Every month the Bureau of Labor Statistics (BLS) releases their initial unemployment estimates and these figures headline many newspapers and news broadcasts. The unemployment percentage moves markets and provides endless opportunity for comment and analysis.

TrimTabs, an investment research company, argues in a concise four page report that the employment numbers from the BLS are inaccurate. There have been many critics of the BLS methods in the past but they have generally focused on the “birth/death” adjustment that the BLS uses to account for the creation of jobs by new businesses. In the past several years, these BLS adjustments have proven to be fairly accurate. But the BLS does not appear on CNBC to refute their critics.

This TrimTabs report focuses on several other flaws in BLS methodology. The methods that the BLS uses were developed many years ago when the majority of employment in the U.S. was in manufacturing. Today, manufacturing represents a small portion of the U.S. economy. Yet the BLS continues to survey mostly manufacturing companies and government institutions to come up with their unemployment number. Of the private companies surveyed by the BLS, most of them are large despite the fact that smaller companies, those with 500 employees or less, make up 50% of the economy and most of the economic growth in the U.S.

The BLS reports a statistical 90% confidence in their estimate numbers, resulting in an error of + or – 100,000 jobs lost to their monthly estimate, a relatively small error out of a workforce of 140 million. But market trading is based on pre-estimates of monthly jobs lost by “analysts” as well as competing estimates like that of the payroll processing company ADP. If the BLS figure of jobs lost is 300,000 and concensus pre-estimates were 400,000 jobs lost, the stock market often rebounds. Yet, statistically, the BLS estimate could be the same as pre-estimates. The market conveniently forgets the sampling error of both analyst estimates and the BLS estimates and trades on the estimates as though they were hard data.

The BLS does not report on the hard data, actual state unemployment insurance claims, till it has received and compiled all the state reports. This is done about 12 months later.
In a 12 month period during 2005-06, TrimTabs research shows that the BLS had underestimated job growth by 750 million when the estimates were compared with actual data. While the BLS does a remarkable job gathering data from almost 400,000 companies in a month to arrive at their estimates, the problem of accurately assessing the employment activity of this country is enormous.