Moderator analysis: regression x partial corr.

#1
Hi,

I hope someone can answer my question soon. I am helping my friend with research. She has independent variable X and dependant variable Y and potential moderator Z. I did bivariate correlation for XY and then partial correlation XY controlling Z in SPSS. Then I compared zero order and first order correlation by eye and confidence interval. Her supervisor told her (before I used partial corr.) to use multiple regression. My question is: is my method valid for this situation? And what's the main difference between using partial correlation and moderator analysis using regression? There is a set of Xs and Y and I don't wanna compute regressions unless it's neccessary. All moderator regression analysis tutorials seemed time demanding and her deadline is coming.

Thanks a lot
 

kiton

New Member
#2
Correlation analysis allows you to see if the variables are related, but it does not establish causality. Also, in order to test for the moderating effect of Z on the relationship b/w X and Y, one needs to estimate and examine the coefficients of at least two models:

y = a + x + z + u (1)
and
y = a + x + z + x1*z (2)

If (A) z is significant in equation (1), and (B) interaction term is significant, then moderation is present. Yet a couple other things will have to be tested as well. This is actually relatively simple and not time demanding in principle.