Re: st: Specification of Parametric Hazard Models in Stata

[Steve Samuels <sjsamuels@gmail.com>]

There is no problem here.

I presume you refer to Table 1 in the Manual entry for -streg- (p. 383
Stata 13). It shows neither hazard function probability density
function for any of the distributions (except the exponential, for which
lambda is the hazard function). So there's no evidence in the table for your
statement about what Stata does or does not enter into either type of
function. In fact, the Manual presents the hazard functions in the
following sections, showing they do indeed contain auxiliary parameters,
when present. And, so do the pdfs, since pdf = Survival x hazard.

You might be thinking that the "Parameterization" column shows the
hazard function. It doesn't. It shows how the contribution of covariate
is parameterized in each model. The lambdas in that column are parameters
in the models, not functions of time.

A minor point: there's no "p" in the parameterization of the log-logistic
distribution; I presume you meant the Weibull.

Steve


> On Sep 18, 2013, at 4:03 PM, Pedro Gardete wrote:
> 
> I've been playing with the hazard models in Stata, and have been
> comparing the results to some code of my own. What follows is for the
> log-logistic, but I assume it applies to the Weibull distribution as
> well.
> 
> In building the likelihood function, it appears Stata introduces the
> parameter "p" of the distribution in the lambda(.) function (as
> defined in Table 1 of the [st] manual) that goes into the survival
> function, but not into the pdf or the hazard functions. I can't find a
> good justification for this...

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