Hierarchical Regression assumptions

In my study, i would like to use hierachical regression, but I can't find a proper book on the assumptions that should be met. Is is true that the variable to-be-controlled-for should not be related to both the predictor and the dependent variable?

I think my problem might a a more methodological than a statical one, but I hope someone can help me.

The idea is, I want to predict children's aggression against their peers by the following predictor:

Overestimation. This is the discrepancy between a child's self-perceived social competence and peer-rated social competence. I generated this variable myself.

I expect that overestimation will only predict aggression in children who are not liked by their peers. Therefore, I create a dummy variable (liked or not-liked) based on the same measure (peer-rated social competence). I call this "rejection". Children who are rated by their peers 1 SD below the mean are considered rejected.

My idea was to create an interaction term overestimation*rejection after controlling for rejection. The idea behind this is that rejection by peers is related to a child's aggression. If the interaction term is significant, can i conclude that the remaining variance in aggression can be explained by overestimation in rejected children?
(I am not a statistician)

Have understood you correctly

IV1 = continuous variable created from two rating measures, self-perceived and peer-rated social-competence

IV2 = binary dummy variable created from peer rated social competence rating

DV = peer directed aggression measure

My first pass thought it is that I would try and fit the two original ratings measures rather than create a new measure out of the two of them. It seems simpler to me, both in terms of analysis and in terms of interpretation.

What does the combined measure gain you over the two original measures?

What is the correlation between the two ratings measures? Would there be problems of collinearity?
Thanks for your thoughtful reply! There is one thing I did not explain in the text:

The reason why i use the combined measure is because I am not interested in self perceived competence nor in peer perceived competence but in the discrepancy between these measures. I want to show in my study that children who overestimate their social competence are more aggressive to their peers.
Because these variables were measured on a different scale, I couldn't simply take the difference score. Therefore, I regressed Self perceived over Peer perceived and saved the standardized residual score of the regression equation. These residual scores represent the self-perceived competence of a child that could not be explaned by peer-rating.
About the relatedness of the two variables: yes, self-perceived and peer-perceived are related and should be related if i want to create this measure in the way I did it.