Key points are not available for this paper at this time.
On the basis of some numerical studies we argue that Bartlett adjustment of the likelihood ratio statistic for discrete data does not in general yield an improvement in the asymptotic error rate of the chi-squared approximation. Also we argue that the small-sample effect of the Bartlett adjustment is unclear and that it is in no way comparable with the improvement obtained for continuous data.
Frydenberg et al. (Sun,) studied this question.