I am using the Consumer Expenditure Survey put out by the US Bureau of Labor Statistics. They have a whole series of weights for estimating means, and "half sample weights" for
constructing "balanced" replicates to do sort of a bootstrap for
estimating variances of estimators.
But I'm not sure how to best combine this with a method that uses individual fixed effects. I suppose the issue, since this is not a balanced panel, is whether to count someone who appears in the
survey for 4 periods more or less than someone who aeppears in the survey for 3 periods. It seems to me a fair way would be to use aperson's weight in the first period they appear, as suggested below I think. Probably doesn't matter much for the results.
I am also confused about how to estimate the variance of a variable itself, rather than the variance of an estimate. Any references to recommend on this?
BTW, the BLS had this to say about the ability if Stata to deal w balanced repeated replicate weighting:
"We looked into software last year that could easily calculate variances using our weights. We looked at STATA, WESVAR, and SUDAAN. We found that the latter two supported the BRR (Balanced Repeated Replication) technique for which our weights we designed for. Our recollection is that STATA did not support BRR, but you may want to double check with STATA."
-Anyone confirm or deny this?
thanks.
This message may have been composed with voice recognition.
Please forgive any errors .
David Reinstein
UC Berkeley, Economics
Need a new email address that people can remember
Check out the new EudoraMail at
http://www.eudoramail.com
*
* 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/
