Closer asymptotic approximations for the distributions of the power divergence goodness-of-fit statistics (Q760107)

The members of the power divergence family of statistics \(2(\lambda (\lambda +1))^{-1}\sum^{k}_{j=1}X_ j[(X_ j/n\pi_{0j})^{\lambda}-1]\) all have an asymptotically equivalent \(\chi^ 2\) distribution. An asymptotic expansion for the distribution function is derived which shows that the speed of convergence to this asymptotic limit is dependent on \(\lambda\). Known results for Pearson's \(X^ 2\) statistic and the log-likelihood ratio statistic then appear as special cases in a continuum rather than as separate (unrelated) expansions.