Full text of Monthly Labor Review : April 2021

View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

5/3/2021

To work from home or not to work from home? That is the question : Monthly Labor Review: U.S. Bureau of Labor Statistics

U.S. BUREAU OF LABOR STATISTICS
Bureau of Labor Statistics

Publications

Monthly Labor Review

HOME

RELATED CONTENT

Related Articles
Employment trends in retail trade, 2010–19, Beyond the Numbers, January 2021.
Employment growth accelerates in 2018, extending a lengthy expansion, Monthly Labor Review, May 2019 2005.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

Productivity measures for retail trade: data and issues, Monthly Labor Review, July 2005.

Related Subjects
Firm size

Employment

Expansions

Economic and Social Statistics

Recession

Industry studies

April 2021

The U.S. productivity slowdown: an economy-wide
and industry-level analysis
Labor productivity—defined as output per labor hour—has
grown at a below-average rate since 2005, representing a
dramatic reversal of the above-average growth of the late
1990s and early 2000s. The productivity slowdown during
these years has left many economic observers wondering
why this situation has occurred and what factors may have
contributed. To clarify potential sources of the productivity
slowdown, this article presents an analysis of labor
productivity and its component series—multifactor
productivity, contribution of capital intensity, and
contribution of labor composition—at both the economywide and industry levels, complemented with a survey of
Shawn Sprague
sprague.shawn@bls.gov

the contemporary productivity literature.
The figure—$10.9 trillion—represents the cumulative loss in
output in the U.S. nonfarm business sector due to the labor
productivity slowdown since 2005, also corresponding to a

Shawn Sprague is an economist in the Office of
Productivity and Technology, U.S. Bureau of
Labor Statistics.

loss of $95,000 in output per worker.[1]
These figures show that, when there is consistently belowaverage productivity growth, year after year, a substantial
effect can result over an extended period. How could this situation have occurred, in a modern and technically
advanced economy such as in the United States? Well, not only has the productivity slowdown been one of the
most consequential economic phenomena of the last two decades, but it also represents the most profound
economic mystery during this time, and though many economists have grappled with the issue for over a decade
and even created some innovative research approaches to address the question, we still cannot fully explain what
brought on this situation.
One of the more perplexing aspects of the current slowdown is its genesis: that it came immediately following a
historic productivity boom in the United States, and represented a swift rebuke of the popular idea of that time that
we had entered a new era of heightened technological progress. The suddenness and size of the reversal were
difficult to comprehend. For some background, in the late 1990s, when that much-cited productivity boom had
begun, U.S. labor productivity growth had accelerated to rates of change that had not been seen since the late
1960s and early 1970s. This late 1990s surge surprised many economic observers, who had become accustomed
to the below-average productivity growth rates of the mid-to-late 1970s through the early 1990s. In addition, the

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

situation in the United States was even more startling due to the fact that the rest of the more-developed
economies of the world were not similarly experiencing a speedup in growth rates.[2]
A debate ensued among economists: Was the tremendous productivity growth of the late 1990s here to stay—a
fundamental change generated by the computing and internet-related innovations that were all around us—or was
it a temporary phenomenon that would pass? The fact that the productivity speedup persisted through the
recession of 2001, and then became even more pronounced in 2002, convinced many observers that perhaps
something had changed.[3] The acceleration of U.S. productivity growth is shown in figure 1, illustrated by the
growth rates during 1998 through 2005, which rise above the long-term average rate since 1947, denoted by the
dashed blue line. Over these high-growth years, U.S. labor productivity grew at an average rate of 3.3 percent,[4]
which is markedly higher than the cumulative 2.1-percent average rate from 1947 to 2018.

This high-growth period came to an end during the mid-2000s, when U.S. labor productivity growth rates began to
stumble, and in 2006 receded below the long-term average trend line for the first time in a decade. And,
notwithstanding 2 years of high growth in 2009 and 2010 following the Great Recession, productivity growth rates
have remained stubbornly low in subsequent years. Many economic observers were yet again surprised, in this
case at just how drastically growth rates slowed, given the recently observed high rates of growth and the
continued technological innovations that were proliferating throughout the economy. In the years since 2005, labor
productivity has grown at an average annual rate of just 1.3 percent, which is lower than the 2.1-percent long-term

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

average rate from 1947 to 2018. The slow growth observed since 2010 has been even more striking: labor
productivity grew just 0.8 percent from 2010 to 2018.
As the slowdown in labor productivity growth has steadily held on throughout the past decade, economic observers
have been trying to understand this phenomenon, which has the effect of placing downward pressure on economic
growth, worker compensation gains, profits growth, and gains in living standards of Americans. Many observers
began to wonder: Why has U.S. labor productivity growth been so consistently low in recent years, and why is it so
markedly different from the strong growth observed relatively recently? This article presents two approaches to
address these questions, with each approach including an analysis of the U.S. Bureau of Labor Statistics (BLS)
productivity data and a review of the contemporary productivity literature. First, the economy-wide slowdown in
labor productivity growth is analyzed by breaking out the series into its three component series: multifactor
productivity (MFP) growth, the contribution of capital intensity, and the contribution of labor composition. Second,
industry-level productivity data are used to identify the industries that made notable contributions to the economywide labor productivity slowdown.

Economy-wide analysis of the U.S. labor productivity slowdown
This section presents an analysis of the economy-wide slowdown in labor productivity growth that decomposes the
series into its three component series: multifactor productivity growth, the contribution of capital intensity, and the
contribution of labor composition. This section also presents a dollar and time cost analysis of the slowdown, and
an analysis of how U.S. regions impacted the economy-wide slowdown.

Decomposition of labor productivity growth
Labor productivity is a measure of economic performance that compares the amount of goods and services
produced (output) with the number of labor hours used in producing those goods and services. It is defined
mathematically as output per hour of work, and growth occurs when output increases faster than labor hours. For
example, if output is rising by 3 percent and hours are rising by 2 percent, then labor productivity is growing by 1
percent.
Labor productivity growth is vitally important to present and future prospects for economic growth, because it
represents the only path by which economic growth can rise above what would be possible by simply increasing
labor hours (as, by definition, economic growth can only come from either hours growth or labor productivity
growth). The economic gains brought about by labor productivity growth make it possible for an economy to
achieve higher growth in labor income,[5] profits and capital gains of businesses, and public sector revenue; these
economic gains also hold the potential to lead to improved living standards for those participating in an economy,
in the form of higher income, greater leisure time, or a mixture of both. In addition, as labor productivity rises, all of
these factors may increase simultaneously, without gains in one coming at the cost of one of the others.
Given the importance of labor productivity growth, it is worth delving into the measure in more detail to see what
underlying factors are making this growth possible. As such, in addition to labor productivity growth being defined
as a residual—the difference between output growth and hours growth—we can also analyze it as a sum, built up
from the contributions of its three component series.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

Components of labor productivity growth
The following equation allows us to quantify the sources of labor productivity growth and helps us better
understand the measure by looking into its three component series:
labor productivity growth = multifactor productivity growth + contribution of capital intensity + contribution of
labor composition
Multifactor productivity (MFP) growth represents the portion of output growth that is not accounted for by
the growth of capital and labor inputs and is due to contributions of other inputs, such as technological
advances in production, the introduction of a more streamlined industrial organization, relative shifts of
inputs from low to high productivity industries, increased efforts of the workforce, and improvements in
managerial efficiency. Similar to labor productivity growth, MFP growth can also be defined as a residual—
output growth minus the growth of the combined inputs of labor and capital.
The contribution of capital intensity is defined as the capital-weighted change in the capital-labor ratio.
The measure is computed as capital’s share of current dollar costs multiplied by the growth in capital
services per labor hour. The contribution of capital intensity—also called capital deepening—reflects
businesses’ decision-making process between hiring more workers and purchasing more or higher-quality
equipment, or of substituting equipment for workers or vice versa.* In cases in which firms increase their
usage of capital relative to labor, or where capital costs rise relative to labor costs, there will be an increase
in the contribution of capital intensity to labor productivity growth.
The contribution of labor composition is defined as the labor-weighted change in a measure—labor
composition—which reflects shifts in the level of skills and experience of the workforce. It is computed as
labor’s share of current dollar costs multiplied by labor composition.** The contribution of labor composition
helps us gauge the productive capacity of the workforce at a given point in time. When firms hire more
workers with higher skills and more experience or lay off workers with lower skills or less experience, or
when labor costs rise relative to capital costs, the contribution of labor composition to labor productivity
growth increases.
* Specifically, the contribution of capital intensity is defined as
, where wk is the
2-year average cost share of capital and Kt and Lt are capital services and labor hours at a given time t.
** The BLS labor composition methodology can be found at https://www.bls.gov/mfp/mprlabor.htm.

The three components of labor productivity growth are displayed in figure 2 for the slowdown period (2005–18), the
speedup period (1997–2005), as well as other selected post-World War II (WWII) periods and the long-term
historical average.[6] Labor productivity growth, corresponding to the purple dots, represents the sum of the three
stacked bars of MFP growth,[7] contribution of capital intensity, and contribution of labor composition. It is apparent
that the labor productivity growth rate (1.4 percent) of the slowdown period has slackened relative to the rate of the
4

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

speedup period and is also below the long-term historical average. Furthermore, we can see from the diminished
red and dark-blue stacked bars of the slowdown period relative to the bars of the speedup period that MFP growth
and the contribution of capital intensity are the sources of the U.S. labor productivity slowdown. (The contribution
of labor composition was approximately the same as that of the speedup period and did not contribute to the
slowdown.[8]) MFP grew 0.4 percent during the slowdown period, which is less than one-fourth the growth of the
speedup period and is also well below the long-term historical average. In addition, the contribution of capital
intensity in the slowdown period, 0.7 percent, is around half that of the speedup period and is also below the longterm historical average.

