Beyond Average Social Mobility: The Distribution of Economic Opportunity

Studies of heterogeneity in social mobility typically compare a few predefined groups. This paper instead develops the Conditional Mobility Forest, a nonparametric estimator of each child’s expected income rank at the 25th parental percentile. In the PSID, this measure of upward mobility varies by about 20 percentile ranks across children. A Shapley-value decomposition identifies race as the most important characteristic, followed by childhood health and parental age at birth. Parental education, homeownership, and household structure, despite strong raw correlations, contribute comparatively little once correlated characteristics are accounted for. These results can sharpen the targeting of social mobility policies.