After some reading on- and off-line, I'm exploring the use of -clogit- to
counter this (although I incur a substantial df cost since -clogit- models
only produce estimates for groups and not individuals). As a test, I've
fitted this model:
note: multiple positive outcomes within groups encountered.
note: 66 groups (263 obs) dropped due to all positive or all negative
outcomes.
[...]
Conditional (fixed-effects) logistic regression Number of obs = 1249
Wald chi2(12) = 175.60
Prob > chi2 = 0.0000
Log pseudolikelihood = -374.99176 Pseudo R2 = 0.3333
(standard errors adjusted for clustering on newparln)
----------------------------------------------------------------------------
| Robust
con01ge | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-----------+----------------------------------------------------------------
con99le | 1.965858 .2322615 8.46 0.000 1.510634 2.421082
lab99le | -1.366049 .3604063 -3.79 0.000 -2.072432 -.6596655
ldm99le | -.2783222 .2425517 -1.15 0.251 -.7537148 .1970703
con00le | 2.268708 .3634956 6.24 0.000 1.55627 2.981147
lab00le | -.6086171 .5684727 -1.07 0.284 -1.722803 .5055689
ldm00le | .0275779 .3858629 0.07 0.943 -.7286996 .7838553
toutle99 | .0192292 .0396104 0.49 0.627 -.0584058 .0968643
tout00le | -.1089908 .0429779 -2.54 0.011 -.1932259 -.0247558
gecpst01 | .4763015 .1203856 3.96 0.000 .24035 .7122531
gecxpc01 | .1244878 .0745309 1.67 0.095 -.0215901 .2705657
ghinps01 | .1295617 .0822547 1.58 0.115 -.0316546 .290778
ghinxp01 | .0834748 .0889073 0.94 0.348 -.0907802 .2577299
----------------------------------------------------------------------------
All of these variables are extremely well-behaved as far as collinearity
are concerned. The main findings here are that voting Conservative at
local council elections 1 or 2 years earlier before the general election
(GE)raised the odds of voting Conservative again at the GE, as we would
expect, net of all other effects.
The key question is this: are these valid findings from fitting such a
model? To ask a 'supplementary' question: if using -clogit- _is_ valid, am
I right in saying that including 'alternative-specfic constants' (Liao,
1994: 61-2) wouldn't make sense in estimating the above model? (it
certainly produces garbage output!). Thanks.
CLIVE NICHOLAS |t: 0(044)7903 397793
Politics |e: [email protected]
Newcastle University |http://www.ncl.ac.uk/geps
Liao TF (1994) INTERPRETING PROBABILITY MODELS, Sage QASS Series Paper
07-101, Thousand Oaks, CA: Sage.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
