Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: choosing parametric model

From   Maarten Buis <>
Subject   Re: st: choosing parametric model
Date   Fri, 1 Jul 2011 09:53:25 +0200

On Fri, Jul 1, 2011 at 6:51 AM, Yusvita Triwiadhian S. wrote:
> Parametric models in survival specifies a particular shape for the
> hazard rate i.e the time dependency, maybe exponential,weibull, etc..
> How do we choose the correct parametric model to assume about the
> shape of the time dependency?
> can I use AIC by comparing the AIC scores for different parametric
> models and then the smallest is better? if "yes", The AIC scores that
> is used when there are no covariates or when there are covariates?
> which one is better?

When making such decisions you should not rely on a single statistic.
I would look at residuals and see if something weird is happening. I
would inspect the consequences of your model graphically using
-stcurve- and see if they make sense. I would also look at measures
like AIC and BIC. In an ideal world they all point clearly in the same
direction, but often you need to make a judgement call. We cannot help
you with making judgement calls as that necessarily must be made based
exact knowledge of the data and how it was collected, the aim of the
study and how important/detrimental different types of errors are, the
folk wisdom within your specific sub-sub-discipline, etc. etc. As to
AICs, small is better. Whether or not you want to base it on a model
with or without covariates just depends on which models you want to
compare. If you want to compare models that have covariates, than the
AICs should be based on the models with covariates.

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index