Residential Mobility Constraints and Immigration Restrictionism

Whereas voice involves directly expressing grievances to other group members, exit is defined by withdrawal from the group.

According to Hirschman, loyalty to the group plays an important role in “activating voice” and leaving “exit at bay.”.

It is important to note that the existing theory focuses on predicting responses in areas where demographic change has occurred. If we apply the general theory described by Hirschman, exit costs should only be relevant under conditions of organizational decline. Similarly, characteristics of potential moving destinations should only be relevant to the Fight-Flight decision in the presence of demographic change. The evidence presented across the three studies suggest that while the marginal effect of surrounding immigrant composition increases as a function of \(\Delta\) immigration, it is not zero in places that have either remained the same or lost immigrant populations. Still, explaining these additional cases (which comprise approximately 15% of the sample) could introduce additional complexity at the cost of parsimony. Since the focus of this paper is on contexts undergoing demographic change, the theory is constrained in its ability to explain these cases. Future studies could expand upon the existing model by exploring why surrounding immigrant composition often has the opposite effect in communities experiencing influxes of native-born residents.

A strict view of symbolic politics would argue that racial attitudes are entirely symbolic and a function of early childhood socialization experiences. Much like the rest of the literature on context effects, the existing theory is inconsistent with a strict view of symbolic politics. However, a more flexible view of symbolic politics might suggest that while racial attitudes are formed due to early childhood socialization experiences, individuals’ local environments can weaken or bolster those attitudes. Since the question of why individuals might be compelled to view “fight” or flight as responses to demographic change is not explicitly defined in the theory, both symbolic and realistic concerns could motivate these responses, and thus, the theory can accommodate this more flexible view.

Mobility data from the Census and the Panel Study of Income Dynamics (PSID) are two alternatives to the 2000 SCCBS. However, a complete test of the theory is not possible with either data set. Although the Census provides information about mobility patterns, they are aggregated at the county level, and thus, are not well equipped to test individual-level hypotheses. Moreover, political outcomes are not measured by the Census. While the PSID can be used to test individual-level mobility patterns, it does not include any measures of political attitudes or behavior tied to immigration.

The use of a ten-year interval is common in the immigration literature (Hopkins 2010; Alexseev 2006; Newman and Johnson 2012).

Though the Internal Revenue Service (IRS) collects fine-grained mobility data, these data are only released at the state level. Therefore, they are not helpful for assessing community-level demographics.

Models of residential mobility often consider static characteristics of surrounding communities as predictors of moving (Segal 1979). Thus, the use of static levels of surrounding immigrant composition is consistent with extant theories. Moreover, I consider the interaction between local and surrounding changes in immigrant populations, and fail to find evidence of conditional effects (see Online Appendix A).

A thorough search for data sets measuring mobility intentions using the Inter-university Consortium for Political and Social Research (ICPSR) database revealed that virtually no studies asked about residential mobility and political attitudes. While this measure is not ideal since it does not explicitly ask about mobility intentions, it still captures whether respondents plan to remain in their communities.

While this measure departs from the original formulation of voice described in Hirschman (1970), it is possible to recast this measure as an indicator of voice, given that the Social Capital Community Benchmark study was conducted in partnership with community organizations, and thus, respondents could have interpreted the survey context as one where they could relay their concerns to other community members. Still, even if the use of this measure might introduce some conceptual slippage, existing theories in the racial context literature do not explictly incorporate the role of residential mobility in political behavior, and using EVL as a general framework produces a more nuanced view of context that synthesizes research across multiple disciplines.

The correlation between percent Hispanic and \(\Delta\) Immigrant is .31 (Spearman’s rho).

Missingness for individual-level variables ranges between zero (e.g., gender) and 10% (e.g., income). The largest source of missingness in the data set involves contextual data, where roughly 13% of the observations lack information about local area demographics due to changes in ZIP codes over time. I perform listwise deletion in all of the analyses, given that multiple imputation is only preferable over listwise deletion when data are missing at random (Pepinsky 2016); an assumption that seems unlikely in this case.

It is important to note that the findings presented using multi-level models also hold in completely pooled models. I also present a set of simplified models in Online Appendix C that show that the key interaction holds across various specifications.

This is because mobility considerations are measured using a dichotomous item, and anti-immigration expression is measured on an ordinal scale.

The estimated coefficients are still statistically significant if two-tailed tests are used.

Instead of assuming that Hispanic and immigrant composition are equivalent, immigrant composition is modeled as a bundled treatment consisting of immigrants from various ethnic backgrounds. This assumption does not imply that Hispanic immigration is not a relevant determinant of exit or voice, but rather that immigrant status should outweigh ethnicity.

These models are estimated independently, and therefore, these groups of residents need not overlap.

The 2010 Census dropped questions about foreign-born status. Therefore, the 2011–2015 American Community Survey is used.

The correlation between pct. Hispanic and \(\Delta\) Immigrant is approximately  − .069 (Spearman’s rho).

The percent of missing values for individual-level variables hovers between zero for gender and 31% for Trump support. This is due to the fact that non-voters were not included in the analyses. As for contextual variables, the percent of missing values falls between .002% for variables such as Median Home Value and 3% for \(\Delta\) Immigrant.

The findings presented here also hold in completely pooled models.

The results also hold in the full sample.

For example, Kaufmann and Harris (2015) find limited evidence of white flight in response to growing immigrant populations. Understanding the conditions under which this dynamic operates should yield important insights about relationships between residential mobility and political behavior across contexts.