Date:         Mon, 9 Dec 2002 11:26:54 -0500
Reply-To:     painter@email.unc.edu
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         painter@email.unc.edu
Subject:      Re: Deviation contrast in Logistic regression
In-Reply-To:  <7D25CCC35D3CD31192BF006008BFB1FE04F5737B@nsofs14.uchc.edu>
Content-Type: text/plain; charset=us-ascii; format=flowed

Hello,

Many thanks to Joe for answering my initial question. Of course an answer
usually leads to another question, in this case two.....

First, assuming that the Deviation contrast is the correct procedure for
addressing my research question. What conclusion is one permitted to draw
about the one group that is held out (and why)?

Second, I have the situation where the groups vary considerably in size.
I'm   now concerned that if the Deviation uses the Unweighted average of
all groups then varying group size is not being taken into account. Would
it be appropriate to use group size as a covariate in my logistical model?


TIA,
John



--On Wednesday, December 04, 2002 2:59 PM -0500 "Burleson,Joseph A."
<burleson@up.uchc.edu> wrote:

> This phenomenum threw me, also. The reason is that while SPSS asks you to
> think in terms of the BASIS or DESIGN matrix, it provides the feedback on
> the output for the CONTRAST matrix. Some of these are, by coincidence, the
> same, while others, like INDICATOR, are different, as they should be.
>
> Hence, don't worry; they are correct as you requested, and not how you
> might have interpreted the output matrix.
>
> See David Nicols white paper, "What kind of contrasts are these?," No. 63,
> 1997 on:
>
> www.spss.com/tech/stat/articles/contrast.htm
>
> -Joe Burleson
>
> -----Original Message-----
> From: painter@email.unc.edu [mailto:painter@email.unc.edu]
> Sent: Wednesday, December 04, 2002 2:00 PM
> To: SPSSX-L@LISTSERV.UGA.EDU
> Subject: Deviation contrast in Logistic regression
>
>
> Hello,
>
> Using Logistic regression I want to compare members of various categories
> to the grand mean. Normally the Deviation contrast would be my choice.
>
> However upon examing the Categorical Codings Table, it appears that when
> using the Deviation contrast results in a matrix that looks like the
> matrix for a Simple contrast. When I use the Simple contrast I get what
> appears to be a matrix for Deviation contrast.
>
> I'm confused! And I'm also relatively new to logistic regression, so
> comments and / or references are welcome!
>
>
> Thanks in advance,
> John