Talking With Erica Groshen, Former BLS Commissioner

When labor data become a political issue

Not long ago Donald Trump fired the head of the Bureau of Labor Statistics because he didn’t like the jobs numbers. There’s clearly a push on to place the BLS, one of the crown jewels of U.S. statistical agencies, under political control. So I managed to arrange a chat with Erica Groshen, who headed the BLS from 2013 to 2017 — and who warned in advanced about political interference — about what the agency does, how it does it, and what’s at risk. Transcript follows.

. . .

TRANSCRIPT:
Paul Krugman in Conversation with Erica L. Groshen

(recorded 9/10/25)

Paul Krugman: Hi everyone, Paul Krugman again. This week I'm talking with Erica Groshen, former commissioner of the Bureau of Labor Statistics. Someone who was actually warning in advance that there might be an assault on the statistical agencies, the BLS in particular, and so it has come to pass, with probably a lot more still ahead of us.

Erica Groshen: That's right.

Krugman: I thought we'd talk about some current events, but also some background on what the BLS and other agencies do and why it matters. So, hi Erica.

Groshen: Hi Paul, glad to be here.

Krugman: Among the things I didn't think would become exciting and controversial would be data collection by the federal statistical agencies, but everything is now controversial and political. Tell me a bit about why the BLS exists and why we need it?

Groshen: What does the BLS do? It collects information, it produces statistics and it disseminates them, all to inform the decision makers in the country. The areas in which it does this are employment issues, wages, inflation, productivity, working conditions, et cetera. It was formed in 1884. It's the first independent statistical agency formed in the federal government, and that was at a time of industrial unrest when we had a lot of concerns about tariffs and trade and immigration—that's surprising, topics still in the news today. Nascent unions and employers were killing each other in the streets and the policymakers thought we'd be one step closer to peace if they were working from a common set of trustworthy information.

Krugman: I was shocked at how old the BLS is. It really does go back to pre-McKinley. It was intended to inform policymakers, at least initially?

Groshen: Yeah. Well, not just policymakers, because there was a sense that negotiations would be going on between workers and employers, so it was also intended to inform both sides that were in conflict with each other.

Krugman: These days the private sector and markets hang on each BLS report. People certainly treat the information as if it's extremely valuable and important.

Groshen: Absolutely. In statistics you often see examples of Say's law all the time. Say's law is this concept that sometimes having a supply of something creates its own demand. It turns out that a lot of economic statistics are kind of like that. So while the BLS was created to measure the cost of living and what was going on with wages, employment, the state of the labor market for those negotiating purposes, and for fiscal policy type inquiries, guess what? It became the dual mandate of the Federal Reserve System. Modern monetary policy was possible because this data was available. We could move off the gold standard to modern monetary policy because we had trustworthy information on employment and inflation that we didn't have before.

Krugman: There are multiple surveys, but the two that seem to attract the most attention are the nonfarm payrolls and consumer prices. We just had a lot of news surrounding nonfarm payrolls. We're recording this on Wednesday, the day after there was a big revision. So tell me how the payroll and employment numbers are put together. I think that's really important for people to understand.

Groshen: The official name for this survey is the Current Employment Statistics Program, the CES. The goal here is to find out what's going on with the number of jobs in the country. This program is designed to try and meet two different goals that sometimes are in conflict with each other, timeliness and accuracy. There's a need for very timely information and everybody wants it to be as accurate as possible, but the more timely it is, the less certainty we have on the estimates. But two or three years after the fact, the BLS has a very good estimate of the number of jobs in the economy because it pays all of the state unemployment insurance agencies to curate their information on all of the employers in their states. They send that to the BLS and that becomes what's called the Quarterly Census of Employment and Wages. Virtually all employment in the US is covered by unemployment insurance and that's 96% of jobs in the US. So very detailed, very accurate, but you have to wait a long time for that. So in-between what the BLS does is run a survey drawn from a sample from the QCEW, and it asks them in a more real-time way, “how many people do you have on the payrolls during the pay period that contains the 12th of the month.”

