Nonfarm Payroll Response Rate Warning Continues — 7 Million Jobs Too High?
The most important feature of the Nonfarm Payroll report continues to go unnoticed. We should be watching the response rate in the second revision of the data. That is where failed and shuttered business show up.
This is important because the BLS methodology (and they are doing as well as possible with the tools at hand) assumes that missing reports from closed businesses are offset by new businesses. This usually is true but count me as a skeptic in the current environment. Here is the latest table of response rates, abbreviated to focus on recent years and the third release.
This version shows the continued post-COVID increase in missing responses. The first and second responses have returned to normal levels. Why is only the third response lower than usual? The shuttered business hypothesis is an obvious possibility.
The recent rates are 4 – 5% lower than the average of the prior eight years. If the failed firms are average in size, the NFP jobs estimate could be as much as 7 million jobs too high.
The uncritical acceptance of the data by so many is supporting the viewpoint that the economy does not really need additional stimulus. Eventually we will see the final verdict on this topic, but it will not be until January with the relevant edition of the Business Dynamics report.
Those who make the facile comment that these are good response rates completely miss the point. The response rate if fine for inferences about the characteristics of the sample. The problem is that you cannot infer the sampling frame or universe from the sample results.
Past posts here and here provide useful background drawn from more than 100 articles I have written on the BLS and nonfarm payroll reporting.
OldProf, what do you think of New Deal Democrat’s argument that comparison to the Household survey indicates that the Establishment survey is NOT missing a significant number of job losses?
# of shut businesses is not the same thing as # of jobs, but I believe total # of jobs is the primary question here.
Article on SA from August:
https://seekingalpha.com/article/4367877-julys-headline-jobs-number-miss-business-closures-and-overcount-job-gains
Current FRED chart:
https://fred.stlouisfed.org/graph/?g=vrg2
FYI, the response rate of the Household CPS survey also bottomed in June, but is coming back up
January 82%
February 82%
March 73%
April 70%
May 67%
June 65%
July 67%
August 70%
@thumbsoup
The two surveys count different things —- jobs versus people. The BLS has a procedure for attempting to compare them by making a number of adjustments. In short, you cannot just take raw numbers as NDD did.
Second, the difference in the two lines of his graph could easily allow for an error of the range that I am suggesting. He is just showing that they (roughly) move together.
Response rate in the household survey does not constitute a similar problem. It is used to identify characteristics of the group, not the size of the universe. Put another way, the HH survey is used like a traditional survey, including tracking people who move out to the “seeking employment” group at various levels.
My friend NDD does not answer the two key questions:
What is an example of a survey that is used to determine the size of the universe?
Do we really believe that new business formation is equal to business closures during this period?
Jeff