Statalist


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

RE: st: AIC and BIC to compare parametric and non-parametric survival models


From   jverkuilen <jverkuilen@gc.cuny.edu>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: AIC and BIC to compare parametric and non-parametric survival models
Date   Sun, 17 May 2009 22:51:59 -0400

If the likelihoods aren't comparable---you would need to check the equations to be sure---no. 

One way things go off the rails is if the normalization terms are dropped and tw different families are compared. Example: To compare say the gamma and lognormal by AIC, you need all the 2*pi and whatever even if they don't affect the estimates in any way.  

JV

-----Original Message-----
From: "Tom Trikalinos" <ttrikalin@gmail.com>
To: statalist@hsphsun2.harvard.edu
Sent: 5/16/2009 3:05 PM
Subject: Re: st: AIC and BIC to compare parametric and non-parametric survival 	models

Maarten thanks very much- precise and clear instructions, as always.

Out of curiosity, though:
is it theoretically correct to use BIC or AIC to compare fit between
Cox and a parametric model (e.g., exponential)?

t



On Fri, May 15, 2009 at 4:13 PM, Maarten buis <maartenbuis@yahoo.co.uk> wrote:
>
> --- On Fri, 15/5/09, Tom Trikalinos wrote:
>> To compare non-parametric and parametric survival
>> analysis models, can I use the AIC and BIC?
>> Specifically, I fit Cox PH models and exponential and
>> weibull parametric regressions. It was pointed out to
>> me that AIC & BIC-based comparisons may not be valid
>> (because Cox uses partial likelihood).
>>
>> PS. I am performing survival analyses to inform a decision
>> analysis. For this reason I strongly prefer to fit
>> parametric models - will make life easier and restore the
>> smile on me face.
>
> You could try estimating a piecewise constant model. The
> idea is very similar to the idea behind -stcox-: estimate
> a flexible baseline hazard and the explanatory variable
> multiplicatively move this baseline hazard up or down.
> Alternatively you could model the baseline hazard with
> some other flexible curve, like a restricted cubic
> spline. See the example below:
>
> *---------------- begin example --------------------------
> sysuse cancer, clear
> gen long id = _n
> stset studytime, failure(died) id(id)
> stsplit t, every(1)
> gen t3 = floor((t)/3)
>
> // piecewise constant
> xi: streg i.t3 i.drug age, dist(exp)
> adjust _Idrug_2=1 _Idrug_3=0 age , by(t3) exp gen(haz_piece)
>
> // restricted cubic spline
> mkspline tsp=t, cubic knots(5 10 20 30 35)
> xi: streg tsp* i.drug age, dist(exp)
> adjust _Idrug_2=1 _Idrug_3=0 age , by(t) exp gen(haz_cubic)
>
> twoway line haz* studytim, sort c(J) ///
>    legend(order(1 "piecewise" "constant"  ///
>                 2 "restricted" "cubic spline")) ///
>    ytitle(hazard)
> *----------------- end example ------------------------
>
> Hope this helps,
> Maarten
>
> -----------------------------------------
> Maarten L. Buis
> Institut fuer Soziologie
> Universitaet Tuebingen
> Wilhelmstrasse 36
> 72074 Tuebingen
> Germany
>
> http://home.fsw.vu.nl/m.buis/
> -----------------------------------------
>
>
>
>
>
>
> *
> *   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/
>

*
*   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/
 

*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index