I am testing for the independence of a specific set of variables with respect to a common event. Despite having a fairly decent sample size (n=236), the event I am testing for is relatively rare (2% of the sample). Consequently, when constructing my contingency tables, many cells for various variables have an expected count of less than 5.
As far as I recall, whenever this count is greater than 20%, there's a violation of chi-squared assumptions, and therefore, another test should be used instead. I have been taught that Fisher's exact test is well-suited for such instances.
I would like to conduct some measures of strength of association for those variables that exhibit a significant association with the event. However, I am unsure whether coefficients of effect size such as the contingency coefficient or Cramér's V, which rely on the value of chi-square, can be used when chi-square assumptions aren't met.
Although I have already computed the relative risk (as I am conducting a cohort study), I am concerned that comparing variable strength of association based solely on their relative risks could become convoluted. Thus, I believe Cramér's V or a similar measure might be better suited for the task.
I would highly appreciate it if you could confirm whether this measurement can be used and, if not, suggest alternative approaches.
