On Feb 19, 2007, at 2:24 PM, Daniel Davies wrote:
hmmm not completely fair there Doug. The household survey does at least in principle capture corporate births and deaths, which the payroll survey has to deal with via an ad hoc adjustment. And the point he makes about the sampling error is actually a good one; the accuracy of the household survey is known because it ought to follow sampling theory, whereas the payroll survey is nonrandom with an unknown bias.
But the household survey is benchmarked, if you can call it that, once every ten years, while the payroll is every year (and now with the quarterly publication of the CWES, you can check it every three months). It's been seven years since the last census, and we have only approximate ideas of what's happened to the population since - which matters more than usually now, because of post-9/11 immigration changes. From 1996 to 2005, the average absolute value of the benchmark adjustment was 0.2%; the wide miss of 2006 was a real outlier, and the reasons for it aren't clear yet. Doug
