Population-Averaged Inference: Generalised Estimating Equations for Income Dynamics and Upward Mobility

This chapter develops a population-averaged approach to intergenerational mobility using generalised estimating equations (GEE), contrasting it with subject-specific estimators and motivating working correlations (independence, exchangeable, AR(1)) for longitudinal dependence. We implement Gaussian and logit GEEs in SAS PROC GENMOD with REPEATED to model annual income and an upward-mobility indicator. Using a synthetic LAD-style panel, we apply diagnostics and select models via quasi-likelihood under the independence model criterion, retaining time-invariant parental covariates central to mobility research. Empirically, education and employment dominate income and mobility; parental quintile predicts mobility but not income. Results are robust across working correlations, with independence/exchangeable often preferred.