A note on misspecification in general linear models with correlated errors for the analysis of crossover clinical trials