Krugman: I was actually shocked. Recently I learned the fact that the number for July is not the number for July. It's for the week in July that includes the 12th of the month. That sometimes feels like, “hey, that's not close enough.”

Groshen: It’s balancing what's easy for the employers to report. So they'd have to ask, “how often do you pay your workers, is it weekly? Is it monthly?” Instead, just look at the pay period that contains the 12th of the month and see how many people you paid. That's easier for them than asking them to make some new calculations. So they're asked that the last week of the month.

Not all employers know how many people were on their pay period during the 12th of the month because they may pay monthly, or they haven't gotten around to it because they're under turmoil, or “Sally is sick,” or whatever, so the BLS has this panel. It's not all employers, but it accounts for about a third of employment. It's 120,000 employers, 613,000 work sites. They're going to get three times to report that number for let’s say July. First is the last week of July, then the last week of August, and then the last week of September. They're going to be asked the same question about July. If they can report it at the end of July, that's great. Two thirds of them reported that week, that first time. A month later, they have another chance to report it. So they can even correct it if they made a mistake, that does happen. But mostly it's people replacing the missing value with the actual value. Then they even have another chance to report it. You get up to about 90% on the second try and then about 95% on the third try.

Krugman: So in the end, it's voluntary. They don't have to reply, but basically almost everybody does reply within two months.

Groshen: All of the ones who have been initiated, that's right. What I haven't told you is that to get into the survey, first you're invited, and then if you agree, you go through a few months of trying it out and making sure that you actually report it. Then if you turn out to be a reliable reporter, then what you're reporting actually starts going into the statistics. That initiation, that's another part of the response rate. The initiation rate for the CES has fallen by a half in the past 10 years.

Krugman: So in fact we're getting a relatively incomplete sample, hopefully not too biased, but I guess that's one of the issues.

Groshen: The lower the response rate, the more chance there is of bias.

Krugman: You have a sample, it's a big one. The scale of it is actually boggling. Last week I talked to scientists Peter Hotez and Michael Mann, and I said something like, “unlike physical sciences, economics is cheap.” But I was thinking about how I don't need a lot of resources, but the government statistical collection is not cheap. That's actually a major project, and it's part of the reason we need this.

Groshen: Yes, but then on the scale of the government, it's actually not that expensive. The fiscal agencies are a pretty cheap rounding error, generally. (laughs)

Krugman: Yeah, I know. Well, then there's an additional problem. So tell me about “birth-death”.

Groshen: So to go back to this idea that we have this QCEW, we have this administrative data that tells you how many jobs there are. Then you're doing this survey in between to see how those companies are doing: are they adding jobs or losing jobs, et cetera, for the monthly report? But what you can't do with that kind of survey is track firm employer births and deaths well because the births aren't in there and even the deaths aren't easy to detect right away because employers die oddly, they slow down, right? So in order to really get a good estimate of what's going on, the BLS estimates how many jobs have been added or lost due to employer births and deaths, that's done on a monthly basis.

Krugman: The problem is you know how many employees each firm that answered has, but you don't know how many firms there are out there in total. Then, in effect, you have to multiply those numbers to get a total employment number. You have to make an estimate.

Groshen: Right. That's been refined over time, it's gotten much better. But I remember my first year at BLS, they came out with a major upgrade of the birth-death model and it clearly performed better than what they'd had before. But one of the things that they're always trying to deal with is the noise in these numbers, right? So any particular number has just sort of statistical noise around it. One month, the changes could be a change in trend, or could just be noise, right? If you smooth it too much to get rid of the noise, then you're not going to pick up the turning point. But if you hang on every bit of noise, you may predict many more turning points than you actually have. So they had a range to choose from on how sensitive to make it, they chose one that seems reasonable but it's going to smooth things a little bit more than you might want in retrospect.

Krugman: It's often said that the numbers really do tend to be a few months off at missing turning points. That when the economy has turned sharply down, we don't really see that in the monthly numbers until they're revised.

Groshen: Right, that's because of this effect. In order to not overreact during normal times, you have to do some smoothing right around the turning point. If you knew when it was there, you wouldn't do it, but you don't know it's there.

Krugman: So this really is a mammoth issue.

