Afford Anything - BONUS First Monday: How Did the BLS Get the Jobs Report So Wrong?

Episode Date: August 4, 2025

Special bonus episode. The Bureau of Labor Statistics issues massive job revisions on Friday morning. The revisions wipe out nearly 90% of previously reported gains for May and June. This raises fundamental questions about how our most trusted economic data gets calculated. In this episode, we break down how the system works. We examine why the revisions are so large. We explore what this means for understanding the real economy. Friday arrives. The BLS delivers what appears routine: 73,000 new positions added in July. But the revisions tell a different story. May's initially reported 144,000 job gains become 19,000. June's seemingly solid 147,000 drops to just 14,000. These represent 87-90% overestimates. They fundamentally alter the economic picture for those months. The BLS surveys 560,000 businesses each month. They use payroll data from the 12th of the month. But only 60-73% of those businesses respond by the initial release deadline. The remaining portion gets filled through statistical modeling. The models rely on historical patterns. This approach typically produces revisions in the 20,000-50,000 range. But throughout 2025, average monthly revisions reach 66,000. That's triple the normal size. The statistical models aren't capturing current economic conditions effectively. The problem becomes clear when economic conditions shift rapidly. Historical patterns become unreliable guides. The 2024 annual revision was the largest since 2009. What happened in 2009? The Great Recession. Another period when traditional forecasting tools struggled with rapid change. ADP is a private payroll processor. They serve 460,000 companies. They provide useful comparison data. For May, their 37,000 private-sector job estimate aligns reasonably well with BLS's revised 19,000 total. For June, ADP reports a 33,000 job loss. BLS shows a 14,000 gain. ADP's independent data helps validate the revised numbers while highlighting the magnitude of the initial errors. These numbers drive real decisions. Federal Reserve officials use employment data for interest rate policy. Investors allocate capital based on these reports. Workers make career decisions based on perceived labor market strength. When the initial data misses by 90%, everyone operates with fundamentally flawed information. The revisions expose how fragile our economic measurement systems become when conditions change faster than models can adapt. Learn more about your ad choices. Visit podcastchoices.com/adchoices

Welcome to a very short and completely spur-of-the-moment bonus episode that I'm referring to as First Monday. If you are a longtime listener of this show, you know that on the first Friday of every month, we cover the latest economic news. And we lead those Friday episodes by talking about the jobs report, which is produced by the Bureau of Labor Statistics, the BLS. And our first Friday conversations have always been wonky, data-driven, a little in the weeds, a little for the nerds. But that's who we are. We're not an outrage-bate show. We like data. We like evidence.

We like nuance. And part of the reason why, ever since the first Friday segments began, we have focused so heavily on jobs data from the BLS is because, as I have said, time and time again, if you want to understand how the economy is doing, Don't look at the stock market. Look at two things. Look at jobs data and look at the bond markets. Look at the number of investors who are flooding into or out of treasuries. Because between those two things, between jobs data and the bond markets, that's how you get a much better understanding of the economy,

far, far better than anything the stock market is ever going to tell you. So if you're a long-time listener, and if you listen to our first Friday episodes, you know that we've been covering this for a very long time. And so it was a surprise over the weekend when the BLS ended up in the headlines. For those of you who have not heard, on Friday, the BLS issued the jobs report, as they do on the first Friday of every month. That report showed that the U.S. gained 73,000 new jobs in the month of July. It also revised the estimates for May and June downward by a very substantial margin,

which we will talk more about in a moment. after the first Friday report came out, the administration fired the head of the BLS, accusing the BLS commissioner of politicizing the numbers. In a post-on truth social, President Trump wrote, quote, in my opinion, today's jobs numbers were rigged in order to make the Republicans and me look bad, end quote. And so in this first Monday bonus episode,

