspect that seeing Spanish will not influence self-reported income, vote choice or other covariates.
19. The interval n was determined by the researcher at each site based on voter traffic and then fixed. Such randomization is commonly employed in exit polls, and relies on an assumption of no systematic ordering in how voters leave the polling place. The treatment would be confounded, for example, if everyone voted with her spouse and the more conservative partner always exited first. Randomization checks confirm that there were no such problems. Treated respondents were no more or less likely to vote for McCain, for example.
20. There, our exit poll found 72% support for Barack Obama, when in actuality 62% of voters voted for Obama. At the other three sites, we over-estimated Obama's support by four percentage points, five percentage points, and nine percentage points. National exit polls have seen comparable overstatements of Democratic support in recent years (Edison Media Research and Mitofsky International Citation2005).
21. Of the eight available covariates, those who received a survey with Spanish were more likely to be Black (16.0% versus 11.3%; p = .039 from a two-sided t-test). They were also slightly less educated, reporting 14.6 years of education as compared to the control group's 15.0 years (p = .024).
22. Ceiling effects can operate at levels of support or opposition as low as 67% so long as binary immigration attitudes have considerable fundamental variability.
23. Using ANOVA, we confirmed an inter-group difference among Obama voters at p = .04.
24. Similarly strong results hold when we condition on self-reported Hispanics or remove the 69 Hispanic Obama voters from the analysis. They hold as well if we estimate a similar model on the full data set with an interaction between the treatment and the respondent's vote choice. In models of the full data set, 5.6% of respondents are deleted due to missing data.