25A more efficient approach still would be to employ a random effects estimator which optimally combines the variation in the cross-section with that over time, but the key features of the power advantage of panel data are readily observable in the comparison of these two estimators.