I want to address some of the questions that you posed to me over the weekend. People in this community have posed to me over the weekend. Again, I heard from many of you on Twitter and on Instagram, so I wanted to address some of the questions that I know are on your mind. And again, I want to do so in the most fair, nonpartisan, and nuanced manner possible. Let's start with the question. How did the BLS get the jobs report so wrong? For context, on Friday,

the BLS issued two massive revisions. They had originally announced that we added 144,000 jobs in the month of May. We actually added only 19,000 jobs. They had also stated that we added 147,000 jobs. We actually added only 14,000 jobs. So those numbers, the numbers for May and June, were revised down by approximately 90%. And that leads to the numbers. to the question, how did they get that so wrong? To understand that, let's take a look at how the BLS report is put together. The BLS sends out something that's called the Establishment Survey. This is a survey of 560,000 work sites representing about 150,000 businesses and government agencies, all of which are chosen to represent a sampling of employers weighted by industry, size,

and location, and it uses payroll data as of the 12th of the month for each month. Survey responses that are gathered by the 12th of the month are included in the initial release date. However, of the 560,000 that the BLS surveys, only a fraction of them report by that initial release date. I've seen differing numbers somewhere between 60% to 73%. Now, outside of that, the BLS makes up the rest with statistical guesswork that is based on historical patterns. And then, as more responses come in, those additional responses are included in the revision. Now, historically, the BLS has been off by a range of between 20,000 to 50,000 jobs per month is typical or normal. And if we ask the question, what's the average?

the answer is going to depend on what length of time we're looking at and also on whether we're looking at monthly revisions versus, and this is important, the annual benchmark revision because the BLS actually revises their numbers multiple times. They issue monthly revisions as they just did for May and June. They also create an annual benchmark revision where they change job numbers by fairly substantial margins. For example, in the 2024 revision,

the BLS very significantly revised down the number of jobs that were added in the previous year. 2024 had a total net revision of over 800,000 jobs, making it the biggest revision since 2009. Meanwhile, the monthly revisions for May and June that we just saw, those numbers were revised downwards by 90%. And so what we see is that right now, their infill method, which is based on historical patterns, that doesn't work. And it likely doesn't work because so many things have changed in the past few months and are continuing to change. Remember, as I said, the annual revision for 2024 was the biggest revision since 2009. Well, what happened in 2009? It was the Great Recession. It was a time of turmoil, a time of rapid change. And what we see is that in times of

rapid change, that statistical infilling often gets it wrong because it can't adjust for how quickly the job's landscape is changing. That's why those revisions are so important, but that's also why those revisions are so massive. The revisions are important for getting more accurate data. They're massive because the level of accuracy that we can gather from some hastily compiled preliminary reports from a small sample size is just not that good compared to the data that can be seen when we have a much bigger sample size, which comes in over time. It takes a while for survey respondents to respond. Now, there are different sets of data that you can look at to round

out the picture. What we've seen in the data is that even though there are huge revisions to the jobs numbers, the unemployment reporting has been incredibly steady at between 4% to 4.2%. It is notable that despite these huge, huge fluctuations and huge revisions in jobs numbers, there's almost no fluctuation in unemployment numbers. Why is that? Those are two incredibly different data sets and different data gathering methodologies, and the unemployment numbers are a lot more sure, while payroll data is just a lot harder to assemble. The BLS, however, is not the only entity that assemblies. payroll data. There are private companies that are payroll processors, one of which is a company called ADP. ADP is a payroll processing company and they also release a monthly jobs report based on

private data from 460,000 companies, which are their clients, and those 460,000 companies represent 25 million U.S. workers. Every month, ADP comes out with payroll numbers based on their clients. Now, they only serve the private sector. Their payroll numbers are based purely on private sector data. They don't have any public sector employment included in their report, which means that necessarily it's an incomplete picture. But if you want to double-check BLS data against other sources of information, you can always check the BLS data against the ADP data, and that can really round out the picture that you see. Now, ADP found that. in the month of May, the U.S. added 37,000 private sector jobs.

So again, let's just compare those numbers side by side. The BLS in their revision stated that the revised jobs report for the month of May was 19,000 jobs. ADP says 37,000. Look at how close the BLS's revised May number of 19,000 jobs is to ADP's number of 37,000 jobs. 37,000, 19,000, those are very, very close. In a country of 330 million people, we have two different entities giving us a jobs report that are within a margin of 18,000 jobs of one another. That gives a great deal of confidence that May's revised numbers are pretty close to being accurate.

The BLS definitely made a huge error in their initial estimate that we grew by 144,000. jobs in May, no doubt about it. But the revised number of 19,000 jobs in May is so close to ADP's number of 37,000 jobs in May that we just have a huge degree of confidence in those May numbers. All right, looking at that, let's move to June. As you recall, the BLS, initially they said that we added 147,000 jobs in June. In their revision, they said that we actually added only 14,000 jobs in June. Now, what did ADP say about the month of June? ADP said that we lost jobs. ADP reported a loss of 33,000 private sector jobs in the month of June. So if you compare the BLS report to the ADP report for the month of June, the BLS report is actually

making the administration look good compared to what the private sector report is saying. the BLS report paints for the month of June paints a much, much rosier picture than the ADP report does. The BLS in June is at least reporting a gain as opposed to ADP, which is reporting a loss. And so that leads us to the next question, which is, is the BLS politically biased?

