The signal the unemployment rate provides can change a lot over time: EPI Macroeconomics Newsletter
In 2019 the unemployment rate was below 4% for the second straight year, the first time this has happened since 1968 and 1969. Despite the current stretch of low unemployment, by many other measures the labor market does not seem particularly tight. Most obviously, wage growth has been accelerating a bit, but is still disappointing relative to what wage growth we would expect at this level of unemployment.
Productivity growth has firmed up slightly in recent years, but employers still aren’t acting like labor costs are something they’re particularly worried about containing through investments in capital equipment or better processes.
|The late 1990s is an obvious reference for highlighting how unresponsive wage and productivity growth have been to low unemployment in recent years. In these years, low unemployment coincided with notable accelerations in both wage and productivity growth. In this newsletter, we highlight some reasons why the headline unemployment rate measured in the late 1990s does not provide quite the expected apples-to-apples comparison with the unemployment rate of today. Key findings are:
The rest of this brief highlights evidence on these three points.
A lower unemployment rate is needed to signify labor market tightness with an older and better-educated workforce
All else equal, workers with more experience and education credentials have lower rates of unemployment. The economic intuition for this is that more experienced and more educated workers have skills that are in greater demand by employers at any given level of economy-wide slack. This demand premium for more experienced workers holds in the aggregate despite the fact that age discrimination afflicts many workers, i.e., the unemployment/age gradient is clearly downward sloping.
Lower unemployment among more experienced and educated workers means that a given unemployment rate (say 4%) achieved in two different years can signify different things about the labor market if the composition of the workforce has changed. An unemployment rate of 4% might signal a moderate degree of slack for a highly educated and more experienced workforce, but may signal a very tight labor market for a workforce that is younger and with fewer credentials. Figure A shows the actual unemployment rate and the composition-adjusted unemployment rate for two time periods: 1997–2000 and 2016–2019. Both periods saw unemployment below 5%. In the first period, the difference between actual and composition-adjusted unemployment is trivial (essentially by construction—we fix the demographic composition of the workforce at its 1995 level, as described in the note to the figure). By the 2016–2019 period, the composition-adjusted unemployment rate is nearly 0.3 percentage points higher. In essence, after controlling for age and education, the unemployment rate today has to be roughly 0.3 percentage points lower to signify the same level of labor market slack as it did during the late 1990s recovery. We also adjusted unemployment by race, ethnicity, and gender (not shown in the figure), but this changed the composition-adjusted unemployment rates only trivially compared with the effects of age and experience.
Unemployment rates, actual and adjusted, 1997–2000 and 2016–2019
The data underlying the figure.
Notes: The adjusted unemployment rate is calculated by holding age and five categories of educational attainment (less than high school, high school diploma only, some college attendance, four-year college completion, and advanced degree) at their 1995 levels and recalculating unemployment going forward, using the reweighting procedure in John DiNardo, Nicole M. Fortin, and Thomas Lemieux, “Labor Market Institutions and the Distribution of Wages, 1973–1992: A Semiparametric Approach,” Econometrica 64, no. 5, 1996.
Source: Author’s analysis of Current Population Survey Basic microdata.
Nonresponse to the survey used to calculate unemployment is growing and not random
The monthly unemployment rate is calculated from the Current Population Survey (CPS), a survey of roughly 50,000 households. Household nonresponse to the CPS—people refusing to fill out the survey—has grown in recent decades. In their 2006 Center for Economic and Policy Research report, John Schmitt and Dean Baker compared the employment rate from the decennial census (which has near-universal response) and the monthly CPS for 2000, drawing on information from “a unique dataset that matched respondents to the February, March, April, and May CPS for 2000 with their Decennial Census forms” to correct for incomparability in reported employment rates between the CPS and the census. In 2000, household nonresponse to the CPS was roughly 10%, and Schmitt and Baker estimated that bias in nonresponse likely understated employment rates by roughly 1.4 percentage points compared with the near-universal census. Since 2000, nonresponse has risen by another 7–8 percentage points. This suggests significant scope for rising nonresponse since 2000 to pull down measured rates of unemployment.
More recently, a 2019 paper by Hie Joo Ahn of the Federal Reserve Board and James D. Hamilton of the University of California, San Diego, finds that a number of “inconsistencies” in CPS tabulations of unemployment—many likely driven by nonresponse—lead to understatements in the unemployment rate (and in labor force participation). Figure B shows the extent of understatements in the unemployment rate in the early 2000s relative to today. According to Ahn and Hamilton, these understatements have increased between 2001 and 2018 by roughly 0.3 percentage points (moving from 1.6 to 1.9 percentage points). Again, all else equal, the unemployment rate today likely needs to be adjusted 0.3 percentage points upward to be more usefully compared with measured unemployment rates in the late 1990s.
Estimated percentage-point understatement of CPS-derived unemployment rate, 2001–2002 and 2017–2018
|Estimated percentage point understatement|
|July 2001–September 2002||1.6|
|January 2017–March 2018||1.9|
The data underlying the figure.
Notes: The data start in July 2001 and end in March 2018. The chart compares the first 15 months of data with the last 15 months. The adjustments include all inconsistencies identified by Ahn and Hamilton (see source below), not just nonresponse.
Source: Hie Joo Ahn and James D. Hamilton, Measuring Labor-Force Participation and the Incidence and Duration of Unemployment, March 16, 2019, revised December 13, 2019, supplemental data
|The ever-more-fuzzy line between unemployed and not in the labor force
To be classified as “unemployed,” adults must be both jobless and actively searching for work. Those who are jobless and not actively searching—even if they would like a job—are classified as “not in the labor force.” For economists assessing the state of labor market slack, it would be more convenient if the line between unemployed and not in the labor force was bright and stable over time. For example, if adults labeling themselves “not in the labor force” were quite adamant that they were not going to work in the near future—either because of responsibilities (like child or elder care) or out of simple preference—then they would be solidly outside the scope of “unemployment.”
In the immediate post–World War II period, this line likely was pretty bright and unchanging for a relatively privileged subset of middle-class two-parent families: Men were expected to always be working or looking for work, while women were not expected to be part of the paid labor force. Over time, however, the line has gotten ever fuzzier. For example, newly employed workers who transition into employment from being out of the labor force in the previous month have always constituted a majority of the newly employed, but their share has reached historic highs in recent years. Between 1997 and 2000, this share averaged 64.4%, but between 2016 and 2019, it averaged 71.3%. Figure C shows the changing share of newly employed workers transitioning from out of the labor force in the last three decades.
Share of newly employed workers who transitioned into employment from being out of the labor force in the previous month, Dec. 1990–Dec. 2019
The data underlying the figure.
Note: Data in the figure represent a 12-month rolling average.
Source: U.S. Bureau of Labor Statistics, “Labor Force Flows Employed to Unemployed: 16 Years and Over” and “Labor Force Flows Not in Labor Force to Employed: 16 Years and Over,” Labor Force Status Flows from the Current Population Survey, retrieved January 2020 from Federal Reserve Bank of St. Louis Economic Data (FRED)
Further, in their 2014 working paper, David Blanchflower and Adam Posen of the Peterson Institute for International Economics have found that the downward pressure on wages stemming from adults not in the labor force (but in effect part of the pool of workers competing for jobs) has increased significantly after 2002 relative to the 10 years before.
Both of these measures—the quantity-side measure of how many new jobs are filled by those from out of the labor force and assessments of how much downward wage pressure is exerted by those not in the labor force—indicate that workers out of the labor force are more substitutable for the unemployed than they were in the past. In turn, this means that compared with previous historical periods, any given unemployment rate today obscures the slack embodied in adults not in the labor force who are nevertheless quite likely to be tomorrow’s potential workers.
Be careful when comparing unemployment rates over time
It is by now well recognized that the relationship between the unemployment rate and wage growth is different in recent years than it has been in the past. Part of this divergence is likely due to genuine structural changes in the economy, particularly the degraded bargaining power and leverage of typical workers due to policy choices. But some of the divergence is likely simply because a given unemployment rate is not providing the same signal about labor market tightness over time due to compositional changes in the workforce and difficulties in deriving an aggregate unemployment rate from a household survey. It is worth noting that many of these issues that blur signals from the unemployment rate will apply to other quantity-side measures of labor market slack, like the prime-age employment rate. Finally, the line between the officially unemployed versus potential workers currently classified as not in the labor force has become ever muddier over time.
All of this implies that policymakers—particularly those at the Fed—must continue to be extremely modest in how well they think they can forecast coming wage and price pressure stemming from unemployment. Instead, they should let the data speak and worry about wage and price pressure when it comes—and not rely on historical relationships between unemployment, wages, and prices to make policy going forward.