Some thoughts on the latest U.S. jobs numbers

The gap between the household and payroll surveys, the weird collape in dentist productivity, and more.

Thanks for joining me! Normally, pieces like this will be exclusively for paid subscribers. But this is my first day and I want to make sure people can get a sense of what I’m offering. The paywall goes up next week.

While there are lots of interesting nuggets in Friday’s jobs numbers, they don’t all add up to a single coherent picture. So instead of trying to write a unified narrative of everything, I’ll be focusing on what I think is interesting—and what you hopefully haven’t seen elsewhere.

The first question is: were the latest numbers good or bad?

From one perspective, more jobs were added in June (850,000 on a seasonally-adjusted basis) than in any month since August 2020 (1.35 million excluding Census hiring). But from another point of view, June (-18,000) was the first month of job losses since April 2020.

The truth is probably somewhere in between. And considering that there are probably around 9 million missing jobs, the current pace of progress is probably too slow to get back to full employment before the end of next year at the earliest.

The Bureau of Labor Statistics runs two surveys to assess the job market. The establishment survey asks businesses how many people they paid in the pay period that includes the 15th of the month. The household survey asks individuals whether they have jobs and, if not, whether they are looking for one.

In general, these surveys tend to tell similar stories about what’s going on in the economy, but they can diverge on a month-to-month basis by large amounts.

For one thing, the two measures of employment have different definitions. If someone is working two formal jobs she will be counted twice in the payroll survey and once in the household survey. But if she’s self-employed (like me) or working on a farm, she’ll be counted once in the household survey and zero times in the payroll survey.

The household survey also has a much smaller sample size than the payroll survey, which means it’s often more volatile on a month-to-month basis. And then there are the revisions. Every year the payroll survey gets benchmarked against tax data. Older household data isn’t revised to be internally consistent over time, which is why there are often weird jumps in population estimates at the start of each year.

So what’s happening now?

The payroll survey is saying that job growth is incredibly strong, while the household survey shows the first outright decline in employment in over a year.

There are two pandemic-specific factors that seem to playing a big role right now: the past misclassification of furloughed workers and the seasonality of employment at public and private schools.

Since 2000, the BLS has been tracking the number of Americans who say they are “nonagricultural wage and salary workers with unpaid absences”. Normally there is a reliable seasonal pattern caused by summer vacations. Starting in 2003, the BLS began publishing a series on the number of Americans who are “with a job, not at work”. This includes a lot more people because it counts workers taking paid vacations, paid family leave, and sick days, but it still follows the same seasonal pattern.

I confess that I didn’t know about either of these series until the pandemic. But then it turned out that millions of people who were being furloughed—or staying home to avoid catching or spreading the virus—were wrongly self-identifying as being employed with an unpaid absence. The BLS first pointed this out last March. As they put it:

There was also a large increase in the number of workers who were classified as employed but absent from work. Special instructions sent to household survey interviewers just before data collection started for March called for all employed persons absent from work due to coronavirus-related business closures to be classified as unemployed on temporary layoff.

However, it is apparent that not all such workers were so classified. Such a misclassification is an example of nonsampling error and can occur when respondents misunderstand questions or interviewers record answers incorrectly. If the workers who were recorded as employed but absent from work due to “other reasons” (over and above the number absent for other reasons in a typical March) had been classified as unemployed on temporary layoff, the overall unemployment rate would have been almost 1 percentage point higher than reported.

However, according to usual practice, the data from the household survey are accepted as recorded. To maintain data integrity, no ad hoc actions are taken to reclassify survey responses.

By April 2020, there were about 6.3 million Americans claiming to have jobs even though they weren’t showing up to work and weren’t getting paid, up from an average of about 1.7 million in the preceding Aprils. Similarly, there were 11.5 million people employed but not at work in April 2020 compared to a pre-pandemic average of about 4.5 million.

Things have improved a lot since then, but even as recently as April and May 2021 there were about 500,000 American workers who were probably misclassified as employed when they should have been listed as unemployed on temporary layoff.

Things changed last month. There were just under 2.4 million workers with an unpaid absence in June 2021, compared to the 2000-2019 average of 2.5 million and the 2017-2019 average of 2.6 million. The absolute drop in the number of absent workers (paid or unpaid) compared to the prior average for Junes was even bigger, although basically the same in percentage terms.

If you correct for the changes in the number of misclassifed workers, the gap between the June changes in the payroll and household surveys shrinks dramatically. Instead of falling by 18,000, June household employment rose by about 620,000. As it happens, that’s really close to what the BLS estimated when they tried to match the household survey data to the payroll definition of employment (604,000 jobs added in June).

That explains a lot, but 620,000 is still a bit lower than 850,000. That’s where seasonality comes into play. Public and private schools have been hit hard by the pandemic, with about 1.5 million job losses between February and May 2020 on a seasonally-adjusted basis. Things have been picking up more recently, but it’s been hard to tell because of the timing of the initial losses. After all, students aren’t around over the winter or summer. Schools normally cut employment sharply between April and July, with smaller cuts between November and January.

The seasonal-adjustment algorithm corrects for this pattern in normal times, but it could be overstating the recent improvement. On a seasonally-adjusted basis, payroll employment at public and private schools rose by 268,000 jobs in June. But in raw terms employment fell by 607,000 jobs.

