Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: xtreg fixed effects
From
Nick Cox <[email protected]>
To
[email protected]
Subject
Re: st: xtreg fixed effects
Date
Wed, 23 May 2012 14:14:25 +0100
This is the simple-minded way I think about it.
0. Getting the mean structure average(responses | predictors) right is
more important, usually, than secondary assumptions about variability
or error structure.
1. Each panel contributes at least one data point to the -xtreg-.
That's information!
2. Each singleton panel contains no information about within-panel
variability, or rather implies that within-panel variability is zero
in some cases. But that is only part of the data's properties.
Nick
On Wed, May 23, 2012 at 2:07 PM, Alexander James
<[email protected]> wrote:
> 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: [email protected]
>> To: [email protected]
>>
>> 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
>> <[email protected]> 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?
*
* 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/