RE: st: Perfect predictors of an occurrence (survival analysis)

 From "Luis Ortiz" To Subject RE: st: Perfect predictors of an occurrence (survival analysis) Date Thu, 13 Mar 2008 15:31:46 +0100

```Thanks, Maarten.

Indeed, my data look like this.

The problem is that some of my covariates, seemingly, do not change their
value during t. The type of contract does not change while the individual
holds a job for which s/he is over-educated.

In other words, the type of contract perfectly predicts my outcome

As you'll see next, whenever I run the model with sex and age dummies,
everything is OK. Yet, if I add to the model the type-of-contract
dummies....

. cloglog dest_match_3 sex age1 age2 age3, robust cluster(nid) nolog

Complementary log-log regression                Number of obs     =
340100
Zero outcomes     =
339264
Nonzero outcomes  =
836

Wald chi2(4)      =
404.03
Log pseudolikelihood = -5553.2628               Prob > chi2       =
0.0000

(standard errors adjusted for clustering on
nid)
----------------------------------------------------------------------------
--
|               Robust
dest_match_3 |      Coef.   Std. Err.      z    P>|z|     [95% Conf.
Interval]
-------------+--------------------------------------------------------------
--
sex |   .3627809   .0734651     4.94   0.000      .218792
.5067698
age1 |   1.911637   .1246839    15.33   0.000     1.667261
2.156013
age2 |   2.289894    .119044    19.24   0.000     2.056572
2.523216
age3 |   1.295038   .1402895     9.23   0.000     1.020076
1.570001
_cons |  -7.635593   .1219848   -62.59   0.000    -7.874679
-7.396507
----------------------------------------------------------------------------

...the result changes.... STATA reports that type-of-contrac dummies
"predict failure perfectly"

. cloglog dest_match_3 sex age1 age2 age3 tipcon1 tipcon3, robust
cluster(nid) nolo
> g
note: tipcon1 != 0 predicts failure perfectly
tipcon1 dropped and 81273 obs not used

note: tipcon3 != 0 predicts failure perfectly
tipcon3 dropped and 28763 obs not used

-----Mensaje original-----
De: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] En nombre de Maarten buis
Enviado el: jueves, 13 de marzo de 2008 13:55
Para: statalist@hsphsun2.harvard.edu
Asunto: RE: st: Perfect predictors of an occurrence (survival analysis)

Your data should look something like this

id t leave
1  1  0
1  2  0
1  3  1
2  1  1
3  1  0
3  2  0
3  3  0
etc.

In case of panel data this is often called long format, in case of
survival data this is often requires the use of -expand-. This is
discussed in detail in lecture 6 of:
http://www.iser.essex.ac.uk/teaching/degree/stephenj/ec968/

Hope this helps,
Maarten

--- Luis Ortiz <luis.ortiz@upf.edu> wrote:

> Dear Maarten,
>
> Thanks for the quick response.
>
> Initially, since I only have discrete time data, I was using cloglog,
> but,
> as I said, the analysis was... rejected, since type-of-contract
> dummies turn
> out to be perfect predictors of the outcome.
>
> Thanks for your attention again
>
> Luis Ortiz
>
>
> -----Mensaje original-----
> De: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu] En nombre de Maarten
> buis
> Enviado el: jueves, 13 de marzo de 2008 13:38
> Para: statalist@hsphsun2.harvard.edu
> Asunto: Re: st: Perfect predictors of an occurrence (survival
> analysis)
>
> What commands are you using?
>
> --- Luis Ortiz <luis.ortiz@upf.edu> wrote:
>
> > Dear Statalisters,
> >
> > I'm analysing the occurrence of the following event: getting out of
> > employment if you are over-educated. I'm doing so for a number of
> > over-educated individuals, using panel data. The spells are the
> > periods in
> > over-education.
> >
> > I am especially interested by the type of contract. This is one of
> my
> > main
> > independent variables. Therefore, I have split the initial variable
> > in the
> > corresponding dummies, in order to include them in the analysis.
> Yet,
> > I have
> > found that each one of these dummies perfectly predict the
> occurrence
> > of the
> > event I'm studying. It might be because the spells are not very
> long
> > and the
> > job, no matter how over-qualified it is, goes associated with a
> given
> > type
> > of contract. In other words, the type of contract does not change
> > during the
> > spell.
> >
> > In such a case, dummy covariates on type of contract are not
> feasible
> > for
> > the analysis. They perfectly predict the occurrence of the event.
> >
> > Yet, this does not mean that over-educated workers are equally
> likely
> > to get
> > out of employment REGARDLESS of their type of contract. Moreover, I
> > suspect
> > that the type of contract may be associated to a different
> likelihood
> > of
> > getting out of the employment.
> >
> > How could I overcome the problem I have mentioned for running
> > survival
> > analysis? Is there any alternative way to estimate the relative
> > likelihood
> > of temporary / permanent over-educated workers quitting their jobs
> > that you
> > come up with?
> >
> > I do appreciate your attention
> >
> > Best regards
> >
> > Luis Ortiz
> >
> > *
> > *   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/
> >
>
>
> -----------------------------------------
> Maarten L. Buis
> Department of Social Research Methodology
> Vrije Universiteit Amsterdam
> Boelelaan 1081
> 1081 HV Amsterdam
> The Netherlands
>
> Buitenveldertselaan 3 (Metropolitan), room Z434
>
> +31 20 5986715
>
> http://home.fsw.vu.nl/m.buis/
> -----------------------------------------
>
>
>       __________________________________________________________
> Sent from Yahoo! Mail.
> The World's Favourite Email http://uk.docs.yahoo.com/nowyoucan.html
> *
> *   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/
>
> *
> *   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/
>

-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands

Buitenveldertselaan 3 (Metropolitan), room Z434

+31 20 5986715

http://home.fsw.vu.nl/m.buis/
-----------------------------------------

__________________________________________________________
Sent from Yahoo! Mail.
The World's Favourite Email http://uk.docs.yahoo.com/nowyoucan.html
*
*   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/

*
*   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/
```