Most of the news reports on the January employment report expressed confusion over the seeming contradiction between the 0.4 percentage point plunge in the unemployment rate shown by the survey of households and the weak 36,000 job growth reported by the establishment survey. The drop in unemployment in the household survey was the result of a reported increase in employment of 589,000, after adjusting for changes in population controls. This difference is actually not very confusing to people familiar with the data.
The household survey is always erratic. It effectively is measuring the level of total employment in the economy. Even if it is off by just 0.2 percent, this implies an error of almost 300,000. If it errors by this much on the high side one month and then by an equal amount on the low side the following month, it would imply a drop in employment of 600,000 in a context where there was no actual change in employment. Looking back over the last two decades it is easy to find months with large changes in employment that did not coincide with any obvious upturns or downturns in the economy.
For example, in April of 2007 the survey showed a drop in employment of 724,000 at a time when the economy was still showing healthy growth. In May of 2000, also a period of healthy growth, the survey showed a fall in employment of 640,000.
For some reason, probably associated with the difficulty of seasonal adjustments, January is especially prone to show such out of line numbers. In January of 1992, 1994, and 1997, the household survey showed gains in employment (not counting any population control effects) of 512,000, 502,000, and 438,000, respectively. The economy was growing in each of these months, but certainly not at a pace that would be consistent with the creation of 5-6 million jobs a year. In January of 2000, the employment gain (adjusted for the change in controls) was 784,000, which would imply an annual rate of job creation of more than 9 million.
This is why economists familiar with the two surveys tend to rely much more on the establishment survey. This survey is benchmarked every year to the state unemployment insurance data, which is a virtual census since it covers nearly all employers in the country. The establishment survey effectively measures changes rather than levels. There are reasons that it can be inaccurate as well (most importantly in picking up jobs in newly created firms), but the error is likely to be measured in the tens of thousands, not hundreds of thousands.
Those desiring a third source of data on labor market could have looked to the data on unemployment insurance filings. The 4-week average stands at 430,000. The economy did not start generating jobs on a consistent basis following the last recession until claims fell below 400,000. A weekly average of 430,000 new claims is certainly inconsistent with the sort of extraordinary job growth implied by the household data.
National Public Radio deserves special blame for misleading its audience on this one. It told listeners that the economy was actually generating jobs quite rapidly, except for the jobs that involve outdoor work where growth was constrained by the weather. The data don't support this picture. While construction (an outdoor occupation) did lose 32,000 jobs, finance lost 10,000 and information services loss 1,000. Even health care showed very weak growth, adding just 10,600 jobs, less than half the average increase over the last year.
NPR ironically chose to focus on the courier industry as an especially affected outdoor industry, since it reportedly lost 44,800 jobs in January. However, the reporter apparently missed the fact that the number of jobs in this industry had jumped by an anomalous 46,100 in December. This means the job loss shown in January was most likely just reversing some unusual circumstances that led to a surge the prior month and had little to do with January's weather.