Doug Thomas wrote:
>Sujet: st: stcox discrete versus continuous
>Date: Wed, 15 Oct 2003 10:58:05 -0600
>De: "Douglas Thomas" <[email protected]>
>A: <[email protected]>
>
>I have a question about stcox with multiple records. I have a dataset with variables
that are measured every year for 10 years on an an annual basis (e.g., sales for each
year). I then have my failure measured on a daily basis. In other words, the failure can
occur anytime during the year and I know the specific date of the failure. Is it
inappropriate to use stcox, a continuous hazard model? Should I be using a discrete
model instead? Any suggestions? A reviewer on the paper indicated the following:
>
>I have several reservations regarding the empirical analyses and interpretation of
the results. Foremost, it appears that your research design confounds discrete time
and continuous time event-history analyses. Cox's partial likelihood method is a
continuous time method. Therefore, creating and pooling discrete 1-year event
windows (which would be consistent with a discrete time analysis) appears not to be
appropriate.
>
>
>
>Thanks,
>
>Doug Thomas
I'm a beginner in survival analysis, so my answer is likely to be shaky, but I give it
nevertheless.
The difference between discrete and continuous time specifications is that in the
latter events are not constrained to happen in predefined periods. However, the way I
understand the distinction is that it applies to the underlying process generating the
events more than to the characteristics of the observed events. If this is true, you can
use a continuous time specification to analyse observations that are in fact discrete.
The advantage is that, if you are interested in duration dependence, a continous time
specification takes this into account while a discrete time doesn't.
However, the reviewer seems to suggest that you should take days as units of
analysis rather than years, in which case a continuous time specification would no
doubt be appropriate.
fabrizio
----------
Fabrizio Gilardi
Institut d'Etudes Politiques et Internationales
Universit� de Lausanne -- BFSH2
1015 Lausanne (Switzerland)
Bureau 5123
Phone +41 21 692 31 48
[email protected]
http://wwwpeople.unil.ch/fabrizio.gilardi/
*
* 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/
