You described the mismatch between people's self-reported view of the economy and the claims of the Biden administration as merely due to "negative narratives" that "distorted people's understanding" and "entice us into believing things that aren't true". It was, you said, "asymmetric warfare".
There was another very obvious possibility at the time: that people's perceptions were correct and the US data was wrong. This scenario was so obvious that I raised it in your comments section:
"The other possibility is that the western economy is highly correlated and that the USA is actually doing more like Europe, but the US data has gone bad in some undetected way." and "Something big missing from the data is the usual kind of issue that causes divergence between people's overall impression and what they should be feeling."
Now we have a huge revision in the numbers downwards. The "vibes" were apparently not so inaccurate after all. It would be reasonable to reflect on the fallibility of macroeconomics and governments vs the wisdom of the crowds.
Thank you for pointing this out, and keeping receipts. The "vibecession" phrase was always infuriating to me. BLS makes unfavorable revisions to its data after the fact every. single. month. The revisions are almost always in the same direction; that is, worse than the headline print.
Why is their data ALWAYS held in higher esteem than lived experience? And what reputational harms are liberals self-inflicting by insisting that people are delusional up to and after data revisions prove we're not?
It looks like in 2023 and 2024 about 80-90% of revisions have been downward, as you say. But in 2021 and 2022 about 90% of revisions were upward. And pre-pandemic it seemed to be closer to 50-50.
My guess is that the sudden step change in remote work is causing problems for their methodologies. At least in tech there are widespread reports of ghost jobs being advertised that don't exist to engage in candidate farming, and it's possible they're struggling to deduplicate jobs being advertised in multiple geographies that are in reality remote jobs.
If this is what's going on the problem would affect Europe less, because you're not so likely to see remote jobs advertised in every country in Europe in parallel due to the less unified labor market, whereas a single job being advertised in every state or major city is a reasonable thing to do in the USA now (for some kinds of work).
Yeah. We should regulate that, like make a law thar for every job posting you must hire an external candidate within 60 days of the posting or pay a fine equal to the annual salary, or something similar. I’m one of these discouraged people that is looking for a job but unwilling to put out hundreds of applications into the void and have largely given up.
Sooo every month that they've been trying to play the soft landing narrative, it's been revised unfavourably after the fact. Got it. Thanks for clarifying.
If by "giant conspiracy of lizard people" you mean "the exceedingly ordinary case of government overstating its efficacy in order to amass more power," then sure.
The takeaway here shouldn't be that the old data is wrong and the new data is correct. The right takeaway is that the US statistical authorities struggle to measure and communicate their own epistemic uncertainty. Seeing major revisions to jobs numbers should decrease your certainty in all economic data from the US, and that revision in your priors should be retroactive.
Jeremy has a worthwhile observation. The last two years have been good ones, and this is the time span over which deleterious effects of a poor employment environment, if they exist, should have been evident.
The revisions to the data don't actually change the growth environment - if anything they suggest productivity growth was higher than previously thought. The magnitude of the vibecession was way beyond that - it looked like an actual recession, which didn't happen.
Minor but important quibble. IF the US pop is growing at 145k/month and the age structure is stable, you need to add E/P*145k jobs to keep E/P stable (P is entire population not 16+ as in official stats) That's more like 70k/month
> But more importantly, there’s just no motive for doing that.
One blind spot of economists is explaining the behavior of government agencies. Sociology roughly divides society into happiness-pursuing citizens, profit-seeking corporations, and power-oriented states.
Modern economics is modeling the first through utility function, and the second through profit maximization, but it lacks a model for government behavior. Accordingly, economists like Noah Smith are not equipped with the tools to explain government action.
Generally, an excellent predictor of government behavior is: does a particular law or regulation give the government more power? If yes, the government is typically supportive. It will be very hard to find a single regulatory body in human history that ever lobbied for its power being reduced. And no authority ever admitted when it became obsolete.
I'm an economist myself so I don't dare speculating about the motive in this case. But I'm smart enough to not trust other economists to be good at explaining government behavior.
"Modern economics is modeling the first through utility function, and the second through profit maximization, but it lacks a model for government behavior. Accordingly, economists like Noah Smith are not equipped with the tools to explain government action."
I'm fairly critical of mainstream economics, but even I know that's wrong. There's an entire subfield of economic research called "public choice" theory that attempts to explain the behavior of government bureaucracies and political agents. I happen to disagree with a lot of the stuff that comes out of the public choice subfield, but it's absurd to claim that economists haven't tried their luck here.
"Generally, an excellent predictor of government behavior is: does a particular law or regulation give the government more power? If yes, the government is typically supportive. It will be very hard to find a single regulatory body in human history that ever lobbied for its power being reduced. And no authority ever admitted when it became obsolete."
