Table 3.
Effects of College Completion on Proportion of Time in Low-Wage Work: Covariate and Propensity Score–Based Partitioning
| (a) |
(b) |
(c) |
|||||||
|---|---|---|---|---|---|---|---|---|---|
| IPW | NN Matching | Causal Forest (grf) | IPW | NN Matching | Causal Forest (grf) | IPW | NN Matching | Causal Forest (grf) | |
|
| |||||||||
| Propensity score strata: (a) = low, (b) = mid, (c) = high | −.226*** (.028) | −.205*** (.032) | −.208*** (.036) | −.137*** (.024) | −.132*** (.028) | −.134*** (.026) | −.098** (.030) | −.083** (.032) | −.101*** (.031) |
| Parental income terciles: (a) = low, (b) = mid, (c) = high | −.262*** (.037) | −.316*** (.034) | −.233*** (.046) | −.180*** (.026) | −.199*** (.033) | −.194*** (.043) | −.114*** (.024) | −.083 (.075) | −.091*** (.023) |
| Mothers’ education: (a) = less than HS, (b) = HS, (c) = college or higher | −.269*** (.034) | −.261*** (.034) | −.220*** (.042) | −.127*** (.026) | −.156*** (.030) | −.161*** (.038) | −.140*** (.025) | −.124*** (.031) | −.123*** (.025) |
| Ability terciles: (a) = low, (b) = mid, (c) = high | −.278*** (.048) | −.267*** (.040) | −.287*** (.063) | −.140*** (.029) | −.168*** (.030) | −.146** (.041) | −.128*** (.019) | −.118*** (.024) | −.120*** (.020) |
| Race: (a) = black, (b) = Hispanic, (c) = white | −.218*** (.032) | −.280*** (.053) | −.195*** (.048) | −.235*** (.028) | −.186*** (.046) | −.210** (.060) | −.165*** (.022) | −.189*** (.025) | −.157*** (.031) |
Note: Data are from the National Longitudinal Survey of Youth 1979 cohort. The sample is restricted to individuals who were 14 to 17 years old at the baseline survey in 1979 (n = 5,582), who had completed at least the 12th grade (n = 4,548), and who had no missing data on the outcome (n = 4,382). College completion is measured as a 4-year degree completed by age 25. Propensity score strata and parental income and ability terciles are 1 for low, 2 for mid, and 3 for high. For mothers’ education, 1 indicates less than high school, 2 indicates a high school degree, and 3 indicates some college attendance or more. For race, 1 indicates black, 2 indicates Hispanic, and 3 indicates white (these categories were based on an ordering of the probability of college completion). Estimates are based on NN matching with four control units per treated unit on the linearized propensity score and on causal forest (grf) estimates applied to each partition. grf = generalized random forest; HS = high school; IPW = inverse propensity weighting; NN = nearest neighbor.
p ≤ .01 and
p ≤ .001 (two-tailed tests).