Cressie and Read(1984)introduced the class of multinomial goodness-of-fit statistics, R
a, based on measures of the divergence between discrete distributions. All R
a have the same chi-squared limiting null distribution. The power of R
a is usually approximated by a noncentral chi-squared distribution that is also the same for all a. In this paper, we propose a new approximation of the power of R
a. The new power approximation, NT, is a normal approximation based on a normalizing transformation. The NT approximation is numerically compared with the other approximations. As a result of the comparison, we find that the NT approximation is superior to the other approximations when a=0(the loglikelihood ratio statistic)and is effective when a<0.
抄録全体を表示