I know a single counterexample isn't gonna disprove your argument (and I agree with the broader assertion that government bureaucracies have a tendency towards self-expansion), but there are a few credible examples of government agencies that "admitted" of their own obsolescence. The Civil Aeronautics Board comes to mind: the head of the CAB even conceded as much and helped usher in the deregulation of the airline industry. I actually happen to think that airline deregulation was a mistake, but the CAB's complicity in its own demise cannot be overlooked.
From your linked text on public choice theory: 'Public choice rejects the construction of organic decision-making units, such as “the people,” “the community,” or “society.” Groups do not make choices; only individuals do.'
I'm doubtful whether the dissection of decision taking processes captures the essence of government entities. Economists have found good ways to model companies as abstract entities without having to resort to analyzing the behavior of individual managers and employees. And while economics has developed metrics for happiness and wealth, it lacks a metric for power.
In other words: even public choice theory cannot offer models that capture the behavior of government agencies as a whole or a way to measure power in non-monetary terms.
Thanks for the example of the CAB. I wasn't aware of that.
Noah - The Bloomberg comment about unauthorized immigrants having a large impact on the revision is likely wrong:
1) Immigrants, who enter the country illegally but claim asylum, are eligible to apply for work authorization 180 days after filing their asylum applications. So they can work legally.
2) All employees who are authorized to work are eligible for Unemployment Insurance.
So the revision is probably mostly because of the Birth/Death model adjustments.
There are several ledes that have been buried here.
First, that has been a very large divergence over the last two years between employment growth shown by the household survey vs the payrolls survey. Employment growth in the household survey has been anemic and lagged the growth in the workforce, which is why the unemployment rate has gone up.
The other issue is that household survey doesn’t seem to be capturing the huge increase in workers who have arrived illegally over the past several years - the growth in the labor force (according to the survey) is a fraction of some of the numbers people have tossed around.
Meanwhile the initial release of each monthly payrollls survey has shown very robust growth.
Divergence between the two surveys is fairly common, though this one is significantly large and persistent . It is generally thought that the payrolls survey is more accurate as it is a larger survey. Over time, the two data series generally converge.
Each survey has to make population estimates. If you are contacting “x” number of firms for payrolls and “y” number of households for employment status, you need to estimate what proportion of total firms and households “x” and “y” represent. The BLS makes assumptions about the birth/death rates of firms for the payroll survey and uses census estimates for households, if I remember correctly.
Both surveys get a little squirrelly after major crises or exogenous events (GFC- see 2009 revision of 2008 data and 2022 revision of 2021 data) because there is more creative destruction of firms, more people go self-employed, and the seasonal adjustments generally get screwed up for years because of unusually large changes over a couple of quarters during the crisis.
This sort of revision we just saw, relating to a non-crisis period, is very unusual. What is more unusual, and ignored by Noah, is that in addition to the benchmark revisions there are ordinary month to month revisions (two of them) after each months preliminary release.
Since the start of 2022, 14 of the 17 months of payroll data (the headline releases) have been revised downward, and by a significant amount. Both the scale of the direction and the fact that it is nearly unidirectional over an extended period of time are unusual, looking at the BLS data (they keep revision stats back to 1979 on BLS.gov). And this is before the additional 70k jobs a month reduced by the new revision through March.
It is pretty clear that the BLS have a survey or sampling error and/or are making inaccurate assumptions about firm behavior.
Powell himself a couple of months ago alluded to the payrolls data being overstated. Do I think the BLS is cooking the books? I doubt it. Do I think they haven’t addressed these errors with the sort of urgency they would have if, say, the data was consistently undercounting jobs in Biden’s term leading up to an election year? If I were Sec of Labor I would have been lighting a fire under their butts to fix the problem of it was a consistent undercount.
This is inexcusable. The Fed is essentially flying blind with poor quality employment data. I’m not sure I trust either the payrolls or household survey. Normally I would trust payrolls more, and I had been thinking the weak household report was likely wrong, but it seems that most of the adjustment is occurring in the consistently poor quality payroll survey. Even so, the household survey doesn’t seem to be capturing all of the illegal immigrant population growth, so it could be way off, too.
Hence, for my forecasting I have been relying more on initial claims (also subject to revisions and quirky seasonals), which have been softer but not radically so.
I dont envy the Fed having to deal with really bad data. And no, the BLS should not be “commended” for doing the same old revision process they have been paid to do for decades. They should be called onto the carpet for poor survey data and asked what they are doing about it and what they think is behind it.
As I said, we shouldn’t jump to any conspiratorial conclusions or accuse them of cooking books- I agree with Noah that are likely not doing so, but given the politicization of the census, IRS, FBI, DOJ (all mostly in one direction), we shouldn’t wave away the possibility.
