Date:         Wed, 4 Dec 2002 19:25:52 -0500
Reply-To:     Rob Penfold <penfoldr@sympatico.ca>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Rob Penfold <penfoldr@sympatico.ca>
Subject:      conditional logistic fails - nested solution in SPSS?
Content-Type: text/plain; charset="iso-8859-1"

I've run a conditional logistic regression and completed some tests of the
IIA assumption (deleting alternatives and looking at the consistency of the
parameters). In 2/119 cases the coefficients are inconsistent when the
alternative is deleted (i.e. I estimated the model 119 times since each
individual chooses from 120 alternatives). I don't know how to get the
covariance matrix from SPSS in order to do the Hausman test. I can't afford
STATA ($500 US for a grad student!!!!).

Is there a way to trick SPSS into doing a nested logistic regression? If
not, can I estimate separate conditional logit models - one for each of the
branches (groups 1, 2, 3) as an equivalent procedure?  My tree is below.


Store
                                                                       /
\
                                                           Cluster
Spatial Monopoly (zero competitors within 400m)
                                                         /      |     \
  (# competitors within 400m)          1       2      3
                                                     /|\        /|\        \
                                                  L1...L22   L3, L8, L21
L1             (locations where this happens - i.e. choices)

Thanks a million,

Rob

Rob Penfold
PhD candidate
Department of Geography
University of Toronto
________________________________________________

Scientia Est Potentia