I have four value variables (self-transcendence, achievement, conformity (dual item) and conformity (single item)).

I have a moderator variable (P-O Fit).

I ran four regressions with two steps (step one is main effects of one value and the moderator; step two is the interaction between that value and the moderator).

However, just for exploration, I ran a regression with all values and moderator in step one and all interactions in step two.

I had more significant results in the larger regression than the individuals. This is because, I assume, of the argument of crop yield (i.e., the more factors which contribute to the yield, to more likely they are significant contributors as the noise is controlled for).

However, given the focus of my research hypothesized each value individually as a contributor I ran separate hierarchical moderated regressions for each value rather than accounting for any orthogonality between the values.

Should I have done the individual regressions or the all-inclusive regression for my research?