The deceleration in MFP growth—the largest contributor to the slowdown—explains 65 percent of the slowdown
relative to the speedup period; it also explains 79 percent of the sluggishness relative to the long-term historical
average rate. The massive deceleration in MFP growth is also emblematic of a broader phenomenon shown in
figure 2. We can see that throughout the historical period since WWII, the majority of the variation in labor
productivity growth from one period to the next was from underlying variation in MFP growth, rather than from the
other two components. While the contribution of labor composition varied only between a range of 0.1 percent to
0.3 percent during the entire post-WWII era and the contribution of capital intensity varied between 0.7 percent and
1.3 percent, MFP growth varied within a wider range, between 0.0 percent and 2.0 percent.[9]

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

At the same time, in addition to the notable variation in MFP growth during the recent periods, something
unprecedented about these recent periods was the additional contribution from variation in the contribution of
capital intensity. The contribution of capital intensity had previously remained within a relatively small range (0.7
percent to 1.0 percent) during the first five decades of post-WWII periods, but then in the 1997–2005 period, the
measure nearly doubled, from 0.7 percent up to 1.3 percent, followed by nearly halving to 0.7 percent in the 2005–
18 period. This unprecedented variation in the contribution of capital intensity was the factor that combined with
the variation in MFP growth to bring about such historic speedup and slowdown periods in recent years, increasing
the size of the overall labor productivity slowdown to rival the widely noted 1970s slowdown. The contribution of
capital intensity accounts for 34 percent of the labor productivity slowdown relative to the speedup period and
explains 25 percent of the sluggishness relative to the long-term historical average rate.

The slowdown in MFP growth
Now let us take a deeper look into the two contributors to the labor productivity slowdown. For MFP growth, let us
start out by noting the inherent difficulty in attempting to quantify the sources, components, or causes of MFP
growth or lack thereof. This difficulty arises because MFP growth itself cannot be measured or identified on its own
but can only be ascertained as the leftover output growth that remains after all measurable inputs to production—in
this case, labor and capital[10]—have already been taken into account. With MFP growth, we are actually
measuring something that is unidentifiable, similar to how cosmologists can measure the extent of dark matter and
its influence on the universe even as they do not know what this matter comprises.[11] This is the reason that the
question of what is driving the slowdown in MFP growth has puzzled so many economic observers in recent years
and still remains incompletely explained following more than a decade of work on the issue.
However, there are a few approaches that can be taken to help us gain a foothold on what might be happening to
MFP growth, both by using BLS data as well as by looking at some clever approaches of the numerous
researchers working on this issue. As a first step in our analysis, let us look at the BLS series used in calculating
MFP growth, in order to provide some background and context on the economy in which the MFP slowdown took
place. As just noted, MFP growth is a residual: output growth minus the growth of the combined inputs of capital
and labor. Figure 3 reveals that, although both output growth and combined input growth atrophied in the
slowdown period relative to their higher rates of the speedup period, output growth succumbed to a much more
serious retrenchment. Namely, while combined input growth slowed by 0.4 percentage point, output growth slowed
4 times as much, by 1.6 percentage points. The fact that output growth retreated so much further than combined
inputs during the slowdown period is reflected in the notably low MFP growth rate of 0.4 percent during this period,
and is a key fact connected with the productivity slowdown. For this reason, as we begin our investigation of the
MFP slowdown, we will first be looking closely at the historically weak output growth and getting a sense of the
state of the economy during this period.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

Historically weak output growth of the post-2005 slowdown period
The rate of output growth during the 2005–18 slowdown period (2.1 percent) is a historically weak growth rate. Not
only does this rate pale in comparison to the 3.7-percent growth of the speedup period, but it also represents a
historically slow rate for the entire post-WWII period, well below the historical average growth rate of 3.4 percent
(see figure 3). Of course, a large portion of the below-average output growth in the slowdown period reflects the
fact that this period encompasses the global financial crisis and Great Recession of 2007–09 and the subsequent
recovery. It might surprise some to discover that the post-2007 business cycle, which contains this historic
downturn, not only had slower cyclical growth than all previous business cycles since WWII, but it even recorded a
slower overall growth rate than the Great Depression of the 1930s (see figure 4). In the case of the Great
Depression, output plummeted by 26 percent—a much more severe decline than the 3-percent decline during the
Great Recession.[12] However, the recent cycle exhibited a much weaker recovery than that of the Great
Depression, with output growth from 2009 to 2018 being less than one-third of the 7.2-percent rate posted during
the peacetime recovery from 1933 to 1940.[13] Because of this extended weak recovery, as of 2018, the
post-2007 cycle’s growth rate came in slightly below what had occurred during the 1930s—even more striking
when one considers that the population growth rate was the same in these two periods.[14]

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

The historically low output growth of the post-2007 business cycle—and particularly the anemic recovery—was not
wholly unexpected. Valerie Cerra and Sweta Chaman Saxena, Carmen M. Reinhart and Kenneth S. Rogoff, and
Carmen M. Reinhart and Vincent R. Reinhart show that a permanent loss of output and a lack of rebound to the
long-term growth trend often follow financial crises such as the one the United States experienced in 2008.[15]
Similarly, the International Monetary Fund (IMF) asserts that output losses arising from banking crises are usually
substantial, stating that “typically, output does not recover to its precrisis trend. On average, output falls steadily
below its precrisis trend until the third year after the crisis and does not rebound thereafter,” although the IMF also
clarifies that following this permanent output loss, “medium-term growth rates tend to eventually return to the
precrisis rate.”[16] Daisuke Ikeda and Takushi Kurozumi offer a story that may underlie this phenomenon,
suggesting that an “adverse financial shock tightens firms’ financing and thereby dampens their activities, which in
turn has a significant negative impact on the economy as a whole by decreasing activities not only on the demand
side but also on the supply side of the economy. The effect on the supply side, such as the sectors of research and
development (R&D) and technology adoption, induces a persistent decline in [MFP] and thus can cause a
permanent decline in output relative to a pre-shock balanced growth path.”[17]
James H. Stock and Mark W. Watson look even further back in time, to before the financial crisis, claiming that
more than half of the weakness of the recovery is from slower long-term trend growth—due to changing
demographics—that was already apparent before the Great Recession.[18] However, they also note that
particularly slow government spending (specifically from the phaseout of the American Recovery and

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

Reinvestment Act and the budget sequestration of 2013, as well as from slow state and local government hiring
during the entire recovery) and faltering international demand following the recession also played a role. Ray C.
Fair also cites sluggish government spending following the Great Recession, asserting that it was the central factor
underlying the weak recovery.[19]
In addition to researchers citing weak fiscal stimulus, other researchers have pointed to limitations on monetary
stimulus. Robert E. Hall offers that the zero lower bound on interest rates following the financial crisis presented a
limiting factor that had not been operative in prior U.S. recessions.[20] Robert J. Barro also claims that monetary
stimulus was insufficient to spur a vigorous recovery.[21]
The drag on MFP growth from the Great Recession
Now that we have an understanding of the recent trends in the measures used to compute MFP growth (in
particular the unusually slow output growth of recent years), our first task is to attempt to use this basic information
to help us better understand the slowdown in MFP growth. Specifically, we may ask: Could the state of the
economy during the slowdown period, as indicated by the atypically low output growth, especially with it running
below its potential and capacity during and following the Great Recession, have helped to cause the low MFP
growth?
Yes, according to several authors, the weakened state of the economy during much of the 2005–18 slowdown
period could be one factor underlying the low MFP growth. As noted earlier, Ikeda and Kurozumi argue that when
an economy is operating below its potential, firms may pull back on investment in R&D and new technology.[22]
David M. Byrne, Stephen D. Oliner, and Daniel E. Sichel concur, noting that one possible explanation for the
productivity slowdown is that “the economy has taken a long time to recover from the financial crisis and Great
Recession, as the repair of balance sheets has proceeded slowly and as uncertainty about the pace of the
recovery has held back investment.”[23] Romain Duval, Gee Hee Hong, and Yannick Timmer agree, postulating
that “the combination of pre-existing firm-level financial fragilities and tightening credit conditions made an
important contribution to the post-crisis productivity slowdown,”[24] and also that “while most forms of physical
capital can be pledged as collateral to get a loan, intangible assets such as R&D or workforce training cannot.
Furthermore, investments in intangible assets tend to translate more slowly into sales and to be riskier. Therefore,
[their] hypothesis is that credit-constrained firms cut their investment in intangible assets, contributing in part to a
sharper productivity slowdown after the crisis.”[25]
Waning dynamism: reduced responsiveness to productivity gains at the firm level
At the same time, although the Great Recession and its aftermath have substantially affected recent economic
trends, the data clearly show that the productivity slowdown started before the global financial crisis and Great
Recession.[26] Looking back at figure 1, we can see that labor productivity grew at a successively lower rate in
each consecutive year from 2002 through 2006, descending to well below the long-term average trend by 2006.
So, a second question emerges: What might have led to this initial slowing and commencement of the slowdown
period, or, more broadly, what factors might have contributed to the productivity slowdown of the entire 2005–18
period, other than the Great Recession and its deficient recovery?

