Separating different individual effects in a panel data model

Summary In this paper we consider a panel data model with individual effects that are arbitrarily correlated with the explanatory variables. The effects are composed as the sum of two different interpretable components, such as inefficiency versus heterogeneity in a production frontier setting, or ability versus socioeconomic background in an earnings function, or genetics versus environment in an epidemiological analysis. We wish to predict the two components separately. This is made possible by assuming that there are observables that are correlated with the first component but not with the second, and other observables that are correlated with the second component but not with the first. This can be true in terms of either simple correlations or partial correlations.