Abstract
Several methods for analyzing untransformed data from negative multinomial distributions are presented with examples. These include a test of fit, a comparison of mean values, and an analysis of linear models. The techniques are generalizations of existing univariate methods used to analyze data from generalized Poisson distributions such as the negative binomial. Constrained minimum chi square estimators are derived separately for the marker parameter and the proportions for each variate in the negative multinomial distribution. The statistics required for performing these tests are based on sample factorial cumulants and sample zero frequencies for each variate in each distribution under consideration. The goodness of fit test is free of disadvantages associated with the Pearson chi square test and is more easily computed through weighted least squares. The advantages of analyzing mean values and linear models on untransformed data are discussed along with a history of the negative multinomial distribution.
Case, Robert Joseph (1981). Methods for analyzing untransformed data from the negative multinomial distribution. Texas A&M University. Texas A&M University. Libraries. Available electronically from
https : / /hdl .handle .net /1969 .1 /DISSERTATIONS -648505.