When assessing whether a model warrants fixed- or random-effects, threats posed by, among other factors, autocorrelation and heteroscedasticity typically loom large. A recent discussion of this topic emerged on the StataList (click here) and considers various approaches.
Comments