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

One major finding is that the businesses that have been spurring recent innovations are having difficulty
expanding, and thus, their innovations are failing to make a bigger impact on the economy as a whole than would
otherwise be the case. Ryan A. Decker, John C. Haltiwanger, Ron S. Jarmin, and Javier Miranda show that,
despite the broad slowdown in productivity growth, many firms are actually still seeing strong productivity gains,
with innovation having continued to occur at the “productivity frontier” during the slowdown period since the early
2000s, among the firms that are the productivity leaders in their respective industries.[27] The authors reveal this
indirectly, by observing that productivity dispersion in the United States has expanded in recent years, which
means that a wider gap exists between these leading firms and the laggards. The authors claim that, within the
framework of Michael Gort and Steven Klepper, this increased productivity dispersion implies that there has not
been a declining pace of innovation.[28]
So then, why are the innovations that have been sparking at these higher-productivity firms not translating into
solid economy-wide productivity gains? The answer has to do with how these firms have responded to their
productivity windfall. Namely, Decker et al. observe decreased responsiveness to these firm-level productivity
bursts as a potential source of the aggregate productivity slowdown, as evidenced by falling rates of job
reallocation among these firms.[29] In other words, many of the firms that have been innovating have not similarly
been able to scale up and hire more employees commensurate with their improved productivity.
This slump in firm-level reallocation coincides with the timeline of the aggregate productivity slowdown, with
Decker et al. finding that “reallocation has declined in all sectors—particularly the high-tech sector—since the early
2000s.”[30] In terms of quantifying the impact of this phenomenon, the authors observe that “counterfactual
exercises imply that the decline in responsiveness yields a significant drag on aggregate (industry-level)
productivity, as much as 2 log points in high-tech manufacturing and more than 5 log points economy-wide in
recent years.”[31] (Note that we will further explore in depth the slowdown in high-tech manufacturing in the
“Industry-level analysis of the U.S. labor productivity slowdown” section of this article, particularly the dramatic
slowdown in the computer and electronic products industry.) As for the underlying sources, which may be resulting
in these lower rates of reallocation, Decker et al. cite several potential factors: rising adjustment costs,
globalization, increased regulation, and declining competition.
The factor of declining competition—as potentially having a stultifying effect upon productivity-enhancing job
reallocation—has been of particular interest in the literature, with Jan De Loecker, Jan Eeckhout, and Gabriel
Unger outlining this phenomenon in the United States, offering as evidence that average markup costs have nearly
tripled—from 21 percent as of 1980 to a level of 61 percent in 2016—in addition to the rate of profits expanding by
8 times its size, from 1 percent to 8 percent.[32] The authors claim that “because passthrough [of cost savings
from productivity growth to lower prices for consumers] is lower in the presence of higher market power, the rise in
market power will give rise to [a] lower degree of adjustment of the variable inputs, including labor, for the same
[productivity] shock process. The rise in market power thus can rationalize the decrease in labor reallocation
across firms, even if the observed shocks to firm productivity [have] remained constant.”[33]
Gustavo Grullon, Yelena Larkin, and Roni Michaely echo De Loecker et al.’s data and analysis on increasing
market power, stating that such findings “are robust to the inclusion of private firms and factors that account for
foreign competition, as well as the use of alternative measures of concentration. Overall, [their] findings suggest
that the nature of US product markets has undergone a structural shift that has weakened competition.”[34]

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

Furthermore, undergirding these findings regarding expanded market power are findings of not only rising
concentration across firms but also rising concentration in ownership across firms, with a few large shareholders in
multiple companies in a given industry.[35] Other related findings include relaxed antitrust enforcement, increased
mergers and acquisitions, and other restraints on competition, including increases in occupational licensing by
states, the growth of land use restrictions, a greater scope of intellectual property law, and increases in lobbying
and political rent seeking.[36]
Additionally, it might be of interest that, in a slightly different analysis of the widening productivity dispersion of
high-growth and low-growth firms found by Decker et al., Dan Andrews, Chiara Criscuolo, and Peter N. Gal have
proposed that “stalling technological diffusion” may be a possible source of this widening productivity dispersion,
theorizing that low-growth firms may be having a difficult time integrating new technologies.[37] However, Decker
et al. caution and clarify that “while the diffusion hypothesis could play a role, [their] estimates of [MFP] persistence
suggest that the group of ‘frontier firms’ is sufficiently fluid to somewhat limit the diffusion story’s explanatory
power. Increased adjustment frictions is an alternative, but not mutually exclusive, explanation.”[38]
Income inequality
Escalating income inequality may also be a factor underlying the productivity slowdown, according to Jason
Furman and Peter Orszag.[39] Namely, though low productivity growth may be leading to rising inequality, it may
also be that rising inequality is reducing productivity growth, by stifling “the ability to harness the talents of potential
innovators across the income spectrum.” The authors caution, however, that “any plausible magnitude for such an
effect would fall well short of explaining the 1.0- to 1.5-percentage-point drop in productivity growth.”[40] Furman
and Orszag further qualify that rising income inequality and low productivity growth may both “have a common
cause, namely that reduced competition and reduced dynamism—in part caused by specific policy changes—have
contributed to both issues.”[41] Some empirical support for Furman and Orszag’s hypothesis that income
inequality may be a potential factor in sluggish productivity growth has been offered by Ruchir Agarwal and Patrick
Gaulé, who examine income disparities on a global scale and observe that “talented individuals born in low- or
middle-income countries are systematically less likely to become knowledge producers.”[42]
The debate over innovation possibilities in the 21st century
At this point, it should be noted that a weighty qualifier exists with regard to all the foregoing material regarding the
MFP slowdown, on the basis of a notably different perspective on the recent data taken by several economists,
such as Robert J. Gordon and John G. Fernald.[43] These authors have hypothesized that a productivity
slowdown has not occurred per se in recent years. Rather, they contend that a productivity reversion to the “new
normal” of lower productivity growth, established in the early 1970s, has occurred.
More specifically, these economists assert that the information technology (IT)-based innovations of recent
decades are no match for the world-changing impacts of widespread electricity, the internal combustion engine,
and indoor plumbing that emerged in the late 1800s and early 1900s. They claim that productivity growth cannot
be expected to sustainably continue on the same high-growth trend that previously had been seen as of the
mid-20th century. Furthermore, they regard the productivity speedup of the late 1990s and early 2000s as the true

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

outlier and the subsequent low productivity growth as merely the expected case in this relatively lower innovation
era.
One underlying rationale for this potential story is provided by Joseph A. Tainter.[44] This author offers that, in
general, as complexity in a society increases following initial waves of innovation, further innovations become
increasingly costly because of diminishing returns. As a result, productivity growth eventually succumbs and
recedes below its once torrid pace: “As easier questions are resolved, science moves inevitably to more complex
research areas and to larger, costlier organizations,” clarifying that “exponential growth in the size and costliness of
science, in fact, is necessary simply to maintain a constant rate of progress.” Nicholas Bloom, Charles I. Jones,
John Van Reenen, and Michael Webb offer supporting evidence for this view regarding the United States,
asserting that given that the number of researchers has risen exponentially over the last century—increasing by 23
times since 1930—it is apparent that producing innovations has become substantially more costly during this
period.[45]
However, at the same time, some other economists have a few qualifiers of their own regarding the hypothesis that
we have long been in an essentially low-productivity era. Chad Syverson reminds us that productivity slowdowns
did occur between the waves of innovation during the late 1800s and early 1900s, as seminal technologies such
as electricity and the internal combustion engine emerged.[46] Ana Paula Cusolito and William F. Maloney add that
“while this prior diffusion hardly implies that a second IT wave is imminent, it does show that productivity
accelerations from general-purpose technologies do not have to be one-off events. Just because their resultant
productivity growth sped up in the late 1990s and early 2000s does not mean it cannot speed up again.”[47]
To sum up this section, from an economy-wide perspective, we can identify several plausible explanations for the
slowdown in MFP growth, including
• declining rates of productivity-enhancing job reallocation,
• rising market power and industry concentration,
• greater restraints on competition,
• growing income inequality,
• the drag from the global financial crisis and Great Recession and its weak recovery,
• diminishing returns to innovation relative to that of the late-19th and early to mid-20th centuries, and
• a historically wavelike tendency of innovations.
It appears likely that the MFP slowdown is coalescing from a combination of these factors, though it may not be
possible to place them into an integrated framework or decomposition, given that they address the issue of the
MFP slowdown from widely different perspectives—some cyclical, some noncyclical, and some more qualitative in
nature than quantitative—and as there may be other factors, which have not yet been discovered. However, these
factors do provide us with an overall sense of what may be undergirding the slowdown.
It should also be noted that the story of the MFP slowdown does not end here. In the “Industry-level analysis of the
U.S. labor productivity slowdown” section of this article, we will analyze the slowdown from an industry-level
perspective, investigating the
• large negative contribution from the computer and electronic products industry,

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

• large negative contributions from the retail and wholesale trade industries, and

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

• small negative contributions from most other industries.

The slowdown in the contribution of capital intensity
Alongside the slowdown in MFP growth is the other contributor to the labor productivity slowdown: the contribution
of capital intensity. This measure has exhibited an unprecedented variation and, from 2005 to 2018, was cut by
nearly half relative to the 1997–2005 speedup period. As noted previously, the contribution of capital intensity grew
just 0.7 percent during the 2005–18 slowdown period, which is lower than the 1997–2005 speedup period (1.3
percent) and the historical average rate (0.9 percent). As was also noted, the measure accounts for 34 percent of
the labor productivity slowdown relative to that of the speedup period, explains 25 percent of the sluggishness
relative to the long-term average percentage rate, and had the lowest growth among all the selected post-WWII
periods (see figure 2).[48]
Given that the contribution of capital intensity is calculated as the difference in growth rates between capital and
labor inputs, multiplied by the capital cost share, we can determine how much each of these three underlying
factors contributed to its slowdown. As shown in figure 5, capital services grew during the slowdown period at a
rate of 2.5 percent, which is well below both its rate for the speedup period (4.5 percent) and its long-term average
(3.9 percent). Labor hours grew at a rate of 0.7 percent, which lies between its rate for the speedup period (0.4
percent) and its long-term average (1.2 percent). Capital’s cost share was 38 percent during the slowdown period,
which is higher than during the speedup period (33 percent) and the long-term average rate (34 percent).

