One Language, Two Meanings: Partisanship and Responses to Spanish

Abstract

The growth and dispersion of America’s immigrant population exposes increasing numbers of non-Hispanic Whites to Spanish. Yet the political impacts of that exposure depend on whether Democrats and Republicans respond in similar ways. To address that question, this article first presents survey experiments showing that exposure to Spanish increases restrictive immigration attitudes only among Republicans. To confirm the external validity of that result, the article then presents an analysis of California’s Proposition 227 indicating that support for ending bilingual education was higher in heavily White, Republican block groups with Spanish-language ballots. No such pattern appears in Democratic block groups. Together, these findings demonstrate that Spanish is a politicized symbol, provoking different responses among Whites depending on their partisanship. To the extent that other immigration-related cues produce similar effects, the salience of immigration seems likely to reinforce existing partisan divisions rather than undermining them.

Keywords:

• immigration
• Spanish
• partisanship
• survey experiments
• Voting Rights Act

Acknowledgments and Funding

The first survey experiment was made possible by a Presidential Authority Award from the Russell Sage Foundation, and it was approved by the Georgetown Institutional Review Board (2010-121). Elements of this research were presented at the 2009 Conference on Empirical Legal Studies, the Georgetown Public Policy Institute, the George Washington University American Politics Seminar, the Triangle Political Methodology Group at the University of North Carolina, the 2010 annual meeting of the Midwest Political Science Association, the Dreher Colloqium at Ohio State University, and the 2012 Experimental Political Science Conference at New York University. John Bullock, Rafaela Dancygier, Todd Hartman, Jonathan M. Ladd, Marc Meredith, Melissa Michelson, and Abigail Fisher Williamson provided valuable advice. In addition, the author acknowledges Ileana Aguilar, Belisario Contreras, Franco Gonzalez, Benjamin Hopkins, Pablo Leon, Alejandro Gonzalez Martinez, Eusebio Mujal-Leon, Elizabeth Saunders, Grace Soong, and especially Randy Bell and Stefan Subias for assistance in video production and experimental implementation. Zoe Dobkin, Katherine Foley, Douglas Kovel, Patrick Gavin, Anton Strezhnev, and William Tamplin provided insightful research assistance. The author also gratefully acknowledges the efforts of Political Communication editor Shanto Iyengar and the anonymous reviewers, efforts which considerably strengthened this article.

Notes

1. The three journals are the American Journal of Political Science, the American Political Science Review, and the Journal of Politics.

2. As Berkowitz and Donnerstein (1982) point out, “the generalizability of laboratory findings in comparison with the results of naturalistic investigations is an empirical matter” (p. 249).

3. Defined broadly, priming effects occur when a stimulus activates certain preexisting mental considerations, which in turn shape one’s perceptions or evaluations (Althaus & Kim, 2006; Bargh, 2006).

4. Alternately, exposure to Spanish might also operate through more conscious modes of mental processing, with those who encounter the language using such encounters as informative signals. For instance, seeing Spanish might be perceived as indicating the size of local immigrant populations or the views of the public officials using the language.

5. A cue is defined as “a piece of information that allows individuals to make inferences without drawing on more detailed knowledge” (Druckman et al., 2010, p. 137).

6. Following the National Election Study, the full question read “please rate each group from zero (0) to one hundred (100). The higher the number, the warmer or more favorably you feel toward it. The lower the number, the colder or more negatively you feel toward it.”

7. For one exception, see Cutler (2002).

8. KN began address-based recruiting in the spring of 2009. Approximately 25% of the sample employed here was recruited through address-based sampling.

9. Specifically, Imai et al. (2011) note that “initially, researchers focused on the estimation of causal effects in studies that quantified the effects of media cues on policy attitudes and the effects of incumbency on electoral outcomes. Once a certain level of consensus emerged about the magnitude of causal effects, scholarly attention shifted to the question of causal mechanisms” (p. 767).

10. Here, I ignore a separate randomization to immigrants of varying skin tones, as it shows no strong attitudinal impacts (Hopkins, 2014). All respondents were told to turn on their computer’s sound prior to the video and were asked immediately after if they had seen and heard it. A total of 4,648 respondents from the KN panel were invited to participate in the first experiment, yielding an RR3 panel response rate of 44.3%. Given the original panel recruitment rate (17%) and the panel questionnaire completion rate (62%), the overall response rate was 5%. Similar response rates are common both for online surveys and for most contemporary telephone surveys.

11. All experimental manipulations are publicly available at http://www.youtube.com/user/immigrationsurvey

12. For this experiment, the relevant response rate to the initial request to join the panel was 16%, 64% of these respondents successfully joined the panel, and 44.3% of the 2,564 invited panelists completed the survey. The overall response rate is 4.5%.

13. The survey had no more than 15 questions and had approximately eight questions in the average administration.

14. To test for evidence of attrition related to treatment assignment, I predicted the binary indicator of satisficing as a function of the treatment indicators, core variables (such as partisanship), and their interactions. The only significant relationship uncovered is that self-employed people are more likely to be satisficers (p = .02). In no case do I recover a significant treatment effect or interaction, suggesting that the treatments did not induce satisficing.

