significance of regression

#1
Hi,

I have a question about the regression.
If I do the regression as y=x1, x1 is significant; If y=x1+x2, then x1 and x2 are both significant; but if I do the regression as y=x1+x2+x1*x2, then x2 and x1*x2 are not significant. Anybody can tell me why? It's so confusing. Thank you so much!
 

Blaz

New Member
#3
It could be due to nonessential multicolinearity, have you centralized the variables before computing the interaction term? If not, the correlation between the interaction, and individual variables remains spuriously high, reducing individual partial contributions which could result in this situation of yours.
 
#5
thank you guys. well, there are a large number of data. i think it has enough power. there might be correlations between x1 and x2. but why centered them can decrease the correlation?