problem with logistic regression in SPSS

[h1]

I've just tried to do logistic regression in SPSS - followed all instructions in a good textbook so I'm happy that bit is k. I have a sample of 600 and entered 10 variables all of which were significant in univariate analysis at the p<0.01 (or better).
I have a bizarre output though. One of the variables which in univariate analysis predicted increased likelihood of the outcome at a p<0.0001 level has come out as prediciting decreased likelihood of the outcome in the regression (B-1.628 p<0.0001 Exp B 0.196 95% CI 0.97-0.398).
The categories are definitely coded correctly - ie 0 - don't have the variable and 1 - they do.
What am I doing wrong??

Thanks so much in advance for any help
h

[Swank, Paul R]

You need to look to how all the predictors are related to each other. Remember that with multivariable prediction you are noting how x predicts when all the other predictors have been partialed out of X. Thus, x can have a very different impact in combination with 9 other variables than it does by itself.

Dr. Paul R. Swank,
Professor and Director of Research
Children's Learning Institute
University of Texas Health Science Center-Houston


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of h1
Sent: Wednesday, April 29, 2009 1:17 PM
To: [hidden email]
Subject: problem with logistic regression in SPSS

I've just tried to do logistic regression in SPSS - followed all instructions
in a good textbook so I'm happy that bit is k. I have a sample of 600 and
entered 10 variables all of which were significant in univariate analysis at
the p<0.01 (or better).
I have a bizarre output though. One of the variables which in univariate
analysis predicted increased likelihood of the outcome at a p<0.0001 level
has come out as prediciting decreased likelihood of the outcome in the
regression (B-1.628 p<0.0001 Exp B 0.196 95% CI 0.97-0.398).
The categories are definitely coded correctly - ie 0 - don't have the
variable and 1 - they do.
What am I doing wrong??

Thanks so much in advance for any help
h
--
View this message in context: http://www.nabble.com/problem-with-logistic-regression-in-SPSS-tp23298368p23298368.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
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

=====================
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

[Hector Maletta]

It is a perfectly normal outcome if the effect of that variable is
"explained" by the other predictors in your equation. Once the effect of the
other predictors is controlled for, then X=1 (compared to X=0) may decrease
the odds of the event.
Hector

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of h1
Sent: 29 April 2009 15:17
To: [hidden email]
Subject: problem with logistic regression in SPSS

I've just tried to do logistic regression in SPSS - followed all
instructions
in a good textbook so I'm happy that bit is k. I have a sample of 600 and
entered 10 variables all of which were significant in univariate analysis at
the p<0.01 (or better).
I have a bizarre output though. One of the variables which in univariate
analysis predicted increased likelihood of the outcome at a p<0.0001 level
has come out as prediciting decreased likelihood of the outcome in the
regression (B-1.628 p<0.0001 Exp B 0.196 95% CI 0.97-0.398).
The categories are definitely coded correctly - ie 0 - don't have the
variable and 1 - they do.
What am I doing wrong??

Thanks so much in advance for any help
h
--
View this message in context:
http://www.nabble.com/problem-with-logistic-regression-in-SPSS-tp23298368p23
298368.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
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

=====================
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

[SR Millis-3]

I think that you need to several things to figure out what might be going on:

--examine the degree of collinearity among your covariates/predictors by looking at the condition indexes and variance decomposition proportions.

--regarding the specific variable you mentioned, it may be functioning as a suppressor variable.  The presence of a suppressor variable is identified by the pattern of regression coefficients and correlations of each covariate with the response variable. You might have a suppressor variable if (1) the absolute value of the simple correlation between the covariate and response variable is substantially smaller than the coefficient for the covariate or (2) the simple correlation and coefficient have opposite signs.  Unfortunately, I not aware of any statistical test that assesses how much different a coefficient and simple correlation need to be in order to identify suppression.

Scott R Millis, PhD, ABPP (CN,CL,RP), CStat, CSci
Professor & Director of Research
Dept of Physical Medicine & Rehabilitation
Dept of Emergency Medicine
Wayne State University School of Medicine
261 Mack Blvd
Detroit, MI 48201
Email:  [hidden email]
Tel: [phone]
Fax: [phone]


--- On Wed, 4/29/09, h1 <[hidden email]> wrote:

> From: h1 <[hidden email]>
> Subject: problem with logistic regression in SPSS
> To: [hidden email]
> Date: Wednesday, April 29, 2009, 2:17 PM
> I've just tried to do logistic regression in SPSS -
> followed all instructions
> in a good textbook so I'm happy that bit is k. I have a
> sample of 600 and
> entered 10 variables all of which were significant in
> univariate analysis at
> the p<0.01 (or better).
> I have a bizarre output though. One of the variables which
> in univariate
> analysis predicted increased likelihood of the outcome at a
> p<0.0001 level
> has come out as prediciting decreased likelihood of the
> outcome in the
> regression (B-1.628 p<0.0001 Exp B 0.196 95% CI
> 0.97-0.398).
> The categories are definitely coded correctly - ie 0 -
> don't have the
> variable and 1 - they do.
> What am I doing wrong??
>
> Thanks so much in advance for any help
> h
> --
> View this message in context:
> http://www.nabble.com/problem-with-logistic-regression-in-SPSS-tp23298368p23298368.html
> Sent from the SPSSX Discussion mailing list archive at
> Nabble.com.
>
> =====================
> 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

=====================
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