Puzzling Regression issue.

First post :D Wondering if anyone could lend a hand, I can't figure this out.

When I perform two separate regression analyses ('promotion_A' predicting 'promotion_b'
and 'cond' predicting 'promotion_b'), the beta/sig. values differ from
when I enter both IV predictors in step 1 predicting 'promotion_b'. The
values differ quite a lot (in that the second method produces
significant/near sig. results) - and I'm wondering if anyone knows which
method should technically be used? And why the values differ?


Mean Joe

TS Contributor
I think that technically you should use the method that produced significant results, because the other method is insignificant.

Let's say you get beta=1.6 when you run 'promotion_A' predicting 'promotion_b'. This means that promotion_b increases by 1.6 on average (over all people) when promotion_A increases by 1.

Let's say on the other hand you get beta=1.2 for promotion_A and beta=2.0 for cond when you run both predictors. This means that promotion_b increases only by 1.2 on average for a given value of cond, when promotion_A increases by 1. For a different value of cond, then promotion_b increases by 1.2 for each +1 of promotion_A plus increases by 2.0 for each +1 of cond.
When you ran the model with just promotion_A, it was averaging the change in promotion_b over all people regardless of cond's value.