Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

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

Re: st: which -cmp- option to use for poisson model with count data?

From   Nick Cox <>
Subject   Re: st: which -cmp- option to use for poisson model with count data?
Date   Tue, 1 May 2012 11:49:35 +0100

I don't think anyone can advise you easily, or rather well, as the
nature of your data and the flavour of your problem are emerging only
very slowly. But

1. 0 is not treated as missing by -poisson-. That is a fundamental
misunderstanding.0 is a possible value for -poisson- and treated quite

2. Evidently you are counting, that so many experts 0 <= k <= K is the
response variable, but K can vary as well. That to me sounds like a
binomial distribution.

3. Your problem does not sound like -oprobit- at all.


On Tue, May 1, 2012 at 11:16 AM, Laura R. <> wrote:
> @ Stas: I did not know that -gllamm- does a similar thing, I will do
> some research on how -gllamm- works and if it is suitable.
> @ Austin: In my sample, the number of experts ranges from 0 to 5, and
> I want the 0 in the estimation, it is not a missing value. But, using
> other samples, the number of experts can be more than 5. So, in
> theory, there cannot be less than 0 experts, but more than 5 experts.
> This is truncation to the right, I think, because censoring would mean
> that it is not fully observed. Hence I would like a -poisson-
> estimation. Maybe -oprobit- would work aswell.
> If I understand you correctly, by default 0 is treated as missing in
> -poisson- (not so in -oprobit-, I think)? So, maybe I can create a new
> dependent variable like this:
> experts_created=(experts)+1
> Then I would not have a experts from 0 to 5, but 1 to 6, and just have
> to keep in mind that outcome 1 means 0 experts.
*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index