The robust US recovery has expanded prime-age employment and broadened the labor supply. But wage growth in recent months has stopped accelerating across different measures. I show that the plateauing in prime-age wage growth is not due to direct compositional effects, but due to softer rises among the continuously-employed. In fact, while compositional effects on prime-age wages are generally negative, they have been somewhat less so in the last few years. But this does not rule out other ways in which shifts in employment or the labor supply may be weighing down on wages.
More than a decade after the official end of the Great Recession, the US labor market is still expanding.
Employment among Americans 25–54 years old — “prime-age” workers — has risen by about 665,000 over the last year. That’s a healthy clip, far higher than the 257,000 necessary to keep prime-age employment on pace with population growth.
What’s more, as employment rises, nonemployment is falling fastest among the ranks of Americans long thought to be the least-attached to the labor force. That 665,000 rise in prime-age employment was associated 1:1 with a 665,000 fall in nonemployment, of which 205,000 came from the disabled, representing almost a third of the nonemployment fall.
The continued good news on the labor market seems at first glance to stand in stark contrast with recent wage developments. Average hourly earnings for all private nonfarm workers (regardless of age) over the last three months were 3.1% higher than a year earlier. That’s a passable outcome but actually slower than in the recent past; just last August wage growth was a bit higher at 3.4%. This weakening acceleration in wage growth in the face of above-trend labor supply growth, coupled with the lack of inflation pressure, is strong evidence that the US is not at full employment yet.
One possibility is that these two phenomena are in fact directly related: that at this late stage in the recovery, new workers are coming into the types of jobs that pay below-average wages, and in doing so they pull down the overall average, depressing wage growth. Economists call these compositional effects.
Notice how compositional effects might give a skewed read of labor market health. An economy that brings in low-wage or marginally-attached workers is likely one that is doing well and has strong momentum. But these very workers might lower the measured average wage and make the labor market look a bit worse on its face.
Compositional effects are a real issue to be aware of in labor market data. But a problem with blaming them for the most recent wage sogginess is that alternative measures of wage growth that adjust for compositional effects — such as the Atlanta Fed Wage Growth Tracker and the Employment Cost Index— have also slowed down.
We can see this too for prime-age workers specifically. The Current Population Survey (CPS) allows us to hone in on wages by age group. I decompose the 12-month growth in average wages for prime-age workers into two effects: wage growth among those who were continuously-employed over the year on the one hand, and the drag or boost that came from inflows into and outflows from employment (that is, compositional effects) on the other. 
The summary results below show that the recent deacceleration in wage growth appears to be driven not by new workers pulling down the average, but by existing workers seeing slower raises. In fact, while ongoing compositional wage effects remain negative, they have been gradually shrinking over the last seven years.
Moreover, I can go a step further and break down these ongoing compositional effects by nonemployment reason: that is, the net effect on wages from job finders from a nonemployment margin and job leavers into that same margin. The chart below takes the employment inflow/outflow effect shown above and decomposes it by these nonemployment margins.
What this shows is first that disability flows typically exert positive pressure on prime-age wage growth. That’s because while both workers who leave employment for disability and workers who find jobs out of disability tend to make below-average wages, the job finders tend to make higher wages than the job leavers did, creating a positive swap for wage growth.
More broadly, the softer ongoing compositional effects on wage growth appear to be due largely to evaporating pressure from the long-term unemployed. This makes sense: I’ve defined each of these nonemployment effects to be net of both flows into and out of employment. But unique among these categories, the net flow effect from the long-term unemployed is almost entirely from job finders, since virtually all job leavers who go into unemployment become short-term unemployed by definition. So as the number of long-term unemployed Americans keeps falling, there are fewer and fewer job finders from their ranks — again, by definition — and therefore they exert smaller and smaller downward wage pressure.
But all of this doesn’t mean there still aren’t direct or indirect compositional effects on wages. As more and more marginally-attached workers find jobs, they will gradually shift the population of continuously-employed workers over time, which may affect that component of wage growth. However, the Employment Cost Index shown earlier ought to control for many of these effects as well, and it too is showing signs of slow down, though it is calculated from a survey with a smaller sample size than the CPS or the payroll survey.
There may also be a labor supply effect on wages where, as nonparticipants rejoin the labor force, there is less pressure on firms to raise wages. Wages could therefore only rise in fits and starts as labor supply expansions have a dampening effect on wage growth.
But in the meantime, to the extent job finders and leavers are affecting average wages, it appears to be a somewhat lower drag than normal. The recent wage deceleration therefore is likely driven by something else.
 This analysis uses population and employment estimates from the Current Population Survey (CPS) throughout. Head-count level figures use year-on-year changes in 12-month moving averages; I adjust the underlying population data for the annual Census population controls using rolling regressions so that the entire series is on a January 2020 Census population estimate basis.
 In this analysis I classify nonemployed Americans by their self-described labor market status in the CPS. Workers I call “discouraged” are those who are not actively looking for work but say they want a job, regardless of their self-described reason for nonparticipation. All other categories of nonparticipants are among those who say they do not want a job.
 Strictly speaking, a net fall in a nonemployment margin does not necessarily mean that those Americans transitioned directly into employment. For example, some of those who left disability status in the CPS may have become unemployed, or given a different reason for nonparticipation, and meanwhile other Americans could be transitioning into disability status. But in this example too, ultimately a fall in the disabled population indicates that outflows must be exceeding inflows over time, and a rise in employment indicates that job finding must be exceeding job leaving.
 To do so, I follow the broad contours of the methodology laid out in Daly & Hobijn (2016). That is, I create a series of longitudinally-linked panels of CPS workers across 12 months, treat wage growth among the continuously-employed as my benchmark, and then calculate the net effect on overall wage growth from employment inflows and outflows. Unlike Daly & Hobijn who decompose movements in median weekly wages, however, I analyze (log) average hourly wages. This makes the decomposition far less complex and volatile.
For hourly workers, I use their hourly wage as reported in the CPS. For salaried workers, I calculate the hourly wage as the usual weekly wage divided by usual hours. I impute hours for those whose usual hours vary. Per the Economic Policy Institute’s methodology, I impute top-coded wages assuming weekly wages are Pareto-distributed above the 80th percentile, and I remove outlier real wages above $100/hour and below $0.50/hour in 1989 dollars.
 There are a very small number of CPS flows from employment to long-term (that is, greater than 27 weeks) unemployment. Economists generally regard these flows as specious.