R &R at Journal of Politics
Abstract: American politics scholars disagree on the extent to which voters use policy information to evaluate politicians, versus relying on partisan cues to evaluate policies. We demonstrate the coexistence of both of these perspectives by studying the degree to which objective facts (measured with local Covid-19 cases) and partisan cues (measured with President Trump's tweets about the virus) influence differences in the social distancing behaviors in Democrat and Republican counties in 2020. We find that both factors play an important role in social distancing, but that the relative importance between cues and facts favors the latter. Furthermore, the importance of these signals declines over time, suggesting a crucial but under-appreciated dynamic of how partisan positions evolve in a Bayesian framework.
Abstract: Existing work on the domestic politics of immigration in the United States has documented important relationships between political outcomes and the growth in foreign-born populations, focusing on an ex post response to changing local demographics. Less understood are the ways in which local governments anticipate future changes in immigrant population, strategically adjusting policies to constrain the movement of immigrants across jurisdictional borders before it can occur. This paper estimates these ex ante changes in local policies by predicting variation in the proposal of anti-immigrant legislation as a function of proximity to nearby immigrant populations. We find that a 10 percentage point increase in the share of foreign-born populations in a city's surrounding areas almost doubles the likelihood that a city considers a restrictive policy in our data, which we argue reflects an attempt to prevent the inflow of immigrants into its jurisdiction in the future. We test three competing theories for why cities pursue such policies, finding that our results are most consistent with a cultural threat hypothesis.
Not So Sanctuary: Electoral Incentives and Policy Outcomes in Sanctuary Jurisdictions
Abstract: In the wake of President Trump's election, a growing number of local jurisdictions adopted a variety of reforms that attempt to protect the immigrant residents, labeled as sanctuary policies. However, the word "sanctuary" does not have a legal definition, and the policies that are declared as sanctuary take many different forms. Because the term has no legal meaning, the local government's decision on whether to self-identify as one is a political one. I hypothesize that a declaration of sanctuary jurisdiction is a political tactic exercised by local politicians who seek to secure greater electoral support or to meet the demands made by their majority constituents. I examine whether the sanctuary jurisdictions (irrespective of the specific policies they enact) actually reduce the number of detainees compared to non-sanctuary jurisdictions after the adoption of sanctuary status. The null result of this analyses would support the claim that the sanctuary jurisdiction declaration is likely to be a function of a political strategy rather than of a real intention towards protecting the immigrant residents that are at risk of deportation. I then examine the evidence of the political motivation by exploiting the timing of the declaration of sanctuary policies. Specifically, I test whether the announcement of sanctuary policies coincides with the timing of local elections, and if so, whether such declaration boosts the electoral support for those local officials involved in the policy decision.
Predicting Individual Ethnicity Using Image Recognitions
Abstract: American political scientists often rely on Bayes's rule to infer ethnicity from a surname and geography. However, this approach suffers from its low accuracy rate for racial minorities and ignores the rich information embedded in the first names of individuals. I develop two ethnicity classifiers: full name-based classifier based on name embeddings and image-based classifier with google image search of first and last names. The prediction accuracy rates of these two approaches are evaluated against the conventional Bayesian approach. Ensemble Bayesian Model Averaging is used to improve prediction by pooling estimates from the three models to generate ensemble predictions.