This paper examines a modification of Pearson'sX2test statistic when used in the presence of small cell expectations in the one-way multinomial. This modified test, denoted asT2in this study, maintains type I error rates that are very much closer to the nominal values than those ofX2. Simulated power results also show that bothX2andT2behave very much alike and that differences in power values are not too apparent. ConsequentlyT2may provide a better statistic thanX2in terms of attained #x3B1; levels in small samples situations.
展开▼