Date: Mon, 7 Oct 2002 19:07:46 -0500
Reply-To: Ben Earnhart <firstname.lastname@example.org>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Ben Earnhart <email@example.com>
Subject: Re: Moderator variables
Content-Type: text/plain; charset="us-ascii"; format=flowed
I googled for you (
) and came up with some pretty helpful stuff. Basically, the way I
understand it, if it's a moderator, then you use the interaction term, as
one variable conditions the effect of the other; with a mediator, you do
two models to see if the effect of x1 goes to zero when controlling for x2,
or better yet, use a SEM.
I'd think that the person would really want to clarify theoretically which
approach to use a priori, before starting to model.
At 05:45 PM 10/7/2002 -0600, Susan Crawford wrote:
>One of our graduate students is trying to use a logistic regression
>approach to determine whether or not parenting stress is a predictor of
>attachment (secure or insecure) in children. She thinks that social
>support will act as a buffer or moderating variable that may decrease
>the effects of parenting stress.
>I'm not sure of what approach we should take here. Should we enter the
>social support variable on the first step of the logistic regression,
>followed by the three parenting stress scores on the second step? Or
>should we be looking at interactions between the parenting stress scores
>and social support?
>I'm getting confused on the issue of whether or not to consider social
>support as a mediator variable or a moderator variable.
>Many thanks for your help.
>Susan Crawford, M.Sc.
>Behavioural Research Unit
>Alberta Children's Hospital
*Computer Consultant and ICPSR Data Assistant
*Department of Sociology and College of Liberal Arts
*University of Iowa