Dear Political Methodologists:

I am trying to estimate a model that explains the decision to appoint a
particular type of government official.   Using somewhat similar data in
the past, I was able to used a cox proportional hazard model because I
had a large number of observations (over 600) and a fairly large number
of "failures" (around 100).   I used this type of model rather than a
logit/probit because of its flexibility and the ability to include fixed
effects to capture heterogeneity in the data without "loosing" a bunch
of observations.

But for the current paper I have just 350 observations and only around
20 "failures" (6% or so of the data).  Without a sufficiently large
number of observations and "failures," the event history model will not
give me estimates of some standard errors and gives me "bloated"
z-scores for other variables.   Bootstrapping would seem one way around
the problem, but when I do so I get observed hazard ratios that are
either massive (i.e. some coefficient E+24) or miniscule (i.e. some
coefficient E-14).   Another way out could be use a rare events logit,
which, if estimated, provides me with "reasonable" coefficients and
standard errors.   

So, what do you think is going on here?   How would you proceed?

I am inclined to try the rare events approach, though would prefer to
still use an event history approach. 

Thanks in advance for any insight you can offer.  


Jeffrey Chwieroth
Lecturer in International Political Economy
Department of International Relations
London School of Economics
Houghton Street
London WC2A 2AE
United Kingdom
Tel: +44 (0) 20 7955 7209
Fax: +44 (0) 20 7955 7446
Website: http://personal.lse.ac.uk/chwierot/


Please access the attached hyperlink for an important electronic communications disclaimer: http://www.lse.ac.uk/collections/secretariat/legal/disclaimer.htm

**********************************************************
             Political Methodology E-Mail List
   Editors: Melanie Goodrich, <[log in to unmask]>
            Delia Bailey, <[log in to unmask]>
**********************************************************
        Send messages to [log in to unmask]
  To join the list, cancel your subscription, or modify
           your subscription settings visit:

          http://polmeth.wustl.edu/polmeth.php

**********************************************************