What I don’t get is why doesn’t every single American have to fill out this form every 3 months or so. It’s 2024, we have the internets. I’ve never gotten the American Community Survey. Just make it a law that you have to go online and report your employment status honestly and audit a percentage each month pretty easy peasy these days
Setting aside the fact that this data would ignore people with bad internet access and multiple jobs/limited free time (possibly the most important segments to track in this case), this proposal is a dystopian nightmare policy.
1. How would you “force” Americans to take the survey?
2. How would you enforce “honesty”?
3. What would the agency budget be for investigating compliance and honesty? What powers of investigation and prosecution would they wield against ordinary Americans?
4. How would people be punished for failure to comply, or reporting inaccuracies (intentional or otherwise)?
5. How would the court system have to be buffered to handle the large increase of cases for non-compliance and data irregularities?
This kind of proposal is exactly why technocrats should not be pulling the levers of government.
Uh… revised all in the same direction. At this magnitude? I think this is all unprecedented. Statistically, you would expect there to be ups and downs in revisions. They were ALL downward. unemployment will be 5% by the election.
As a statistician, what are your thoughts on the stark fact that nearly all of these revisions (and certainly the overwhelming net weight of them) occur in one direction: favorable press print, unfavorable revision after the fact?
The observation holds true across every major reporting metric... jobs, GDP, you name it.
It's almost exactly like the data are exaggerated in the direction of the government narrative at the time, no matter what time period you examine. What a coincidence!
"It shows whatever you WANT the data to show" is the most useful thing a statistics professor ever said to me.
I don’t think that anyone is alleging this. Fudging the numbers once does not mean that every adjustment is also a fudge. The previous adjustments were much smaller.
Having a statistic based on phoning people up might not be the best way of getting data, given that most people won't answer phone calls from numbers they don't recognize.
If you can come up with a better way, then it might not be the best. But as far as I can tell, it’s actually the best way that exists, even though it has difficulties.
I wonder why the Social Security administration can’t just document exactly how many people are paying FICA payroll taxes every month, which should be a concrete number covering every job (even held by illegal aliens if they are using an actual SSN) without any polling or interpolation needed. Or are those taxes actually not paid to the government for each paycheck but held in escrow by employers?
Well a survey of corporate payroll departments which is what is done now will also miss self-employed and contract employees. And since much of the labor force is in small businesses (restaurants, gas stations, auto dealers) those won’t get counted either since they don’t have payroll departments either, maybe they survey ADP or other bulk processors. I assume they make some kind of projection for those workers.
Question. How are so called gig workers who are often independent contractors treated in employment statistics? There is a raging war over job classifications of employee versus independent contractor
Comment. The are 6 definitions of employment U-1 a U-6. I believe the official rate is U-3. I kind of like U-4
I think they're included in the household survey but not the payroll survey (unless the payroll survey is adjusted to match the household survey concepts). So they're reflected in the unemployment rate and the employment rate, but not in the jobs numbers.
Does it matter how people are primed to answer? For instance, if a gig worker usually works for three days at a time and is in between jobs for a day, is the answer employed or unemployed?
Regarding the conspiracy to overstate job growth, but failing to keep it a secret till after the election. Did you really come up that scenario, or did you just write up what Trump was ranting about when he called into Fox News last night?
People remember the headline number month to month, not the two successive downward revisions in succeeding months followed by the benchmark downward revision. As a PR strategy, overstating good news first and dribbling out bad news later is a time tested method, I don’t think there is a conspiracy or that the numbers are deliberately cooked, but your counter-conspiracy argument is a weak one. Moreover, the data for most of this year won’t be fully revised until next year, after the election.
What also has gotten little coverage is how many jobs the payroll survey says has been added since Jan 2022 vs how few the household survey says have been created. The household survey usually gets less credence (although it is where the unemployment rate comes from)’and is usually subject to more revision, but for the last 18 months the payrolls have been revised down to be closer to the household results.
The thing about recessions is that we usually don’t know we are in them until some of the old data is revised, and employment is a lagging indicator. One of the big revisions in the series (I think) was in 2005, sharply revising up the Mar 2004-Mar 2005 payrolls total. If you recall, back then Ben was talking “helicopter money” and people were very concerned about growth and potential deflation. Turns out that things in 2004 were not quite as bleak as Ben thought (presaging the 2005-2006 boom?). No, I don’t think the numbers were undercounted in ‘04 to hurt GWB’s election- it is just that there are a lot of assumptions built into these data series and it take awhile for new trends to be fully reflected.
The fact that the household survey has been weak in terms of job growth and the payrolls survey continually revised sharply downward is concerning as a potential trend.
A good analysis of what can be deduced from federal data and where uncertainty remains. I completely agree. OTOH, you may have missed that Trump really did call into Fox News and say that the Biden administration had purposely overstated the employment numbers, but that the truth had leaked out early, instead of after the election. I don't believe that for a minute, but Trump really did claim that.
