Does policy threat mobilise? 287(g) and Latino voter registration in North Carolina and Florida

ABSTRACT

Immigration enforcement affects political attitudes and trust even among individuals and communities not directly targeted for deportation. Literature from across the social sciences finds that restrictive local immigration policy has chilling spillover effects on citizen Latino political attitudes, trust, and interactions with state institutions – but few studies have extended this framework to formal political behaviour. In this article, I contribute to the literature on spillover effects of immigration enforcement policies with a new measure of mobilisation and a credible identification strategy. Specifically, this study identifies the effects of restriction on Latino voter registration by leveraging the county-level selection process for the 287(g) Program in North Carolina and Florida. Contrary to expectations, I find little convincing evidence that acceptance into the 287(g) Program decreased Latino voter registration. These null results are consistent across different 287(g) Program types and modelling strategies that relax temporal assumptions about policy effects.

KEYWORDS:

• Mobilization
• threat
• Latino
• registration
• immigration

Acknowledgments

The authors grateful for helpful comments from Alyssa Browne, Clau Dermont, Dalston Ward, David Attewell, Dominik Hangartner, Emily Wager, Jacqueline Hagan, Leah Christiani, Rahsaan Maxwell, Ruy Manrique, Stefan Schütz, as well as the editors and anonymous reviewers at the Journal of Ethnic and Migration Studies.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 This study focuses on counties in North Carolina and Florida. Floridian and North Carolinian undocumented populations differ on a number of dimensions, most notably legal status and country of origin. Whereas 43% of immigrants are undocumented in North Carolina, only 20% are undocumented in Florida. Moreover, the majority (61%) of North Carolina's undocumented immigrants are Mexican nationals. Mexican nationals are also the largest national-origin group among undocumented residents in Florida (28%), but the undocumented population is more diverse. Though undocumented populations are far from homogeneous across the two states, undocumented populations in North Carolina and Florida are both predominantly from Mexico, Central-, and South-America (78% for Florida and 84% for North Carolina), suggesting that a program targeting Latino immigrants has the potential to effect Latino citizens in both states. See: https://www.americanimmigrationcouncil.org/research/immigrants-north-carolina and https://www.americanimmigrationcouncil.org/research/immigrants-florida.

2 Additionally eight independent city or multi-county districts in Virginia as well as one Commonwealth district applied.

3 Complete North Carolina voter rolls were downloaded from ncsbe.gov on 30 November 2016. These files include all individuals registered to vote on early November, 2016. Complete Florida voter files were collected on June, 2017, from flvoters.com. These files include all individuals who registered in May, 2017.

4 Quasi-Poisson, rather than Poisson, due to over-dispersion in the data. I shows the results of regression-based tests for over-dispersion per Cameron and Trivedi (1990) in Table B2 in the Appendix. For all models, dispersion parameters when using Poisson regression are greater than zero at p<0.05, evidence that alternatives to Poisson models are more suitable.

5 Subsequently referenced tables are presented in the Appendix.

6 The ideal denominator would be Latino citizen voting age population, but the census has only produced annual estimates of this at the county level since 2011, with the first estimate centred on 2007.

7 See Tables B7 – B8 for group placebo tests with African-American registration outcomes. See Tables B9 – B10 for group placebo tests with White unscaled and scaled registration. For both populations, I fit quasi-Poisson models to the raw counts as well as OLS models to the scaled outcome.

8 See: http://www.ncsl.org/research/elections-and-campaigns/voter-list-accuracy

9 Empirically this simulation does not attempt to distinguish between types of attrition. In other words, I do not model whether attrition of ‘unobserved’ registrations in my data is due to out-migration of Latino registrants or removal of inactive Latinos registrants.

10 See Appendix B2.1 for more information on simulation.

11 Hybrid Models combine Jail Enforcement and Task Force models, but I chose to exclude Hybrid Models from the Jail Enforcement category to maintain a narrower comparison between Jail Enforcement and non-Jail Enforcement counties.

12 The fifth lag averages over all post-treatment periods after the second year.

13 This should also increase confidence in the parallel trends assumption in analyses above.