COLLEGE DIRECT AND INDIRECT EFFECT HETEROGENEITY

Ahearn, Brand, and Xiang Zhou (Harvard University) use data from the National Longitudinal Study of Youth 1979 and 1997 to examine how socioeconomic status and family outcomes mediate the effect of college attainment on voting participation, and how the total and mediated effects vary by propensity to complete college. Using machine learning causal mediation analyses, they find that the total effects of college on voting follow a negative selection pattern, such that individuals with the lowest propensity to complete college experience the highest gains. They use boosted regression trees, a flexible machine learning method with strong predictove accuracy to fit the three outcome models for voting. Although college-induced increases in socioeconomic status and family outcomes account for a non-trivial portion of college’s effects on voting, the mediating effects do not vary by propensity to complete college. Thus, their findings suggest that the pattern of negative selection observed in the total effect is driven by the direct effect of college on voting participation.