FRBSF Economic Letter
2002-35; November 22, 2002
Recent Trends in Unemployment Duration
The recession that began in early 2001 probably has ended, as national
output grew moderately during the first three quarters of 2002. Unemployment,
however, remains a problem. Between late 2000 and early 2002, the national
unemployment rate increased by about 2 percentage points, from 3.9% to
about 6%; this represents about 2.8 million additional individuals looking
for work. Thus far in 2002, payroll employment has been flat to down nationwide,
and the unemployment rate has stayed stubbornly close to 6%, raising the
specter of a "jobless recovery" from the 2001 recession. Persistent
labor market weakness implies that the amount of time spent unemployed
(unemployment duration) is likely to increase, which in turn has important
implications for household well-being.
In this Economic Letter, I discuss the concept of unemployment
duration, the various measures available, and the evidence regarding the
pattern of unemployment duration in the current cycle compared to past
cycles. Reliable measures of the expected length of unemployment spells
indicate that although duration increased more than expected in recent
months, it has not been especially long during the recent economic downturn.
Underlying this may be the improved labor market conditions of the 1990s
expansion, which acted to offset a long-term trend toward rising duration
Unemployment duration refers to the amount of time that an individual
remains unemployed. In the 1970s and 1980s, scholars and policymakers
debated whether the typical unemployment spell in the U.S. is best described
as "long" or "short." This distinction is critically
important for assessing the economic efficiency and equity aspects of
unemployment. The short view emphasized the dynamic nature of unemployment,
focusing on job turnover and implying that the pool of unemployed typically
is dominated by a large number of individuals who experience relatively
short spells of unemployment (a month or two at most). This view generally
is consistent with voluntary search activity by unemployed individuals
and employer reliance on temporary layoffs for cyclical employment adjustments.
In contrast, advocates of the long view argued that the pool of unemployed
typically is dominated by individuals who experience relatively long spells
of unemployment (three months or more) and are best described as "involuntarily"
unemployed, often through permanent job loss. Thus, the two views pose
the extremes of a well-functioning market for matching workers and employers
in which the burden of unemployment is widely dispersed and a situation
in which a relatively small number of workers bear the burden of a persistent
shortage of appropriate jobs.
Of course, the truth about unemployment lies somewhere in between the
extremes of short and long durations. No matter what duration structure
characterizes unemployment under typical labor market conditions, however,
the deterioration in labor market conditions that occurs during a recession
implies a cyclical increase in the incidence and share of long spells
of unemployment. As recessions persist, rising unemployment rates are
accompanied by rising unemployment durations. Although households can
rely on savings to tide them over during short spells of unemployment,
their ability to do so declines as unemployment spells lengthen. As such,
rising unemployment duration during economic downturns can have adverse
consequences for household spending and financial solvency and may act
to stifle the recovery. Policy responses to recessionary unemployment
therefore tend to focus on long spells—for example, in March 2002 Congress
extended unemployment insurance benefits from 26 weeks to 39 weeks (up
to 52 weeks in states with high unemployment rates). Such measures can
reduce the hardship associated with long spells but may have the adverse
side effect of lengthening unemployment by reducing the costs of job search.
A variety of different measures of unemployment duration at the national
level are available on a monthly basis, each based on data from the monthly
household survey administered by the U.S. Census Bureau and Bureau of
Labor Statistics (BLS). Figure 1 displays three of these measures for
the period January 1967 to October 2002, plotted along with the unemployment
rate as a standard indicator of the cyclical status of the labor market.
Each of these measures is corrected for the changes in survey design implemented
beginning in January 1994. These changes affected the measurement of unemployment
duration and other labor force variables. To make the series used below
consistent over time, I applied the adjustment factors described in Polivka
and Miller (1998).
Perhaps the most commonly cited measure of unemployment duration is "average
weeks unemployed," which is tabulated and released each month by
the BLS. It measures the average duration of unemployment spells sampled
while in progress, at the time of the survey. As seen in Figure 1, this
series exhibits pronounced cyclical swings, rising from about 11-12 weeks
at business cycle peaks to above 20 weeks shortly after the end of severe
Although it measures average duration for unemployed individuals at
a point in time, the average duration series does not represent the completed
duration of unemployment that a newly unemployed individual can expect
to face ("expected duration"). Because the average duration
is calculated based on all in-progress spells, it contains both an upward
and a downward bias with respect to the measurement of the expected duration
for a newly unemployed individual. The upward bias occurs because longer
spells, purely by virtue of their length, are more likely to be in the
monthly unemployment sample than are shorter spells. The downward bias
arises because the use of in-progress spells precludes measurement of
completed spell durations. For example, an employee laid off owing to
a plant closure might expect to be out of work for many weeks. However,
in the person's initial phase of unemployment, the household survey will
record an unemployment spell of just a few weeks. On average, the duration
is measured about halfway through the spell.