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

Thus, we can say that it was the combination of a large deceleration in capital services growth with a slight
acceleration in labor hours growth that drove down the change in the capital-labor ratio in the slowdown period.
This dual effect overwhelmed a slight increase in the capital cost share and diminished the contribution of capital
intensity to 0.6 percentage point below what it had been in the speedup period.
At the same time, relative to its long-term trend rate, the sluggishness in the contribution of capital intensity in the
slowdown period was comparatively modest—just 0.2 percentage point below its long-term trend rate (see figure
2). Furthermore, the contribution of capital intensity in the slowdown period was not as much of an outlier as it had
been during the speedup period, in which it was 0.4 percentage point above the long-term trend during these highgrowth years. Nonetheless, both periods exhibited rates that were outside the norm. So, a question arises: What
led to such a dramatic acceleration in the contribution of capital intensity and then led to a similarly sized
deceleration not just back to normal during the slowdown period, but to slightly below the norm?
To answer the first part of this question, we must look back at the capital and labor components of the contribution
of capital intensity during the 1981–97 pre-speedup period (see figure 5). What this reveals is that the vast majority
of the acceleration in the contribution of capital intensity from the pre-speedup period to the speedup period was
not due to capital services growth expanding but to labor hours growth shrinking. While capital services growth
sped up slightly, from 4.2 percent to 4.5 percent, labor hours slowed substantially, from 1.8 percent to 0.4 percent.
This drop in labor hours growth is not altogether surprising, given that the speedup period contained both the
recession of 2001 and the “jobless recovery” of the early 2000s.
So, it can be said that the change in the capital-to-labor ratio—and thereby the contribution of capital intensity—
during the speedup period was boosted up to such a high degree from unusually low labor hours growth, with
capital services growth lying not far outside the norm. In contrast, for the slowdown period, it was the inverse: most
of the slowing in the measure came from unusually low capital services growth, with labor hours growth not far
outside the norm.
We can identify the contributions toward the recent slowdown in capital services growth, which are found in the far
right stacked bar in figure 6. As previously noted, capital services growth slowed from 4.5 percent to 2.5 percent
during the slowdown period. Of this 2.0-percentage-point deceleration, 0.8 percentage point was from a massive
slowdown in computer IT equipment, which shrunk from providing a contribution of 1.0 percentage point in the
speedup period to just making a 0.2-percentage-point contribution in the slowdown period. The other two notable
contributors to the slowdown were from non-IT equipment (0.5 percentage point) and intellectual property products
(0.4 percentage point). In addition, 0.1-percentage-point contributions were from rental residential capital,
structures, communication IT equipment, and inventories.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

So, what factors may have undergirded such below-average capital services growth during the slowdown period?
As noted earlier, Byrne, Oliner, and Sichel assert that the fragile financial condition of the economy following the
Great Recession may have hindered investment during the recovery.[49] Robert E. Hall agrees, stating that “at the
end of 2013 [the capital stock] was 13.2 percent below its trend path. The crisis and Great Recession, including
amplification mechanisms, appear to be responsible for the shortfall.”[50] Also, the work of Ravi Bansal, Mariano
Max Croce, Wenxi Liao, and Samuel Rosen indicates specifically that uncertainty during and following the Great
Recession may have also played a role and, especially, hurt innovative and productivity-driving firms, observing
that “volatility shocks are more disruptive for innovation-oriented ﬁrms both in terms of market valuation and
contraction in their investments. According to the data, when uncertainty increases, there exists a relative
reallocation eﬀect that penalizes investments in R&D-intensive ﬁrms, that is, investments that are important to
sustain long-term growth.”[51]
Taking a slightly longer view and analyzing the entire 2005–18 slowdown period, Germán Gutiérrez and Thomas
Philippon point out that the U.S. business sector has underinvested relative to “measures of profitability and
valuation, particularly Tobin’s Q, and that this weakness starts in the early 2000’s.”[52] In terms of a theoretical
underpinning that could explain this phenomenon, the authors specify that although it is possible for firms to
underinvest either because of a low Q or despite a high Q, the data do not support the first case. So instead, the
authors focus on the latter case, in which they find evidence of three main drivers: “rising intangibles, decreased
competition, and changes in corporate governance that encourage payouts instead of investment.”[53]

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

Regarding the rise of intangibles (intellectual property including software, R&D, patents, trademarks, and goodwill)
as a share of overall capital investment, Gutiérrez and Philippon estimate that this component “can explain a
quarter to a third of the observed investment gap.”[54] This is because these assets are both difficult to measure,
which may lead to their undercounting, and also “difficult to accumulate, due to higher adjustments costs,” leading
to “a higher equilibrium value of Q, even if intangibles are correctly measured.”[55]
The remainder of the underinvestment likely comes from some combination of decreased competition and
changes in corporate governance, according to Gutiérrez and Philippon. Regarding the former, the evidence
indicates that “industries with more concentration and more common ownership invest less, even after controlling
for current market conditions. Within each industry year, the investment gap is driven by firms that are owned by
quasi-indexers and located in industries with more concentration and more common ownership.[56] These firms
spend a disproportionate amount of free cash flows buying back their shares.”[57] And, in terms of the latter,
Gutiérrez and Philippon cite increased shareholder oversight, particularly in guarding against managers’ desire to
expand capital investments beyond an amount that would be in shareholders’ best interests, as well as shorttermism, in which “stock-based compensation incentivizes managers to focus on short-term capital gains” via
share buybacks rather than making long-term capital investments in their firm.[58]

Annual contributions to the productivity slowdown
In addition to analyzing the labor productivity slowdown from a full-period perspective, as we have done thus far,
we can also look at the individual years of the slowdown period itself, to determine how the path of each
component developed over time. Figure 7 illustrates how the three series underlying labor productivity growth—
MFP growth, the contribution of capital intensity, and the contribution of labor composition—progressed over the
course of the slowdown period, from 2005 to 2018.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

In the first year of the slowdown period—2006, which in figure 7 indicates the growth observed from 2005 to 2006,
as we are displaying the growth from one year to the next—MFP growth and the contribution of capital intensity not
only had the same rate of growth but they were also similarly below their respective long-term rates. A 0.5-percent
increase in MFP in 2006 was below its long-term average of 1.1 percent, and a 0.5-percent increase in contribution
of capital intensity was below its long-term average of 0.9 percent. However, over the next few years, it is
remarkable how each of these measures diverged as the Great Recession began and wore on. In each year from
2006 through 2009, the contribution of capital intensity incrementally expanded, reaching a series-high 3.2 percent
in 2009 and composing the majority of the above-trend labor productivity growth in that year. This acceleration in
the contribution of capital intensity was due to an outsized decline in underlying labor hours; although capital
services growth slowed from 3.7 percent in 2006 to 1.1 percent in 2009, labor hours growth plummeted from 2.3
percent in 2006 to –7.2 percent in 2009. This difference in magnitudes makes sense, given that it is easier for
businesses to lay off workers than sell capital equipment during a recession.
At the same time as the contribution of capital intensity was expanding during the Great Recession, MFP growth
was stagnating. After posting a below-average value in 2007 (0.5 percent), the measure sank well into negative
territory in 2008 and then edged barely back into positive territory in 2009. What may have brought about this
below-average MFP growth during the Great Recession? John G. Fernald notes that “Factor utilization . . .
‘explains’ the plunge and rebound in [MFP]. Utilization fell below the range of historical experience in the
recession, [and] then recovered rapidly during the recovery.”[59] Nicholas Bloom, Max Floetotto, Nir Jaimovich,
Itay Saporta-Eksten, and Stephen J. Terry cite uncertainty, noting that “plant-level [MFP] shocks increased in

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

variance by 76 percent during the recession” and that “bad times, defined in terms of low growth rates of output,
are also uncertain times in terms of increased cross-sectional dispersion of [MFP] shocks.”[60] In addition, Lucia
Foster, Cheryl Grim, and John Haltiwanger claim that the Great Recession was an atypically detrimental recession
in terms of its effect on MFP growth, in that there was not the usual boost from increased reallocation, which most
recessions offer; the authors show that “the intensity of reallocation fell rather than rose and the reallocation that
did occur was less productivity enhancing than in prior recessions.”[61]
Following the end of the Great Recession in 2009, we observe another year of strong labor productivity growth (3.4
percent) in 2010, though with a sudden reversal in the underlying contributions: MFP growth and the contribution
of capital intensity were virtually mirror images of one another in those 2 years, with MFP growth accelerating from
0.2 percent in 2009 to 2.7 percent in 2010 and the contribution of capital intensity slowing from 3.2 percent in 2009
to 0.3 percent in 2010. What might have brought about this result? As noted earlier, Fernald cites increased
utilization as a potential explanation for the rebound in MFP growth during this early phase of the recovery.[62]
Also, it is often the case when emerging from a recession—especially one as severe as the Great Recession—that
firms may still remain apprehensive about hiring until the recovery begins in earnest, with a lag in employment
recovery relative to output recovery. This was the case with the recovery from the Great Recession, with labor
hours falling for an additional quarter more than output, in the third quarter of 2009, and then remaining virtually flat
for two additional quarters—while output was simultaneously rising—and hours not beginning to recover in earnest
until the second quarter of 2010.[63] This lag in the labor hours recovery contributed to both the dramatic increase
of MFP growth in 2010 as well as the diminution of the contribution of capital intensity in that year.[64]
In the years following 2010, labor productivity growth stagnated, with an average rate of just 0.8 percent—well
below the 2.1-percent long-term average rate since 1947. These early years of the recovery were particularly weak
for the underlying measures, with the contribution of capital intensity receding into slightly negative territory in both
2011 and 2012, and MFP posting a decline in 2011 and a 0.1-percent increase in 2013. More recently, the situation
has improved somewhat compared with those early years, with labor productivity rising above 1.0 percent during 3
of the last 4 years, though still remaining below the long-term historical trend and thus extending the productivity
slowdown for over a decade.[65]
A noteworthy fact about the growth during the historically weak recovery period—specifically, after 2011—is how
consistent and steady the growth rates have been during these years, with labor productivity growth staying within
a historically narrow range of between 0.3 percent and 1.4 percent. This phenomenon reflects the combination of a
steadily weak output recovery and a consistently moderate labor hours recovery. And, in terms of labor productivity
growth’s underlying series, MFP growth was below average during these years as weak output growth was paired
with moderate combined-input growth from labor and capital. And, the contribution of capital intensity was also low
throughout these years, as moderately increasing labor hours were paired with similarly moderate capital services
growth.
Also, note that, in contrast to its typical steadiness during the full periods (see figure 2), the contribution of labor
composition exhibited some within-period variation during the slowdown period from 2005 to 2018 (see figure 7)
and thereby slightly amplified the swings in labor productivity during this period. Specifically, the contribution of
labor composition rose at above-average rates during the Great Recession, with a high of 0.5 percent in 2008, and
then grew at below-average rates since, of 0.1 or 0.2 percent. These shifts in the contribution of labor composition
over the slowdown period—particularly within the post-2007 business cycle—are not surprising, given that lower19

