Gary Chamberlain has worked on the use of sibling data to measure economic returns to education in the presence of unmeasured ability variables. He has developed methods for using longitudinal data to deal with omitted variable bias. He has applied quantile regression methods to provide richer descriptions of the structure of wages, and developed Bayesian methods to provide predictive distributions based on longitudinal earnings data. He has worked on asymptotic efficiency in estimation with conditional moment restrictions, in semiparametric models with censoring, and in panel data models with correlated random effects. He has applied finite-sample decision theory to instrumental variable models that address selection bias, and to panel data models. He has used Bayesian decision theory to develop rules for individual treatment choice that condition on characteristics of the individual and on a sample of other individuals with data on characteristics, treatments, and outcomes. He is currently interested in methods motivated by education data sets, including the measurement of teacher effects.