I don't see any evidence of that. Okay, let's look at May, June, and July. Let's compare BLS to ADP. The May numbers, they're both close, but ADP paints a slightly rosier picture. The June numbers, they're both close, but BLS paints a slightly rosier picture. The July numbers, at least as of where they currently stand and they will be revised later, but in the July numbers, they're both close, but ADP paints a slightly rosier picture. So there's a flip-flopping back and forth between BLS and ADP as to who paints the rosier picture.

Also, BLS numbers get revised not just downwards but upwards. Most recently, the last upward revision happened in December of 2024. And then finally, the other piece is, of course, the unemployment data, where we have actual unemployment claims. And there's no debate about those numbers. It's consistent and it fits with what we're seeing. So is there evidence of politically based manipulation? No, I don't see that.

Now, you can certainly argue that the BLS should not be making such big revisions. I think it's reasonable to make an argument. Not that the BLS is biased, but I think if you wanted to make the argument that the BLS is incompetent, okay, I think you could reasonably make that argument, sure. Now, on the flip side of the coin, you could also argue that even the most competent people would have a tough time making reasonable estimates right now, given how quickly the economic landscape is changing. So if the question is, are they competent?

I can understand the argument in both directions. I think you can make a reasonable argument on either side of the question of, are they competent. The statement they are incompetent, I believe is a reasonable position, and the statement they are competent is also a reasonable position I'll grant you that. But that is an argument about skill, not an argument about political ideology. There is plenty of evidence to suggest that they may be incompetent. But there is no evidence to suggest that their output is biased or manipulated. Now, to be fair, the White House put out a

statement on Friday that states, quote, BLS has lengthy history of inaccuracies, incompetence. That's the title. Within that statement, it cited the fact that the BLS had to revise down both the May and June jobs reports by a combined 258,000 jobs. And it also cited the fact that the Fed, which met last Wednesday, made a decision to hold interest rate steady, and that decision was at least in part influenced by these erroneous job numbers. It also cited the fact that the annual benchmark revision for 2024 was revised downward by 818,000 jobs. That's for the year ending in March of 2024. Now, that is the second largest benchmark revision on record and the biggest one since 2009.

And the report by the White House talks about some of the other major downward revisions that the BLS has made, including July of 2024. They revised May's numbers down by 54,000 jobs. They also revised April's numbers down by 57,000 jobs. In February of 2024, they revised the jobs added in December of the prior year by 43,000, and the jobs in January of that year down by 124,000. So these are big revisions, even in the context of the history of the BLS, these are larger than normal revisions.

And it's notable that the official White House press statement, which you can find on whitehouse.gov, the official statement does not accuse the BLS of being politically biased. So what the president says on truth social is different from the official White House position. The president on truth social accuses them of political bias, but the official White House position is simply that they are incompetent. I do not see any evidence to back the president's position on truth social. I do think you could make a reasonable argument to back the White House's position. I also think you could make a equally reasonable argument that they are competent and that they are simply dealing with a particularly difficult economic landscape. But I want to draw a distinction between the president's social media post and the official

White House position because it's very important to draw that delineating line between is the BLS political or is the BLS simply incompetent? Those are two very, very different questions with very different implications. And that's why it's critical to draw a distinction between what gets posted on social media versus what officially is the actual allegation here. I normally don't chime in on politics or on headlines, but we've talked about the BLS on so many first Friday episodes and a number of you reached out to me over the weekend. And so I wanted to make sure that I answered your questions. And specifically, somebody asked, when did they last underreport

on jobs? They last underreported on jobs in December of 2024. They revised the jobs report upward by 67,000 jobs. A couple of people asked me, wait, so does that mean they were lying? No, lying implies intent. And in the BLS's case, it wasn't intentional. They simply miscalculated. So no, they were not lying. Their estimates were simply miscalculated. That is a bonus first Monday, and I hope that provides some nuance, provide some context, and provides what I hope is a fair, informed, and nuanced take on what is happening with the BLS right now. Thank you for being part of the Afford Anything community, and I'll see you tomorrow on Tuesday for our regularly scheduled Q&A episode.