Re: st: Hazard model with multiple entry AND time varying controls?

[Steve Samuels <sjsamuels@gmail.com>]

Nick:

What is the "it" that analyzes only the first failure? To paraphrase the FAQ: say exactly what you typed and what Stata replied. Which model on the multiple-failure FAQ page are you trying to implement?  We'll need to see an excerpt of the your data lines for one country (one covariate please) to check if your -stset- and analysis commands are appropriate. Some general description of the study--number of countries, time periods, would also be helpful.

Although I know little about your data yet, I'll speculate that it should be treated as grouped, or discrete, not continuous.

Have you reviewed the subject-matter literature to see what kind of approaches others have used? Google searches for "stata transition models"  and "stata semi-markov regression" turn up many references.  

And, what is the package you refer to in your last sentence?
 
Steve
sjsamuels@gmail.com




On Feb 2, 2012, at 2:54 PM, Nick Eubank wrote:

Hi Steve,

The difficulty I'm having is that if I split countries out to one line
per observation (variables all change by year) and identify a country
by its country id, then it only analyzes the first failure. If I split
the country out and give it a different id for each time it goes
through a failure cycle then I'm gonna get issues since STATA doesn't
recognize that its the same country failing repeatedly. Am I just
missing an obvious command to adjust this, or is it a fundamental
problem with stcox and stset?

Austin, that package looks very promising! Thanks!


On Thu, Feb 2, 2012 at 4:58 AM, Steve Samuels <sjsamuels@gmail.com> wrote:
> 
> 
> Although the examples in the FAQ have one record per spell and person, you can use Stata's usual techniques for accommodating time-varying covariates. Here is a line of data from Section 3.2.4 of the FAQ entry:
>        id     group   time0   time   status   number   size   str |
>   4. |  4   placebo       0     10        0        5      1     1
> 
> If covariates change between 0 & 10, you  split the follow-up time into spans within which they are constant. See the multiple-record examples in the manual entry for -stset-; the final section "Using survival time data in Stata";  the examples in the entry for -stsplit-; and the sections on time-varying covariates in the manual entry for -stcox-.
> 
> On Feb 1, 2012, at 7:56 PM, Nick Eubank wrote:
> 
> Hi all,
> 
> Trying to run a hazard model with multiple ordered entry and time
> varying controls. The only help I could find is this:
> http://www.stata.com/support/faqs/stat/stmfail.html , but it doesn't
> seem able to accommodate data in my format (where each individual has
> different control variables for each year and can't be collapsed to a
> single year).
> 
> To be specific, I have a time series of country regimes, and am
> modeling transition from democracy to autocracy. Some countries "fail"
> repeatedly after returning to democratic status. Control variables
> (like income) are
> time varying so I can't collapse to one line per event with an entry and
> exit variable.
> 
> Any suggestions?
> 


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/