Transformation of a Discrete Distribution to Near Normality

Utilizing an information number approach, we propose an objective method for the normalization of either discrete distributions, or sample counts, by means of a power transformation. Approximations are also given to the original known probabilities. Next, we derive the large sample distribution of our estimate of the power transformation. We compare our methods with the Box-Cox procedure, applied to observed counts, and conclude that their technique often provides good approximations even though their underlying assumption of normality is clearly violated. Two examples illustrate our methods.