Joining Panel Data with Cross Sections for Efficiency Gains

Under the classical linear regression model assumptions, fixed effects estimates properly control for time-invariant unobservables and produce unbiased estimates. However, they often rely on limited data variability and present high standard errors. We present an innovative methodology that complements longitudinal data with other sources of unpaired data to increase estimation efficiency. The methodology assumes that the are no time varying unobservables correlated with the observables and with the fixed effects. We apply the methodology to three sets of Leaving Standard Measurement Study data from Nicaragua and estimate a household consumption model. We find that, if the correlation between observables and unobservables does not vary across time, our methodology has the potential to lead to unbiased and more efficient estimates.