Economic segregation and public support for redistribution

ABSTRACT

This study assesses whether local economic segregation, or the degree to which people live among others of similar economic status, influences the American public’s preferences for government redistribution. To test this proposition, we combine unique measures of economic segregation at the local level (using zip codes), covering nearly the entire U.S. population, with individual-level opinions on government spending and taxation. Multilevel regression analysis with random intercepts is used to assess whether the public’s preferences for redistribution are shaped by local economic segregation. Our findings suggest that residents living in highly segregated areas are less likely to favor redistributive government policy. Additionally, the results show that the influence of economic segregation on public support for redistribution is particularly strong among the affluent. This research not only contributes to our understanding of the consequences of economic change, but it also demonstrates the importance of considering local context when studying the attitudes of the American public. While the expansion of income inequality is certainly a global phenomenon, the political, economic, and social environments that make up the communities where people live are bound to have an influence on public opinion.

KEYWORDS:

• Economic segregation
• Inequality
• Class
• Public policy

Supplementary Materials

Supplemental data for this article can be accessed on the publisher’s website.

Highlights

• Less support for redistributive policy for those in economically segregated areas.

• Also less support for redistributive policy when the rich are isolated from the poor.

• Influence of economic segregation on attitudes is strongest among the affluent.

Notes

¹ Bjorvatn and Cappelen (2003) is theoretical in nature and while Minkoff and Lyons (2019) empirically examine the relationship between neighborhood income diversity and preferences for redistribution, their study focuses on a single U.S. city (i.e., New York City) and is therefore not generalizable outside of this particular context. As we discuss in more detail below, our analysis is much broader in scope and considers nearly the entire U.S. population.

² The relationship we propose between economic segregation and policy attitudes based on intergroup contact theory is also consistent with research showing that when the affluent are isolated from lower-income groups they are more likely to have optimistic views of general social conditions (Thal, 2017). These biased views of society could certainly lead to less support for government programs designed to address social problems since the affluent will have less concern for issues they are mostly unaware of or do not see as problematic. Of course, more contact between the rich and the poor would be one mechanism that would lead the affluent to have less biased perceptions and be more conscious of social problems. While we do not argue that intergroup contact will have a large influence on preferences for redistribution among the poor, it is possible that economic context can shape the policy attitudes of the poor through other mechanisms. For instance, when the poor live among those with higher incomes they may become more aware of their class position in society, less likely to view economic outcomes as meritocratic, and as a result more supportive of government redistribution (see Newman et al., 2015).

³ For instance, recently developed measures by Reardon and Bischoff (2011) and Watson (2009) are restricted to metropolitan areas. To get reasonable estimates of our economic segregation measures, we only calculate measures for zip codes that have at least 100 households. Overall, the measures are calculated for over 29,000 of the 33,000+ zip code areas created by the Census, which covers approximately 99% of the U.S. population.

⁴ This approach is similar to the one used by Johnston and Newman (2016), but our measures use income quintiles to account for rich and poor populations while their research defines low and high incomes based on absolute income thresholds. The Herfindahl–Hirschman Index (HHI) is a similar measure that has been used in previous research to account for a neighborhood’s income diversity (Minkoff & Lyons, 2019), but this measure requires the use of all income categories in its calculation. While this may be ideal for some research questions, our central focus is on the living patterns of rich and poor residents. Isolating the rich and poor, as we define these groups here, in order to study the effects of potential exposure between these two groups would not be possible using the HHI measure of diversity. Therefore, we believe our interaction approach is a better fit for our particular study.

⁵ Returning to the previous example, in a place with 90% rich and 10% poor the original rich insulation index would be 2.2 (i.e., log(.90/.10) = 2.2) as would a place with 9% rich and 1% poor. The population weighted measure, however, takes these differences into account. The weighted measure for the first example would not change since the proportion of rich and poor is 1 (i.e., 2.2 × 1 = 2.2), but the latter example would be substantially lower at 0.22 since this neighborhood is composed of only 10% rich and poor (i.e., 2.2 × .10 = 0.22).

⁶ The total sample sizes for each survey, without accounting for missing responses, is: 32,800 (2008), 55,400 (2010), 20,150 (2011), 54,535 (2012), 56,195 (2014), 64,600 (2016). There was also a CCES survey fielded in 2009, but the questions we rely on to measure policy attitudes were not asked in this year so it cannot be included in the analysis.

⁷ Responses to the questions tend to cluster around multiples of 25 (i.e., 0, 25, 50, 75, and 100) for both variables in all survey years, which can be seen in the distributional plots of the variables on the 0–100 scale (see Figure A1 in the supplementary materials section). Because of this, we re-estimate the main models presented below using five-category dependent variables in place of the 0–100 variables as a robustness check. The results can be found in supplementary materials Tables A2 and A3, and are substantively consistent with the results presented in the main text.

⁸ The zip code measures of the Gini coefficient, Black (non-Hispanic) population, and Hispanic population were obtained from the Census Bureau’s five-year ACS estimates. The urban indicator is a dichotomous variable that is equal to 1 for zip codes in urban areas and 0 otherwise. The urban status data is available on the Census’s Urban Area Relationship Files page (https://www.census.gov/geo/maps-data/data/ua_rel_download.html).

⁹ Support for intergroup contact theory holds when considering a variety of methodological approaches (e.g., experimental, quasi-experimental, and observational studies) and across a number of different group definitions (e.g., race, ethnicity, religion, age, and sexual orientation). See Pettigrew and Tropp (2006) meta-analysis for more details.

¹⁰ The intraclass correlation (ICC) for responses at the zip code level are all approximately 0.03 for the tax increases over spending cuts variable and all around 0.04 for the income tax versus the sales tax variable. These are fairly low correlations, which is one indicator suggesting that modeling the data using random intercepts at the zip code level may not be required. We still use multilevel models for our estimates since our data are structured at multiple levels and low ICCs do not necessarily indicate that multilevel analysis cannot improve estimates (see Nezlek, 2008). In any case, we replicate the models presented in Tables 1 and 2 in the main text using standard OLS regression and present these results in Tables A4 and A5 (see the supplementary materials section). The OLS results are substantively consistent with the results presented in the main text.

¹¹ In the supplementary materials section, we also present the results of the analyses found in Tables 1 and 2 after standardizing each variable to have a mean of zero and standard deviation of one. These results can be found in Tables A6 and A7.