Accounting for dynamic selection bias in educational transitions: The Cameron–Heckman latent class estimator and its generalizations☆

摘要: Abstract The sequential logit model of educational transitions has long been the dominant modeling framework for study inequality opportunity ever since seminal works Mare, 1980 , 1981 . But conventional applications are known to be biased by ubiquitous presence unobserved heterogeneity. Cameron and Heckman (1998) propose a that allows two or three latent classes if selection bias is solely generated person-specific component stable To evaluate class regression estimator, this makes use simulated data eliminate influences other problems transition modeling. simulation based on five independent pairs large samples from standard distributional assumptions new estimator appears an effective way adjust dynamic when family background effects transition-invariant sample size in order ten thousand above. By contrast, produces results very different generating models. This also considers alternative ways improve statistical efficiency: (1) incorporate crude indicator heterogeneity; (2) pool effect estimates across transitions, variables, estimators smooth out noise under null hypothesis invariance. In addition, examines impact reliability performance models suggests practical guidelines.