Moderating variable - causal relationship between X and Y?

Hi all,





So lets say i did a regression between X and Y and it turned out they aren't related. Is it still possible that I add a third variable (the moderating variable) so that they do relate to each other?





When you test for mediation, there needs to be a relationship between X and Y, and Z and Y and ultimately X and Z in order for the mediation to be valid. But when you test for moderation, what if your result shows there is no main effect between X and Y, can you still get an effect when you add in a possible third variable? From articles they always say it alters the STRENGTH and DIRECTION of the relationship between X and Y. But my question is can it also CAUSE a relationship between X and Y if your initial regression showed there is no main effect?


Thanks alot!|||I would seriously doubt it, but don't forget, we are working with probabilities. Theoretically, if you perform enough studies that look for a relationship between two variables, there are likely to be some studies that will show a significant relationship; that is what probability means, whether you add a third variable or not. Adding a third variable increases the chances that you could find an interaction if there is some theoretical reasons for thinking they are related. It is possible for X to be related to Y and Y to be related to Z, but there be no direct relationship between X and Z. However, X, Y, and Z could have a significant interaction without necessary being related to each other, even with no significant relation between any of the variables. Increasing the number of variables increases the chances that some significant relationship between at least two of the variables will occur, which was one of the problems pointed out with Factor Analyses and similar anlyses that used the "shot gun" approach (when a shot gun is fired at long range, the pellets separate and some may hit the target), but that is the reason why statistics use degrees of freedom in their mathematical terms to calculate statistics; it is meant to reduce the influence of using many variables and reduce the "shot gun" effect.


I hope that helps.