😂 Yes. Guys, it's just a happy accident that BLS data has gotten unfavorable revisions every month for at least 1.5 years! First we get a favorable press print to set the narrative, then an unfavorable revision later, and that may seem funny to you but actually these one-directional revisions are just statistics! Nay, the *best* statistics.
Notice that the previous year and a half was the reverse. This suggests there was something systematically easy to miss in one direction going on in the one period, and something systematically easy to miss in the other direction going on in the other period.
If you want to claim there’s a political message-setting going on here, then how do you explain why it switched between 2022 and 2023?
Is it possible that it was the impending Presidential election?
I am not claiming to know what happened but it is plausible that political appointees had less incentive to set a narrative in 2021-22 when there was no impending Presidential election that might cost their boss his job.
Everyone in DC is very sensitive to the election cycle.
Possibly. It is quite possible that one bad jobs report is deemed better than many bad jobs reports sprinkled throughout the year. It is also not clear to me that all the downward revisions will be reported before the election.
Everyone has biases, including government employees. What biases might government employees have that could lead to bad data?
The BLS also creates the CPI/inflation numbers. Can those figures be trusted? Can you think of any reason why it might benefit the government to underreport their own inflation? I can think of about 35 trillion reasons.
They have consistently updated their methodology to “lower volatility” of the results. However, if you measure 2020-2024 inflation with the same methodology they used in 1980, the inflation rate peaked at over 15%. Perhaps there is an explanation to the ‘vibecession’, people are losing real purchasing power even while fully employed.
Exactly. You think USG is trying to pay bond yields that are higher than the TRUE rate of inflation? No thank you. They manipulate the yield with Fed bond buying and use all sorts of heuristics to manipulate CPI so they never pay much more than 5%.
I’ll provide you with consistently inaccurate information that is often critical to effectively operate your ‘organization’ (for-profit, not-for-profit, fed/state/local govt, etc).
The information will occasionally be significantly inaccurate (though I will not share what level of accuracy over time is deemed acceptable let alone excellent, nor what our current targets are and what is being done to ensure we are achieve those targets)
This inaccurate information ensures you continue making decisions about current and future operations, including but not limited to people, with relatively poor, incomplete and even wrong information.
You (citizens) should feel good about this because this is how it’s been done for a long time - this is how we do it and you live with it.
In fact, it’s more important to value the act of acknowledging consistent job failure than it is to focus on the consistent failure to do the job well or excellent in the first instance - because, you know, it’s a hard job.
Aside from the obvious disinformation uses of modifying job stats : “the economy is great , jobs everywhere… etc” … the motive to now lower job stats is to provide cover for monetary expansion via a rate cut in September,and more inflation in 2025 … expectations tampering alone will not fix stagflation and seignorage … have to boost supply side of real sectors by lowering energy costs dramatically and boosting manufacturing by eliminating regulatory drag … let middle class people get good paying jobs and stop taxing and regulating them to death
“We tried to fake the numbers, and we almost got away with it, but our master plan was foiled by our own regularly scheduled revision process! CURSES!” That is not how things work.”
If Trump’s Plan 2025 were to go forward, gutting federal departments and replacing civil service employees with Trump cronies, that would be how things work.
You should take this more seriously. It was less than a year ago you were bemoaning how voters were believing "vibes" over data.
https://www.noahpinion.blog/p/vibes-vs-data
You described the mismatch between people's self-reported view of the economy and the claims of the Biden administration as merely due to "negative narratives" that "distorted people's understanding" and "entice us into believing things that aren't true". It was, you said, "asymmetric warfare".
There was another very obvious possibility at the time: that people's perceptions were correct and the US data was wrong. This scenario was so obvious that I raised it in your comments section:
"The other possibility is that the western economy is highly correlated and that the USA is actually doing more like Europe, but the US data has gone bad in some undetected way." and "Something big missing from the data is the usual kind of issue that causes divergence between people's overall impression and what they should be feeling."
https://www.noahpinion.blog/p/vibes-vs-data/comment/44699252
Now we have a huge revision in the numbers downwards. The "vibes" were apparently not so inaccurate after all. It would be reasonable to reflect on the fallibility of macroeconomics and governments vs the wisdom of the crowds.
Thank you for pointing this out, and keeping receipts. The "vibecession" phrase was always infuriating to me. BLS makes unfavorable revisions to its data after the fact every. single. month. The revisions are almost always in the same direction; that is, worse than the headline print.
Why is their data ALWAYS held in higher esteem than lived experience? And what reputational harms are liberals self-inflicting by insisting that people are delusional up to and after data revisions prove we're not?
It’s not Every. Single. Month. as you claim.