Groshen: Actually one other thing that contributes to that is these late reports (this is separate from the birth death issue), companies where demand for their goods and services has fallen dramatically, they're often in turmoil and they're more likely to report late.

Krugman: Late reports, I was guessing that, that's really true. So we're busy trying to keep our business alive and responding to the BLS is probably going to take second priority.

Groshen: There's a bias that comes at them. The imputation method that the BLS uses is very simple. For most of these employers, if you don't report, you're a missing value. So that implicitly imputes to you the growth rate of the other people in your industry firm size location cell.

Krugman: So there are other ways of trying to figure out what's happening to employment. There's ADP, which is Automatic Data Processing, which does payrolls for a lot of companies, and they produce numbers. There are these surveys of purchasing managers, which don't actually give you a number, but give you how many people say it's up, how many people say it's down, which tend to map on to it. But everyone I talk to says, “if that was all we had, we'd go with it, but there's nothing like the BLS”

Famously in my circles, the ADP number comes out usually a day before the BLS number and everybody says, “oh, that's interesting, but let's wait for the real number.”

Groshen: Yeah. Something like ADP doesn't have the whole sample. It has just their customers, right? And so they work hard at trying to reweight it and they have some information that the BLS doesn't have, so they're able to divide things up in an interesting way. There are two issues with relying on it instead of the payroll numbers. One is that their sample is not complete. The other one is that the way that they adjust for not having a complete sample rely on the CES, so they're not actually that independent. They would have a hard time if the payroll survey disappeared. They would have a hard time producing their estimate and they're probably working on it now: “how would we do this, if we all of a sudden lost faith in the payroll numbers or they got corrupted?”

Krugman: Yeah, so it might not disappear, but it might be a fiction. I hope that they're working on it right now.

Another controversy is the household survey, the unemployment rate, so that’s the current population survey? That’s a random sample of…

Groshen: Yes, 60,000 households per month.

Krugman: Which sounds like a lot, but once you start to slice and dice into subgroups it's not.

Groshen: It’s the slicing and the dicing that is the problem.

Krugman: That also has a problem, which is that that also yields an employment number—a different one—it’s asking, are you working? But that’s the fraction of the households we've interviewed that are working. But in order to get an overall employment number you need to know how many households there are, which is not necessarily something that a sampling method can get you, right? How do we estimate the number of households?

Groshen: So that's the Census Bureau, that's part of their job. They actually conduct the CPS and the BLS owns the labor market questions within it. The Census Bureau owns the more demographic questions. This grows out of their demographic work. I think the thing to realize is that The CPS is constructed so that the ratios and the percentages calculated from it have a decently small standard error, but their estimates of numbers of jobs and those have huge standard errors.

Krugman: Right — for listeners, what that means, roughly speaking is that there’s a bell curve, but how far out does it spread?

Groshen: Yeah, how much uncertainty do you have about that estimate, right? And so the rule of thumb is when you're looking for macro information, to use the CPS for things like the unemployment rate, the labor force participation rate, and these percentages and ratios rather than for the estimates of the number of jobs. But the BLS nevertheless, every month, goes through an exercise where it adjusts those two numbers to be as similar to each other as possible because it's not just a household issue and the uncertainty, but it's also that the definition of a job in the household survey is different than the definition of a job in the payroll survey. So in the payroll survey, if one person holds three jobs, they're going to be there three times and there's no correction. In the household survey, it's how many people have a job. So if you have three jobs, you're only there once. If you change jobs, you could be there twice also in the payroll survey, but only once in the household survey. So there are a number of adjustments that the BLS does to try and bring those two closer together. They get them much closer than might appear if you just look at them separately. They also try to track changes and in the long run, these things move together. But any particular month, they can go in very different directions.

Krugman: There's been a controversy lately, the times being what they are, about employment growth among native born and foreign born workers and which all of course comes off the household survey but depends very much on your estimate of how many native born and foreign born people there are. I've been told by people who really work with it to basically pay no attention to those recent numbers, that our population controls are just not reliable, is that a fair statement?

