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 on April 23, and its replacement, statalist.org is already up and running.


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

Re: st: reoprob claims to have no observations


From   Nick Cox <njcoxstata@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: reoprob claims to have no observations
Date   Mon, 5 Sep 2011 13:24:51 +0100

-reoprob-'s code suggests that it ignores missings in a standard way,
and it is a wrapper for -oprobit- which would do that too. So,
although it's good that you got results, my hunch is that the
explanation lies elsewhere.

Nick

On Mon, Sep 5, 2011 at 1:06 PM, Benjamin Volland <volland@econ.mpg.de> wrote:
> Dear statalisters,
>
> just in case anyone runs into similar problems, I wanted to post that the
> "solution" to the problem described below, was in the end fairly simple.
> $indepvar contains the lagged dependent variable, which is (naturally)
> missing for the first observation. Once I deleted the first observation for
> each unit of analysis reoprob ran without a problem. Hence, my hunch is that
> reoprob cannot deal with missing values. Hope this helps.
>
> Best, Ben
>
> On 30/08/2011 15:04, Benjamin Volland wrote:
>>
>> Dear statalisters,
>>
>> I'm currently trying to run a dynamic ordered probit random effects
>> panel estimation as suggested by Woolridge (2005, Journal of Applied
>> Econometrics, Vol. 20 (1), pp. 39-54).
>> I therefore make use of the reoprob command written by Guillaume
>> Frechette (2001, Stata Technical Bulletin, Vol. 10 (59), pp. 23-27).
>> I am using Stata version 10.1.
>> The panel is unbalanced (between 5 and 12 observations per unit of
>> observation). The dependent has 12 steps. The independents contain the
>> LDV, a number of controls (mostly dummies), plus the initial value of
>> the dependent and the mean values of all independents (as suggested by
>> Woolridge). I run a simple pooled ordered probit model up front to
>> identify all variables that may cause problems of multicollinearity
>> (e.g. the mean values of sex and race are the same as the actual values
>> of sex and race). These are dropped before the reoprob estimation, which
>> looks like this:
>>
>> . reoprob foodc $indepvar $meanvarfood $ofoodvar firstfoodc, i(pid)
>>
>> The program then (w/out any problem) runs through the constant-only model
>>
>> . Fitting constant-only model:
>>
>> . Iteration 0: log likelihood = -28629.788
>> . Iteration 1: log likelihood = -26312.933
>> . Iteration 2: log likelihood = -26271.232
>> . Iteration 3: log likelihood = -26269.137
>> . Iteration 4: log likelihood = -26269.133
>> . Iteration 5: log likelihood = -26269.133
>>
>> but after the first iteration of the full model, stata tells me that
>> there are no observations
>>
>> . Fitting full model:
>>
>> . Iteration 0: log likelihood = -23495.781 (not concave)
>> . no observations
>> . r(2000);
>>
>> Interestingly, when I specify the - trace - option it also performs
>> Iteration 1 before returning the same error code.
>> - xtreg ..., re - runs w/out problems (so does the user-written command
>> xtabond2 [Roodman]).
>> Does anyone have a suggestion how I could fix this?
>>
>> Thanks so much, Ben
*
*   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