Re: st: RE: intreg with fixed effects and clustered standard errors

[Stas Kolenikov <skolenik@gmail.com>]

In all likelihood, you still have confounded effects of industry,
county and firm that effectively eliminate a firm effect. To find yet
another dummy to drop, you can try something like

foreach x of varlist industry_id* {
  qui regress `x' i.county*firm_code
  if e(r2) == 1 d `x'
}

to find out where you have perfect prediction.

On Wed, Feb 29, 2012 at 2:21 PM, Pamela Campa <pamela.campa@iies.su.se> wrote:
> Thanks for your reply
>
> If I use intreg it does not estimate some standard errors. The data are on
> log emissions; for most of the plants I have point data, but for plants that
> release substances below a certain level I have interval-coded data. This is
> how the output looks like:
>
> . xi: intreg llnemissions ulnemissions L1lndistance_year lidensity_bu5
> industry_id1-industry_id20 industry_id22-industry_id64 i.countyid if
> year>=2001, vce(cluster firm_code)
> i.countyid        _Icountyid_1-1568   (_Icountyid_1 for coun~id==01001
> omitted)
> note: industry_id35 omitted because of collinearity
> note: industry_id36 omitted because of collinearity
> note: industry_id38 omitted because of collinearity
> note: industry_id39 omitted because of collinearity
> note: industry_id41 omitted because of collinearity
> note: industry_id42 omitted because of collinearity
> note: industry_id46 omitted because of collinearity
> note: industry_id49 omitted because of collinearity
> note: industry_id54 omitted because of collinearity
> note: industry_id55 omitted because of collinearity
> note: industry_id56 omitted because of collinearity
> note: industry_id61 omitted because of collinearity
> note: _Icountyid_1232 omitted because of collinearity
> note: _Icountyid_1373 omitted because of collinearity
>
> Fitting constant-only model:
>
> Iteration 0:   log pseudolikelihood = -421061.15
> Iteration 1:   log pseudolikelihood = -420990.91
> Iteration 2:   log pseudolikelihood = -420990.91
>
> Fitting full model:
>
> Iteration 0:   log pseudolikelihood = -407783.52
> Iteration 1:   log pseudolikelihood = -407606.27
> Iteration 2:   log pseudolikelihood = -407606.26
>
> Interval regression                         Number of obs   =   134769
>                                            Wald chi2(1574) =       0.00
> Log pseudolikelihood = -407606.26           Prob > chi2     =     1.0000
>
>                          (Std. Err. adjusted for 24839 clusters in
> firm_code)
> ------------------------------------------------------------------------------
>             |               Robust
>             |      Coef.   Std. Err.      z    P>|z|     [95% Conf.
> Interval]
> -------------+----------------------------------------------------------------
> L1lndistan~r |  -.0103242          .        .       .            .      .
> lidensity_~5 |  -.4127465          .        .       .            .      .
> and the standard errors are also not estimated for some industry and county
> fixed effects
>
> Thanks
> Pamela
>
>
>
>
>
>
> On 2/29/2012 6:36 PM, Nick Cox wrote:
>>
>> -intreg2- qualifies as user-written; you are asked to specify this.
>>
>> However, in this particular case hasn't the functionality been folded into
>> -intreg-?
>>
>> That aside, you are asking for guidance on a key issue without telling us
>> anything precise about the data or showing any output.
>>
>> Nick
>> n.j.cox@durham.ac.uk
>>
>> Pamela Campa
>>
>> I'm trying to estimate a regression model with an interval-coded
>> dependent variable. I have plant-level data for several years, and I
>> regress an interval coded dependent variable on some continuous  X's,
>> industry and county fixed effects, and state by year shocks.
>>
>> I cluster the standard errors by plant. I use the command intreg2. The
>> Stata output gives standard errors for all the variables I put in, but
>> it does not show the Wald chi(2) statistic. I did try to remove plants
>> that appear only once in the dataset, counties for which there is only
>> one plant and industries for which there is only one plant, but that
>> does not fix the problem.
>>
>> Could anyone please suggest a way to deal with this?I'm not necessarily
>> interested in the Wald chi(2) , but I'm afraid that the fact that it is
>> missing signals some misspecification in my model.
>>
>> Moreover, my pseudo likelihood is as low as -590000. Is that worrisome?
>>
>>
>> *
>> *   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/



-- 
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.

*
*   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/