Tolerance Toward Immigrants as a Dimension of Cosmopolitanism: Explaining Attitudes Toward Immigrants in Houston

ABSTRACT

Tolerance toward immigrants can be explained as the development of dispositions toward the acceptance of foreigners as locals influenced by the city’s multicultural contexts. Accordingly, tolerance toward immigrants represents a dimension of cosmopolitanism in metropolitan areas of receiving societies. In this study, I examine whether the proportion of immigrants in communities is directly associated with tolerance toward immigrants and whether there are significant differences in attitudes toward immigrants by occupation and educational attainment in the metropolitan research setting of Houston, Texas. Using data from the Houston Area Survey, I find that the percentage of foreign born in census tracts is directly associated with tolerance, that white-collar workers are not more tolerant than non-white-collar workers, and that the effect of education on tolerance toward immigrants is not always positive. I conclude that, in Houston, tolerance toward immigrants represents a dimension of cosmopolitanism, and cosmopolitanism is not well founded on socioeconomic status.

Acknowledgments

Earlier versions of this paper were presented at the 2014 American Sociological Association Annual Meeting and at the Kinder Institute for Urban Research at Rice University. I am deeply grateful to Daniel Powers, Néstor Rodríguez, Robert Hummer, and Mark Warr for their feedback throughout the writing of this paper. I also thank Jenifer Bratter, Stephen Klineberg, Jie Wu, and the people at the Kinder Institute for Urban Research at Rice University for sharing the Houston Area Survey data and for their questions and comments during my presentation at their research workshop. Furthermore, I thank the editors of Sociological Spectrum and the anonymous reviewers for their comments and suggestions.

Notes

¹Gouldner’s (1979) cosmopolitan new class is one of the most popular references of cosmopolitanism in the United States. To my knowledge, the only study that quantitatively examines attitudes toward immigrants in the United States using a cosmopolitanism framework is the article by Haubert and Fussell (2006), which is founded on new class theory.

²The environmental contact approach presented in this study differs from the traditional contact hypothesis, which has been primarily examined at the individual level with variables such as “knows and feels close to an African American person” and “just knows a Hispanic person.” See the distinction between contextual and individual contact in the study by Stein, Post, and Rinden (2000).

³I assume that a census tract is more multicultural if its percentage of immigrants is greater regardless of the existence of multicultural policies that particularly affect the community (see Bloemraad 2007).

⁴Past rounds of the HAS included respondents from Harris County only.

⁵Although response and cooperation rates were overall low, these rates are relatively high for current telephone survey data (e.g., see O’Neil and Tienda 2010). However, the sample distributions support the confidence in the reliability of the data (Klineberg 2013).

⁶The scale is slightly skewed to the left. This is not a significant problem because the distribution of standardized residuals is symmetric according to the quantile-comparison plot. I also used robust estimators, which are especially efficient when the errors are heavy tailed, because they are resistant to outliers (see Fox 2008).

⁷Past rounds of the HAS included only a few of these questions (usually as binary variables rather than the four-level Likert items included in the 2012 round) that did not lead to a reliable scale. Question 6 was randomly asked as “immigrants” or as “undocumented immigrants” to a respective half of the sample. I grouped all answers as one item. I also worked with a scale of the factor analysis scores and with a Cronbach’s alpha scale that included observations with missing values in any of the scale items. I chose to present the additive scale because its values can be straightforwardly interpreted and because the findings were consistent using any of the dependent variables.

⁸In alternative analyses (not presented in this study), I separately included two additional variables that roughly measured change in the percentage of foreign-born individuals in counties (based on U.S. Census estimates) (1) between 1990 and 2012 and (2) between 2000 and 2012. I also included binary variables that distinguish established versus nonestablished destinations based on county data. In this way, I attempted to capture some of the complexity involved in immigrant reception. However, the coefficients of these variables were statistically insignificant. I opted to not include any of these variables in the regression models presented in this study because I chose census tracts as contextual units. The results of the alternative analyses were consistent with the results presented in this study.

⁹I computed the USA Today Index of Ethnic Diversity (Meyer and McIntosh 1992) using the 2006–10 ACS estimates of percentages of Anglo, Black, Asian, and Hispanic people in census tracts linked to the HAS. Ethno-Racial Diversity = [1 – [(%Anglo/100)²+ (%Black/100)² + (%Asian/100)² + (%Hispanic/100)²]] × 100.

¹⁰Nonetheless, I also fitted regression models (not presented here) including the total sample size, and categorical variables for the missing values of age and education. The coefficients of these alternative regression models are consistent with the results presented in this study.

¹¹I compared the coefficients of an ordinary least squares (OLS) regression with all independent variables (as in Model 4a) with the coefficients of a weighted OLS regression, and they produced different estimates. Then I fitted an OLS regression with all independent variables (as in Model 4a) plus the variable weights plus the interaction terms of the independent variables and weights. An F-test indicated that weights and the interaction terms were collectively significant. Therefore, I opted to use weights in the analysis (Winship and Radbill 1994).

¹²I included logged household income as another independent variable with educational attainment and occupational status in separate regression models not presented here. The effects of logged household income on the scale were statistically insignificant and did not alter the significance or the estimates of the rest of the variables in the regression models. Therefore, I did not include this variable in the regression models presented in this analysis. Moreover, variance inflation factors of independent variables in the regression analyses presented in this study suggest that multicollinearity is not a problem.

¹³To support my choice of individuals exposed to school education in Houston, I fitted another regression model (not presented here) with an alternative subsample that also consists of individuals who likely attended school in Houston: respondents who lived outside of Houston 5 years or less. Net of other effects (the same variables included in Model 4c), the coefficient of high school compared with less than high school was still negative and significant as the findings presented in this study.

¹⁴The first equation with attitudes as dependent variable and all independent variables as in Model 4a except the contextual independent variables percentage high school or less and percentage unemployed in census tract. The second equation with percentage of foreign-born as dependent variable, and the contextual independent variables percentage high school or less, percentage unemployed in census tract, and the index of ethno-racial diversity.

¹⁵For clarity, I did not include in Table 1 the weighted percentages or the unweighted observations of the items included in each version of the dependent variable. Differences in these percentages, however, do not change the idea of a balanced sentiment as it is reflected in Table 1. The scale is also a reliable and valid outcome variable in the subsample analyses.

¹⁶Dummy-variable regression and ANCOVA produce identical predicted values, but the effects of independent variables using ANCOVA are expressed in terms of deviations from the grand mean, rather than deviations from specific reference categories as in the dummy-regression approach (Table 2) used in the main analysis (Fox 2008). The estimated intercepts using the ANCOVA formulation refer to these grand means.

¹⁷Due to the proportion of foreign-born Latinos in Houston, I fitted a regression model without Hispanics in the analytic sample (not presented here) and still found evidence of a positive association between percent foreign-born and tolerance.

¹⁸I fitted a regression model (not presented here) without other (not working, unanswered) in the analytic sample and still found no significant differences between white-collar and non-white-collar.

¹⁹Not just events oriented to Latino audiences, but events that involve individuals and organizations from different ethnicities.

Additional information

Notes on contributors

Cristian L. Paredes

Cristian L. Paredes is an Assistant Professor of Sociology at Loyola University Chicago (beginning Fall 2017). Cristian is primarily interested in the influence of immigration on several social dynamics in U.S. metropolitan areas, and in the multidimensionality of the ethno-racial condition in Latin American contexts of mestizaje. His work has appeared in Ethnic and Racial Studies and Sociology of Race and Ethnicity.