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

skilled or less-experienced workers are more likely to be laid off during recessions and then may gradually be
reintegrated into the workforce as a recovery progresses.[66]
Trend comparisons for MFP and the contribution of capital intensity
In addition to determining the extent to which each component series contributed to labor productivity growth within
each year of the slowdown period, we can also determine how each of these component series tracked during
these years compared with its own previously observed growth trends, particularly its speedup period trend rate
(1997–2005) and its long-term trend rate (1948–2005). This analysis allows us to see how much the movements
after 2005 either stayed on course with, or diverged from, the growth trends that had been previously and
historically observed for these series. We will do this for the two contributors to the slowdown: MFP growth and the
contribution of capital intensity.
The path of the MFP index series over the slowdown period, as well as the long-term trend rate and the speedup
period trend rate for this series, is shown in figure 8. We see that in 2006 and 2007, MFP was already falling
slightly behind both the speedup period and long-term period trend lines, and this gap widened substantially during
the recession. Then, in 2010, 1 year of high MFP growth partially shrank the gap, but subsequently, from 2011 to
2018, below-average MFP growth widened the gap substantially. Strikingly, we can see from the figure that every
annual movement from 2005 to 2018 other than the gain in 2010 acted to widen the gap, either with a negative
annual change or a positive change that was slower than the historical trend.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

The contribution of capital intensity index series took a much different path through the slowdown period, as is
illustrated by figure 9. As just discussed, the contribution of capital intensity took an inverse path relative to MFP
during the Great Recession, with an increase in 2008 and a surge in 2009 that sent the series above both the longterm and speedup period trend lines by 2009. However, the subsequent stagnation in the series, with virtually no
growth over the next 4 years, submerged it below both trend lines by 2013 and widened the gap from that point
onward. At the same time, note that this cumulative gap in the growth of the contribution of capital intensity was, as
of 2018, less than that of MFP, which is cumulatively much further behind its historical trends (see figure 8).

Dollar and time costs of the productivity slowdown
In addition to analyzing the productivity slowdown in terms of percent changes, as we have done up to this point,
we may also wonder: how much of a real-world impact did the labor productivity slowdown have in terms of dollars
of lower output or hours of lost leisure time, for participants of the U.S. economy? Before undertaking this analysis,
we should clarify that it is not possible to know in what combination the additional productivity growth—if growth
had continued at average historical rates following 2005, rather than at the low rates we have observed—would
have translated into greater output and additional leisure time. However, these calculations give us a sense of the
losses that have been incurred by Americans, due to the productivity slowdown.
We will first estimate the loss from the productivity slowdown by assuming that the additional productivity growth
(representing the difference between recorded productivity growth and what productivity growth would have been if

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

rates had continued at average historical rates following 2005) would have all contributed to producing additional
output, and we will then make an analysis assuming that the added productivity growth would have all contributed
to accumulating additional leisure time.
To estimate the total loss in output, we first ascertain how much total output was produced during the slowdown
period. This amount is $175.2 trillion. We can then calculate a hypothetical total output, incorporating a consistent
2.3-percent labor productivity growth rate.[67] This amount is $186.1 trillion. So, the difference in output,
representing the loss due to the productivity slowdown, is $10.9 trillion. Furthermore, as there were, on average,
114.6 million workers in the nonfarm business sector during these years, this result translates into a loss of
$95,000 in output per worker.[68]
The productivity slowdown can also be framed in terms of lost time, specifically the lost leisure time that could
have been available for workers to consume if a slowdown had not occurred.[69] To do this analysis, we first add
up all hours worked during the slowdown period, which comes to 2.51 trillion. Then, assuming that labor
productivity grew at a consistent rate of 2.3 percent throughout the period and that all of the effect of the added
productivity growth contributed to a reduction in hours worked, this calculation would yield a total of 2.37 trillion
hypothetical hours worked. Then, subtracting this hypothetical hours figure from the actual hours figure (2.51
trillion) would result in a hypothetical gain of leisure time of 138.5 billion hours, or 1,209 per worker. And, given that
the average weekly hours during these years would have been 30.7 in this case,[70] this would result in a total of
39.4 weeks of leave lost because of the slowdown in productivity growth during this period or, correspondingly, 3.0
additional weeks of leave per year.

Did the productivity slowdown progress differently in U.S. regions?
Before we move on to the industry-level analysis, it might be of interest to some readers to know that there
was some variation in how the economy-wide productivity slowdown progressed between states and
regions of the United States. BLS publishes data on U.S. state and regional labor productivity for 2007
forward, and we can use these data to illustrate how the slowdown progressed in these areas for this portion
of the post-2005 slowdown period.*
Box figure 1 tracks the progress of the labor productivity series for the private nonfarm sector in the four
U.S. regions during the 2007–18 period. What this figure shows is that the Western United States
outperformed the other three regions (Midwest, South, and Northeast) during these years. The West not
only outperformed the other regions throughout and following the Great Recession, but its outperformance
expanded during the recovery.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

The West realized a 1.6 percent rate of labor productivity growth during these years, whereas the Midwest,
South, and Northeast posted rates of 0.9 percent, 1.0 percent, and 1.0 percent, respectively. Given that the
national rate for these years is 1.3 percent, we can see that, if not for the faster growth of the West during
these years, the overall U.S. slowdown would have been around 0.3 percentage point lower than the
already low rate observed during this period. This insight may offer researchers a potential avenue for future
research, to look into what potential factors may have contributed to the somewhat higher productivity
growth in the Western United States as compared with the rest of the country. At the same time, it is still the
case that all four regions posted below-average growth, compared with the national long-term average rate
since 1948 of 2.2 percent.
As for the individual states, box figure 2 reveals that, in addition to having the highest overall growth, the
Western United States also had the most outliers, both on the high and low end (the figure shows the top six
and bottom six state productivity growth rates). Washington, California, Oregon, and Colorado had four of
the top six rates, which buoyed the overall growth of the West and more than counterbalanced very low
growth in Arizona and Alaska and a decline in Wyoming. North Dakota took the top state rate, at 3.1 percent,
and was the only state that had a growth rate that placed above the long-term average U.S. rate of 2.2
percent. As noted by YiLi Chien and Paul Morris, North Dakota underwent an oil boom during these years,
“bringing a large influx of capital” to the state, with the authors concluding that the above-average labor
productivity growth in this state “is very likely associated with the boom of the oil industry.”** (The turnaround

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

in this industry, as well as the results of other notable industries, will be discussed in detail in the next
section.)

* For more information on BLS state-level productivity data, see Sabrina Wulff Pabilonia, Michael W. Jadoo,
Bhavani Khandrika, Jennifer Price, and James D. Mildenberger, “BLS publishes experimental state-level
labor productivity measures,” Monthly Labor Review, June 2019, https://www.bls.gov/opub/mlr/2019/article/
pdf/bls-publishes-experimental-state-level-labor-productivity-measures.pdf. In addition, the data can be
found at the BLS state productivity home page at https://www.bls.gov/lpc/state-productivity.htm.
** YiLi Chien and Paul Morris, “Slowdown in productivity: state vs. national trend,” The Regional Economist
(Federal Reserve Bank of St. Louis, first quarter 2017).

Industry-level analysis of the U.S. labor productivity slowdown
Up to this point, we have analyzed the U.S. labor productivity slowdown from an economy-wide perspective. We
can also examine more detailed, industry-level data to extend our analysis and identify the industries that
contributed the most to the slowdown in productivity growth from the 1997–2005 period to the 2005–18 period. In

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

breaking out the U.S. labor productivity slowdown into its industry-level components, we will investigate the two
series contributing to the slowdown—MFP growth and the contribution of capital intensity—to determine which
industries contributed most to the economy-wide slowdown via these two factors. This can be done by calculating
Domar-weighted growth rates of these two factors, for all the industries that make up the private nonfarm business
sector.[71] First, we will be looking at the industry contributions to the MFP slowdown, followed by the industry
contributions to the slowdown in the contribution of capital intensity.