Groshen: Well, they're working on them, but they lag. Demographics are pretty easy to predict unless there's a big change because they're based on—the reason they're normally easy to predict is because births and deaths of people are pretty easy to predict. But things like immigration, when they're affected by policies, that's very hard for the Census Bureau to pick up on a real-time basis. Then with the CPS, you also have the influence that response rates there are falling also, and particularly among people who might feel vulnerable reporting this information to the government.

Krugman: There was some evidence that seemed to show that seemed to show a large increase in native-born employment, but somehow or other the same numbers imply that there had been a large increase in the working-age native-born population like last year, which sort of can't be true, unless there were a lot of 24-year-olds suddenly emerging from thin air.

Groshen: Some of that understanding has been people answering the question in a little bit different ways. You can have the proportion of employment that is native born growing with actually not the number of native born people growing, right?

Krugman: Right. Okay, consumer prices, tell me how we get consumer prices.

Groshen: We've talked a little bit about the complexities of collecting employment information. Doing price indexes is much more complicated. Because the concept of employment is actually very similar no matter what kind of employer you are. But pricing of goods and services is very different depending on whether you're talking about medical services or shoes or utilities or carrots. You have to have a different strategy for pricing these different kinds of things, then you also have to weight these different price changes by their share of people's market basket. Starting with the market basket, the BLS has what's known as the consumer expenditure survey, which asks voluntarily people in households to give the BLS information about what they spend their money on every month. That is the most burdensome survey that the BLS does, it's a real pain in the neck. The people who do this are performing an enormous public service for the rest of us. That response rate, not surprisingly, has been falling. The BLS has been trying to modernize it to take advantage of all of these new sources of digitized information, people's credit card records, their bank records, things like that, so it's not relying just on their memory. It has a program for modernizing that survey. It has never been funded despite numerous requests for it. They're working on it very, very slowly because they don't have separate funding for it. Now they're modernizing. The modernization plan probably needs modernization. (laughs)

But anyway, still working with what they have, they say, “okay, this much of the market basket is for carrots and this much is for rent and this much is for medical care and insurance and all these things.” Then they go out and they price these. Now, originally, most of that pricing was done by these field economists visiting places and having conversations. Now, more and more of that is collected electronically in some way or another, sometimes by purchasing from companies that are already doing it. Other times, it's scraping it off the web. It's also getting in touch with particular companies and having them just send in the information from their records.

Krugman: More of those prices are estimated or imputed even in good times than we would like, but now it's a lot of indirect estimates.

Groshen: That was beginning to happen anyway, but with the policies of this administration, they've really had to go into that much more. BLS staffing is probably down 20% from what it was before. A third of the leadership positions at BLS are now vacant. The BLS now has PhD economists out there collecting prices because they have to do it. They want to get the numbers out, right? So they're not working on modernizing anything, they're just trying to collect the data.

BLS has suspended collection of price data in three cities already. It's cut back on what it's collecting. It doesn't have the bandwidth to go back and get to prices that weren't reported on time, so you get to this situation. This is still the best estimate out there. It still meets publication quality, but it's just not as good as it used to be. Some of this was a slow moving train wreck and some of it is a problem being caused right now.

Krugman: You have to think that the modern world also just makes it harder. We sort of knew what a loaf of bread was and how to judge its price. We don't exactly know, you know?

Groshen: Amen. Yeah, quality changes are always problematic and with electronic equipment and goods, quality is a real challenge.

Krugman: So the news the day before we talked was, as it does every year, the BLS revised based on the QCEW and at least the initial estimate was that employment in March was 900,000 less than the previous number. Those numbers in turn get revised.

Groshen: Because it doesn't have the adjustment for birth deaths in it yet.

Krugman: Yeah. Last year, I think it started at minus 800 and ended up at minus 500, something like that. But still, the White House did say that this shows that the BLS is broken. But this is normal procedure, right? Although the number was unusually big. I guess the question is, do we have a story about that?

