Tomas M wrote:
>I want to correct the standard errors for potential correlation between
observations. Is there a cluster() option equivalent? It appears that
there is not, according to the Stata help files.<
You could always bootstrap by cluster, using the bootstrap prefix
command. I'm not aware of pitfalls that might hit in this case, but
quantile regression isn't a "nice" smooth function of the data (as it is
using L1 loss), so there might be problems.
>Also, is there a command to identify the intraclass correlation
coefficient, to determine if there is indeed clustering? Or will it be
okay if I bootstrap anyway, regardless?<
I'm not sure if there's an equivalent or even if one makes much sense
but one way to get a rough feel for it is to use by cluster
bootstrapping and then bootstrap without the clustering and see how
different things are.
>Also, what is a typical rep number, 1000? Or will 100 suffice?
I'd say that depends on the quantile you're estimating but I'd err on
the high side rather than the low side. How long is each replication
taking to run? Quantile regression should be pretty fast these days but
obviously it's not as fast as least squares.
JV
*
* 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/