It’s entirely plausible that schools are recovering at a brisk pace, and I don’t want to imply that the seasonally-adjusted numbers are inherently wrong. But it happens to be the case that the changes in payroll employment excluding public and private education have been much better fits with the household survey data over the past few months.

Here’s what happens if you compare the monthly changes in the payroll survey to the monthly changes in the household survey, along with the various modifications I mentioned above:

So, on to the next thing!

What is happening to labor productivity at dentists’ offices and restaurants?

In absolute terms, the biggest and most persistent job losses were at restaurants, hotels, bars, and live entertainment. Employment in leisure and hospitality is still down about 2.2 million compared to February 2020, with bars and restaurants accounting for 1.3 million of that shortfall.

In percentage terms, however, the layoffs at dentists’ offices in March and April 2020 were even bigger, with payroll employment down by 56% at the trough. But unlike restaurants, museums, hotels, movie theaters, and concert halls, dentists’ offices were quick to rehire. In fact, employment at dentists passed the pre-pandemic peak back in February 2021 and is still rising at a decent clip.

While I have nothing against dentists, I don’t understand how this works. Most people could safely enjoy restaurants even before vaccines thanks to delivery, takeout, and outdoor dining. But being stuck inside a waiting room before having someone spend a lot of time in your mouth seems like a risky activity during a respiratory pandemic. Anecdotally, I know plenty of people (including me) who have spent at least as much money at restaurants since the pandemic began than before. I don’t know anyone who has spent more time and money at the dentist.

And, in fact, the underlying detail tables of the National Income and Product Accounts confirm my suspicions. Inflation-adjusted consumer spending at dentists’ offices in May (the latest month for which we have data) was 16% below the level in February 2020 on a seasonally-adjusted basis. By contrast, inflation-adjusted spending at bars, restaurants, and other eateries is higher than before the pandemic, with bigger gains if you focus only on workers at sit-down restaurants and fast food establishments.

In other words, there is a large and growing disconnect between the number of people working in these two labor-intensive industries and the real volume of business that they’re pulling in. Food services and drinking places are, in the aggregate, generating about 13.5% more real output per worker than before the pandemic, while workers at dentists’ offices are about 17.5% less productive.

(Note that job losses in March 2020 in the chart above are understated because the BLS surveys take place in the first half of each month, whereas the BEA spending data cover the entire month.)

I have no good explanation for this chart.1 If anyone has any ideas, please leave a comment or email me mck[at] And speaking of restaurants…

What’s going on with average hourly earnings of leisure and hospitality workers?

Since January, average hourly pay for nonsupervisory workers in the leisure and hospitality industry has grown at a 24% annual rate, rising from $14.81 in January to $16.21 in June. That’s the most dramatic pay hike on record. Before the pandemic, the fastest annualized pay rise over a 5-month period was 14%—and that was in 1968.

Understandably, this has some people worried about incipient inflation or difficulties faced by business owners. But it’s helpful to have some context. From the BLS handbook of methods (my emphasis added):

Average hourly earnings are on a “gross” basis. They reflect not only changes in basic hourly and incentive wage rates, but also such variable factors as premium pay for overtime and late-shift work and changes in output of workers paid on an incentive plan.

And then there’s this (my emphasis added):

Data measure usual hourly and weekly earnings of wage and salary workers. All self-employed persons are excluded, regardless of whether their businesses are incorporated. Data represent earnings before taxes and other deductions and include any overtime pay, commissions, or tips usually received.

If a lot of people working at bars, hotels, and restaurants stopped getting tips because of the pandemic—and then started getting tips again as vaccinated consumers came back—a chart of average hourly earnings relative to the pre-pandemic trend would probably look at a lot like this:

Put another way, wages don’t look high compared to what would have been expected if there hadn’t been a pandemic. That may seem odd given how many workers are still jobless, but it makes more sense given how strongly restaurant revenues have recovered during the reopening.

It’s possible that there’s an alternate universe where restaurant sales and employment are up even more—and wages up by less—but it’s not obviously a better outcome than what we have now.


Why is growth in the rest of the economy so slow?

Outside of the rebounding education, personal care, and leisure and hospitality sectors, job growth has been anemic, with payroll gains averaging just 48,000 over the past three months. At the same time, employment in the majority of the economy remains about 3% below the pre-pandemic peak.

Construction and temporary help employment, which both tend to be sensitive to the state of the business cycle, have both been essentially flat for six months, while the broad health care sector has been flat for almost a year. Motor vehicle manufacturers shed 12,000 jobs in June. Employment in the industry is still down 11% compared to February 2020, which is all the more remarkable given the sustained strength in consumer demand. Local governments other than school districts have cut about 50,000 jobs since April.

It’s too early to say that this is something to worry about, but it’s definitely something to keep an eye on. We’ll get more insight into the job market on Tuesday when the BLS publishes the Job Openings and Labor Turnovery Survey (JOLTS) for May.

Hope everyone has a great weekend!



The growth of takeout over the past 15 months means that restaurants can produce and sell more meals with fewer workers, which would show up as a productivity increase in the data. Delivery is different, because the reduced need for waiters is at least partly offset by rising demand for couriers. But couriers aren’t generally counted as employees of the restaurants.

Moreover, none of this explains why the gap between employment and inflation-adjusted sales—a gap that was basically zero as recently as February—has ballooned at precisely the time when mass vaccination has brought back indoor dining across the country.

And that still leaves the open question about dentists.