Hello,

I am relatively new to using regression methods, so please bear with me if this is a silly thing I am overlooking..

In my regression model, I have two predictor variables (X1, X2) and an interaction term (X1*X2). My dependent variable (Y) is a measure of of engagement; X1 is a continuous variable (working memory span) and X2 is a dichotomous variable (activity type; coded 1= high activity, 0=low activity).

I am trying to use SPSS to fit this model hierarchically so that in Step 1, I enter the main effects (Y= B0+B1X1+B2X2), and in Step 2, I enter the interaction effect (Y= B0+B1X1+B2X2+B3X1*X2).

As expected, In step one, neither of the main effects are significantly different from zero. Similarly, as expected, in step 2, the interaction term is significant. What is unexpected, and what I don't know how to explain, is that from step 1 to step 2, my X1 main effect (Working memory) goes from non-significant, to significant at the .01 level.

How do I interpret this finding, that adding the interaction term in step 2 changed the main effect from step 1 to 2? I don't think it is an issue of multicolinearity, my tolerance values were no less than .75.

Is my Beta weight for X1 the slope for the equation when activity type is low (=0), and it is significant? While B1+B3 would be my slope for the regression when activity type is high (=1), and that slope is also significant?

Thanks!