The effect of misspecifying the random-effects distribution in linear mixed models for longitudinal data