It looks like in 2023 and 2024 about 80-90% of revisions have been downward, as you say. But in 2021 and 2022 about 90% of revisions were upward. And pre-pandemic it seemed to be closer to 50-50.
https://www.bls.gov/web/empsit/cesnaicsrev.htm
My guess is that the sudden step change in remote work is causing problems for their methodologies. At least in tech there are widespread reports of ghost jobs being advertised that don't exist to engage in candidate farming, and it's possible they're struggling to deduplicate jobs being advertised in multiple geographies that are in reality remote jobs.
See here: https://www.resumebuilder.com/3-in-10-companies-currently-have-fake-job-posting-listed/
If this is what's going on the problem would affect Europe less, because you're not so likely to see remote jobs advertised in every country in Europe in parallel due to the less unified labor market, whereas a single job being advertised in every state or major city is a reasonable thing to do in the USA now (for some kinds of work).
Yeah. We should regulate that, like make a law thar for every job posting you must hire an external candidate within 60 days of the posting or pay a fine equal to the annual salary, or something similar. I’m one of these discouraged people that is looking for a job but unwilling to put out hundreds of applications into the void and have largely given up.
Sooo every month that they've been trying to play the soft landing narrative, it's been revised unfavourably after the fact. Got it. Thanks for clarifying.
Is this "they", the giant conspiracy of lizard people? Just asking.....
If by "giant conspiracy of lizard people" you mean "the exceedingly ordinary case of government overstating its efficacy in order to amass more power," then sure.
Even with the downward revisions it’s been a soft landing. Or is your point that we’ve been in recession for the last two years?
The takeaway here shouldn't be that the old data is wrong and the new data is correct. The right takeaway is that the US statistical authorities struggle to measure and communicate their own epistemic uncertainty. Seeing major revisions to jobs numbers should decrease your certainty in all economic data from the US, and that revision in your priors should be retroactive.
Jeremy has a worthwhile observation. The last two years have been good ones, and this is the time span over which deleterious effects of a poor employment environment, if they exist, should have been evident.
The revisions to the data don't actually change the growth environment - if anything they suggest productivity growth was higher than previously thought. The magnitude of the vibecession was way beyond that - it looked like an actual recession, which didn't happen.
Minor but important quibble. IF the US pop is growing at 145k/month and the age structure is stable, you need to add E/P*145k jobs to keep E/P stable (P is entire population not 16+ as in official stats) That's more like 70k/month
And I don't like the overly tight link of chronological age and leaving the labor force.
> But more importantly, there’s just no motive for doing that.
One blind spot of economists is explaining the behavior of government agencies. Sociology roughly divides society into happiness-pursuing citizens, profit-seeking corporations, and power-oriented states.
Modern economics is modeling the first through utility function, and the second through profit maximization, but it lacks a model for government behavior. Accordingly, economists like Noah Smith are not equipped with the tools to explain government action.
Generally, an excellent predictor of government behavior is: does a particular law or regulation give the government more power? If yes, the government is typically supportive. It will be very hard to find a single regulatory body in human history that ever lobbied for its power being reduced. And no authority ever admitted when it became obsolete.
I'm an economist myself so I don't dare speculating about the motive in this case. But I'm smart enough to not trust other economists to be good at explaining government behavior.
"Modern economics is modeling the first through utility function, and the second through profit maximization, but it lacks a model for government behavior. Accordingly, economists like Noah Smith are not equipped with the tools to explain government action."
I'm fairly critical of mainstream economics, but even I know that's wrong. There's an entire subfield of economic research called "public choice" theory that attempts to explain the behavior of government bureaucracies and political agents. I happen to disagree with a lot of the stuff that comes out of the public choice subfield, but it's absurd to claim that economists haven't tried their luck here.
https://www.econlib.org/library/Enc/PublicChoice.html
"Generally, an excellent predictor of government behavior is: does a particular law or regulation give the government more power? If yes, the government is typically supportive. It will be very hard to find a single regulatory body in human history that ever lobbied for its power being reduced. And no authority ever admitted when it became obsolete."
I know a single counterexample isn't gonna disprove your argument (and I agree with the broader assertion that government bureaucracies have a tendency towards self-expansion), but there are a few credible examples of government agencies that "admitted" of their own obsolescence. The Civil Aeronautics Board comes to mind: the head of the CAB even conceded as much and helped usher in the deregulation of the airline industry. I actually happen to think that airline deregulation was a mistake, but the CAB's complicity in its own demise cannot be overlooked.
https://www.econlib.org/library/Enc1/AirlineDeregulation.html
From your linked text on public choice theory: 'Public choice rejects the construction of organic decision-making units, such as “the people,” “the community,” or “society.” Groups do not make choices; only individuals do.'
I'm doubtful whether the dissection of decision taking processes captures the essence of government entities. Economists have found good ways to model companies as abstract entities without having to resort to analyzing the behavior of individual managers and employees. And while economics has developed metrics for happiness and wealth, it lacks a metric for power.
