Econintersect: A blast from Chief economist for the National Federation of Independent Business (NFIB) William C. Dunkelberg towards the BLS' Jobs report released today.
All in all, we received what can only be characterized as a strange report from the volatile Household survey. Giving me advice about our doctoral research at the Survey Research Center at Michigan, one of my professors posited a theorem: ‘Nature never jumps.’ The corollary is this: ‘If it’s interesting, it’s wrong.’ The Household survey is volatile, and averaging its job number over the year produces a more reasonable figure.
Econintersect analysis of the data was presented in post entitled BLS Jobs Situation Disturbing in September 2012 which to a large extent is based on analysis of the unadjusted data. In addition Econintersect Managing Editor John Lounsbury has weighed into the analysis with the following observations that add some more detail to Dunkelberg's observations:
- The 90% confidence level for non farms payroll (NFP) data is around +/- 100,000 (according to the BLS). This means that the changes in NFP for September of 114,000 has a 90% confidence interval going from 14,000 to 214,000 payroll jobs gained. The difference between 114,000 and the 115,000 consensus estimate (1,000) is such a small fraction of the confidence interval that the consensus prediction was actually essentially perfect.
- The number reported (114,000) could have been close to zero (or over 200,000) in the statistical reality of the survey sampling process.
- To reduce the uncertainty of the reported numbers it is useful to use a moving average. Using a 3 month moving average reduces the uncertainty by the Square-root of 3 (equals 1.7). Thus for three months the 90% confidence interval is reduced to +/- 60,000. The average for NFP for the last three months is 146,000 +/-60,000. That's between 86,000 and 206,000 jobs gained on average each month with 90% confidence.
- The uncertainty in the Household Survey is much larger than in the Establishment Survey. The BLS keeps continually assessing the uncertainty and I have seen it reported as low as +/-260,000 and as high as +/-400,000 over the past 5-6 years. The average over time is close to +/-300,000 and I will use that number in this discussion:
- The decline in the number employed in August (119,000) cannot be distinguished statistically from no change.
- The decline in the labor force in August (366,000) is significant, but not by a large margin.
- The increase in the number employed in September (873,000) is clearly a statistically significant change.
- The increase in the labor force in September (418,000) is also statistically significant, but not by the impressive margin for the number employed.
- To get the uncertainty in the Household Survey data down close to +/-100,000 requires a 9-month moving average. I have published in the past that a 6-month moving average (+/-122,000) is a reasonable procedure to implement.
- Bottom line is that people try to use the BLS data in ways that are not justified by the clearly defined statistical limitations of the data. The reported increase from 8.0% to 8.1% for August had no significance, yet many people discussed it as if it did. The decrease from 8.1% to 7.8% in September is a statistically significant change. It is all the more important because it came when the labor force was increasing by a statistically significant amount. All too much of the decline in the unemployment rate in the last few years has come because of a declining labor force more than from an increase in the number employed.
It is likely that pundits and politicos will have their fun with the jobs data this month. Econintersect continues to point out that one cannot rely on the BLS jobs report in real time - but eventually it over time through backward revision and averaging - the data will be correct. To take one month of data in isolation will often turn out to be a statistical mistake.
Going back to the NFIB statement which was the lead for this post - here is their news release in its entirety.
WASHINGTON, D.C., October 5, 2012 — Chief economist for the National Federation of Independent Business (NFIB) William C. Dunkelberg, issued the following statement in response to the jobs report issued today by the Department of Labor.
“Today’s employment report was pretty much what we expected—114,000 new jobs in the Payroll survey. That is consistent with the NFIB survey findings of tepid job growth and most, if not all, reports on the health of the economy. Small businesses were largely sidelined in this jobs report as those businesses in retail, services and construction continue to add only few jobs, certainly not typical of a strong recovery.
“The Household Survey, however, was a shocker, finding over 870,000 new jobs—the biggest gain recorded since June 1983. But in 1983, real GDP was growing at an eight percent pace, not the 1.3 percent it is growing now. GDP growth was consistent with the Household job numbers posted at that time. The less than two percent pace of growth in the current economy does not generate 800,000 new jobs.
“Payroll revisions did add 86,000 jobs for July and August, but virtually all were government jobs, not jobs produced by a healthy private sector; in the private sector numbers were actually revised down a bit. And 580,000 of the new jobs in the Household survey were part-time workers, a rather hollow way to reduce the unemployment rate. However, a more comprehensive measure of unemployment (U6) remained at 14.7 percent. Again, this is not at all consistent with substantial improvement in labor markets.
“All in all, we received what can only be characterized as a strange report from the volatile Household survey. Giving me advice about our doctoral research at the Survey Research Center at Michigan, one of my professors posited a theorem: ‘Nature never jumps.’ The corollary is this: ‘If it’s interesting, it’s wrong.’ The Household survey is volatile, and averaging its job number over the year produces a more reasonable figure.
“As the data come in, we’ll continue to assess the credibility of these latest numbers.”
source: NFIB and Econintersect
Written by Steven Hansen