multinomial logistic regression

[Scott Roesch]

Hi all:

A colleague is running a multinomial logistic regression with a sample
size of 131. The dependent variable is composed of three groups and she
has three binary predictors. When the multinomial procedure is run she
gets a subpopulation value in the case processing summary of 7 even though
she has only 1 case that is missing, and thus we assumed the sample size
should be 130. I was curious if anyone has encountered this and what the
problem might be, and specifically what the subpopulation value
represents. SPSS appears to be deleting about 95% of the case and I have
not found figured out why. There are some cells (over half) empty or low
numbers in cells when you consider the 3 X 2 X 2 X 2 contingency table,
but this still should not result in losing 95% of her data. Any help would
be greatly appreciated!

Scott

*****************************
Scott C. Roesch, Ph.D.
Associate Professor
Department of Psychology
San Diego State University
5500 Campanile Drive
San Diego, CA 92182-4611

office phone: [phone]
fax: [phone]
*****************************

[Reutter, Alex]

NOMREG isn't deleting any valid cases, it simply aggregates them into subpopulations.  From the "Data Aggregation" topic in the NOMREG algorithms:
"
Observations with negative or missing frequency weights are discarded. Observations are aggregated by the definition of subpopulations. Subpopulations are defined by the cross-classifications of either the set of independent variables specified in the command or the set of independent variables specified in the subpopulation command [actually, that should say SUBPOP subcommand].
"

Under normal circumstances, you would have 2 x 2 x 2 = 8 subpopulations.  If you post the contingency table, someone might be able to point out why you're only getting 7.

Alex



-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Scott Roesch
Sent: Thursday, August 09, 2007 6:48 PM
To: [hidden email]
Subject: multinomial logistic regression

Hi all:

A colleague is running a multinomial logistic regression with a sample
size of 131. The dependent variable is composed of three groups and she
has three binary predictors. When the multinomial procedure is run she
gets a subpopulation value in the case processing summary of 7 even though
she has only 1 case that is missing, and thus we assumed the sample size
should be 130. I was curious if anyone has encountered this and what the
problem might be, and specifically what the subpopulation value
represents. SPSS appears to be deleting about 95% of the case and I have
not found figured out why. There are some cells (over half) empty or low
numbers in cells when you consider the 3 X 2 X 2 X 2 contingency table,
but this still should not result in losing 95% of her data. Any help would
be greatly appreciated!

Scott

*****************************
Scott C. Roesch, Ph.D.
Associate Professor
Department of Psychology
San Diego State University
5500 Campanile Drive
San Diego, CA 92182-4611

office phone: [phone]
fax: [phone]
*****************************