Statalist


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

Re: st: areg vce(cluster id) for panel intraclass correlation on individual level?


From   Nils Braakmann <nilsbraakmann@googlemail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: areg vce(cluster id) for panel intraclass correlation on individual level?
Date   Thu, 4 Feb 2010 13:49:14 +0100

Dear Björn,

I'd definitely cluster the standard errors on the level of the
individual. The reason basically is the one stated by Lahiri/Li: With
repeated person-data, you basically have person-year-observations
clustered within individuals. The fixed effects account for the
time-constant part of the unobservables. However, for a lot of
questions you can imagine that unobserved random shocks that hit an
individual at time t might also influence his behavior at t+1 thus
leading to correlated errors within persons. Clustering your standard
errors takes care of that. Additionally, the usual robust standard
errors are biased in the fixed-effects case while the clustered errors
are robust to heteroskedasticity (see Stock/Watson, 2008,
Heteroskedasticity-Robust Standard Errors for Fixed Effects Panel Data
Regression, Econometrica 76(1), pp. 155 - 174). In fact, using -xtreg,
fe robust- leads Stata to suppy standard errors based on -xtreg, fe
cluster(id)- since an update some time ago.

Hope this helps,
Nils

2010/2/4 Björn Bünger <13bjbu@wiwi.uni-muenster.de>:
> Dear Statalister,
>
> I'm working with data from a household panel (GSOEP). When doing a fixed effects regression ( "areg, absorb(id)") the question is whether to use the "clustered data" option (vce(cluster id)) if I want to analyze data "only" on the level of individuals.
>
> Lahiri/Li (2009), 3430 say: "In health studies, observations are often collected on the same subject over time leading to correlated observations.", while textbooks like Wooldridge
> (2002) give the example of, e. g. the grouping of individuals
> in households. Therefore I'm quite certain that the "cluster" option would be necessary if I included the problem that normally more than one individual is a member of the same household.
>
> But what about the problem that observations of one individual at different points of time form one cluster, so intraclass correlation is to be expected? Is this already accounted for by the use of xtreg, fe or areg or is the additional use of the cluster-option sensible?
>
> Lahiri, P./Li, Yan (2009), A new alternative to the standard F test for clustered data, Journal of Statistical Planning and Inference 139, 3430 - 3441.
>
> Wooldridge, Jeffrey M. (2002), Econometric Analysis of Cross Section And Panel Data, MIT Press, Cambridge, MA/London.
>
> Thanks for your consideration!
>
> Best wishes,
>
> Björn Bünger
> PhD-Student
> Institute of Public Economics
> Münster
> Germany
>
> *
> *   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index