Association for Public Policy Analysis & Management

A growing body of econometric literature examines the effects of a variety of tuition and aid policies on the experiences of undocumented migrants residing in the United States. Although government surveys do not currently ask residency status directly, the resultant data are known to capture responses from undocumented migrants. Researchers interested in undocumented populations are, thus, confronted with a dilemma: engage admittedly noisy data to offer imprecise, but best-available results or develop data sources that might more accurately and consistently capture information from this population. Given the low likelihood that governmental data would ever be able to isolate with perfect precision, residency-status information from a population which would be vulnerable to a buffet of adverse actions should this information be known, scholars have acceded to the former option.

In a preliminary attempt at mitigating this limitation, scholars have invariably identified foreign-born non-citizens of Mexican or Latino ancestry as a treatment proxy for likely undocumented migrants. As a first pass, this makes sense as an ameliorative strategy: a majority of undocumented migrants come to the U.S. from Mexico and Central America as a predictable function of geography. However, such identification improperly reduces undocumented-migrant status to Mexican and Latino origin. This creates two distinct, but related problems. One, it requires the empirical assumption that all Mexican or Latino non-citizens are undocumented. Two, it denies the well-known reality that many undocumented migrants are neither Mexican nor Latino. In tandem, these assumptions confound our ability to understand phenomena affecting undocumented migrants with phenomena affecting individuals because of their racial, ethnic, or national origin.

In this paper, I address these assumptions by introducing weighting across the panel of governmental data as an alternative strategy. In extension and elaboration of a crude weighting strategy I used in a previous paper, here I propose calculating as a stabilized weight, the ratio of the marginal probability of a survey participant being undocumented given their national-origin ancestry to the inverse probability of the same participant being a member of the intended tuition or aid-policy treatment group conditional on observed covariates.

To example this strategy in action, I revisit seminal papers on in-state-resident tuition- and aid-policy effects on undocumented migrants’ postsecondary educational attainment. Preliminary results suggest that refinement of undocumented-migrant identification strategies yields meaningful differences in policy estimates. As hypothesized by several scholars, previous estimators attenuate policy effects in terms of both magnitude and statistical significance. While the proposed weighting strategy cannot fully decompose measurement biases, adoption of the same in evaluative models should yield more precise estimates of policy effects on undocumented migrants. This strategy will also enables the study of national-origin variation among undocumented populations, with implications in both regards for research, policy, practice, and advocacy.