Date: Mon, 11 Oct 1999 13:55:34 GMT
Reply-To: Joel Winter <jpwinter@UMSL.EDU>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Joel Winter <jpwinter@UMSL.EDU>
Organization: University of Missouri - St. Louis
Subject: Re: regression by n groups
On Sat, 9 Oct 1999 15:47:55 -0700, "hsifei" <firstname.lastname@example.org>
: I have a question about regression analysis:
: I ran a regression model with the same independent variables by n groups.
: I'd like to know if the coefficients of each variables are statistically
: significant differece between the n groups. What kind of test I should use?
: For example, the model is:
: travel time = a + b1* (household income) + b2 * (education background)
: 5 racial groups: white, black, ...
: Ho: b1(asian) = b1 (white) = b1(black) = b1 (hispanics) = b1 (others)
: Same for b2:
: Ho : b2(asian) = b2 (white) = ....
: What kind of command I should use by SAS or SPSS?
I suggest some steps:
1) Do some reading on moderation analysis or using interaction
terms in multiple regression. Nearly all regression textbooks
2) Get Aiken & West's text. They lay out the procedure and give some
sample SPSS code.
If you already have done some reading, then:
3) Dummy code your group variable
4) Compute product terms between these dummys and the variables of
5) Add these product terms and see if they incrementally improve fit.
I HTH. Your hypotheses actually seem closer to a multi-group SEM type
of analysis, but you asked about regression.