Industry contributions to the MFP slowdown
When we break out the economy-wide MFP slowdown into its components, it is instructive for us to disaggregate
the data into both sectors and industries, because each of these approaches provides a slightly different
perspective and can deepen our understanding of the issue. As such, we will first look at contributions at the 14sector level and then at contributions at the 60-industry level.[72]
The sector-level contributions to the economy-wide MFP slowdown are shown in figure 10, along with the
corresponding contributions of the overall goods sector and services sector. First, we notice that the goods sector
made a contribution (0.63 percentage point [ppt.]) to the MFP slowdown that is larger than the contribution made
by the services sector (0.49 ppt.). The fact that the goods sector made a larger contribution is at first glance
somewhat surprising, given that it is much smaller than the services sector and produces just 25 percent of private
nonfarm business output. The potency of the contribution from the goods-producing sector can be isolated to the
manufacturing sector and, most prominently, durable goods manufacturing, which itself contributed 0.51 ppt., or
nearly half the overall contributions to the private nonfarm business sector (1.11 ppt.).[73] When the durable goods
sector contribution is combined with the nondurable goods sector contribution, the total manufacturing sector
accounted for 65 percent of the private nonfarm business MFP slowdown.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

For the services sector, the largest contributors to the slowdown were retail trade (0.22 ppt.), wholesale trade (0.20
ppt.), and transportation and warehousing (0.10 ppt.). These three sectors together more than explain the overall
contribution to the MFP slowdown coming from the services sector (0.49 ppt.), as they made a combined
contribution of 0.52 ppt. It is also worth pointing out that two sectors had notable productivity speedups, especially
considering their small size—natural resources and mining (0.12 ppt.) and utilities (0.11 ppt.); natural resources
and mining makes up just 2.7 percent of the private nonfarm business sector, and utilities makes up just 2.3
percent. Also, though the financial services sector and the professional and business services sector made
relatively flat contributions to the slowdown, there were a number of industries within those two sectors that made
notable contributions to the slowdown (these industries are discussed below). However, when these industry-level
slowdowns are summed with speedups in other industries within the same sector, only a small overall contribution
to the slowdown remains for these two sectors.
The industry-level contributions to the economy-wide MFP slowdown are shown in figure 11, specifically for a
selection of the largest contributions (positive or negative). (Also, see appendix, table 1, for a full list of industry
contributions to the MFP slowdown and industry contributions to the slowdown in the contribution of capital
intensity, in the private nonfarm business sector.) It is not surprising that the largest industry-level contributor to the
slowdown (computer and electronic products) is from within the largest sector-level contributor to the slowdown
(durable manufacturing). Computer and electronic products incurred a massive slowdown, with a contribution to

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

MFP growth of 0.45 ppt. from 1997 to 2005 dwindling to 0.10 ppt. from 2005 to 2018. A startling fact about this
industry is that, even after having the largest MFP slowdown among all the industries, computer and electronic
products still possessed a positive contribution and, in fact, the third-largest contribution among all 60 industries
during the 2005–18 period, behind only real estate (0.12 ppt.) and oil and gas extraction (0.12 ppt.).[74] The MFP
slowdown in computer and electronic products represents 66 percent of the slowdown in durable manufacturing
and 31 percent of the slowdown in the private nonfarm business sector.

Numerous researchers have focused on the historic acceleration and subsequent moderation in the growth of the
computer and electronic products industry as being a major driver of the economy-wide productivity slowdown.
Many had been aware of the remarkable growth of computer and electronic products in the late 1990s and early
2000s, with Stephen D. Oliner and Daniel E. Sichel observing that the trend at that time was already apparent and
strong. They note that “the multifactor productivity contributions from computer and semiconductor producers
moved up sharply during 1996–99, reaching 0.26 and 0.39 percentage point per year, respectively,” and that “the
increases largely reflect the faster decline in the relative prices of computers and semiconductors . . . and the
rising output shares of computer and semiconductor producers.”[75] David M. Byrne and Carol Corrado also

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

emphasize the rapidly declining prices during those years: “the greatest computer [price] declines, and the greatest
gap, occurs [sic] in the 1994 to 2000 period, when [microprocessor unit] prices were falling especially fast.”[76]
Then, however, during the mid-2000s, the pace of microprocessor unit price declines began to stall.[77] And, more
recently, price declines have shrunk in size even more, with Byrne and Corrado citing “extremely small declines of
late, after having gradually lost force since 2004.”[78] At the same time, Byrne, Oliner, and Sichel argue that there
has not been a slowdown in technological progress that the paltry price declines might indicate, clarifying that
“technical progress in the semiconductor industry has continued to proceed at a rapid pace.”[79] So, might there
be some mismeasurement occurring with regard to these prices? Yes, there was, argue David M. Byrne, John G.
Fernald, and Marshall B. Reinsdorf, who say that in “IT-related hardware and software . . . mismeasurement is
sizable.”[80]
However, these authors also caution that this measurement was not a new issue in the mid-2000s and that IT price
mismeasurement was evident even before the productivity slowdown, clarifying that they “find no evidence that the
biases have gotten worse since the early 2000s.”[81] In fact, they point out that, if one were to consistently adjust
for mismeasurement across time, it would actually make the labor productivity slowdown worse, given that
“mismeasurement of IT hardware [was] significant prior to the slowdown,” and also given that “the domestic
production of these products has fallen, [and thus] the quantitative effect [of mismeasurement] on productivity was
larger in the 1995–2004 period than since, despite mismeasurement worsening for some types of IT.”[82]
Chad Syverson concurs with Byrne, Fernald, and Reinsdorf, agreeing that mismeasurement is unlikely to be a
driver of the productivity slowdown,[83] and observes that “the productivity slowdown has occurred in dozens of
countries, and its size is unrelated to measures of the countries’ consumption or production intensities of
information and communication technologies [ICTs],” further contending that “if measurement problems were to
account for even a modest share of this missing output, the properly measured output and productivity growth
rates of industries that produce and service ICTs would have to have been multiples of their measured growth in
the data.”
So, what factors might have led to the slowdown in the productivity growth of IT goods? Decker et al. point out that
a dwindling of the “marginal employment growth response of businesses to idiosyncratic productivity draws . . . is
especially large in the high-tech sector, with the responsiveness of young firms in the post-2000 period only about
half (manufacturing) to two thirds (economy-wide) of the peak responsiveness in the 1990s.” The authors conclude
that “the timing of reallocation and responsiveness patterns in high-tech is consistent with the timing of the
productivity slowdown, which evidence indicates was driven by ICT-producing and using industries.”[84]
The waning responsiveness of young high-tech firms that is cited by Decker et al. could potentially be explained, at
least in part, by the work of Mordecai Kurz, who finds growing market power in the IT sector, which may be stifling
the entry and growth of young firms.[85] Kurz reports that “declining or slow growing firms with broadly distributed
ownership have been replaced by IT based firms with highly concentrated ownership,” and that “IT innovations
enable and accelerate the erection of barriers to entry and once erected, IT facilitates maintenance of restraints on
competition.”[86] Foster, Grim, Haltiwanger, and Wolf also reference the concentration within high-tech industries,
noting that, in contrast to the late 1990s, when “the productivity surge in the high-tech sectors [had] a high
contribution of increased within-industry covariance between market share and productivity . . . the productivity
slowdown in the post-2000 period in high tech is due to both a decrease in within-firm productivity growth but also

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

a decrease in this covariance.”[87] Titan Alon, David Berger, Robert Dent, and Benjamin Pugsley offer further
evidence to support this finding, noting that “over the last three decades, the U.S. business sector has experienced
a collapse in the rate of new startups alongside an enormous reallocation of economic activity from entrants and
young firms to older incumbents.”[88] Alon et al. clarify that this finding is not just particular to high-tech industries
but is “widespread across industries and geographic markets,”[89] so that while this could be relevant in high-tech
industries, it could also help explain the productivity slowdowns in other industries. And, more generally, Grullon et
al. observe that “more than 75% of U.S. industries have experienced an increase in concentration levels over the
last two decades.”[90]
Beyond the concentration argument, David Autor, David Dorn, Gordon H. Hanson, Gary Pisano, and Pian Shu
propose another potential contributor to the productivity slowdown in IT, particularly in IT-intensive manufacturing
industries, which is a reduction in U.S. innovation caused by increased foreign competition from China.[91] The
authors observe that “despite accounting for less than one-tenth of U.S. private non-farm employment, U.S.
manufacturing still generates more than two-thirds of U.S. R&D spending and corporate patents,” and claim that
“increased imports from China ramped up competitive pressure on publicly listed U.S. firms” and that “this increase
in competitive pressure caused U.S. firms to decrease their output of innovations as measured by patent
grants.”[92] However, this phenomenon is more long-term and may not apply only or specifically to the 2005–18
slowdown period, although it could potentially be a contributor.
Now, shifting gears a bit, let us look beyond the large slowdown in the computer and electronic products industry
and examine some other sizable contributors, located in the trade sector. Specifically, retail trade and wholesale
trade contributed 0.22 ppt. and 0.20 ppt., respectively, to the MFP slowdown, and when combined, they actually
exceed the size of the slowdown in computer and electronic products. These trade sectors transitioned from
making sizeable positive contributions to MFP growth during the speedup period to being virtually flat during the
slowdown period.
Might the size and coincidence of these slowdowns in the trade sectors and those in the IT-related industries be
related? The answer is likely yes, according to several researchers, at least regarding the retail trade sector. Lucia
Foster, John Haltiwanger, and C. J. Krizan assert that “the retail trade sector underwent a massive restructuring
and reallocation of economic activity in the 1990s. Retail businesses changed their ways of doing business with
intensive adoption of advanced information technology, including everything from improvements in inventory
control to the introduction and widespread use of scanners and rapid credit card processing technologies.
Structural changes occurred with entering establishments from large multiunit national firms displacing singleestablishment firms.”[93]
These changes were widely seen in the economy, with the proliferation of “big box” stores such as Wal-Mart,
Home Depot, and Best Buy that swept the country and displaced many small businesses that could not compete
with the advanced IT that these corporations were using.[94] Emek Basker argues that the effect was particularly
powerful regarding Wal-Mart, “because Wal-Mart competes with retailers across many categories, including
general merchandise stores, drugstores, apparel stores, and grocery stores.”[95]
Moreover, Foster et al. contend that, in their firm-level analysis of the dispersion and reallocation dynamics within
the retail trade sector, “virtually all of the productivity growth in the retail trade sector over the 1990s is accounted
for by more productive entering establishments displacing much less productive exiting establishments,” which

