Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: Using Bayesian Information Criterion (BIC)


From   "Lorena Barberia" <barberia@fas.harvard.edu>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: Using Bayesian Information Criterion (BIC)
Date   Wed, 23 Apr 2008 23:30:56 -0300

Thanks to Rodrigo and Stas for your helpful comments. 

I was trying to implement the BIC per the recommendation of an article by
Beck and Katz titled "Time-Series-Cross-Section Issues: Dynamics, 2004" that
advocates and calculates the BIC for different specifications of a pooled
OLS regression with panel corrected standard errors (e.g. AR1 model vs.
lagged dependent variable model with standard errors estimated with XTPCSE).
In the article, the authors argue that the BIC performs better than the
F-test for testing if fixed effects are necessary.  The paper can be
downloaded from Nathaniel Beck's website at NYU. It should be noted that the
authors do not advocate or apply the BIC to GMM.  I had also read the BIC
note in Stata. 

I will look into this further. Your comments have been very helpful.
 
Lorena

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Stas Kolenikov
Sent: Tuesday, April 22, 2008 6:36 PM
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: Using Bayesian Information Criterion (BIC)

On Tue, Apr 22, 2008 at 12:15 PM, Rodrigo Alfaro A. <ralfaro@bcentral.cl>
wrote:
>  You are right, there is not e(ll) in the output of -xtpcse- I am not sure
about the model
> behind that command, but it seems to me that you need the matrix e(Sigma)
along
> with the degree of freedom (in this case e(df) is the scalar), and the
sample size: e(N).

BIC is not applicable to neither panel models nor GMM, at least in
direct way. It is derived as an approximation of the Bayes factor for
model selection, in a number of situations it gives a poor
approximation, and at any rate it needs a concept of the likelihood to
work with, and those models do not provide for likelihood.

Entertain a very simple question: even if you want to construct an
analogue of BIC for panel data, should your n be the number of panels
or the number of observations?

-- 
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: Please do not reply to my Gmail address as I don't check
it regularly.
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   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