After reading this survey of methods for controlling FDR and considering my problem at hand (https://genomebiology.biomedcentral.com/articles/10.1186/s13059-019-1716-1), IHW seems like a good choice.

In order for IHW to perform well, it needs an "informative covariate which is independent of the p-value under the null hypothesis". In other words, something meaningful that isn't related to the p-value when the null hypothesis is true. My understanding is based on this: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4930141/.

In my case, my hypotheses correspond to tests for association on 2x2 contingency tables. It seems like maybe the count of positives (sum of the first row in each contingency table) might be a good choice? I'll of course make the diagnostic plots to see if the choice makes sense. But I still don't have a good grasp of what the covariate is supposed to represent and how to choose one, even after looking at some examples.

Can anyone help with my understanding?


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.