Principal components in the nonnormal case: The test of equality of q roots (Q793465)

The problem considered is that of the limiting distribution of the likelihood ratio statistic, \(W_ q\), for testing the equality of q characteristic roots of the covariance matrix when normality does not hold, though the statistic is derived on normality assumptions. This limiting distribution is derived as a linear combination of, central or non-central, chi-squares with one degree of freedom, with coefficients depending on the fourth cumulants. The case of an elliptical parent distribution is investigated in detail; that is one with density \(c| \Sigma |^{-frac{1}{2}}g(x^ t\Sigma^{-1}x)\), for \(\Sigma>0\) and a non-negative function g. It is suggested that \(W_ q/({\hat \kappa}+1)\) be used as a test statistic, on the basis of theoretical investigations, where \({\hat \kappa}\) is a consistent estimate of the kurtosis parameter. Sampling experiments investigate this procedure and suggest it to be reasonable for sample sizes of 50 and greater.