Results change based on reference?

I am trying to detect differentially abundant species based on body location (for ex: face vs arm vs neck).

Just a simplied example of what I am encountering:
If I choose face as the reference, let’s say I see 2 species different between face and arm and 4 between face and neck. Next, I switch to neck as the reference, for some reason, I may get 2 different between neck and face. This is confusing to me because when I chose face as the reference, I saw 4 different species. Why is this number changing? Is it something to do with the reference chosen? What is the “true” value?

Thank you!


Apologize for the late response but changing the reference will change the question that you are trying to model. As such coefficients can change and more importantly the p-value distribution can change between your multiple tests which will impact the q-values during FDR correction. I would double check to see if the coefficients are similar (but in opposite directions).

Jacob Nearing