15. I also performed randomization checks using 12 key covariates, including income, Internet access, ideology, party identification, education, age, prior contact with Spanish, and various measures of employment status (e.g., retired, self-employed, unemployed). When comparing the Spanish treatment to either English-language treatment in the full sample of non-Hispanic Whites, only partisan identification is imbalanced (p = .02, two-sided test). For the reduced sample without satisficing respondents, there are no statistically significant differences across any of the 12 variables, although partisanship remains close (p = .06, two-sided test). Nonetheless, this potential imbalance is addressed by stratifying on partisanship.

16. Given the split sample, these differences are not quite statistically significant in two-sided t tests, with p values of .15 and .25, respectively.

17. For the two experiments, the p values estimated through two-sided t tests are .49 and .75, respectively.

18. Republican partisans are markedly less supportive of a pathway to naturalization (, ), but the interactions between partisanship and the accented English treatment (, ) or the fluent English treatment (, ) indicate reduced relationships in those experimental groups. As a point of comparison, Table 1A in the online Appendix presents these results alongside models with only partisanship interactions. The inclusion of several other interactions between the treatment indicators and other variables does little to change the partisanship interaction. If anything, the interactions of interest grow slightly stronger when including interactions between the treatments and covariates such as education and ideology.

19. 19. The specific survey question was asked after the experimental treatments, and it read: “In your day-to-day life, how frequently do you hear Spanish spoken? Never or almost never, less than once a month, 1–3 times each month, at least once a week, or every day?”

20. 20. This result holds as well when removing ideology from the models, with a two-sided p value of p . Similarly, when estimating the results with those respondents who did not watch the video or took more than 30 minutes to complete the survey, the two-sided p value is p .

21. 21. On priming effects before and while voting, see also Berger et al. (2008) and Bryan et al. (2011).

22. 22. The 1992 determinations are available on page 35,371 of Volume 58, Issue 125, of the Federal Register. The full list of counties included in those determinations and that meet the other criteria described below includes Alameda, Fresno, Inyo, Kings, Lake, Monterey, San Benito, Santa Clara, and Tulare. Of these, only Kings County is additionally covered by Section 4(f)4 of the Voting Rights Act, which also requires foreign-language election materials.

23. 23. In fact, due to the significant heterogeneity within counties, the Pearson’s correlation between Section 203 coverage and precinct-level Republican support is small but positive, at .08. Put differently, among these heavily non-Hispanic White precincts, it is the Republican precincts that more commonly use Spanish-language ballots.

24. 24. San Francisco County was not mandated to provide Spanish-language ballots, but did so anyway. Here, my focus is on the variation in Spanish-language ballots induced through a federal mandate, so I code San Francisco County (accurately) as not being required to provide Spanish-language ballots. This is a conservative choice, as it means that some of the control group was exposed to a variant of the treatment. The county’s 178 block groups account for 0.6% of the data set, and the conclusions are robust to its exclusion as well.

25. 25. McCue (2008) describes how precinct-level election outcome data were disaggregated to the level of census block groups. This process does induce some measurement error in the dependent variable examined below—precinct-level voting for Proposition 227—but the independent variables from this source require no such disaggregation.

26. 26. RDD has seen increased use in recent political science research (e.g., Eggers & Hainmueller, 2009; Green et al., 2009; Meredith, 2009; Trounstine, 2011), in large part due to its ability to recover unbiased estimates of the local average treatment effect from observational data.

27. 27. The cubic term for the number of citizens who speak Spanish and limited English is correlated with the squared term at .993, so it is excluded.

28. 28. These measures generally behave as expected, with the block-group-level percentage of Democratic registrants strongly correlated with opposition to Proposition . The percentage of residents with a bachelor’s degree is as well .

29. 29. The interaction grows stronger still when excluding Inyo, Lake, Kings, and Merced counties, counties whose Spanish-language mandates come from VRA requirements not subject to the Section 203 thresholds such as Section 4(f)4. A total of 1,054 observations come from these counties, accounting for 3.8% of the total neighborhoods and 13.8% of neighborhoods with Spanish-language ballots.

30. 30. A less highly interacted model, with only an interaction between percent Republican and Spanish-language ballots, returns similar results . This same model returns a similar effect when estimated for the 3,203 block groups in seven counties within 3 percentage points of the percentage-based threshold.

31. 31. I reach the same conclusion using an OLS model with county fixed effects and an interaction between the block group’s percent Republican and a county-level language mandate .

32. 32. As an example of this model dependence, the core result holds up using the percentage Republican as the interacted variable when removing any one of the 40 counties. If the interaction is instead specified using the percentage Democratic, however, it becomes substantively smaller and statistically insignificant when removing either San Mateo or Colusa County.

Additional information

Notes on contributors

Daniel J. Hopkins

Daniel J. Hopkins is Associate Professor, Department of Government, Georgetown University.