Employment & Productivity

When will it ever end, aka, potential subscribers take note.

The market is currently abuzz with news that Robert J. Gordon, longtime member of NBER's Business Cycle Dating Committee, has provided solid evidence supporting a tight link between the peak in initial claims and, dare we say it, the end of recessions.

Back on April 2, we sent this to our subscribers:

Here's a rare note of possible cheer: could the behavior of initial claims be signaling that an end to the labor market contraction is in the offing?

Graphed below is the yearly change in initial claims for unemployment insurance. Note that this measure typically peaks a little before the end of a recession. No surprise, perhaps. But what is a little surprising is how similar current behavior  looks to earlier end-of-recession patterns.

Claims-yty-5-28-09

And as the table below shows, the peak in claims growth occurs, on average, four months before the official NBER trough month, though the range is one to seven months. Note too that employment doesn't necessarily turn positive until a few months later, with the trough in yearly job growth typically coming five months after the peak in claims growth, and the trough in the employment level coming eight months after. In 1970s, yearly job growth hadn't picked up even a year after the peak in claims. But even then, the bottom was getting close.

Months to Trough in Employment Employment Growth, year-to-year
  peak in claims, yty cycle level growth claims peak year later swing
Dec 69-Nov 70     Apr 70 7 7 7 1.8% -0.4% -2.3%
Nov 73-Mar 75 Nov 74 4 5 7 0.4% -0.7% -1.2%
Jan 80-Jul 80 May 80 2 2 2 0.7% 1.0% +0.3%
Jul 81-Nov 82 May 82 6 7 5 -1.3% -0.5% +0.8%
Jul 90-Mar 91 Feb 91 1 3 3 -0.6% -0.4% +0.2%
Mar 01-Nov 01 Oct 01 1 22 4 -0.7% -0.7% +0.0%
average 4 8 5 0.1% -0.3% -0.4%

Since there's a good chance that the job market in this cycle will resemble the slow recovery after the 2001 recession, it could be 2011 before employment starts growing again. But if we think of this as a sort of leading indicator of a leading indicator, we could start seeing signs of the recession's end by mid-summer. We hope.

The Envelope Please: NBER Study finds ratio between Establishment and Household Surveys to the Cyclical

NBER has just released a working paper, "Exploring Differences in Employment between Household and Establishment Data," that presents research and analysis carried out by the Census Bureau and the BLS concerning the unusually large gap between the two major employment surveys that developed in the late 1990s. By matching individual Unemployment Insurance records with individual respondents in the Household Survey, the authors unearthed characteristics of the workers most likely to show up in one survey yet be missed in the other, and conclude that most of these workers are on the margins of the income and education spectra. For example, poor recent immigrants working under the table and highly educated consultants might both be missed by the Establishment Survey but included in the Household Survey.

But that example should not give Household Survey enthusiasts false hope. The study demonstrates that divergences between the Current Employment Survey (aka the Establishment Survey or Nonfarm Payroll) and the Current Population Survey (aka the Household Survey) are "cyclical phenomenon," with the CES outpacing the CPS during business-cycle expansions, and then falling back during recessions and the early stages of recoveries. The 60-year history of the ratio between the two surveys graphed below shows this clearly. (Take that, Kudlow & Co.) Also note that the ratio failed to rise during the most recent recovery, which seems to underscore the ongoing weakness in terms of employment growth.

Ratio_Est_HH

Based on characteristics of respondents discovered in their study, the authors contend that tight labor markets create a growing number of marginal jobs that often go unreported in the Household Survey, e.g., establishments hiring short-term workers to cover busy periods, which begins to lift the Establishment Survey. As economic conditions continue to improve, workers tend to drop informal jobs (which would be reported in the CPS but not the CES) for formal jobs (that would be reported in the CES), thus widening the gap between the two surveys. These trends then reverse as economic activity falls off, with establishments laying off workers who then turn to informal employment, moving from the CES to the CPS. The graph of the two series since 1994 directly below illustrates this process.

HH-vs-Est

The unusually large and long-lived gap between the two surveys began in 1998 as the Establishment Survey rose well above the Household Survey, and reversed in 2001, when the Household Survey remained stable as the Establishment Survey fell.  The relative pick-up in the Household Survey that got so much attention in the press was the unwinding of the prior trend and not the beginning of a new one. To put some numbers on it, in comparing UI and CPS data the authors found that jobs counted in the UI but not in the CPS grew by 2.3 million between 1996 and 2001, while jobs counted by the CPS and not by the UI grew by just 600K. Between 2001 and 2003, jobs counted by the CPS but not the UI grew by 800K, while jobs counted in the UI but not the CPS fell by 500K.

 So next time any of us, understandably, seeks solace in the Household Survey’s strength when the Payroll Survey disappoints, we need to remember that NBER researchers, who are responsible for officially calling recessions, have determined that such relative strength is indicative of a weakening economy.

NBER Working Paper No. 14805, issued in March 2009, "Exploring Differences in Employment between Household and Establishment Data," by Katharine G. Abraham, John C. Haltiwanger, Kristin Sandusky and James Spletzer, available here (subscription required): http://www.nber.org/papers/w14805.pdf

Calculating the Unemployment Rate

Recently several news pieces have made the claim that if the unemployment rate were calculated as it was during the Great Depression, the current rate would be close to double what it is, and creeping toward the formidable rates back in the 1930s.

Unempl-1929-2009

The first problem with this statement is that there was no official unemployment rate until the 1940s. The ones we use today were reconstructed after the fact. As unemployment ballooned during the Great Depression a number of ad hoc attempts were made to calculate the rate, and the widely divergent results led private researchers and some state and local governments to experiment with various sampling methods. In 1940 the WPA began publishing statistics on those working (the employed), those looking for work (the unemployed), and those doing something else (hiding under the bed perhaps?) and so not in the equation.*

The second problem with the statement is that it's just not true. Although the BLS has refined their surveys and made questions more specific, conceptually the unemployment formulas have not changed, and the BLS's own analysis of test data shows that the impacts of several sets of changes on the overall numbers are minor.

In 1962 high unemployment and two recessions in three years led to the formation of The Presidential Committee to Appraise Employment and Unemployment Statistics, led by Robert Gordon, and tasked with reassessing the concepts used in gathering labor-market data. The Committee gave high marks to the BLS's integrity and suggested some improvements. For several years the BLS tested new survey techniques before instituting a number of changes in 1967.

Among the most important of these were the requirement that workers must have actively sought employment in the last four weeks in order to be classified as unemployed.  A contact at BLS agrees that some discouraged workers were probably counted as unemployed before this change was made, but the effect of this migration is small. As they generally do, the BLS ran the new definitions alongside the old, in this case for 2.5 years, before adopting the new.  Although the test series is not entirely comparable with the new series, the overall unemployment rate in the new series dropped by just one-tenth of a percent and, within that, the rate for adult men was down 3/10th, up 4/10th for adult women, and off a full point for teenagers.  (Maybe they were just being teenagers: the requirement that they give a concrete example of their job search may well have reminded them of their parents and got the blank stare.) The Committee also recognized the need for more detailed data on persons outside the labor force, who are highly sensitive to changes in labor demand, and the BLS began collecting information on those who wanted a job although they were not looking for work in 1967.

In 1976, in order to provide more information on the hidden unemployed (who would presumably be part of the labor force in a full-employment scenario), the BLS first published the original U1 to U7 tables, which break out marginally attached workers.  These tables were revised in the 1994 redesign (becoming U1 to U6) and the controversial requirement that discouraged workers must have sought work in the prior year was added. This change halved the number of discouraged workers, resulting in a complete break in the time series.

But those workers can still be found in the U-6 series, which is the broadest measure of labor underutilization, and it ain't a pretty sight. Up 4.8% over the year, U-6 currently includes an ugly 13.5% of the labor force. Update: In February U-6 unemployment rose to 14.8%. There's no need to fool around with the official unemployment rate (U3) to get an accurate picture of how quickly our labor market has deteriorated: the U1 to U6 tables tell the story.

Update 03/14/2009
In response to a reader's comment:

There are three unemployment series available for the early 1930s: Stanley Lebergott’s, Michael Darby’s reworking of the Lebergott series, and the G.H. Moore series, available through NBER. (Michael Darby is the economist who pointed out that the Lebergott series included those on work-relief as unemployed. His series moves them to employed.) We used Moore’s series, which pretty much splits the difference between the other two. When you combine different series, usually necessary for long-term views, the series breaks themselves produce spikes or dips. Splicing the Darby series to the official BLS data makes it look like the unemployment rate jumped in 1940, which we did not want, and Lebergott’s inclusion of those on work relief as unemployed was in line with 1940 census practice.

Here are the yearly averages for the three series:

  Moore Lebergott Darby
1929   3.2% 3.2%
1930   8.7% 8.7%
1931   15.9% 15.3%
1932   23.6% 22.9%
1933 23.4% 24.9% 20.6%
1934 19.1% 21.7% 16.0%
1935 17.6% 20.1% 14.2%
1936 14.2% 16.9% 9.9%
1937 12.2% 14.3% 9.1%
1938 18.4% 19.0% 12.5%
1939 16.3% 17.2% 11.3%
1940   14.6% 9.5%

Basically, if you want to evaluate the effect of government work programs, compare the Lebergott series to the Darby series. If you want a more readable trend line (while avoiding accusations of playing politics) use the Moore series.

For more information and some notes on definitions, please see “Employment and Unemployment in the 1930s,” by Vanderbilt economist Robert A. Margo, available here: http://fraser.stlouisfed.org/docs/MeltzerPDFs/maremp93.pdf

Philippa Dunne and Doug Henwood

*There is currently a bit of a fracas over the reconstructed unemployment rates for the period prior to official series. Stylish Stanley Lebergott, the BLS economist who put together the most widely used series, categorized workers on emergency relief as unemployed. In the 1980s data reclassifying these workers as employed were released, a definition in line with current practice and more widely accepted. In the past month or so, those wishing to show the WPA programs did little to alleviate unemployment have been relying on the unrevised Lebergott series, and those taking the opposite view the revised data. Of course, if you compare the two series it appears that between 1934 and 1941 WPA projects took 2 to 3.5 million workers off the unemployment roles, and shaved the rate by 4 to 7 percentage points.


Comments on April Employment

Originally published May 14, 2008

Though headline job losses were smaller than expected in April, the details of the report were weaker than the first impression would encourage. And even though the household survey was stronger than its payroll counterpart, as advertised by the decline in the unemployment rate, a look under its surface also uncovers weakness.

Total employment was off by 20,000, though plus signs were very hard to come by as you scan the sectoral breakdown. Construction fell by 61,000, about evenly divided between residential and nonres; manufacturing shed 46,000, almost all of it in durables. Private services gained 81,000, but many major sectors declined. Wholesale trade was off 11,000; retail, -27,000; information, -2,000. Finance, somewhat mysteriously, gained 3,000. The biggest gains were in health care, up 37,000; bars and restaurants, +18,000; computer systems design, +10,000; and administrative and support services, +13,000. The last sector got no help from its temporary help component, which fell 9,000. Government added 9,000, thanks to an unusually large gain of 4,000 in federal employment. State and local employment was up just 5,000, about a quarter its average over the last year. Budgetary pressures may finally be taking their toll on public employment.

Does it fizzle? Does it sizzle

No. It just lays there.

Originally published March 6, 2008

Surely all memory of the 1950s send-up of the original “kerplunk goes the tablet that gives the fizz” ad has long faded, but its apt description of the American job market is recently won.

In recent years our job market has indeed lost much of its fizz. This point is best made by the BLS’s quarterly business employment dynamics (BED) series. It comes out with a long delay—figures for the second quarter of 2007 were only released on February 14. But the BED series is very useful in analyzing longer-term trends.

Net changes in total employment over time are a function of gross job losses and gains. For example, in 2006, there was a net gain of 1.7 million jobs in the private sector, according to the BED program. (This number differs from the establishment survey.) But that net gain came from a gross gain of 30.8 million jobs, and gross losses of 29.1 million jobs. That’s quite a furious pace of turnover under a rather placid surface.