Method Revisions for the Missing Workers Indicator

Hardcore fans of EPI’s labor market indicators will notice a change today. Our estimate of the number of “missing workers”—potential workers who are no longer classified as in the labor force but who will likely be working or looking for work if the labor market improvement continues—has been revised.

Our earlier estimates were built in large part upon projections for labor force growth contained in a paper published by the Bureau of Labor Statistics in 2007. These projections examined labor force participation rates for age-specific groups of both men and women between 1986 and 1996 and between 1996 and 2006. The paper then projected age- and gender-specific labor force participation rates for 2016.

We used these projected rates to see what labor force participation “should” be in each month between 2006 and 2016, and interpreted shortfalls between the actual participation rate and these projections as how much participation was depressed due to economic weakness, as opposed to structural changes in the labor force, like the retirement of baby boomers. We chose to look at pre-2008 projections precisely because we wanted these projections to be free of any cyclical drag imposed by the Great Recession.

But looking again at these projections recently, we noticed some slightly worrying features. For one, the labor force participation rate for men 25-34 fell significantly in both the 1986-1996 and 1996-2006 periods, yet was projected to rise substantially between 2006 and 2016. Further, the unemployment rate in 1986 was 7.0, the unemployment rate in 1996 was 5.4, and the unemployment rate in 2006 was 4.6 percent. This means that the trends estimated in the BLS projections may be buoyed up by cyclical effects. The BLS projections made no attempt to parse trends in participation rates that were driven by long-run trends versus cyclical weakness in the economy.

In this re-examination, we want to be conservative in our “missing workers” estimate, thereby creating what we see as a lower bound on the number of missing workers. To that end, we decided to make sure that our projected labor force participation rates are largely free of cyclical effects. We take the trends from 1989, 1996, and 2004—years when the overall unemployment rate was 5.3, 5.4, and 5.5 percent, respectively. This should largely purge the trends of cyclical impacts. In fact, because the unemployment rate is rising (albeit slowly) in each successive end point, our missing worker trends are likely slightly weighed down by cyclical weakness.

We weight the 1989-1996 annualized trend change in age- and gender-specific labor force participation rates by one-fifth, and the 1996-2004 trends by four-fifths. We then project these trends forward and interpret them as the structural trend in labor force participation, as before. And also as before we interpret shortfalls between actual participation rates and these structural trends as an estimate of “missing workers” – potential workers who we think are likely to return to the labor force as the recovery strengthens.

The resulting number for the missing workers is 3.33 million. This is down substantially from our earlier estimate of 6.2 million. However, we should be very clear just how conservative the resulting number from our new methodology is. We’re essentially projecting the participation rates one should expect at 5.5 percent unemployment given structural trends. But participation rates that would prevail should the unemployment rate reach 4.5 percent are likely to be higher than when unemployment rates are 5.5 percent. And there is no evidence that 4.5 percent unemployment is likely to lead to economy-wide overheating or is in any way unsustainable on its face.

In future blog posts, we’ll show what the missing workers number would be with some slightly different assumptions about which years to use as end points for measuring trends. It turns out that they are quite sensitive to these decisions, and any estimate of missing workers will come with some substantial uncertainty. Further, extrapolating linear trends is always wrong in some sense. For example, prime-age male labor force participation declined between 1989 and 1996 and again between 1996 and 2006. Hence it declines a lot in our projection through 2016. But a simple linear extrapolation means that we’re implicitly saying one day that prime-age males won’t participate in the labor force at all, which is obviously absurd. The same problem holds in reverse for women. Prime-age female labor force participation rose in both periods before 2004. So we forecast a slight rise by 2016. Of course, we don’t think that one day prime-age female labor force participation is going to be 100 percent. That said, we do think it’s at least worth examining how actual data looks when held up against pre-existing trends.

It’s important to note that our new missing workers methodology also affects our estimate of the jobs gap and the number of months before we return to pre-recession labor market health. Reductions in these measures do not indicate that substantial changes have happened recently in the labor market, but rather changes in our estimation technique.

That this uncertainty cannot be avoided when trying to measure labor market slack very much reinforces the strategy we’ve urged policymakers to adopt—complement quantity measures of labor market slack (like “missing workers”) with price measures like growth in nominal wages. Nominal wage measures indicate substantial labor market slack still exists. This would imply that either there are a substantial number of missing workers (that is, that labor force participation remains cyclically depressed) or the natural rate of unemployment (the unemployment rate consistent with stable rates of inflation) is much lower than conventional estimates would imply, or a combination of both.

Finally, as a policy matter, our revised method for calculating the number of “missing workers” doesn’t imply that we think the problem of excess labor market slack is noticeably smaller this month than last. The combined weight of evidence on the quantity and price side signals pretty clearly that substantial slack remains, and there is no reason for policymakers to seek the slow the pace of recovery anytime soon.