Changes in immigration policies concerning specific visa holders can affect both the outcomes for those targeted and selection into different types of immigration. Understanding these effects can be challenging given that publicly available datasets do not include an individual’s visa status, which is necessary for isolating who would be impacted by specific immigration policies and what the policy’s individual-level impacts are. This paper explores the prospect of causally identifying the effects of visa policy changes using survey data. Specifically, I evaluate the effects of a 2015 change in immigration policy in the U.S. that allowed H-4 visa holders (spouses of high-skilled H-1B holders) access to employment authorization (EAD). This setting uniquely enables me to study both the policy’s direct effects on H-4 visa holders and its indirect selection effects. I use natural language processing (NLP) and unsupervised machine learning (ML) to identify likely H-1B and H-4 visa holders in the American Community Survey. After validating my classification through a series of checks, I estimate matched and within-tenure event studies to isolate the policy’s effects. I find that one year after the policy was implemented, the employment rate of H-4 visa holders increased by 44.5 percentage points relative to naturalized citizens, and the fertility rate increased by 6.3 percentage points relative to other noncitizens. I use descriptive evidence to suggest that the H-4 EAD policy did not change the composition of H-1B or H-4 workers, nor did it alter marriage formation decisions. Overall, this paper uses the example of the change in policy impacting H-4 visa holders to show how a simple machine-learning-based approach to visa classification can be used to analyze the effects of immigration reforms on economic, family, and selection outcomes.
