On asymptotic distributions of normal theory MLE in covariance structure analysis under some nonnormal distributions (Q1284054)

The asymptotic distribution of the MLE parameters in a covariance structure model based on a multivariate normality assumption has been studied in the past. In these papers, the authors examined the case when the true sampling distribution is far from normal, especially when the sampling distribution is asymmetric. The asymptotic distributions of the MLE parameters are derived for two classes of non-normal distributions. It is shown that the marginal skewness does not affect the asymptotic distribution of the MLE for both classes. Moreover, existing results from the family of elliptical distributions are verified. Normal theory can be used for skewed data sets sampled from a pseudo normal distribution. In addition normal theory standard error is still valid even when the sampling distribution is far from normal indicating that the asymptotic distribution of the MLE of the parameters derived from the normal theory can give correct inference. A factor analysis from non-normal populations is considered as an example. It is shown that the asymptotic results lead to valid conclusions.