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

they clarify is due to a combination of “selection effects and post-entry learning effects. That is, establishments that
enter might be immediately more productive than the establishments they are displacing, or it may take time for the
productivity gap to widen or emerge.”[96]
Also, looking beyond the trade sectors and the computer and electronic products industry, we see several other
notable downward contributors to the slowdown in MFP growth: Federal Reserve banks and credit intermediation
(0.13 ppt.), securities, commodity contracts, and investments (0.11 ppt.), and broadcasting and
telecommunications (0.08 ppt.).
In addition, a few industries worked in the opposite direction of the overall slowdown and posted accelerations in
MFP growth during this period: oil and gas extraction (0.15 ppt.), real estate (0.13 ppt.), utilities (0.11 ppt.), and
rental and leasing services and lessors of intangible assets (0.10 ppt.). The increase for the oil and gas extraction
industry during the slowdown period catapulted it from being the 56th ranked industry among all 60 industries as of
the speedup period, to a rank of 2 as of the slowdown period. This astounding turnaround, note David Popp,
Jacquelyn Pless, Ivan Haščič, and Nick Johnstone, may reflect technological innovations in this industry, in which
the “rise of hydrofracturing lowered fossil fuel prices so much that natural gas is now the primary fuel for electricity
generation in the U.S.”[97] The real estate industry had a similarly extraordinary turnaround in its MFP growth
contribution, rising from a rank of 51 to 1.
Given that there were both negative and positive contributors to the MFP slowdown, with sizable contributors on
both sides, it may be of interest to look at the distribution of these industry-level data (see figure 12). The first item
to note is that there were more large negative contributors (those slowing by 0.05 ppt or more) than large positive
contributors (those expediting by 0.05 ppt. or more). The net slowdown of the large-contributing industries was
0.74 ppt., with 1.39 ppts. of the large contributors on the negative side and 0.66 ppt. on the positive side. The
second item worth noting is that the remaining 0.38 ppt. of the slowdown comes from the small contributors—
specifically, many more small negative contributors existed than small positive contributors. While just 6 industries
contributed between 0.01 and 0.04 ppt., 24 contributed between –0.01 and –0.04 ppt. These negative small
contributors had a combined slowdown of 0.44 ppt., greatly outweighing the positive small contributors, which had
a combined speedup of just 0.08 ppt. So, we can say that, although there were numerous large contributors to the
overall slowdown on the negative side (particularly computer and electronic products and the trade industries),
there was also a widespread, generalized negative slide among the vast majority of the industries, which also
helped bring about the historic decline in MFP growth.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

Industry contributions to the slowdown of the contribution of capital intensity
The slowdown in the contribution of capital intensity came more from the services sector than the goods sector.
This finding can be seen in figure 13, which shows the sector-level contributions to the overall slowdown in the
contribution of capital intensity. The services sector accounted for 0.45 ppt. of the overall slowdown in this
measure, with the goods sector contributing 0.26 ppt.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

The largest sector contribution to the slowdown was from the financial services sector, with a contribution of 0.23
ppt. There were also noteworthy contributions from professional and business services (0.12 ppt.), durable
manufacturing (0.13 ppt.), and nondurable manufacturing (0.08 ppt.).
Figure 14 shows the large industry-level contributions to the slowdown in the contribution of capital intensity,
including those with contributions of 0.05 or greater in either direction, positive or negative. Five of the six large
contributors were negative and were of similar sizes. The four largest outliers each had slowdowns of 0.06 ppt.;
these were rental and leasing services and lessors of intangible assets, retail trade, computer and electronic
products, and insurance carriers and related activities. Miscellaneous professional, scientific, and technical
services slowed by 0.05 ppt. The broadcasting and telecommunications industry had a speedup of 0.05 ppt.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

The distribution of industry contributions to the economy-wide slowdown in the contribution of capital intensity
skews heavily negative (see figure 15), with 31 negative contributors and just 4 positive contributors. At the same
time, the net contribution to the slowdown from the large contributors (0.24 ppt.) was lower than the net
contribution from the small contributors (0.46 ppt.), indicating that there were relatively few outliers with regard to
the slowdown in the contribution of capital intensity, especially when compared with the case of the MFP
slowdown, which had some substantial outliers. To summarize this section, the bulk of the slowdown in the
economy-wide contribution of capital intensity came from relatively small slowdowns in this measure that occurred
in a substantial number of industries.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

Conclusion
At this point, we have a general understanding of the many factors—at both the economy-wide and industry levels
—which may underlie the productivity slowdown since the mid-2000s. However, some questions remain, perhaps
the most central one being: Is the U.S. economy now, as some researchers have suggested, in an intermittent lull
in between waves of high growth, or, as others contend, in a “new normal” of lower growth that has resulted from
fundamentally diminished returns to innovation? The answer is not known at the moment, and only time—and
additional data in our time series—will tell us.
At the same time, one thing that we can say for certain about our present situation is that the productivity
slowdown of the past decade and a half has left the U.S. economy in a weaker position—yielding a sizable loss of
potential output during these years—and perhaps even more importantly, it has also left the economy in a weaker
position going forward. This is because the productivity slowdown has resulted in a lower base of output from
which to grow onward from here, relative to the more elevated starting position that the economy would instead
now have if productivity had continued to grow at the long-term historical trend after 2005.
Thus, it will be important for participants of the U.S. economy to keep an eye on productivity data in coming years,
to determine whether the slowdown since 2005 simply represented a periodic variation in trend, which can be
explained from recent cyclical and noncyclical factors, as some observers have claimed, or whether it comes to be
seen as a continuation of the low-growth economy of the last few decades of the 20th century. BLS productivity

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

data, including labor productivity, multifactor productivity, and capital and labor data, at both the economy-wide and
industry levels, will continue to shed light on this issue.

Appendix. Full list of industry contributions
Table 1. Industry contributions to slowdown in private nonfarm business labor productivity growth, from
1997–2005 period to 2005–18 period, percentage points
Contributions from industry MFP

Contributions from industry

growth

contribution of capital intensity

Industry

Forestry, fishing, and related activities
Oil and gas extraction
Mining, except oil and gas
Support activities for mining
Utilities
Construction
Food and beverage and tobacco products
Textile mills and textile product mills
Apparel and leather and applied products
Wood products
Paper products
Printing and related support activities
Petroleum and coal products
Chemical products
Plastics and rubber products
Nonmetallic mineral products
Primary metal products
Fabricated metal products
Machinery
Computer and electronic products
Electrical equipment, appliances, and
components
Motor vehicles, bodies and trailers, and parts
Other transportation equipment
Furniture and related products
Miscellaneous manufacturing
Wholesale trade
Retail trade
Air transportation
Rail transportation
Water transportation
Truck transportation
Transit and ground passenger transportation
Pipeline transportation
Other transportation and support activities
Warehousing and storage

1997–2005

2005–18

period

Difference

1997–2005

2005–18

period

Difference

0.011
–0.029
0.014
–0.004
–0.067
–0.091
0.019
0.007
0
0.001
0.003
0.022
0.019
–0.001
0.027
0.003
0.027
0.007
0.027
0.445

–0.001
0.122
–0.020
0.016
0.042
–0.122
–0.020
–0.001
–0.002
0.004
0
0.009
–0.021
–0.075
–0.002
–0.004
0.004
–0.015
–0.004
0.104

–0.012
0.15
–0.034
0.02
0.109
–0.031
–0.039
–0.008
–0.002
0.003
–0.004
–0.013
–0.040
–0.074
–0.028
–0.006
–0.023
–0.022
–0.030
–0.341

0.002
0.004
0
–0.004
0.044
0.036
0.022
0.004
0.007
0.002
0.01
0.006
0.014
0.103
0.016
0.005
0.004
0.014
0.023
0.09

0.002
–0.035
0.014
–0.001
0.029
0.014
–0.001
0.001
0
0.001
0.003
0.003
0.018
0.071
0.002
0.003
0.004
0.006
0.007
0.032

0.001
–0.039
0.014
0.003
–0.015
–0.023
–0.023
–0.004
–0.007
–0.001
–0.007
–0.003
0.004
–0.032
–0.014
–0.002
0.001
–0.007
–0.016
–0.058

0.008

0.003

–0.006

0.009

0.003

–0.006

0.076
0.009
0
0.023
0.181
0.216
0.042
0.014
–0.005
–0.008
0.003
0.009
0.029
0.015

0.001
0.015
–0.003
0.006
–0.015
–0.007
0.033
0
0.008
–0.007
–0.003
0.007
–0.035
–0.003

–0.074
0.006
–0.002
–0.017
–0.196
–0.223
–0.009
–0.014
0.012
0.001
–0.007
–0.002
–0.065
–0.017

0.037
0.01
0.004
0.012
0.1
0.099
0.009
0.003
–0.003
0.007
0.003
0.004
0.001
–0.001

0.001
0.009
0.002
0.007
0.059
0.041
0.005
0.003
0.001
0.008
–0.001
0.005
–0.004
–0.002

–0.036
–0.001
–0.002
–0.005
–0.041
–0.058
–0.004
0
0.004
0
–0.004
0.001
–0.005
–0.001

See footnotes at end of table.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

Table 1. Industry contributions to slowdown in private nonfarm business labor productivity growth, from
1997–2005 period to 2005–18 period, percentage points
Contributions from industry MFP

Contributions from industry

growth

contribution of capital intensity

Industry

Publishing industries, except internet (includes
software)
Motion picture and sound recording industries
Broadcasting and telecommunications
Data processing, internet publishing, and other
information services
Federal reserve banks, credit intermediation,
and related activities
Securities, commodity contracts, and other
financial investments and related activities
Insurance carriers and related activities
Funds, trusts, and other financial vehicles
Real estate
Rental and leasing services and lessors of
intangible assets
Legal services
Miscellaneous professional, scientific, and
technical services
Computer systems design and related services
Management of companies and enterprises
Administrative and support services
Waste management and remediation services
Educational services
Ambulatory health care services
Hospitals and nursing and residential care
facilities
Social assistance
Performing arts, spectator sports, museums,
and related activities
Amusements, gambling, and recreation
industries
Accommodation
Food services and drinking places
Other services, except government

