Small sample considerations in estimation from panel data

摘要: Abstract Recent interest in statistical inference for panel data has focused on the problem of unobservable, individual-specific, random effects and inconsistencies they introduce estimation when are correlated with other exogenous variables. Analysis this always assumed variance components to be known. In paper, we re-examine some these questions finite samples must estimated. particular, uncorrelated explanatory variables, show that (i) feasible Gauss-Markov estimator is more efficient than within groups all but fewest degrees freedom its never 17% above Cramer-Rao bound, (ii) asymptotic approximation similarly true remains significantly smaller moderately large sizes, (iii) estimators do not necessarily yield estimators.