Although the two biases can in principle cancel out, in recessions the
upward bias tends to outweigh the downward bias, leading to an overstatement
of expected duration for a newly unemployed individual. This dominance
of the upward bias can be seen in part by examining the series representing
the percentage unemployed for at least 27 weeks, also displayed in Figure
1. This measure of long-term unemployment exhibits especially wide cyclical
swings, varying from around 10% during cyclical peaks to as high as 30%
during severe recessions. The sharp increase in the measured incidence
of long spells during recessions raises the level of the average duration
measure well above the expected duration for newly unemployed individuals.
Estimating the expected duration for newly unemployed individuals is
possible, however, through use of additional BLS unemployment duration
figures. Valletta (1998) demonstrated a simple method based on a transformation
of the percentage of all unemployed who have been unemployed fewer than
five weeks in the survey month (new monthly entrants to unemployment).
The cyclical properties of this constructed series are similar to those
for estimates of expected duration that are based on more detailed and
precise tabulations of individual duration experiences (see, e.g., Baker
1992). The simple estimate of expected duration, tabulated according to
the technique in Valletta (1998), is displayed in Figure 1. It exhibits
less pronounced cyclical swings than do the other duration measures, ranging
from a low of about 10 weeks to a high of about 19 weeks. The upward bias
in the average duration measure (as a measure of expected duration) can
be seen in this figure: average and expected duration are close when the
labor market is tight, but average duration substantially overestimates
expected duration during periods of high unemployment.
Is duration high or low in 2002?
Despite the persistence of the current slowdown, unemployment measures
suggest that it is not as severe as the preceding three major recessions
(1975-1976, 1982-1983, and 1991). The recent peak unemployment rate of
6.0% is well below the peaks reached in those earlier recessions. Consistent
with this, each of the duration measures has remained below its peak in
the three preceding recessions (Figure 1). For example, the expected duration
of unemployment hovered around 15 weeks from June to October of this year,
well below its peaks of about 18-19 weeks in the past two recessions.
Given the recent up-and-down movements in each of the three duration series,
the direction of their movement in coming months is uncertain. However,
the recent movement contrasts with a steady increase in preceding months,
which suggests that a turning point may have been reached and unemployment
duration will soon begin to decline. If so, the current slowdown in retrospect
probably will be regarded as mild in terms of its effects on unemployment
Another way to assess the severity of unemployment duration in the current
slowdown is in purely relative terms. In particular, given the relatively
low unemployment rate in the current recession compared to past recessions,
it is useful to ask whether unemployment duration is long relative to
the current unemployment rate and any long-term trends in unemployment
duration. If the unemployment pool has a relatively large share of individuals
with long spells, even with a low unemployment rate the uneven burden
of unemployment can hamper economic recovery.
Figure 2 addresses this issue by plotting the expected duration of unemployment
based on observed data, as defined above, against its predicted value.
To focus more clearly on the recent recessionary pattern, the figure displays
values for the period October 1989 to October 2002. The prediction model,
however, is based on data for the period January 1967 to May 1998 (the
same sample as used in Valletta 1998). Thus, the predictions for the period
June 1998 to October 2002 are "out of sample" forecasts based
on the pre-existing relationships between the variables in the prediction
model. This model incorporates only the unemployment rate and a trend
over time as explanatory factors, so that the differences between the
actual and forecast values represent a break from the pre-existing relationship
between current expected duration and the unemployment rate or the long-run
trend in expected duration. Measured independently of the unemployment
rate, the long-run trend in duration is upwards; as reported by Valletta
(1998), expected duration increased by about 17% (approximately 2 weeks)
between 1976 and 1998.
The figure indicates that actual expected duration was below its predicted
values during much of the late 1990s expansion and most of the recent
slowdown to date. The excess of the predicted values over the actual duration
indicates that unemployment duration was shorter than we would expect
based on the level of unemployment achieved and long-term upward trend
in duration. Thus, the economic expansion of the late 1990s may have produced
favorable labor market conditions (such as effective job matching) that
acted against a pre-existing trend toward longer unemployment spells.
On the other hand, since April of this year actual expected duration has
been running a bit above its predicted values, suggesting recent deterioration
in job prospects for unemployed workers. This may be only a temporary
development, or it may signal that recovery in the labor market will be
further delayed; only time will tell.
Baker, Michael. 1992. "Unemployment Duration: Compositional Effects
and Cyclical Variability." American Economic Review 82 (March) pp.
Polivka, Anne E., and Stephen M. Miller. 1998. "The
CPS after the Redesign: Refocusing the Economic Lens." In Labor Statistics
Measurement Issues, eds. John Haltiwanger, et al. Chicago: University
of Chicago Press.
Valletta, Robert G. 1998. "Changes
in the Structure and Duration of U.S. Unemployment, 1967-1998."
FRBSF Economic Review 3, pp. 29-40.