Last week, I did a couple of interviews with Brian Edwards-Tiekert of KPFA: one about what the latest BLS employment data is telling us, and one about the Fed’s decision to lower interest rates by a quarter point – in part in response to that data. Having given this some thought for the interviews, I thought it might be worth putting something down in a blog post.
I should stress: I am not a labor economist, or an expert on labor-market data. If I add any value here, it’s from applying a more systematic Keynesian perspective. I also know there are people who read my posts but don’t read more specialized economics content, who would appreciate a discussion of recent employment data. If people find this post useful, perhaps I’ll do more like it in the future. Note that many of the figures here come from Brian Dew’s excellent chartbook.
The headline numbers were relatively weak employment growth, and a downward revision to employment numbers from earlier this year. Employment was up a bit less than 1.5 million over the year ending in August, by the establishment survey. This is not just slower than the post-pandemic recovery, but well below the consistent two million jobs per year that were being added prior to the pandemic. The headline in the Financial Times was typical: “US companies put brakes on hiring after Donald Trump’s tariffs hit.”
The article begins:
The American industries most exposed to trade turmoil have slammed the brakes on hiring and in many cases begun to lay off workers, causing growth in the US labour market to grind to a halt.
It’s worth pausing over this sentence for a moment, to call attention to something that should be obvious, but is not always foregrounded: Hiring decisions are made by employers.
Businesses choose a level of output based on current or anticipated demand for their products. They then choose a level of employment based on how many workers they need to produce that output. The “labor market” is not a textbook market, in which the quantity sold depends on both supply and demand.1 The quantity of employment depends solely on demand. Where labor market conditions matter is for wages and working conditions, though this is better thought of in terms of bargaining power than labor supply.
The point here is that if we are trying to explain why the employment numbers are what they are, we need to think about demand for current output, and perhaps business expectations about future demand. Labor supply explains nothing about why employment is growing so slowly today, or why it was growing more rapidly last year. At the same time, it is true that the working-age population is now growing more slowly than it was a year or two ago. 2024 now appears to have had the greatest number of immigrants, in absolute numbers, in US history.2 Much lower immigration this year, along with the secular fall in natural growth, means the working-age population is probably growing quite a bit slower now; but because of the way the BLS constructs its population estimates, this will not show up in the official data until next January. Again, slower population growth has no direct effect on actual employment, which depends strictly on demand the side. But it does mean that slower employment growth implies a tighter labor market than it otherwise would. I’ll come back to this point later on.
The fist question that the KPFA interviewer asked me was about the revisions to the earlier 2025 employment numbers, which reduced job growth for the 12 months ending this past March by 911,000 — in absolute terms, the largest revision since the data has been collected in its modern form. I did not have a definitive answer to this one. What I do think, as I said on the show, is that first, US economic data, despite all of DOGE’s efforts to date, remains probably the best in the world, and the revisions are an integral part of maintaining its quality; and second, that there is plenty of other evidence we can look at that shows a clear picture of a weakening labor market, so there is no need to try to read something into the revisions.
The one thing to add to that is that the revisions refer to the establishment data. BLS data comes from two fundamental sources — a survey of households that is used to produce statistics on people’s employment status, and a survey of businesses that is used to produce statistics on total employment and its distribution across industries. Since both of these are surveys, they need to be combined with some independent measure of the size of the relevant population to turn shares into numbers. The same survey results combined with different estimates of the overall population, will result in changes in the absolute numbers of people employed. But will not affect the share of individual demographic groups employed or unemployed, or the distribution of employment across different types of employers. Guy Berger has a thorough discussion of these issues on his blog. I’ve reproduced one of his figures below.
For the establishment survey, the independent measure of the population size comes from Quarterly Census of Employment & Wages (QCEW), which generates counts of establishments and employees based on payments into the unemployment insurance systems which almost all private employers must participate in.
While the establishment survey does a good job capturing changes in the number of workers at a given establishment (along with changes in pay, hours, and so on), it cannot, by its nature, tell us anything about changes in the total number of establishments. That’s where the benchmark from the QCEW is needed, which — given the lags in collecting and assembling the data — comes some months after the initial BLS data is released. What this means is that revisions are likely to reflect changes in the numbers of establishments — births and deaths — rather than anything related to employment at individual businesses. The big downward revision for the first half of the year means that either more businesses closed than the BLS model had assumed, or fewer new businesses opened (or both). It’s hard for me to see how this could have any direct connection with the immigration crackdown.
In any case, we have plenty of better evidence that the labor market is weakening. Despite the stable headline unemployment rate over the past year, it is clear that the bargaining position of workers relative to employers is substantially worse than it was a year or two ago.
*
A central fact about the post-covid economy was the extremely tight labor market, which was very favorable to lower-wage and less privileged workers. Overshadowed by inflation, the very favorable conditions for workers aren’t a central part of our narrative of the economy during those years. (They also don’t fit a neat political narrative, since the strong post-covid labor market under Biden was in some ways a continuation of a strong pre-covid labor market under Trump I.) They mainly show up in the negative form of employers’ complaints.
But what looks like a labor shortage from one side looks like an abundance of jobs from the other. As Arin Dube and his coauthors pointed out in an important paper, the three years after 2020 reversed a full quarter of the rise in wage inequality of the previous four decades.
Yet this period was hardly perceived as an economic success story. The obvious reason is inflation: while wages did rise faster than prices at the bottom of the distribution, most of the compression described by Dube and his colleagues took the form of wages higher up the scale falling in real terms.
