Asymmetrical Information links to Random Jottings, who quotes an anonymous economist on the possibility that the payroll employment survey double-counts jobs whenever the labor market gets so tight that workers take new jobs before their old employers can even update their headcounts. This might have been an important phenomenon in 1998 and 1999.
I raised this possibility back in March. Advantage Econlog!
If the story is true, then during the late Clinton years employment growth was overstated and productivity growth was understated. Conversely, the spectacular productivity growth of the early Bush years was overstated, but job growth was better than the official statistics show.
UPDATE: Another way to look at the state of the labor market is to measure the ratio of employment to population. Randall Parker pointed me to Angry Bear:
If the natural EP ratio is 64%, then the 62.2% EP ratio as of August 2003 suggests we should have had 141.764 million employed. This estimate suggests unemployment was 4.071 million. The 62.4% EP ratio last month suggests we should have had 143.153 million employed. This estimate suggests unemployment has fallen to 3.472 million, that is, a 15% reduction over the last year.
For Discussion. Does the peculiar pattern of productivity data reinforce suspicions about the payroll employment survey?
This is the Payroll survey. I think companies are very good at not giving ex employes monthly checks. The numbers are likely to be fairly accurate. I think this possibility is a red herring.
No it does not. Structurally, Data will appear throughout with standard deviations, and it simply outlines the Wave action of the deviations. At no time could the conceived miscounting, or double-counting, be imagined to have exceeded 35000 Jobs, while the delay miscounts made by Employers could have only hidden the Job losses for Bush.
I have asserted for some time that the Labor Dept. should establish the Survey based upon total number of hours worked for which Wages were paid, then divide by 40. This has the standard deviation of Overtime worked, but which can be estimated accurately. lgl
Payroll tax collections are probably a better indicator of employment than any survey.
Pardon the Naivete, but kindly explain this: If the Payroll survey counts one employed person TWICE -- their "new" job, and then their old job (before payroll records get updated) -- wouldn't that make the employment picture look rosier?
Why ptay tell, aren't we still "overcounting" jobs by this tortured measure?
Reality: One person, one job.
It is just as likely that the survey will be filled out by someone at the new job using an old employee list, and at the old job using new employee data, thus UNDERCOUNTING the payroll by 1.
I suspect that what our productivity data raises suspicions of is the accuracy of our hours worked data, not of our counts of employees. That said, when companies are laying people off and not hiring, they are going to lay off the "marginal" person, thus raising marginal productivity and thus average productivity, and the existing people are going to continue to progress up the learning curve, becoming more productive.
New hires "dilute" the increases in productivity of existing staff, a situation which I believe caused productivity numbers from the mid-late 90s to be lower than what you might expect from the level of capital investment at the time. We caught up during the '00s, resulting in the "unexpected" growth in productivity over the last couple of years.
We know what the E/P ratio is, do we:
http://data.bls.gov/servlet/SurveyOutputServlet