APPLIED MULTIVARIATE STATISTICS: WITH SAS® SOFTWARE

6.9. Concluding Remarks

The models and methods considered in this chapter are limited in that (i) only models which are linear in fixed or random coefficients are considered (ii) the error component was assumed to be additive (iii) normality of error is assumed and (iv) usually some appropriate covariance structures on errors and random effects are assumed. Obviously situations where there is substantial departures from one or more of these assumptions will require some other alternative approaches. Additionally, most of the statistical tests are asymptotic and/or approximate and may not necessarily possess certain desirable properties such as a reasonably high power or the ability to maintain the type I error at the specified level. Thus caution ...