Re: help with comparing regression coefficients

Posted by Bruce Weaver on Nov 03, 2016; 1:54am
URL: http://spssx-discussion.165.s1.nabble.com/help-with-comparing-regression-coefficients-tp5733418p5733421.html

My code computes two version of the test, depending on the value assigned to variable Pool:

* Where Pool=1, a pooled estimate of the MSE is used to compute SE(diff),
* and apart from rounding error, the results will match those of a Pothoff
* analysis on the raw data.

* Where Pool=0, equal residual variances are not assumed when computing
* SE(diff), and therefore, Satterthwaite degrees of freedom are used when
* computing the p-value.

Rich Ulrich wrote

I do not see how there is any importance for the /testing/ as to whether Xa is the same variable as Xb -- the

question is whether the regression coefficient is the same for the two groups.  That is the simple interaction

term in the regression of Y= c + b1*Group + b2*X + b3*Interact .

X would be Xa and Xb respectively, for the two groups.

The interaction may be computed as the centered term,
COMPUTE  Interac= (Group-meanGr) * (X - meanX) .

This ignores the possible difference of variances, which I think Bruce's code does, also, for comparing the b's.

The difference introduced by having two different variables is the argument that is necessary to say that
(a) the differences are trivial (if that is how they look) or (b) that the differences are meaningful, beyond the
fact that Xa may or may not differ from Xb, or Group A and B may differ.

--
Rich Ulrich



________________________________
From: SPSSX(r) Discussion <[hidden email]> on behalf of Dominic Lusinchi <[hidden email]>
Sent: Wednesday, November 2, 2016 4:53 PM
To: [hidden email]
Subject: help with comparing regression coefficients


Greetings fellow SPSS users,



I checked the archives but couldn't find anything that fitted this case, as I could tell. I also searched the web but could not find anything - at least, that I could understand.



I am trying to find out if there is a test to compare two (ols) regression coefficients. Here's the setup. Individuals have been randomized to group A and B. They have been measured on Y (the DV) and group A have been measure on xA and group B measured on xB. To be clear xA and xB are two different (independent) variables. I want to compare in a simple regression (Y=c + xA and Y=c + xB) the regression coefficient for xA and xB (bxA - bxB). Thus, I am dealing with two different models (the same DV but different IVs) AND two independent groups.



So my questions are: Is there a test for such a situation? If so, (2) can it be implemented in SPSS?



Thanks in advance for time and help - Dominic



=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD