[Mr. Bird]

This is very interesting, thanks for the post. A couple of things sort of jumped out at me. One was that the sampling method had a 90 out of 100 chance of falling within 230,000 of the actual number. That, to me, appears to be a compounded margin of error, much higher than either the "90 out of 100" or "within 230,000". I'm not a statistician, though, so what do I know?

Secondly, I can't help but think this whole "interviewing" thing is a completely outmoded and unreliable method of gathering valid data. Just as we frequently wonder why we never get polled for elections (phone calls to our homes), I'd like to know if any Freepers have been contacted by the BLS. I'd hang up on them, and I'm sure many other people who work for a living would too.

Third, how can they issue monthly stats if there is not 100% turnover in the sampled homes? Clearly, a household classified as "unemployed" would be more likely to remain in that status in consecutive months. To retain that household (or an "employed" one, for that matter) for 4 consecutive months essentially reduces the pertinent sample by 25 percent, I think.

That said, if the methodology is consistent, then I think we can still draw some conclusions regarding trends. But it appears that when it comes to hard data, we're dealing with a whole lot of ifs.

[Dave S]

While "within 230,000" may seem large, it is quite small in comparison to the total number of likely unemployed in a month or the total in the workforce which is probably more than a 100,000,000. All they are saying is that if they continually drew different samples from the same universe, 90 times out of 100, they would come up within 230,000 of the actual number. In most published political surveys the polster reports 90 or 95 times out of 100 the sample number is within three to five percentage points of the actual.

My understanding is that MOST of the interviewing is done in person by interviewers knocking on specific doors in specific communities. They probably use telephone when they have made a number of attempts in person and have not been able to reach the potential respondent. So you may hide when the interviewer knocks on your door but I dont think we are biasing the sample much by having people hang up rather than particpate.

The overlap is a slight concern. It would be better if they drew a new sample of 60,000 each and every month BUT it costs a fortune to do this survey they way it is done now. Drawing a completely new random sample of that many households each month is very costly. Very few surveys conducted by US corporations, by the media, or by politicians use a national probability sample. Most are driven more by convenience and cost. Of course the feds dont worry as much about cost, after all they can always just print some more dollars.