Groshen: So if you count the actual estimate of what the change is, then it would be the biggest. But as a percentage of employment, it's actually not the highest we've seen. It was 6/10ths of a 1% adjustment to the total employment in the country. So that makes it sound tiny, and in many ways it is. But if what you're looking at is what's the impact on how much job growth we thought we had, then that becomes much more substantial. It says we've had about half as much job growth as we thought there was over that year. So that is substantial. So why did this happen? Well, BLS actually gave us a little more information this time than they have in the past. We know that there are two components to this. One is that the CES is the Survey of Employers, it's not the administrative data reported to the states, right? It turns out that there can be discrepancies in this and usually those discrepancies more or less cancel out. But this time there was a strong pattern of companies reporting more that they had more jobs than they reported to the state of how many people were covered by you.

There was more growth in employment that they reported to the CES than they reported to the state. This is misreporting because BLS asks them, “tell us how many people are going to be covered by Unemployment Insurance,” because they want them to work off of those records.

So why is this? Well, it could be that they are hiring more people off the books or contract workers and they're adding those in and they shouldn't be, that's the easiest thing to come up with.

Krugman: So it could be that they've got somebody, maybe an undocumented immigrant, and they are honestly telling the BLS, “we've got this guy working for us,” but they're not paying unemployment insurance taxes, and they're not telling the state of New Jersey that they've got this guy.

Groshen: That's right. There's an increase in that activity over what it used to be.

Krugman: It's interesting that they would be honest with BLS.

Groshen: They believe the BLS swears that it will never use any information that it gathers to provide this to any kind of enforcement agency or anything like that.

Krugman: That's a really interesting possibility. I'm not in the business, or contractor world, but you can easily imagine there's a lot of those workers not being reported to the state unemployment insurance agency.

Groshen: Yeah, and then there's another component, which is the bias in non-response. This is a better known phenomenon, but in this case, it seems to have been a little bit stronger. This is more of what you might expect around the business cycle, which is that the companies that didn't respond grew more slowly than the ones that did. The ones that agreed to participate grew faster than the ones that didn't agree to participate. If one of the reasons you don't participate is you're not very well organized, you don't have the bandwidth, you're really stretched out, you're really stressed, then you could see how, particularly if you're entering this kind of turning point, that those already fragile employers would lose more jobs than the ones who had the bandwidth to say, “we'll do this.”

Krugman: I want to get to the harsh questions. Obviously we're all extremely worried about—there's a political assault. We don't know what's going to happen, it would take more than just changing the commissioner to alter the numbers. It would take really quite extreme stuff, but nothing is off the charts now as a possibility. What my readers ask a lot is, “what do we do if we can no longer trust these statistics?”

I really feel bad asking a former BLS commissioner, “what do we do if we can no longer trust the BLS?” But, what would be the things to do? And how would we know?

Groshen: Think about the norms of these surveys and this exercise that the agency goes through every month to produce these things. It's really highly automated. So all of the thought and the judgment has gone into designing the programs and the processes for getting those numbers out the door in trustworthy form. That's the real genius of the process, if you wanted to muck with them, you're going to have to throw a wrench into that process, and that would be visible. Things would come out late, they wouldn't add up. They would have to change the tables, and just changing the tables to add or subtract a line can take two or three months at the BLS because they have to make sure that everything still comes out right and looks right. When the numbers come in, there's a certain amount of checking and then suddenly, they're producing hundreds of tables.

Krugman: We'd probably be hearing from whistleblowers as well.

Groshen: That's the other part. There'd be resistance from the staff. These are the most dedicated, modest data nerds on the face of the earth, right? They choose to do this because they know it's important and they really care about it. That culture is really strong at BLS. You would get leaks, resignations, whistleblowers, refusal to implement these changes.

Krugman: One of the things that would be roughly contemporaneous with your spell as commissioner, there were a lot of inflation truthers out there claiming that the CPI numbers were rigged and that there was really much more inflation than the US government was reporting. It turned out there was a really useful tool which had been developed originally by Argentine economists, the billion prices index. Because they suspected that their numbers were being rigged, which they were. But there was a billion prices index for the US, and I was leaning on it quite a lot to say, “it looks just like the CPI.” Unfortunately, State Street Bank bought it and put it behind a horrendously high paywall now.

