I have two categorical datasets, say, $A$ and $B$, which are sparse. I would like to apply pairwise $\chi^2$ tests to a certain categories, which are sufficiently populated (say, have expected values $>5$) in the following way. Let $k_{A}$ be number of elements of $A$ which are $k$, and $\neg k_{A}$ be number of elements which are not $k$. So, I have a contingency table of the form:

$$ \begin{array}{|r|r|} \hline k_{A}&k_{B}\\ \hline \neg k_{A}&\neg k_{B}\\ \hline \end{array} $$

After applying $\chi^2$ test, I multiply each $p$ by a number of tests I run, and this is a Bonferroni correction. And then I can conclude, that certain categories (having $p<0.05$ after applying correction) are significantly different in datasets $A$ and $B$.

Is this a correct? If, for example, I would simply collapse categories with low expected values into a catch-all category 'others', and run a single $\chi^2$ test, then I don't need to apply any corrections?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.