On Dec 23, 1:42 pm, czytaczg...@gmail.com wrote: > Hi, > > Using a short questionnaire I have interviewed two populations. For majority of questions the answers are of the nominal type only (there are more than two possible answers which are self excluding and complete). Now, I would like to check if these populations are/are not the same with respect to the frequencies given to answers. Therefore, the every possible answer was coded as a distinct integer number. This gives me a probability distribution. Can I use statistical tests to compare these distributions? If yes - would be the outcome of a test robust to the way in which the numbers are adjusted? > > Best, > Gruppo
To compare the groups on a single question, do a Pearson chi-square on the corresponding 2 x K contingency table, where K = the # of possible answers to that question. If you do that for several questions, you should adjust your threshhold for declaring a result "significant". The simplest procedure is the so-called "Bonferroni correction": a test is significant only if the product of its p-value times the number of tests is less than the usual threshhold (e.g., .05).