Groshen: There are people working now to set things up like that, particularly for inflation and labor market estimates. So yes, there are people doing that and I applaud them. I think that's important.

With all of these efforts like ADP and Reveleo, there are groups of people who are really trying to use this burgeoning set of digitized information in the economy to produce new measures. Those are complementary to what the BLS does, they help the BLS in many ways and vice versa. If there were monkeying, I think it would eventually show up in one of these two ways, the disruption caused or divergence from things that otherwise normally track each other.

Krugman: The ones that are available now are a little disappointing. I think Adobe has one, but it's just extremely noisy.

Groshen: But it would be better than something that was monkeyed. If you had a group of them together you would probably pick something up.

Krugman: ADP is problematic, but if ADP was showing much weaker numbers than NFP, then we would really start to wonder—not necessarily believe either one was right, but know that something was wrong.

How are people feeling about the situation right now?

Groshen: I think morale is really suffering and the way they're working is not sustainable and they don't know how much longer they can keep it up. So you've got these PhD economists out there collecting data.

Krugman: So this is aside from the political intervention, this is just plain under-resourced.

Groshen: Yes, that's right. This is collateral damage from what's happening throughout the federal government, not aimed at BLS in particular, but definitely affecting BLS. There are a few things that just happened to be at the same time. BLS moved its building, it was down near Union Station, it's now co-located with the Census Bureau in Suitland. That transition was eased by the ability to work from home, but that has been rescinded for all federal workers. Now all these people who all of a sudden have a much longer commute than they ever had before have to show up at the office. It's just an additional hardship that's been imposed on the staff.

Krugman: Wow.

Groshen: The work from home order is a decision on the part of the administration, but the move was in place long before that. So anyway, it's a confluence that just made things harder for the BLS in particular, so there's that. Then yesterday, the secretary of labor, who you would think would be supportive of the agency, issued a statement that basically condemned the BLS and said that it was broken. To have your own secretary say that about you, that's pretty difficult for the employees to take when they know that they've been working all out to get the numbers out and have them be trustworthy.

Krugman: Yeah. This is a general observation that I've had about federal employees, certainly statistical agencies, that we in general have better federal workers than we deserve. Considering the pay, that for highly educated people is way below the private sector, we have these really dedicated people. I know people who moved from the public sector to the private sector for much more pay and then moved back because they felt, “I want my job to have purpose.” Now we're really sort of trashing exactly that kind of person.

Groshen: These people are thrilled when the releases are in the headlines, but they don't want to be in the headlines. That's really the culture there.

Krugman: It's an amazing thing. I think the US probably led the way in producing high-quality government statistics.

Groshen: Totally, the unemployment rate was invented by the BLS. The CPI was invented by the BLS, right?

Krugman: So the whole world took us as their model, and now we're the ones who seem to be deciding it’s not for us anymore. It must be heartbreaking for you too.

Groshen: It's terrible, it's really difficult. The glimmers of hope are that this is a moment when people are beginning to realize what they might lose. So the uproar and support of the BLS has been really heartening from a lot of sectors, people who have been not willing to speak up about loss of other things, have been willing to speak up about not wanting to lose the statistical system, BLS in particular. That's been heartening.

Then the other part that's been heartening, or perhaps as an opportunity, is that one of the barriers to some of the modernization that this agency really wants to go into to try and bring in these other sources of data, we could have more accurate and more timely payroll data if we had some public-private partnerships for ADP-type data, or maybe through the IRS for payroll information that's submitted, like some other countries do. Australia does that, right? So there are opportunities there. Many of these things would be kind of disruptive. This administration doesn't worry about being disruptive. Maybe if they get behind this, this is an opportunity. When they're gone, this will be an opportunity then to jumpstart, leapfrog, and get to where we should be.

Krugman: Okay, terror for the moment, but possibly hope for the future. I guess that's a good message. Not one you expect to be hearing about labor statistics, but applies to lots of things these days.

Groshen: There's a group of us who are trying to figure out how we would redesign the statistical system from scratch.

Krugman: That's great to hear. Well, thank you so much for talking to me.

Groshen: Thank you for inviting me. This has been a lot of fun.

blah blah