2
$\begingroup$

I am a novice Stata user (forced here from R to conform to coauthor's quirky option choices). I need access to the Pearson residuals from a negative binomial regression. I currently have the regression equations specified in this manner:

xi: nbreg cntpd09 logpop08 pcbnkthft07 pccrunion07 urbanpop pov00 pov002 edu4yr ///
 black04 hispanic04 respop i.pdpolicy i.maxloan rollover i.region if isser4 != 1,
 cluster(state)
est store pd

As I understand it, to get the Pearson residuals I need to undertake the same regression using the glm function. Something like this:

xi: glm cntpd09 logpop08 pcbnkthft07 pccrunion07 urbanpop pov00 pov002 edu4yr ///
 black04 hispanic04 respop i.pdpolicy i.maxloan rollover i.region if isser4 !=1, ///
 family(nb) vce(cluster state)

Some details to note: 1)I use an if statement to select a subset of the total records, 2)I am using robust standard errors clustered by "state"

Sorry for not including a replicable example at this stage, but I am new enough to Stata that I am not sure what each part of these statements is doing exactly and wanted the formatting to be exact.

Is there something basic in the coding that would explain differences in coefficients and standard errors between these methods as I have them written out? Should I be worried about these differences? How can I make the glm version conform to the nbreg version?

A secondary question since I have your attention already, how do I save, aggregate, and export the Pearson residuals assuming I get the glm function to work properly and am able to run it on each of the 6 regression models in our paper (Each model has the same observations, just different dependent and independent variables)? Ideally I would have a table with six columns of residuals in any format readable by R (including dta or other Stata defaults).

Thanks in advance,

Chris

Improve this question
$\endgroup$
3
  • 2
    $\begingroup$ Forced by coauthor's quirky use of STATA?? STATA is good software and especially good for Poisson and negative binomial regression. I beleieve Joe Hilbe wrote some of the code for it. $\endgroup$
    – Michael R. Chernick
    Commented Aug 6, 2012 at 15:14
  • 1
    $\begingroup$ I would echo this sentiment as well. Stata is as mainstream as it comes, and although one must actually pay for it, there is no nonsense such as buying "modules" - you get the whole package when you buy. Plus, an amazing user community (Statalist), incredible documentation, and a long history of very well-supported software can make you very confident about your analyses... $\endgroup$
    – pmgjones
    Commented Aug 6, 2012 at 18:46
  • $\begingroup$ Funny to see what triggers people's defensive mechanisms. I meant no slight to Stata. I happen to be an R user because most of my work is with spatial statistics where Stata has a bit of a blind spot. My comment was meant to indicate how I could be so ignorant as to not know how to output results but still need to fit a negative binomial regression. As it happens the clustered robust standard error option in the code above does not have a direct parallel in R and writing the code to replicate it seemed harder than just getting someone to help me output residuals from Stata. Peace Stata users. $\endgroup$
    – csfowler
    Commented Aug 7, 2012 at 12:12

1 Answer 1

Reset to default
3
$\begingroup$

You can definitely use glm to fit this model. In glm, you can specify family(nbinomial $\#_{k}$) and then search for a $\#_{k}$ that makes the deviance-based dispersion equal to 1. However, you can also use family(nbinomial ml) to estimate $\#_{k}$ with maximum likelihood, which should report the same value as nbreg. On the other hand, nbreg will also give you a confidence interval. The nb link function is $\eta=\ln \frac{\mu}{\mu +k}$, where $k=1$ if you specify family(binomial) without the $\#_{k}$ parameter.

To get your residuals, run your glm command first. Then type predict resid1, pearson. Do that for the other 5 specifications to get resid2-resid6. I am not sure what you mean by aggregate, but you can export the residuals (and an id) as a csv file with outsheet idvar resid1-resid6 using "C:/pearson_resids", comma.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Perfect, hadn't seen any references to the nbinomial ml family. That did the trick. $\endgroup$
    – csfowler
    Commented Aug 7, 2012 at 12:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.