If you’re among those who spend a lot of time scrutinizing each month’s unemployment numbers, you might benefit from a close reading of Gene Epstein‘s latest “Economic Beat” column in Barron’s.

There is a long section in every Bureau of Labor Statistics monthly Employment Situation release called “reliability of the estimates” that all reporters should read. The “weak” number in the September report was the rise in nonfarm payroll employment of 148,000, against a consensus expectation of 180,000. But just for starters, as the BLS notes, that figure is based on a sample. Any monthly number is subject to a sampling error of plus or minus 90,000, with a confidence level of 90%. That means, in this case, a 90% chance that the real number is somewhere between 58,000 and 238,000 — anywhere from quite weak to quite strong.

But even that broad range is too narrow. The figure of 148,000 was just the first estimate, based on incomplete data, and is subject to revision. For example, when the March figure was reported by the Journal on April 6, the gain was 88,000 jobs, a figure that now stands in the record books at 142,000. That is just one example among many of what the BLS calls “nonsampling error.”

GIVEN THE ENORMOUS POTENTIAL for error that plagues these numbers, it is misleading to report on any single month’s worth as having the status of fact. But one strategy is to recognize that, from month to month, errors tend to be offsetting. Thus, as the BLS also points out, one can improve the “precision of estimates…when the data are cumulated over time, such as for quarterly and annual averages.”

Looking at the past six months, we immediately notice that the monthly gains in employment are pretty volatile, ranging from a low of 89,000 in July to a high of 199,000 in April. The six-month average through September has been 163,000, essentially the same as the six-month average through August of 162,000.