On Apr 3, 2011, at 10:19 PM, Paul Bartlett wrote:
> You will of course want to use unadjusted data, generally the more years the
> better. BLS probably has some differences with the Census in their
> customizations regarding specifically employment hiring behaviour. I would
> like to hear from Doug the differences of using CPS data (household survey)
> from BLS data. I would assume that BLS is better in employment data, and
> CPS might be better at characterizing the unemployed.
The BLS puts both of 'em out - you must be thinking of the household survey (which is done by Census for the BLS) vs. the establishment survey. The household survey is best for all the demographic data, but its employment numbers are sometimes better at cyclical turns. Generally, though, the standard error on the est survey is much smaller, given the gigando sample size of 300,000 (vs. 60,000, which is still enormous compared to a typical opinion poll).
Doug