 |
 |
Hi There I have 2 sets of count data A and B that I would like to compare. In both cases there are K outcomes and N1, N2 observations respectively split amongst the K outcomes. Now I am familiar with the Chi^2 tests for homogeneity or goodness of fit. But they compare the whole distribution. I am interested in pairwise comparisons of the K individual outcomes to see which of the outcomes are "different" between A and B. If I assume my data are multinomial the proportions for the individual K outcomes are Binomial. Does it make sense to compare the proportions of individual outcomes between A and B using the large sample Normal Approx to the Binomial dist. For example consider outcome K_i for A and B we have counts a and b so then we would have the following proportions p_A ~ Bin( N1, a/N1), p_B ~ Bin(N2, b/N2). We could then test p_A - p_B using the Normal Approx? And repeat as necessary for each of the K outcomes or is there a better way? Best Regards, Richard