CODEWITHSUNDEEP
pythonstatsmodelsfdr
Davide
Davide

Reputation: 5

FDR of neuroimaging data with Statsmodels: many of the corrected p-values have the same value

I ran a number of Pearson's R correlations and I then wanted to correct for multiple comparisons using Statsmodels. Specifically, I used this function. Given a list of p-values, the function returns whether the hypothesis was rejected or not, based on the provided alpha value, as well as the adjusted p-value, based on the method specified. In my case, I used the method "fdr_bh", which is the step-wise Benjamini & Hochberg (1995) method.

However, when assessing the adjusted p-values, I noticed that there are "groups" of p-values that have all the same value, and I don't understand if this is a normal behaviour or not. Is this related to point 2 of the second answer to this question?

This is an example of my adjusted p-values

This is the code I use, given the list of p val

fdr_p = multitest.multipletests(uncorrected_p, alpha=0.05, method='fdr_bh', is_sorted=False)

Any help is appreciated.

Thanks

Upvotes: 0

Views: 510

Answers (1)

Josef
Josef

Reputation: 22897

Is this related to point 2 of the second answer to this question?

Most likely yes.

All stepwise p-value correction methods have a monotonicity restriction.

This is imposed in the computation so that the corrected p-values are weakly increasing in the original, uncorrected p-values.

Upvotes: 0

Related Questions