Moderation: what if one predictor variable affects another?

Hi guys! We would be very thankful for some insight on our moderation analysis.

We have two predictor variables, one is manipulated, the other is measured. There’s a measured outcome variable.

We would like to do a moderation analysis where the measured predictor variable moderates the effect of the manipulated predictor variable on the outcome variable.

The issue is that we found that the manipulated predictor variable affects the measured predictor variable.

Does this finding invalidate the model? Because otherwise, we found the moderation effect we were searching for.

Thanks in advance for any answers!


Less is more. Stay pure. Stay poor.
Can you draw this out or provide context (variable names), so we can better follow. I am not understanding the issue. Variables affecting each other's impact on the target variable is pretty much the definition of moderation.
It's research in Consumer Psychology.

We have 4 sponsored Instagram posts by 2 Instagram influencers. Both have a post that is "congruent" and one that is "incongruent" to them (categorical manipulated predictor variable), meaning it either fits their personal brand or doesn't.

The measured interval predictor variable is parasocial interaction (PSI = the interpersonal feelings the subject feels towards the influencer).

The outcome variable is the consumer's attitudes towards the ad (like or dislike, interval).

We found that the relationship of congruence and the attitude is moderated by PSI. Those with high PSI care less if the ad is incongruent and thus "dishonest".

The problem: we first showed the subjects the post and THEN asked them about their general relationship (PSI) with the influencer in question. Thus, we allowed the congruence or incongruence of the post to influence the subjects' subsequent evaluation of how much they like the influencer. Those who viewed an incongruent post indicated that they had less of a parasocial relationship with the influencer.

Therefore, our manipulated predictor affected our moderator. So we are wondering if it's even fair to say that moderation has in fact taken place.


Less is more. Stay pure. Stay poor.
Alright, I am sorry I asked. You are making me miss the day of lung cancer risk regressed on smoking status + asbestos status + smoking status * asbestos status.

Correct, you usually stratified by the manipulatable variable. So for absolute simplicity, you could generate the rates of the outcome for the non-manipulatable variable in the two stratified groups and see if they vary.