Using Machine Learning to Estimate the Effect of Undocumented Status on Education-Occupation Mismatch for College Graduates

• In Event: Newcomer Students' Pursuit of Higher Education in the United States: Examining Access and Returns

Abstract

In principle, obtaining a college degree is a form of human capital investment that should lead to more favorable labor market outcomes for undocumented immigrants. However, foreign-born skilled immigrants face difficulty navigating the U.S. labor market and are more susceptible to education-occupation mismatch and wage penalties (Li and Lu, 2023). Research on undocumented immigrants shows that the lack of legal status generates additional penalties through employer exploitation and mismatch (Hsin and Ortega 2018).We use data from the ACS from the years 2013-2019 to examine education-occupation mismatch and associated wage penalties amongst undocumented college graduates. We define a worker as vertically mismatched if they are employed in an occupation that doesn’t match their educational attainment and horizontally mismatched if they are employed in an occupation that doesn’t match their field of study. While the advantage of using the ACS is that it is a large, nationally representative dataset, it does not include a direct measure of whether an individual is undocumented. The empirical challenge of imputing undocumented status is well documented, and each approach has its limitations (Van Hook et al., 2015). More recently, machine learning models such as the K-Nearest Neighbors (KNN) classifier and the Random Forest (RF) algorithm have emerged as innovative approaches to imputation. Following (Ruhnke et al., 2022), we use the second wave of the 2008 Survey of Income and Program Participation (SIPP) as the donor sample, which contains an accurate measure of undocumented status. The machine learning models are trained on the donor sample to classify whether an individual is undocumented, and the model predictions are then applied to respondents in the target sample (ACS) where accurate information on undocumented status is not available. Our regression results show that undocumented skilled immigrants are more likely to be vertically mismatched and horizontally undermatched. We estimate a wage penalty ranging from approximately four to seven percentage points, depending on the undocumented imputation method. For college graduates that are eligible for DACA, we find evidence of a smaller likelihood of mismatch, which provides suggestive evidence that the federal policy removed some occupational barriers for this population. However, wage regression results provide more mixed evidence on the wage penalty, with estimates varying widely across the imputation methods.We then examine the effect of state-level policies related to educational and labor market barriers for undocumented immigrants (Samari et al., 2021). For vertical mismatch, immigration enforcement appears to generate the largest reduction in vertical mismatch versus policies more related to employment and identification. Specifically, vertical mismatch is approximately two to three percentage points lower in states that do not cooperate with federal immigration enforcement. On the other hand, prohibiting E-Verify plays a more important role in reducing horizontal undermatch for undocumented immigrants. After controlling for mismatch, prohibiting federal immigration cooperation has a large positive effect on wages for undocumented immigrants (estimates ranging from three to seven percentage points).

Author

• Mario Alberto Arce Acosta, University of California, Davis
  Presenting Author