Redundancy test for some variables in linear discriminant analysis with two-step monotone incomplete data

We evaluated the null hypothesis such that a set of variables is redundant in linear discriminant analysis using two-step monotone incomplete data. We derived a likelihood ratio test statistic and proposed a corrected likelihood ratio test statistic via the Bartlett correction. Finally, we evaluated the proposed test statistic and compared it with the likelihood ratio test statistic using Monte Carlo simulations. The results show that the proposed test statistic can control type I errors more accurately than the likelihood ratio test.