this is a somewhat random question: is it possible to find clustered standard errors for first difference regressions? (both theoretically and practically in R or Stata).
Thanks!
$\begingroup$
$\endgroup$
1
-
$\begingroup$ I suggest you give more details and/or a specific example of how your clusters might be defined. This sounds more statistics than programming, so might be better on Cross Validated. I'd expect to see posters' code in this forum. $\endgroup$– Nick CoxCommented Jun 7, 2013 at 6:54
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Yes, it is possible but whether it makes sense depends on what you want to cluster on exactly. As an example, let's suppose you have panel data on individuals who are identified via an id variable in the sample. Then one way to do a first-differenced panel regression in Stata would be
regress D.(y x), vce(cluster id)
In this case, clustering the errors on the id variable corrects the standard errors for serial correlation. Austin Nichols and Mark Schaffer have given a whole presentation on clustering errors in Stata. They also explain clustering on id variables for first-differenced models (slide 6).
