Sensitivity of point and interval estimates to distributional assumptions in longitudinal data analysis of small samples
Titel:
Sensitivity of point and interval estimates to distributional assumptions in longitudinal data analysis of small samples
Auteur:
Fellingham, G. W. Raghunathan, T. E.
Verschenen in:
Communications in statistics
Paginering:
Jaargang 24 (1995) nr. 3 pagina's 617-630
Jaar:
1995
Inhoud:
We explore the sensitivity of the point estimates of the fixed effects and variance components, and of interval estimates and hypothesis tests of the fixed effects, to the distributional assumptions about the random effects in longitudinal data analysis using the REML estimation procedure described by Laird and Ware (1982). In the small sample setting we consider distributions for the random effects that differ in tail thickness and skewness from the assumed normality. We investigate the bias, mean square error of the point estimates of the fixed effects and the variance components, the actual coverage of the nominal normal theory (1-α)% confidence intevals for the fixed effects, and α-level hypotheses tests performance under an asymptotic criteria. When random effects are not normal, but symmetric, point and interval estimates of fixed effects are unaffected, but estimates of variance components are highly variable. When random effects are from a skewed distribution, REML performs quite poorly