----- Original Message -----
From: <[email protected]>
To: <[email protected]>
Sent: Thursday, August 21, 2003 3:35 PM
Subject: st: RE: econometric analysis of proportions
>
>
>
>
> Al,
>
> I was thinking about a multinomial logit model. But what about the grouped
> nature of the data? If this were a single proportion wouldn't I want to
> use -glogit-?
>
> For example: cat indexes the categories, year indexes time, count is the
> count by year and category, pop is the sum of count by year, and pct is the
> proportion (count/pop):
>
> . li
>
> cat year count pop pct
> 1. 1 1 2 10 .2
> 2. 2 1 3 10 .3
> 3. 3 1 5 10 .5
> 4. 1 2 3 9 .3333333
> 5. 2 2 3 9 .3333333
> 6. 3 2 3 9 .3333333
>
> .
>
> I thinking to do the following:
> expand count
> mlogit cat year
>
> Would this give me correct standard errors for the trend and constant
> terms?
>
> I just did a simulation and got identical results from doing the above as I
> do with:
> mlogit cat year [fweight=count]
> on the unexpanded data.
>
> In either case I get the same # of obs (19). Is this correct for the
> degrees of freedom calculations, or should it be 6 - the number of
> categories*years?
>
> Regards,
>
> --Alex Cavallo
> Lexecon
> (312) 322-0208 voice
> (312) 322-0218 fax
>
There is FAQ "How can I do logistic regression or multinomial logistic
regression with grouped data?"
(http://www.stata.com/support/faqs/stat/grouped.html) that maybe helpful.
The correct number of obs is 19 - since with frequency weights you are treating
each observation as one or more real observations. See [U]23.16.1 Frequency
Weights
Hope this helps,
Scott
*
* 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/
