Economic anxiety or ethnocentrism? An evaluation of attitudes toward immigration in the U.S. from 1992 to 2017

ABSTRACT

Does “economic anxiety” explain attitudes toward immigration or can we better understand attitudes toward immigration as an outcome of ethnocentrism? This is a long-standing empirical debate in immigration opinion research and a debate that has struggled to distinguish the relative effects. I help settle this debate with a battery of analyses on attitudes toward immigration across the American National Election Studies and Voter Study Group data, spanning analyses on immigration opinion for white Americans from 1992 to 2017 with economic data at levels as granular as the state, county, core-based statistical area (CBSA), and the ZIP code. My analyses are unequivocal that ethnocentrism is reliably the largest and most precise predictor of attitudes toward immigration. Further analyses and simulations from models most consistent with the economic anxiety argument show that a standard deviation increase in ethnocentrism is still a greater or equal magnitude effect than all economic anxiety proxies combined. I close with implications for immigration opinion research.

KEYWORDS:

• Economic anxiety
• Ethnocentrism
• Immigration attitudes

Highlights

• Subjective and objective indicators about the economy have no robust effect on attitudes about immigration.

• Ethnocentrism is reliably the largest and most precise predictor of an anti-immigration opinion.

• Ethnocentrism has larger magnitude effects on attitudes against immigration than all economic indicators combined.

Supplementary material

The supplemental data for this article can be accessed here.

Notes

¹ The “legally” qualifier is a unique word choice in the July 2017 wave. Other prompts in other waves substitute “illegally” for “legally,” making these other waves less suitable for evaluating the labor market competition hypothesis.

² I leverage the temporal metadata available in all analyses. ANES stopped recording the month of the interview after 2008, so I benchmark the economic anxiety indicators for later ANES waves to November of the survey year. ANES post-election interviews usually transpire from November to January after the election.

³ ANES has county-level metadata available for only the first three waves (1992, 1994, 1996) I use in my analyses. State-level unemployment variables will appear in all models but the national-level unemployment statistics will not appear in the VSG analyses since the VSG analyses are cross-sectional.

⁴ The last category—“farming, fishing, and forestry”—is not typically included in analyses on job polarization, but Kolko (2016) argues to include the farming sector in considerations of exposure to automation/outsourcing because of BLS forecasts of employment declines in this sector over the next decade (Bureau of Labor Statistics, 2017). I choose to do this as well because this variable played a large role in his argument that Trump’s vote share correlated with this measure.

⁵ The appendix contains results of the main analyses in this manuscript that rethink these “economic anxiety” indicators in a variety of ways, like three-month, six-month, and 12-month differences in unemployment rates as well as 12-month differences in the tax return data. This approaches “economic anxiety” less as a “level” and more of a phenomenon that is measured as it potentially “increases.” The results of those estimations are broadly in line with the results I provide in the manuscript.

⁶ The ANES did not include a thermometer rating item for Hispanics as a group in 1998, which would exclude the 1998 wave from this analysis.

⁷ The appendix contains the full results from these models estimated without ethnocentrism.