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

st: Model selection by using BIC values after using glm command

From   "Abdul Q Memon" <[email protected]>
To   [email protected]
Subject   st: Model selection by using BIC values after using glm command
Date   Sun, 15 Nov 2009 22:23:07 -0000


This is my first email on the group. I hope some one will be able to give
me valueable comments for which i will be thankful.

My querry is about the BIC (Bayesian information criteria). I have seen
two version of formulas one is using the likelihood and other is using the
deviance of the model.

1. BIC=D+p*LN(N) after glm command
2.BIC=-2L+p*LN(N) after estat command

I am using glm command and negative binomial regression on count data
(data is daily observations of road accidents from 1st Jan 1991 to 31st
dec 2005). Total observations are 5429. I am estimating the value of alpha
(k) and using nbreg command in STATA.

I was expecting that both the formulas will give a same pattern for model
selection( i have 20 models on the data and i am sorting the results by
BIC). Although likelihood model formula is bit easy to explain because the
deviance formulas is giving the lowest BIC for null model as the deviance
of negative binomial regression is small after GLM.

I am still in double minds which BIC formula to use. R software is
probably using likelihhod formula.

Your comments, sugeestions or directions to concerned literature will be
highly appreciated.

Best regards


*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index