Possible superiority of the conditional MLE over the unconditional MLE

Abstract

Bias reduction, comparison of estimators, Kullback-Leibler loss, odds ratio, Pareto distribution, steep exponential distribution,

Estimating a model through the conditional MLE

“…Yanagimoto and Anraku (1989) gave seven models where the conditional MLE is properly superior to the unconditional MLE for a finite sample size and a finite number of strata. Bar-Lev (1984) and Jorgensen (1987) discussed the restricted exponential families of distributions, and studied the conditional MLE.…”

“…As Yanagimoto and Anraku (1989) noted, there is no sufficient evidence showing superiority of the conditional MLE over the unconditional MLE in the logit model. We compare the RKL(~,/2; 8, #) of the two MLE's.…”