Download PDF Version
There is an old joke in economics. (Are there any new ones?) An accountant, a statistician, and an economist are under consideration for a job. The hiring manager asks them each the same question during their interviews: whats 2+2?
The accountants response: 4.
The statisticians response: The answer is between 3.9 and 4.1 with 95% probability.
The economists response: What would you like it to be?
The September job report was not shaped by politics.
I was reminded of this ditty as I read the accusations from some quarters that the September employment report had been manipulated by the Administration. To my friends who work in Washingtons economic data mills, though, there is nothing amusing about such suggestions; they are not political appointees, and they take great pride in the objectivity of what they do.
From where I sit, government statisticians do an outstanding job given the limited resources available to them. While the economy has grown tremendously in size and complexity over the last generation, Federal funding for statistical monitoring has been reduced in real terms. Given the importance of good data to economic policy, this seems like a dramatic underinvestment.
Still, the figures we received last Friday generated a lot of curiosity. Here are some key questions that were raised, and some tentative answers.
Why did the household survey show much greater job creation than the payroll survey?
The Bureau of Labor Statistics (BLS) conducts surveys of both households and businesses to measure employment. In the long term, the outcomes tend to converge, but September showed gains of 873,000 jobs in the household survey but only 114,000 in the payroll survey.
As the chart shows, the household survey (which is based on a sample of 60,000, and which drives the unemployment rate) tends to be much more volatile than the payroll survey. But the household survey typically does a better job of capturing people who are self-employed or who work for start-up firms. Workers in these latter categories often swell during an economic recovery.
Were there specific categories that had an unusually significant impact in September?
The household survey generally does a better job of capturing new firms and the self-employed.
Those of us who invested some of our weekend parsing the data found a couple of elements worthy of exploration, and potentially vulnerable to reversion over the coming months.
Firstly, about two-thirds of the employment gains reflected in the household survey came in the form of part-time jobs that workers took because full-time positions were not available. Part timers are counted in the payroll survey, but only if they work for one of the 486,000 establishments included in the sample. Those latching on with start-up companies will be a source of variance between the two readings.
One can certainly argue that full-time positions are more desirable, but the unemployment rate does not attempt to adjust for the quality of work that people are doing. And part-time jobs are often a gateway to more permanent posts.
A second area that generated a lot of interest was the increase in employment among workers aged 20-24. As shown in the right-hand chart, above, this cohort has historically left jobs during the month of September, potentially reflecting the end of summer work. The seasonal adjustments used by the BLS to improve on raw data are based on this pattern.
This year, by contrast, the 20-24 year old age group gained employment. The raw increase of around 100,000 jobs became 368,000 when adjusted by past seasonal patterns. If, in fact, students are leaving summer work somewhat sooner than before, there will be a temporary distortion of measured job gains.
Of course, there were certainly components of the September household survey that seemed to understate job gains. So conspiracy theorists on either side could have found grist for the mill. The fact is that there is a lot of monthly churn in these reports, moving in both directions.
With all of this confusion, whats the best way to interpret last Fridays report?
There are two general things we always try to stress when it comes to economic data.
Economic data are not perfect, and we should not expect it to be.
First, the numbers are not perfect. Limited samples and shifting paradigms make it difficult measure economic activity. The social sciences are not governed by the same rules that the physical world is and we should not expect the same precision.
Fridays employment statistics seemed to highlight some of the inherent limitations of the reporting process. If the numbers didnt make perfect sense, it was not the result of willful manipulation. Puzzles in the data arise all the time, even when an election is not proximate.
Secondly, one month does not a trend make. Given the revisions and rebenchmarking that regularly affect economic series, we should be careful not to overreact to a single release.
When viewed from this perspective (which tends to reduce the influence of seasonal adjustments and sampling), the US employment picture has been getting gradually better.
We expect last months drop in the unemployment rate to be retraced a bit when revisions are made. And the decline in joblessness will be slow, given that the economy is struggling to reach its potential rate of growth. But Fridays report supports our view that we are on a positive path.
This is not the first time that the BLS has been accused of bias. Richard Nixon was a regular critic, and George H. W. Bush blamed sluggish reported job creation (later improved on revision) as contributing to his loss to Bill Clinton in 1992.
The reality is far less intriguing: employment is not the easiest thing to measure, especially with limited resources. Questioning the integrity of those attempting this feat is no laughing matter.