Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


[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   Austin Nichols <austinnichols@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: which -cmp- option to use for poisson model with count data?
Date   Mon, 30 Apr 2012 11:26:39 -0400

Laura R. <laura.roh@googlemail.com>:
The problem description sounds entirely wrong.  If the outcome must be
positive, then a zero in the data represents an unknown value and
should be replaced by a missing value, or multiply imputed.  If there
is truncation, it must be above 5, or does 5 represent any number
above 4, i.e. data is censored at 5 via topcoding?  These distinctions
are very important in describing the data you wish to analyze.
Censoring and truncation are not the same thing.

On Mon, Apr 30, 2012 at 10:54 AM, Stas Kolenikov <skolenik@gmail.com> wrote:
> You could try to harness -gllamm- to do it. You could still use -cmp-
> as a crude working model approximation, being aware that an ordinal
> regression is a poor substitute for Poisson model. (Not that I
> disagree with David, the respected author of -cmp-, but it does look
> tempting to run it this way).
>
> On Mon, Apr 30, 2012 at 9:32 AM, Laura R. <laura.roh@googlemail.com> wrote:
>> Dear David,
>>
>> I have a dependent count variable ranging from 0 to 5. It's a pity
>> that this cannot be estimated with -cmp-.
>>
>> Does anyone know a similar model with a two-equation-system where one
>> is poisson regression (for count data), the other one is probit?
>>
>> Laura
>>
>>
>>
>> 2012/4/30 David Roodman (droodman@cgdev.org) <DRoodman@cgdev.org>:
>>> Laura, I'm afraid cmp is probably not appropriate. Unless the counts are high, count data can't be realistically modeled as the outcome of a single underlying process consisting of a linear functional plus a normally distributed error.
>>>
>>> The truncated regression model is intended for the same kinds of outcomes as Stata's truncreg command--one's with a continuous but truncated outcome.
>>>
>>> --David
>>>
>>> ------------------
>>>
>>> From      "Laura R." <laura.roh@googlemail.com>
>>> To        statalist@hsphsun2.harvard.edu
>>> Subject   st: which -cmp- option to use for poisson model with count data?
>>> Date      Fri, 27 Apr 2012 20:16:46 +0200
>>>  Dear Stata users
>>>
>>> Can -cmp- be used with count data, which would normally be estimated
>>> using -poisson- in Stata? For example, the number of experts, ranging
>>> from 0 to 5 in the data (but theoretically limited to >0) .
>>>
>>> As this variable would be truncated (at 0, right truncated), I suppose
>>> I should use the option -indicator($cmp_trunc)- when using -cmp-?
>>>
>>> For example
>>>
>>> *---------------------------- Stata example begin---------------------
>>>
>>> cmp (experts=x1 x2 x3, trunc(0 .)), indicators($cmp_trunc)
>>>
>>> *------------------------- Stata example end------------------------
>>>
>>> would this be correct?
>>>
>>> Thank you very much for your consideration.
>>>
>>> Laura

*
*   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/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index