In other words: even public choice theory cannot offer models that capture the behavior of government agencies as a whole or a way to measure power in non-monetary terms.
Thanks for the example of the CAB. I wasn't aware of that.
Noah - The Bloomberg comment about unauthorized immigrants having a large impact on the revision is likely wrong:
1) Immigrants, who enter the country illegally but claim asylum, are eligible to apply for work authorization 180 days after filing their asylum applications. So they can work legally.
2) All employees who are authorized to work are eligible for Unemployment Insurance.
So the revision is probably mostly because of the Birth/Death model adjustments.
There are two surveys, as NS mentioned.
There are several ledes that have been buried here.
First, that has been a very large divergence over the last two years between employment growth shown by the household survey vs the payrolls survey. Employment growth in the household survey has been anemic and lagged the growth in the workforce, which is why the unemployment rate has gone up.
The other issue is that household survey doesn’t seem to be capturing the huge increase in workers who have arrived illegally over the past several years - the growth in the labor force (according to the survey) is a fraction of some of the numbers people have tossed around.
Meanwhile the initial release of each monthly payrollls survey has shown very robust growth.
Divergence between the two surveys is fairly common, though this one is significantly large and persistent . It is generally thought that the payrolls survey is more accurate as it is a larger survey. Over time, the two data series generally converge.
Each survey has to make population estimates. If you are contacting “x” number of firms for payrolls and “y” number of households for employment status, you need to estimate what proportion of total firms and households “x” and “y” represent. The BLS makes assumptions about the birth/death rates of firms for the payroll survey and uses census estimates for households, if I remember correctly.
Both surveys get a little squirrelly after major crises or exogenous events (GFC- see 2009 revision of 2008 data and 2022 revision of 2021 data) because there is more creative destruction of firms, more people go self-employed, and the seasonal adjustments generally get screwed up for years because of unusually large changes over a couple of quarters during the crisis.
This sort of revision we just saw, relating to a non-crisis period, is very unusual. What is more unusual, and ignored by Noah, is that in addition to the benchmark revisions there are ordinary month to month revisions (two of them) after each months preliminary release.
Since the start of 2022, 14 of the 17 months of payroll data (the headline releases) have been revised downward, and by a significant amount. Both the scale of the direction and the fact that it is nearly unidirectional over an extended period of time are unusual, looking at the BLS data (they keep revision stats back to 1979 on BLS.gov). And this is before the additional 70k jobs a month reduced by the new revision through March.
It is pretty clear that the BLS have a survey or sampling error and/or are making inaccurate assumptions about firm behavior.
Powell himself a couple of months ago alluded to the payrolls data being overstated. Do I think the BLS is cooking the books? I doubt it. Do I think they haven’t addressed these errors with the sort of urgency they would have if, say, the data was consistently undercounting jobs in Biden’s term leading up to an election year? If I were Sec of Labor I would have been lighting a fire under their butts to fix the problem of it was a consistent undercount.
This is inexcusable. The Fed is essentially flying blind with poor quality employment data. I’m not sure I trust either the payrolls or household survey. Normally I would trust payrolls more, and I had been thinking the weak household report was likely wrong, but it seems that most of the adjustment is occurring in the consistently poor quality payroll survey. Even so, the household survey doesn’t seem to be capturing all of the illegal immigrant population growth, so it could be way off, too.
Hence, for my forecasting I have been relying more on initial claims (also subject to revisions and quirky seasonals), which have been softer but not radically so.
I dont envy the Fed having to deal with really bad data. And no, the BLS should not be “commended” for doing the same old revision process they have been paid to do for decades. They should be called onto the carpet for poor survey data and asked what they are doing about it and what they think is behind it.
As I said, we shouldn’t jump to any conspiratorial conclusions or accuse them of cooking books- I agree with Noah that are likely not doing so, but given the politicization of the census, IRS, FBI, DOJ (all mostly in one direction), we shouldn’t wave away the possibility.
What I don’t get is why doesn’t every single American have to fill out this form every 3 months or so. It’s 2024, we have the internets. I’ve never gotten the American Community Survey. Just make it a law that you have to go online and report your employment status honestly and audit a percentage each month pretty easy peasy these days
Setting aside the fact that this data would ignore people with bad internet access and multiple jobs/limited free time (possibly the most important segments to track in this case), this proposal is a dystopian nightmare policy.
1. How would you “force” Americans to take the survey?
2. How would you enforce “honesty”?
3. What would the agency budget be for investigating compliance and honesty? What powers of investigation and prosecution would they wield against ordinary Americans?
4. How would people be punished for failure to comply, or reporting inaccuracies (intentional or otherwise)?
5. How would the court system have to be buffered to handle the large increase of cases for non-compliance and data irregularities?
