Date:         Fri, 17 Jun 2005 07:18:59 -0500
Reply-To:     "Peck, Jon" <peck@spss.com>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         "Peck, Jon" <peck@spss.com>
Subject:      Re: Unexpected singularities
Comments: To: ar7347@wayne.edu
Content-Type: text/plain; charset="us-ascii"

To expand on Hector's point, the singularity could be due to other predictor variables in the equation.  Redundancy in a single dummy variable is the easiest to understand, but you have to consider the whole set of predictors.

Imagine that you have a categorical variable x1 with values a, b, c, but when x1 is c, the x2 value is a constant, d.  Then including the dummy variables for a, b, c and for x2 will result in singularity: you cannot distinguish the effect of x1 being c from the effect of x2 being d.

Regard,
Jon Peck
SPSS, Inc.


-----Original Message-----
From: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU] On Behalf Of Rayman Mohamed
Sent: Thursday, June 16, 2005 10:03 PM
To: SPSSX-L@LISTSERV.UGA.EDU
Subject: [SPSSX-L] Unexpected singularities

Hi all,

I have been doing some multinomial logistic regressions.
When I use two particular categorical variables (in
appropriate dummy form -- see later) in my models I get the
following message.

"Unexpected singularities in the Hessian matrix are
encountered. This indicates that either some predictor
variables should be excluded or some categories should be
merged."

This implies that I have redundancy in my dummy variables (I
think). However, I have checked and rechecked but I cannot
see where I am going wrong. One of the categorical variables
for which I am getting this message is political
affiliation. I created three dummy variables (democrat,
republican, and independent/no affiliation) and entered two
into the regression model. (There are other categories that
I ignored when I created the dummy variables, e.g.,
the "don't knows".)

Could anyone think of another reason why I am getting this
message. Or do you think that perhaps I just looked at this
too much so I cannot see some rather obvious mistake in my
syntax to create the dummy variables?

Here is the syntax that I used to create the dummy variables
for political affiliation:

IF (q78a = 1) democrat_1 = 1 .
EXECUTE .
IF (q78a = 2 | q78a = 3 | q78a = 5) democrat_1 = 0 .
EXECUTE .

IF (q78a = 2) republican_1 = 1 .
EXECUTE .
IF (q78a = 1 | q78a = 3 | q78a = 5) republican_1 = 0 .
EXECUTE .

IF (q78a = 3 | q78a = 5) Ind_other_1 = 1 .
IF (q78a = 1 | q78a = 2) ind_other_1 = 0 .
EXECUTE .

Thanks for your suggestions,

R.
Rayman Mohamed
Assistant Professor
Wayne State University
College of Urban, Labor, and Metropolitan Affairs
Department of Geography and Urban Planning
227 State Hall
5143 Cass Avenue
Detroit, MI 48202
http://www.culma.wayne.edu