In my research I am looking at the effect of 3 IV's on a binary DV that is measured repeatedly on the same sample after different treatments. Therefore, for every treatment, I have created 3 binary logistic regression models that each test the effect of a different IV together with a set of control variables. Thus:
Treatment 1: Model 1: IV1 + controls Model 2: IV2 + controls Model 3: IV3 + controls
This for every treatment. For methodology reasons interaction effects are not necessary/possible.
The results have been very interesting and useful until I got to a certain treatment, let's call it treatment X. For this treatment, none of the logistic regression models were significant. I have tried omitting the least significant control variables, but the model does not improve enough. None of the IV's has a significant effect on the DV in this treatment, so that is a conclusion in itself. However, how can I explain in the discussion why it was not possible to create any significant model for this condition?