RE: st: xtreg fixed effects

[Alexander James <alexandre_lille-paris@hotmail.com>]

Hi Chelsea,

 

Thanks for you explanation, I am thinking that I should do as a robustness check run the analysis only with the observations that appear at least twice. I havent seen many papers in my field that use the observations that appear just once. I will think more carefully about it.

 

Best regard,

 

Alexandre 






> Date: Tue, 22 May 2012 10:40:26 -0400
> Subject: Re: st: xtreg fixed effects
> From: chelsealynn.garneau@gmail.com
> To: statalist@hsphsun2.harvard.edu
> 
> Hi Alexander,
> 
> I am running xtreg with children in families, with the majority of the
> sample only having 1 child per family. I have received consultation
> in this issue, as it was initially a concern of mine as well. Though
> I can't describe the answer to you in the most sophisticated
> statistical language, I do believe that using xt commands will account
> for the clustering where it exists, taking into account the variation
> within clusters with more than 1 observation. My model is has a
> random effect for variation within families, thus the clustering or
> dependency in the data is accounted for by the model and the fixed
> effects represent more of a population-average effect which is
> interpreted as differences across families, regardless of the size of
> the level-2 cluster.
> 
> I hope this helps, but my short answer is I believe you are fine and
> it is not a mistake. If you don't want clusters with only 1
> observation in the analyses, you will have to take them out yourself.
> 
> Best,
> Chelsea
> 
> On Tue, May 22, 2012 at 10:26 AM, Alexander James
> <alexandre_lille-paris@hotmail.com> wrote:
>>
>> Dear Statalist members,
>>
>>
>>
>> I am running a a fixed effect models (xtreg) predicting the share of self citations that firms making in their patents. However, there is one thing that is worring me. I have some observations that appear just once in my database, but when I run the fixed effects they are not droped (they are considered in the model) so I get something like:
>>
>> Fixed-effects (within) regression Number of obs = 315
>>
>>
>> Group variable: firm_id Number of groups = 159
>>
>> R-sq: within = 0.3211 Obs per group: min = 1
>> between = 0.0512 avg = 2.0
>> overall = 0.0092 max = 13
>>
>> F(27,158) = 16.53
>> corr(u_i, Xb) = -0.7005 Prob> F = 0.0000
>>
>>
>> Is this a mistake? how fixed effects model can run on observations that appear only once in the data?
>>
>>
>>
>> Best Regards,
>>
>>
>> Alexandre
>> *
>> * 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/ 		 	   		  
*
*   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/