Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Fwd: Panel or pooled data


From   nicola.baldini2@unibo.it
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Fwd: Panel or pooled data
Date   Thu, 24 May 2007 12:17:40 +0200

I don't know what you mean with controlling for inflated standard errors, but looking at   
http://www.kellogg.northwestern.edu/faculty/petersen/htm/papers/se/se_programming.htm
can provide you all the answers about how to correct the standard errors in panel data analysis.
Briefly:
(a) -xtreg- can improve efficiency over -regress- not only for the standard errors, but also for the estimation of the coefficients
(b) fixed effects mean that you use only the variation across years in your data, and not variation across individuals. The bad thing is that you cannot have estimates for time-invariant variables and that you don't use all the information that you have. The good thing is that the "genetic" (individual time-invariant) charactericts are purged away and don't impact on the results. Also, you allow for some endogeneity in the data (i.e. the correlation between such individual effects and the covariates). Given the results for the Hausman test, you should prefer fixed effects. 
(c) random effects estimates use a matrix-weighted average of fixed effects and between effects (the latter is when you use only the variation across individuals in your data, and not variation across years). In this case, the individual ("genetic") characteristcs are not fixed parameters, but independent and identically distributed with zero mean. Also, correlation between invididual effects and the covariates must be zero.
(d) these methods require pooling individual groups and allowing only the intercepts to differ across the individual groups, with the assumption of homogeneity of slope parameters.
(e) clustering is something different. After choosing between the fixed and the random effects, you may need to correct for some violations of the hypothesis. Clustering on the id variable correct for both heteroskedasticity and correlation between observations in the same cluster. You can check for the presence of these violations with -xttest3- and -xtserial- (or -pantest2-), all available from ssc.
 
Nicola

At 02.33 23/05/2007 -0400, Glenn Hoetker wrote:
>I have unbalanced panel data and thus need to control for inflated  
>standard errors.  Three means of doing so suggest themselves:
>
>xtreg ......, i(id) fe
>xtreg ......, i(id) re
>xtreg ......., cluster(id)
>
>The discussion under [R] regress demonstrates all three and indicates  
>that they are alternative means (unless the assumptions of random  
>effects are rejected by a Hausman test, which they are in my case).   
>However, it does not discuss what would lead one to prefer one over  
>the other--an insight I've not found anywhere else either.
>
>Can anyone inform me as to:
>
>1. What would lead one to prefer one method over the other?
>2. Whether there is any test to determine which is preferable?
>
>Many thanks.
>
>Glenn
>
>Glenn Hoetker
>Resident Associate, Center for Advanced Study
>Faculty Fellow, Academy for Entrepreneurial Leadership
>Associate Professor (Business, Law, Institute for Genomic Biology)
>University of Illinois at Urbana-Champaign
>217-265-4081
>ghoetker@uiuc.edu 
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   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   |   What's new   |   Site index