February's Unemployment Rate Didn't Really Decline

Story Stream
recent articles

In its Friday press release the U.S. Bureau of Labor Statistics (BLS) reported that nonfarm payroll employment rose by a healthy 236,000 in February, exceeding gains in December and January, and that the unemployment rate went down from 7.9 percent in January to 7.7 percent last month, the lowest rate in fifty months.

The pickup in payroll jobs, one of the biggest and broad-based increases in the recovery, was welcome news, though the combined job revisions for December and January were slightly negative.

Not mentioned in the press release was some better and some worse news.

Every month the BLS makes a variety of additions and subtractions to adjust its more comprehensive measure of total employment based on household survey to the definition of employer-reported nonfarm payroll jobs. This is done by subtracting agricultural employment, the unincorporated self-employed, unpaid family workers, private household workers, and workers absent without pay from household survey employment, and adding in nonagricultural wage and salary multiple jobholders. The calculations also smooth out the disrupting effects of annual population control adjustments.

Sometimes the result is surprising, as it was for February.

According to the BLS, the increase in the household count of jobs adjusted to the definition of payrolls was a startling 755,000 seasonally adjusted. (To see this result, go to the BLS website and plug the following into the search box: "Employment from the BLS household and payroll surveys: summary of recent trends.")

True, the monthly confidence interval for changes in household-measured employment is larger than for employer reported payrolls. But the household error interval is about plus or minus 400,000 - not small but far short of the adjusted job count. The 755,000 figure is statistically significant and can't be dismissed.

If the truth of what happened on the jobs front is between the two BLS estimates, then we're talking about a half million job gain last month. That's something to cheer about. Whether a strong job market will hold up in the months ahead is uncertain. Output forecasts don't appear to support it unless productivity remains weak.

There's also what looks like bad news - about last month's reported 0.2 point decline in the unemployment rate.

Prior to the end of last year, in the section of its monthly releases on the employment situation that discusses the reliability of unemployment estimates, the BLS stated that the confidence interval for the monthly change in the official unemployment rate was plus or minus 0.19. Starting in the December 2012 release, reporting November data, the error range was changed to 0.20.

When the unemployment rates for the past two months are calculated out to two decimal places, the results are 7.92 for January and 7.74 for February. The monthly decline is 0.18, which is within the BLS error range. (The BLS uses a 90 percent confidence interval for monthly changes.)

Dare it be uttered? It appears that the so-called drop in the February unemployment rate fails the BLS test of statistical significance. In the language of statisticians, the 0.18 recorded change cannot be said to be different from zero. Or more plainly, it can't be said that the unemployment rate declined last month.

One might wonder why BLS has historically reported monthly changes in unemployment rate confidence intervals out to two digits, yet rounded the reported unemployment rate to one digit, and then called monthly changes of 0.2 point significant (saying the jobless rate went up or down) based on the figures rounded to one digit. As the result of this practice, as happened last month, some month-to-month changes that are not statistically significant become significant and possibly misleading.

The combination of a notable improvement in jobs in February, taking into account both the payroll and adjusted household survey results, and no "real" decline in the jobless rate, is atypical. The erosion of labor force participation has been a contributing factor. Also, there is often a short response lag between an increase in job opportunities and its attraction of the jobless into the labor force. Next month's data should be clarifying.

Alfred Tella is a former Georgetown University research professor of economics. 

Show commentsHide Comments

Related Articles