RE: st: Time Series Poisson

["Jacobs, David" <jacobs.184@sociology.osu.edu>]

I think the main reason that this procedure is unusual has to do with the origins of time-series.  This may be a bit of an overstatement, but most of the time-series users (and probably developers) are macroeconomists who rarely if ever seem to work with non-linear estimators.  

Even if I'm wrong about this conjecture, it nevertheless is the case that corrections for serial correlation in count models could be much stronger.  My source for this is THE book on count models by Priven and Trevidi (sp?) who claim that corrections for auto-correlation in count models are not well developed.  That book was published in the mid 1990s, so perhaps this claim is not true now, but the responses to your initial post suggest that it is still the case.

Dave Jacobs

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Richard Williams
Sent: Monday, October 31, 2011 9:09 PM
To: statalist@hsphsun2.harvard.edu; statalist@hsphsun2.harvard.edu
Subject: Re: st: Time Series Poisson

At 08:53 PM 10/30/2011, Richard Williams wrote:
>One of my students (a political scientist of course -- they always 
>bring up these weird problems I have never encountered myself!) has 
>a data set that consists of 45 yearly records for the United States. 
>The dependent variable is a count. It sounded to me like the sort of 
>thing that should be analyzed by a time series poisson model. But, 
>unfortunately, I wasn't even sure that such a thing existed - I was 
>hoping there was a tspoisson command, but no such luck.
>
>However, I found this Stata Technical Bulletin for a very old 
>user-written command called nwest. 
>http://www.stata.com/products/stb/journals/stb39.pdf. It says "This 
>article discusses the calculation of standard errors that are robust 
>to heteroscedasticity and serial correlation for probit, logit, and 
>poisson regression models."
>
>I also found this slightly newer post from 2003: 
>http://www.stata.com/statalist/archive/2003-06/msg00258.html.
>
>What I take from this is that he should -tsset- his data and use 
>-glm- to estimate a Poisson model with Newey-West standard errors, 
>e.g. something like
>
>glm y x1 x2 x3, family(poisson) link(log) vce(hac nwest)
>
>Does this sound right, and if so is this the best he can do, at 
>least with Stata?

Thanks again to everyone who has offered suggestions, both on and off 
list. I thought there might be no suggestions, and instead I am 
feeling a bit overwhelmed by all the articles and ideas that have 
been tossed out. Two followup questions:

1. Is there any particular reason that more of the methods in the 
suggested articles haven't already been programmed into Stata? Is it 
because of lack of demand, or is it because of disagreement over what 
is legitimate or what is best?

2. What about the original idea suggested above, using glm with 
Newey-West errors? Is it a terrible idea, better than nothing, or not 
so bad? It is the one thing I know my student could figure out how to 
run in Stata, whereas I am not so sure with all the other ideas that 
have been suggested.


-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
HOME:   (574)289-5227
EMAIL:  Richard.A.Williams.5@ND.Edu
WWW:    http://www.nd.edu/~rwilliam

*
*   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/