This kind of proposal is exactly why technocrats should not be pulling the levers of government.
Uh… revised all in the same direction. At this magnitude? I think this is all unprecedented. Statistically, you would expect there to be ups and downs in revisions. They were ALL downward. unemployment will be 5% by the election.
It's not unprecedented, there are literally two examples of similarly sized revisions on the graph in the post
As a statistician, what are your thoughts on the stark fact that nearly all of these revisions (and certainly the overwhelming net weight of them) occur in one direction: favorable press print, unfavorable revision after the fact?
The observation holds true across every major reporting metric... jobs, GDP, you name it.
Note that in 2023 and 2024 most of the revisions were downward. But in 2021 literally every revision was upward, and in 2022 almost all were also.
https://www.bls.gov/web/empsit/cesnaicsrev.htm
It's almost exactly like the data are exaggerated in the direction of the government narrative at the time, no matter what time period you examine. What a coincidence!
"It shows whatever you WANT the data to show" is the most useful thing a statistics professor ever said to me.
People who believe in conspiracy theories have never worked in a large organization.
1. Wow that statement is wrong. Like holy cow.
2. “The government tells the truth” is a comically naive prior.
3. Look at Wolf Richter’s analysis to see just how manufactured these numbers actually are.
Why would they want to create a narrative around slower job growth in 2021 and 2022? Your conspiracy makes no sense.
I don’t think that anyone is alleging this. Fudging the numbers once does not mean that every adjustment is also a fudge. The previous adjustments were much smaller.
That's exactly what they're alleging. How else do you interpret this?
"It's almost exactly like the data are exaggerated in the direction of the government narrative at the time, no matter what time period you examine"
"It is easy to lie with statistics, but it is hard to tell the truth without them."
This is what I thought as well when elections are around they’ll just massage the numbers to show a rosy picture for the voters.
But they revised the numbers before the election.
One has to take statistics with a grain of salt. Every one knows statistics can manipulated to suit the political agenda of any political group.
Having a statistic based on phoning people up might not be the best way of getting data, given that most people won't answer phone calls from numbers they don't recognize.
If you can come up with a better way, then it might not be the best. But as far as I can tell, it’s actually the best way that exists, even though it has difficulties.
I wonder why the Social Security administration can’t just document exactly how many people are paying FICA payroll taxes every month, which should be a concrete number covering every job (even held by illegal aliens if they are using an actual SSN) without any polling or interpolation needed. Or are those taxes actually not paid to the government for each paycheck but held in escrow by employers?
Wouldn't that exclude the self-employed and independent contractors that receive 1099 forms instead of W-2 forms?
Most 1099 employees file quarterly taxes, though it might exclude some gig workers
Well a survey of corporate payroll departments which is what is done now will also miss self-employed and contract employees. And since much of the labor force is in small businesses (restaurants, gas stations, auto dealers) those won’t get counted either since they don’t have payroll departments either, maybe they survey ADP or other bulk processors. I assume they make some kind of projection for those workers.
Probably, because many pay annually while others might pay monthly or quarterly.
Or not at all.
Very good overview. One question and one comment
Question. How are so called gig workers who are often independent contractors treated in employment statistics? There is a raging war over job classifications of employee versus independent contractor
Comment. The are 6 definitions of employment U-1 a U-6. I believe the official rate is U-3. I kind of like U-4
I think they're included in the household survey but not the payroll survey (unless the payroll survey is adjusted to match the household survey concepts). So they're reflected in the unemployment rate and the employment rate, but not in the jobs numbers.
Does it matter how people are primed to answer? For instance, if a gig worker usually works for three days at a time and is in between jobs for a day, is the answer employed or unemployed?
Then given the very large number of gig workers what does than mean about the accuracy of the recents jobs number revision ?
Regarding the conspiracy to overstate job growth, but failing to keep it a secret till after the election. Did you really come up that scenario, or did you just write up what Trump was ranting about when he called into Fox News last night?
People remember the headline number month to month, not the two successive downward revisions in succeeding months followed by the benchmark downward revision. As a PR strategy, overstating good news first and dribbling out bad news later is a time tested method, I don’t think there is a conspiracy or that the numbers are deliberately cooked, but your counter-conspiracy argument is a weak one. Moreover, the data for most of this year won’t be fully revised until next year, after the election.
What also has gotten little coverage is how many jobs the payroll survey says has been added since Jan 2022 vs how few the household survey says have been created. The household survey usually gets less credence (although it is where the unemployment rate comes from)’and is usually subject to more revision, but for the last 18 months the payrolls have been revised down to be closer to the household results.
