Q: Hasn't the US monthly jobs figure frequently been revised downwards over recent years, regardless of which administration was in charge?
(edit - see for instance Aug 2024: https://seekingalpha.com/news/4142722-why-was-there-such-a-b... )
(edit - see for instance Aug 2024: https://seekingalpha.com/news/4142722-why-was-there-such-a-b... )
I think we created a new status for Uber/Deliveroo and other workers to put them out of the category three years ago and it fixed a lot of our employment data issues.
These are two separate metrics, they measure different things, and the figures often differ (unsurprisingly).
The BLS "establishment" survey (aka Current Employment Statistics, CES) surveys 120k+ businesses and government agencies, it measures jobs (not people), counting the number of payroll positions. This is "non-farm payroll employment", excluding the self-employed, farm workers, and private household workers.
The BLS "household" survey (aka Current Population Survey, CPS) surveys ~60k households, measuring individuals, whether they are employed, unemployed, or not in the labour force. These data are used to calculate the unemployment rate and labour force participation. This includes farm workers, the self-employed, and domestic workers.
You assume the data gatherers were at fault... <chuckle>
These data gatherers work for their government. How do you ensure they're happy to gather and publish data which is essentially critical of that very government?
We see large corrections in employment numbers when there's rapid changes in the job market that mess with the models, or when the changes are focused towards small companies. Right-wingers have somehow decided that all of this is instead due to the BLS somehow being out to get Trump, despite there being no significant changes to how the jobs report is made since the mid-90s.
You can have non-biased indicators that have error with mean 0.
Maybe a better question, when judging current operations, is how precise the biased estimates are becoming overtime. Is the size of the error increasing or decreasing.
Such simple statistics and data gathering should be simple for a federal organization.
Simple?! "Sweet summer child..."
On a more serious note, how would one ensure that a government department be sufficiently independent that it can publish data (implicitly) critical of its own political leaders without fear of retribution?
Answers on a postcard, please...
This was broken long before DOGE was a thing:
https://seekingalpha.com/news/4142722-why-was-there-such-a-b...
"There's still ongoing chatter about the huge revision to U.S. job growth seen yesterday and what it might signify for the economy and markets. 818,000 jobs were wiped out in the 12 months through March 2024 (or 68,000 per month), resulting in the biggest downward adjustment since the global financial crisis."
https://www.natesilver.net/p/trumps-jobs-data-denialism-wont...
The monthly revisions are historically all over the place, up and down. My 2024 count says six months were revised up and six were revised down.
The BLS (USA) does adjust the numbers every month (for two months after the initial release) and annually. Regardless if the numbers go up or down, this is fairly common with statistics and forecasting in general. When actuals come in, the forecast is adjusted closer to reality.
Anecdotally: It gets lost in the mix of headlines when those adjustments show that the initial projections were on trend, or "close enough the talking heads don't care enough". However, it gets "interesting" when it's off-trend; or confirms prior notable good/bad news. In this case, it confirms* what was suspected, mostly confirms what was reported. As actuals came in, the reality was worse than projected.
*"Confirms" use case here: job growth is poop right now.
It would be good for everyone if the BLS figures were trusted.
Even "not professional economists" might lose trust in figures which are regularly revised downwards ... months after being published.
Because these numbers are so important- to journalists, to the Fed, to financial markets, etc. they wanted a few million dollars extra, over a few year period, to run the new methodology and the old methodology side-by-side for a significant portion of a business cycle, to understand the differences before they switched, and to gain confidence in the system. Because an important part of this particular data set is what it signals to those others, it is important not to move quickly with this data set, but to give time for everyone to understand all the nuances. It's things like, how the market views the meaning of corrections would be different under a different system, and so they want time so that they themselves and all those other people whose jobs depend on understanding it to be fully aware.
Basically, they wanted to run a blue-green deployment strategy for their updates, but couldn't get the budget for it- and their budget has instead been cut so far. So they have prioritized continuing the system that everyone understands rather than experimenting with new things that no one understands. Because these are smart, well educated people who spend their entire lives thinking about these problems, and understand how the data is used, this is something they have thought about a lot and want to do the best job they can.
However, it's extremely common in forecasting to revise the forecast once actuals come in. In the case of the BLS, it's the documented approach for a very long time.
Every month the numbers are adjusted and annually. All of the notes as to why, the method, etc are in the actual reports*.
*I don't recommend reading them or the footnotes unless you have insomnia. :)
** Also, if the source data is inaccurate, corrupted, etc; if the models are non-transparently adjusted, that would be horrible and cause for alarm. At the moment, we don't know if that is the case. Yet.
Though honestly, I wish the terminology were changed to "forecasted" and "actual" to be clearer.
A few notes from an interview on the Odd Lots podcast, interviewing Bill Beach, former head of the BLS:
* Response rates among surveyed employees are roughly:
Month 1 68%
Month 2 83%
Month 3 93-94%
* Large employers tend to respond sooner, and are staffed to handle these requests better.
--------
April 2025 interview: https://podcasts.apple.com/us/podcast/some-of-americas-most-...
August 2025 interview (after BLS head statistician was fired): https://podcasts.apple.com/us/podcast/bill-beach-on-how-trum...
Some notes and a transcript: https://www.crisesnotes.com/bloomberg-odd-lots-podcast-trans...
Different companies react differently as well. Companies that have a steady flow of cash (food is very inelastic - people eat about the same every day) realize they can give smaller raises, and this is a good time to invest in the company by building so they often hire. Companies that make luxury goods for the common man (think small boats - large yachts for the rich are different) tighten their belts because they are the first place people in fear cut spending.