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

Re: st: question about negative binomial regression by Stata

Subject   Re: st: question about negative binomial regression by Stata
Date   Thu, 14 May 2009 19:04:56 +0200

 From an empirical researcher who almost never heard about statistics but has some knowledge of Stata and user-written commands:
(0) Use -xtpoisson-: the likelihood ratio test at the bottom of the results' table is a test of the overdispersion parameter alpha. When the overdispersion parameter is zero, the negative binomial distribution is equivalent to a Poisson distribution. If the test is significant, then alpha is significantly different from zero and the Poisson distribution is not appropriate.
(1) -xtnbreg- will work with unbalanced panels, too (in my panel, Obs per group: ranges from 3 to 15, with a avg of 13,3)
(2) I don't know almost any test for panels in Stata (but a few for -xtreg-) about the problems you mention. Even if they exist and they find a problem, there are almost no alternative panel and count commands to -xtnbreg- that may solve the problem.
(3) I don't know any count version for the -xtabond- command and the like in Stata
Sorry for not being so much useful,

P.S. I'll NOT receive/read any email but the Digest.

At 02.33 13/05/2009 -0400, Weijia You wrote:
>Dear colleagues,
>I have several questions about building negative binomial regression model in Stata, thank you for your comments and suggestion!
>The dependent varibles yit ranges from 1 to 828 with mean of 44.24 and std. Dev. of 52.75. From the baisc info of the data set, I decide to use negative binomial regression. I don't know if the data of this type meets the requirement of using negative binomial regression? 
>A lot of problems rise immediately when I get down to the analysis. 
>1,is unbalanced data acceptable for this type of model(negative binomial regression)? 
>2,should I consider the collinearity/ heterogeneity/ autoregression/ autocorrelation ? 
>for collinearity, is VIF still suitable for this model? which command should I use? Is variate-covariant matrix helpful in figuring out the related independent variable?
>for heterogeneity, it is recommended to use xtgls(xtgls depvar [indepvars], igls panels(heteroskedastic)). Is the result meaningful for xtnbreg?
>for autocorrelation, "xtserial implements a test for serial correlation in the idiosyncratic errors of a linear panel-data model". As I know, xtnbreg is not a linear panel-data model. So I wonder, which command should I use to do this part of work.
>for autoregression, I don't know how to test and how to correct if necessary. "Xtregar is Fixed- and random-effects linear models with an AR(1) disturbance". The linear here also qenches me. 
>3,what about the combination of dynamic panel data model with the negative binomial regression?  I havn't found any clues about that yet.
>Thank you again for your patience.
>Any comment or suggestion will be much appreciated.

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index