Re: st: parametric survival analysis - choosing the probability distribution

[Maarten Buis <maartenlbuis@gmail.com>]

On Tue, Jun 7, 2011 at 1:34 AM, Magdalena Kapelko wrote:
> I am trying to run survival analysis is STATA, analyzing the impact of
> some variables on the probability of firm failure. Because the
> proportionality assumption for cox model does not hold, I run a
> parametric regression model instead. I want to choose between Weibull
> and exponential distributions.

Both the Weibull and the exponential also assume proportional hazards,
so that does not solve your problem.

> Some literature says: compare AIC of
> different models and choose a model with the lowest AIC. But the
> problem is that:
> - for Weibull I obtain very large negative AIC  (aprox. - 20000)
> - for exponential I obtain positive AIC (aprox. 1000).
>
> when choosing the model, shall I look at absolute values of AIC and in
> this way choose exponential, or shall I choose in general a model with
> the smallest AIC that is Weibull.

You take the values as is, so you do not take the absolute values.

> Is it possible that for the model
> with the same independent variables, the assumption of one
> distribution can give a negative AIC and the other a positive AIC?

yes, see e.g.: <http://blog.stata.com/2011/02/16/positive-log-likelihood-values-happen/>

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany


http://www.maartenbuis.nl
--------------------------

*
*   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/