The thing about recessions is that we usually don’t know we are in them until some of the old data is revised, and employment is a lagging indicator. One of the big revisions in the series (I think) was in 2005, sharply revising up the Mar 2004-Mar 2005 payrolls total. If you recall, back then Ben was talking “helicopter money” and people were very concerned about growth and potential deflation. Turns out that things in 2004 were not quite as bleak as Ben thought (presaging the 2005-2006 boom?). No, I don’t think the numbers were undercounted in ‘04 to hurt GWB’s election- it is just that there are a lot of assumptions built into these data series and it take awhile for new trends to be fully reflected.
The fact that the household survey has been weak in terms of job growth and the payrolls survey continually revised sharply downward is concerning as a potential trend.
That is all.
A good analysis of what can be deduced from federal data and where uncertainty remains. I completely agree. OTOH, you may have missed that Trump really did call into Fox News and say that the Biden administration had purposely overstated the employment numbers, but that the truth had leaked out early, instead of after the election. I don't believe that for a minute, but Trump really did claim that.
Another great piece stimulating excellent comments--why i love this blog!!!
😂 Yes. Guys, it's just a happy accident that BLS data has gotten unfavorable revisions every month for at least 1.5 years! First we get a favorable press print to set the narrative, then an unfavorable revision later, and that may seem funny to you but actually these one-directional revisions are just statistics! Nay, the *best* statistics.
Calculate the probability of that one, nerds.
Notice that the previous year and a half was the reverse. This suggests there was something systematically easy to miss in one direction going on in the one period, and something systematically easy to miss in the other direction going on in the other period.
If you want to claim there’s a political message-setting going on here, then how do you explain why it switched between 2022 and 2023?
https://www.bls.gov/web/empsit/cesnaicsrev.htm
Is it possible that it was the impending Presidential election?
I am not claiming to know what happened but it is plausible that political appointees had less incentive to set a narrative in 2021-22 when there was no impending Presidential election that might cost their boss his job.
Everyone in DC is very sensitive to the election cycle.
So these nefarious forces decided to wait until just before the election to announce the negative revision, hurting the party in power?
Typical government. Can't even do a good conspiracy.
It doesn't need to be a conspiracy.
If the numbers made Biden look bad, you can bet there would be a big push to figure out what was missing so they could fix it.
Since they make the admin look good, sure they may be off, but no hurry.
This is just basic professional and organizational dynamics. Happens all the time.
Possibly. It is quite possible that one bad jobs report is deemed better than many bad jobs reports sprinkled throughout the year. It is also not clear to me that all the downward revisions will be reported before the election.
Everyone has biases, including government employees. What biases might government employees have that could lead to bad data?
The BLS also creates the CPI/inflation numbers. Can those figures be trusted? Can you think of any reason why it might benefit the government to underreport their own inflation? I can think of about 35 trillion reasons.
They have consistently updated their methodology to “lower volatility” of the results. However, if you measure 2020-2024 inflation with the same methodology they used in 1980, the inflation rate peaked at over 15%. Perhaps there is an explanation to the ‘vibecession’, people are losing real purchasing power even while fully employed.
Exactly. You think USG is trying to pay bond yields that are higher than the TRUE rate of inflation? No thank you. They manipulate the yield with Fed bond buying and use all sorts of heuristics to manipulate CPI so they never pay much more than 5%.
What a pitch!
Noahpinion:
I’ll provide you with consistently inaccurate information that is often critical to effectively operate your ‘organization’ (for-profit, not-for-profit, fed/state/local govt, etc).
The information will occasionally be significantly inaccurate (though I will not share what level of accuracy over time is deemed acceptable let alone excellent, nor what our current targets are and what is being done to ensure we are achieve those targets)
This inaccurate information ensures you continue making decisions about current and future operations, including but not limited to people, with relatively poor, incomplete and even wrong information.
You (citizens) should feel good about this because this is how it’s been done for a long time - this is how we do it and you live with it.
In fact, it’s more important to value the act of acknowledging consistent job failure than it is to focus on the consistent failure to do the job well or excellent in the first instance - because, you know, it’s a hard job.
Carry on.
Aside from the obvious disinformation uses of modifying job stats : “the economy is great , jobs everywhere… etc” … the motive to now lower job stats is to provide cover for monetary expansion via a rate cut in September,and more inflation in 2025 … expectations tampering alone will not fix stagflation and seignorage … have to boost supply side of real sectors by lowering energy costs dramatically and boosting manufacturing by eliminating regulatory drag … let middle class people get good paying jobs and stop taxing and regulating them to death
“We tried to fake the numbers, and we almost got away with it, but our master plan was foiled by our own regularly scheduled revision process! CURSES!” That is not how things work.”
If Trump’s Plan 2025 were to go forward, gutting federal departments and replacing civil service employees with Trump cronies, that would be how things work.
“…the U.S. government statistical agencies are generally regarded as the best in the world, either in the public or the private sector.”
@Noah, how can this be? Are US government statisticians paid better than private sector statisticians anywhere in the world?