Another explanation, which I argued for here, is that while those at the bottom of the distribution did well in terms of wages, they experienced a major loss of non-wage income (and economic security more broadly) as the pandemic welfare state was withdrawn. (Stephen Semler recently added some new evidence supporting this view here.)
Another way to reconcile the big relative gains for those lower down the distribution, with the very negative perceptions of the economy: The combination of tight labor markets and rising prices was good specifically for people who sell their labor on a free market. It was bad for people who enjoy the protections of professional credentials, unions, public employment or other sources of stable employment conditions: In this case, protection from the short-term vagaries of the labor market was a negative. It’s worth noting in this context that almost all the outsized wage gains of the post-pandemic period went to people who switched jobs; it’s easier to jump ship for a better offer if you are a janitor or roofer or cashier, than if you are a teacher or lawyer.
Add to that the wave of tech layoffs in 2023 (as companies realized they’d over extrapolated from the pandemic shift to online activity) and you have a period in which the labor market was much worse for the sort of people whose voices are loudest in the public conversation than for those without big megaphones.
In any case, that period is over now. The great compression of wage income ended in late 2023; over the past two years, low-wage workers have lost ground relative to other wage-earners. The great reshuffling has ended too: Quits, new hires and job openings are all back down to pre-pandemic levels. This is still a reasonably strong labor market, but it is not a historically tight one, as it was a couple of years ago. There was a period when Jerome Powell would mention, at every press conference, the need to “rebalance” a labor market that was tilted too far in favor of workers, and against employers. He’s not using that language now, and with good reason.
Quits, in particular, are a strong indicator of workers’ bargaining power. On the one hand, people’s willingness to leave their current job is a vote of confidence in their ability to find another one (or reflects the fact that they already have); on the other hand, the threat of quitting is the main practical leverage that most workers have over their boss. The historically high rate of quits over 2021-2023 was an important sign of a labor market unusually friendly to workers; its subsequent decline suggests one that no longer is.
One way of thinking about the balance of bargaining power in the labor market is the Beveridge curve, which compares the share of jobs vacant with the share of workers unemployed. William Beveridge, who came up with this measure, considered a one to one ratio — shown as dotted line in the figure — a normal, appropriate balance, but there’s nothing sacred about that.
The figure nearby shows this metric for the US for the past 25 years. Points in the upper left correspond to a tight labor market — few unemployed workers and lots of unfilled positions, meaning workers have relatively more leverage. Points in the lower right are the opposite — lots of unemployed workers and few unfilled jobs, so that prospective employers can pick and choose and workers have to take what they can get. Points higher up on the right correspond to what’s known as structural unemployment — lots of people want to work, and there are businesses that want to hire, but for some reason they can’t connect with each other.
The pandemic, not surprisingly, was, by this metric, the outstanding recent period of structural unemployment. (There was also a persistent shift in the structural direction after the financial crisis, suggesting a lasting post-crisis mismatch between jobs and workers in addition to weak demand.) 2022, on the other hand, is way up in the upper left of the figure. With two openings for every unemployed person it was, arguably, the best time to be looking for a job in our lifetimes.
But job openings have been drifting down steadily, and unemployment has been creeping up. A couple of months ago, openings per unemployed person fell below one. There are questions about how reliable the vacancy measure is, especially when comparing over widely-separated periods; but by this metric, it’s a worse market for jobseekers not just than the exceptional recovery from the pandemic, but than the last couple pre-pandemic years as well.
It’s worth noting, as Powell did in his most recent press conference, that the balance in the labor market reflects both weaker demand for labor, and a shrinking (or at least more slowly growing) pool of available workers. The “curious balance” (in Powell’s words) between these two trends is why unemployment has held steady despite what now seems to have been a very sharp fall in employment growth over the past year. Weak employment growth leads to less slack in the labor market if the potential workforce is also growing more slowly.
Does this mean that anti-immigrant policies are working, at least as far as the labor market is concerned? No, it does not.
For a given level of demand, a smaller laborforce does mean a more favorable market for workers. But the condition “for a given level of demand” is key. If immigration is sharply down from 2024 levels, that means fewer people available for jobs. But it also means less spending. Immigrants are a source of demand as well as labor supply, a basic fact that is often forgotten in discussions of the economics of immigration. So while lower immigration may help explain why unemployment has held steady despite much slower hiring, that does not mean that unemployment would be higher in a world where 2024 levels of immigration had continued. In that hypothetical world, businesses would be selling more and hiring would be stronger.
One thing we can say for sure: There is no reason to think that the immigration crackdown has disproportionately benefits native-born workers. There has not been a surge in employment among the US born, despite some claims to the contrary from the administration. In an excellent post, Jed Kolko walks through the data on this, and explains the source of the confusion. The problem is similar to the one we discussed earlier around the revisions to the establishment survey. The BLS surveys households, but updates the total population only once a year, based on the census; in between it uses projections from the previous census. So if fewer people answering BLS questionnaires say they were born abroad — either because there really are fewer immigrants, or because immigrants are more wary of speaking to government officials — then, mechanically, the BLS must assume a greater native-born population. (Since, against he total population numbers are fixed until the next census update.) As Kolko convincingly shows, this statistical artifact explains the entire apparent surge in native employment.
An easy way to confirm this is to look at the employment-population ratio for native-born respondents. (These sorts of ratios are not affected by the population rebasing.) And if we do, we see that the employment-population ratios for US-born workers are lower, not higher, than they were a year ago.
Whatever is going on in the labor market, “native workers are benefiting from reduced competition from immigrants” isn’t it.