I have a dataset including Categorical variables and involves multiple response variable. 20% of the cells in cross tab has less than 5 frequencies and hence I cannot carry out a Chi-square Test. Fisher's Exact test is applicable in such cases but I would like to know if there is any tutorial which shows how to carry out Fisher's exact test in SPSS with multiple response dataset. The tutorials for fisher's exact test in SPSS i found does not include multiple response variables.
The number of cells with low frequencies is unimportant. The chi-square test in this case is generally thought to perform OK when 80% or more of the expected values (not the observed frequencies) are 5+. In these cases statistical power can still be poor (and that's probably the main practical concern).
The Fisher exact test is not a good choice as an alternative as it assumes fixed marginals (which is rarely plausible). To increase statistical power and improve the chi-square approximation you might consider collapsing cells. you could also use the Egon-Pearson correction which, in my view, has a stronger basis than the Fisher exact test. e.g., see https://seriousstats.wordpress.com/2019/09/05/chi-square-and-the-egon-pearson-correction/
Hi Thom Baguley. I have one question about your nice note on the Egon Pearson correction. As you know, Ian Campbell's simulation study examined 2x2 tables only. Are you advocating use of Egon Pearson's correction to Pearson's Chi-square for tables larger than 2x2 as well? I know that one of my statistical gurus (Ray Koopman) once speculated that the correction could be applied to tables of any size. But I do not recall ever seeing any textbooks or articles that directly address that issue. Thanks for clarifying.
Bruce
Good point. It seems plausible as the chi-square statistic is liberal and this makes it more conservative (but not as conservative as Yates' correction). It looks like the obvious thing to do is some simulations ...
- Badges
- Science topic