st: interpreting coefficients in discrete time hazard model

[Daniel Simon <dhs29@cornell.edu>]

Hi - I am estimating a discrete time hazard model, using pgmhaz.  I have 
several covariates that are in the form ln(x).  I'm wondering how to 
interpret the coefficients on these variables.  Specifically, when i 
exponentiate the estimated coefficient, b, what does exp(b) mean? For 
example, if the coefficient b is equal to -5, and exp(-5) is .00674, then 
does this say that a 1% change in x reduces the hazard by .99326?  Or is it 
something different?  Thanks. Daniel

At 10:23 AM 1/9/2003 -0600, you wrote:

> Mar�a Elena Casquel del Campo <casquel@eco.uji.es> asks:
>
> > I�m trying to estimate duration models with Stata. The problem is that I
> > have multispell data in the sense that some individuals experiment failures
> > but later enter another time in the sample. I�d like that when the
> > individual enters after experimenting failure the clock starts clicking
> > another time.
>
> > I have introduced:
> >  stset finish, failure(cens) time0(star) exit(time .) id(pid) origin(star)
>
> >            pid       wave         _t0          _t        _d
> > 120. 130911101          1           0          11         1
> > 121. 130911101          2          12          23         0
> > 122. 130911101          3          24          35         0
> > 123. 151701103          1           0          11         1
> > 124. 151701103          2          12          23         0
> > 125. 151701103          3          24          35         0
> > 126. 151701103          4          36          47         0
> > 127. 151701103          5          48          59         0
> > 128. 213341102          1           0          11         1
> > 129. 213341102          2          12          23         0
> > 130. 213341102          3          24          35         1
>
> > I would like that _t0 start in 0 for the individual 130911101, observation
> > 121.
>
> There is some discussion of how to do this in [R] stsum.  Try,
>
>    . stgen nf = nfailures()
>    . egen newpid = group(pid nf)
>    . sort newpid finish
>    . by newpid: gen newstar = star - star[1]
>    . stset finish, failure(cens) id(newpid) time0(newstar)
>
> Also note: If you plan on fitting any regression to these models, be sure to
> specify -cluster(pid)- since one would usually not consider the multiple
> failures of each subject as independent events.
>
> --Bobby
> rgutierrez@stata.com
> *
> *   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/
Daniel Simon
Assistant Professor
Department of Applied Economics and Management
Cornell University
(607) 255-1626

*
*   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/