I don’t think I was quite precise enough in my last article.

Econopundit posted a few remarks on it, and pointed out a helpful paper published by the Bureau of Labor Statistics discussing the different series and their methodologies and reasons why they produce different measurements. Econopundit summarized his analysis as follows:

Viewed just as a series, differences between the two surveys are actually vanishing on a long-term basis, with a possible interruption in this trend following the last recession. The current uptick is sharp, to be sure, but we've seen at least one downtick (roughly '64-70) apparently just as sharp, and similar but smaller upticks following other recessions.

In short: neither data nor method seem questionable. If the two surveys correlated perfectly it would be a waste of tax dollars to collect them both. These are two useful and perfectly valid methods of measuring the same thing. Neither is "right." Neither is "wrong." The complement and validate each other.

This is true, but ignores the fact that there is also a derivative data series which is politically at play, which is not so much the absolute number of jobs, as the change in employment year-over-year (or administration-over-administration). The manner in which the separation between the two surveys series has recently increased (with the two data series heading in opposite directions) is not irrelevant.

Mathematically speaking: the two main data series (employment in the households survey, and employment in the payrolls survey) over time are themselves closely positively correlated with each other. A look at the graph which Econopundit posted over the weekend (and which I referred to yesterday) suggests something like a +0.8 correlation (just estimating by eye), which of course is quite high. The separation between them has sometimes increased, and sometimes decreased, but the two series have generally moved in the same direction and are highly positively correlated.

But recently the separation between them is increasing because one derivative series (year-over-year change in employment measured by the households survey) is positive, while the other derivative series (year-over-year change in the payrolls survey) is negative. This means that for the politically sensitive derivative measure of year-over-year change in employment, there is negative correlation between the two series. And although it is true that there have been sharp upticks in the separation between the other two series before, it is not quite accurate to say that those previous periods (eg, 1980-1985) are similar: while the separation between the surveys was increasing then, both series were increasing so the derivative series remained positively correlated. I maintain that the recent four-year period 2000-2003 has been unique due to the negative correlation of the two derivative data series, and that this shows some new and fundamental developments in the way work gets done and people get hired.

I actually agree with Econopundit’s last paragraph entirely. My main point in all this is not that the payrolls survey is wrong or should be ignored, but that the overall employment situation has evolved in some subtle ways not entirely or even closely summarized by merely quoting only the payrolls survey. It continues to be a disappointment that the drop in employment from the payrolls survey is all one hears about in the mainstream press—which, after all, is where the main body of voters will get their information when assessing George Bush’s performance in office. The households survey suggests other conclusions than the payrolls survey, and both need to be considered, which politicians don’t seem to be doing to any real extent.