Neither the BLS nor ADP makes any effort to weed out duplicate social security numbers.
The BLS excuse (I asked), is they want to but can’t because they are not allowed to look at social security numbers. This of course leads to questions like why aren’t social security numbers encrypted?
Competent programmers would understand there is not even a need to see the social security numbers at all. This is a simple sort-merge construct where the only requirement is to produce a count of matches. No one has to look at anything.
To get around the problem of duplicates, the BLS goes through an elaborate Birth-Death that likely introduces more errors than it solves.
To be fair, some people, especially the self-employed, do not report every month, but that does not negate the issues pertaining to double counting duplicates.
I suggest, but cannot prove, there is a huge double-counting of persons working more than one job in the establishment survey.
Numbers in Perspective
- In the household survey, if you work as little as 1 hour a week, even selling trinkets on eBay, you are considered employed.
- If you don’t have a job and fail to look for one, you are not considered unemployed, rather, you drop out of the labor force. Searching want-ads or looking online for jobs does not count. You need to submit a resume or talk to a prospective employer or agency.
- In the household survey, if you work three part-time jobs, 12 hours each, the BLS considers you a full-time employee.
- In the payroll survey, three part-time jobs count as three jobs. The BLS attempts to factor this in, but they do not weed out duplicate Social Security numbers. The potential for double-counting jobs in the payroll survey is large.
These distortions and discrepancies artificially lower the unemployment rate, artificially boost full-time employment, and artificially increase the payroll jobs report every month.
Something Amiss
Regardless of why (double counting vs other errors), something is clearly amiss when job gains over the last year allegedly rose by 2.350 million but employment rose only by 1.219 million.
The difference is a whopping 1.06 million. That’s an average difference of 88,000 a month in favor of jobs.
Those are non-farm jobs, thus a subset of jobs whereas employment counts anything and everything. Are we losing 88,000 farm jobs a month? I doubt that.
Something is wrong with at least one of the reports or we are adding 88,000 part-time jobs a month the BLS does not catch.
Compared to last month, the 88,000 per month discrepancy is actually a step in the right direction. BLS revisions took away 75,000 jobs in the past two months while employment allegedly rose more than jobs in May.
For discussion of this month’s numbers, please see Jobs +75,000 vs Employment +113,000: Revisions -75,000.
Expect more revisions.
Mike “Mish” Shedlock
“This is a simple sort-merge construct where the only requirement is to produce a count of matches.”
You don’t even have to do that. You simply have to ask the database how many unique SS numbers are contained in the database table. It will almost instantly report a single number, which is the number of people reported as employed.
Most of the BLS report is based on educated guesses. No one from the government actually goes out and counts real job losses or job creations. It is pure guessing with a large factor of inaccuracy.
SS taxes are likely the best indicator. It is probably still an underestimate but all we need is a relative measure.
You are 100% correct about the match, it’s called hashing. SSN numbers can be hashed and matched without ever seeing the real number. It used in adtech to build look alike audience from vendor data
If the IRS gave the jobs reports, it would be far more accurate.
Tax withholdings are far more accurate. They are up significantly since the 2017 tax cuts.
Maybe another tax cut is a better idea than lowering fed funds rate back to zero?