1997–2005

2005–18

period

Difference

1997–2005

2005–18

period

Difference

–0.006

0.042

0.048

0.073

0.053

–0.020

0.025
0.121

0.021
0.044

–0.004
–0.077

0.018
0.125

0.005
0.173

–0.012
0.048

0.027

0.015

–0.013

0.046

0.063

0.017

0.03

–0.096

–0.126

0.112

0.071

–0.041

0.069

–0.042

–0.111

0.011

0.003

–0.009

0.05
0.008
–0.009

0.046
–0.003
0.124

–0.004
–0.011
0.133

0.073
0.006
0.022

0.018
–0.020
–0.015

–0.055
–0.026
–0.037

–0.098

–0.002

0.095

0.117

0.057

–0.061

0.006

–0.033

–0.039

0.012

0.006

–0.006

–0.052

0.02

0.072

0.078

0.026

–0.052

0.047
0.002
0.08
0.007
–0.015
0.035

0.095
0.023
0.023
–0.003
–0.007
0.034

0.048
0.021
–0.057
–0.010
0.008
–0.002

0.019
0.008
0.046
–0.002
–0.005
0.004

–0.006
–0.003
0.019
–0.001
0.002
0.002

–0.024
–0.011
–0.026
0.002
0.007
–0.002

–0.011

–0.015

–0.003

0.009

0.011

0.003

0.005

–0.003

–0.008

–0.001

0.006

0.011

0.005

0.002

–0.003

–0.014

0.014

0.006

0.004

–0.002

0.007
0.041
–0.023

0.006
–0.007
–0.032

–0.001
–0.048
–0.009

0.015
–0.006
0.009

0.008
–0.008
0.003

–0.007
–0.001
–0.005

Note: MFP = multifactor productivity.
Source: U.S. Bureau of Labor Statistics.

ACKNOWLEDGMENTS: The author would like to thank the following people for their valuable contributions to this
article: Martin Baily, Lucy Eldridge, Ray Fair, Ryan Forshay, Corby Garner, Michael Giandrea, John Glaser, Kendra
Hathaway, Michael Jadoo, Peter Meyer, Sabrina Pabilonia, Thomas Philippon, Susan Powers, Matthew Russell,
Bob Shackleton, Chris Sparks, Jay Stewart, and Leo Sveikauskas.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

SUGGESTED CITATION

Shawn Sprague, "The U.S. productivity slowdown: an economy-wide and industry-level analysis," Monthly Labor
Review, U.S. Bureau of Labor Statistics, April 2021, https://doi.org/10.21916/mlr.2021.4.
NOTES
1 The overall estimated output loss figure ($10.9 trillion) represents the difference between (1) the sum of annual real output amounts
in the nonfarm business sector from 2006 to 2018 and (2) the sum of annual real output amounts during this period assuming that
labor productivity had continued to grow at the same long-term average rate observed from 1947 to 2005 and that all of the additional
gains in labor productivity contributed to higher output rather than higher nonwork time. For more information on the estimation of this
figure, as well as an example in which the additional gains in labor productivity contributed to higher nonwork time rather than higher
output, see the section “Dollar and time costs of the productivity slowdown” in this article. Also, note that the loss for the entire U.S
economy is likely sizably greater than the presented output loss figure. However, because we cannot measure the productivity of the
noncovered sectors that represent the difference (including general government, nonprofit institutions, and private households, which
include owner-occupied housing), since the output data for these sectors are not suitable for productivity measurement, we can only
estimate what the effect would have been for the 76 percent of the U.S. economy covered by the nonfarm business sector. Also note
that the estimated output loss per worker figure ($95,000) does not represent the loss in compensation per worker due to the
slowdown, which would have been sizably less than that figure, given that only a portion of output accrues to workers as
compensation for their labor. For more information on this aspect, see Michael D. Giandrea and Shawn A. Sprague, “Estimating the
U.S. labor share,” Monthly Labor Review, February 2017, https://www.bls.gov/opub/mlr/2017/article/estimating-the-us-laborshare.htm.
2 See figure 2 of Martin Neil Baily and Nicholas Montalbano, “Why is U.S. productivity growth so slow? Possible explanations and
policy responses,” Hutchins Center Working Paper 22 (Washington, DC: Brookings Institution, September 2016), p. 4. Baily and
Montalbano provide an excellent overview of the productivity slowdown and also offer potential policy responses. In addition, other
overviews of the slowdown include Alexander Murray, “What explains the post-2004 U.S. productivity slowdown?” International
Productivity Monitor, no. 34, Spring 2018, p. 81–109; and Emily Moss, Ryan Nunn, and Jay Shambaugh, “The slowdown in
productivity growth and policies that can restore it,” The Hamilton Project (Washington, DC: Brookings Institution, June 2020).
3 In Dale W. Jorgenson, Mun S. Ho, and Kevin J. Stiroh, “A retrospective look at the U.S. productivity growth resurgence,” Journal of
Economic Perspectives, no. 1, vol. 22, Winter 2008, p. 4, the authors offer an anecdote illustrating the changed perspective during
this period: “in just four years, from 1997 to 2001, the Congressional Budget Office more than doubled its ten-year projection of
nonfarm business productivity growth from 1.2 to 2.7 percent!”
4 All percent changes in this article, unless otherwise noted, refer to average annual percent changes.
5 Although labor productivity growth theoretically provides a foundation for potential gains in worker compensation, it has historically
been the case that commensurate gains in worker compensation do not necessarily accompany gains in labor productivity. Since the
early 1970s, real hourly compensation growth has lagged far behind labor productivity growth. For more information, see Susan
Fleck, John Glaser, and Shawn Sprague, “The compensation−productivity gap: a visual essay,” Monthly Labor Review, January 2011,
https://www.bls.gov/opub/mlr/2011/01/art3full.pdf.
6 There are two main ways of dividing time periods when one is doing a historical productivity analysis such as the one in this article:
using business cycles or using variation in trends that are apparent in the data. For the present article, I use variation in trends to
define the speedup period (1997–2005) and the slowdown period (2005–18). I selected these two periods because they correspond
to the periods that are generally discussed in the literature on this issue and because it is apparent from the data that the question at
hand (“What were the sources of the slowdown?”) could not be addressed by operating strictly within the macroeconomic business
cycle framework. One reason for this is that the industry-level sources of the slowdown exhibited their own trends that did not
necessarily fit within the economy-wide business cycles. Another reason is that although using either a cyclical or trend approach for
the slowdown period made little difference (because the business cycle that began in 2007 began just 2 years following the beginning

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

of the trend-based period [2005]), for the speedup period, it appeared to be preferable to use a trend-based approach. This
preference was due to the fact that using the business-cycle break (in 2001) would have split the widely cited speedup period (1997–
2005) roughly in half, disallowing for a one-to-one comparison of a speedup and slowdown period. If a business cycle approach had
been taken instead, then the comparison would have been between two speedup periods (1990–2001 and 2001–07) and one
slowdown period (2007–18), as parts of each of those two preceding periods contain a portion of the speedup in the trend, and the
industry-level and broad-based changes in the economy during the speedup period from 1997 to 2005 would have been split into two
separate periods. It is notable that this disjuncture between business cycles and trend-based periods in terms of the current
productivity slowdown did not occur with the last major productivity slowdown, in the 1970s. This widely cited slowdown, which was
apparent in the data beginning in 1973, represented both a business cycle break and a clear demarcation between extended periods
of high growth (before 1973) and low growth (after 1973). This result contrasts the present slowdown, the period selection of which
was not as clear-cut. As for defining the periods since 1947 that occurred before the speedup period (1997–2005), I use a grouped
business cycle approach (by grouping adjacent business cycles with similar trends) to make the interperiod comparisons in the figures
more concise and make the broad trends over the past 70 years easier to absorb. The fact that the last major productivity slowdown
(in the 1970s) had a match between business cycle breaks and trend-period breaks makes the 1948–1997 period somewhat easier to
break up into periods. Also, for a strictly business cycle analysis of the present slowdown, see Shawn Sprague, “Below trend: the U.S.
productivity slowdown since the Great Recession,” Beyond the Numbers: Productivity, vol. 6, no. 2 (U.S. Bureau of Labor Statistics,
January 2017), https://www.bls.gov/opub/btn/volume-6/below-trend-the-us-productivity-slowdown-since-the-great-recession.htm.
7 Multifactor productivity (MFP) data for the U.S. private nonfarm business sector are available on an annual basis, beginning with
data for 1948. This sector accounted for approximately 74 percent of the total U.S. economic output (gross domestic product [GDP])
as of 2017. As denoted by the term “business” in the series name, three nonbusiness sectors are excluded from GDP: general
government, nonprofit institutions, and private households (including owner-occupied housing). These three sectors are excluded
because their output is measured largely with the use of compensation data, which measure an input to production rather than an
output from production, thus rendering these sectors inappropriate for productivity measurement. Also, farm sector data are excluded
to reduce volatility in the overall measure. And, as denoted by the term “private” in the series name, government enterprises are
excluded, because satisfactory capital measures are unavailable for this sector. In addition, note that the term “MFP” is synonymous
with “TFP” or total-factor productivity, which is used throughout the economic literature and refers to the same measure. (For more
information on this, see the U.S. Bureau of Labor Statistics (BLS) Division of Productivity Research and Program Development
Frequently Asked Questions (FAQs) page at https://www.bls.gov/dpr/faqs.htm#Q01). On another note, although for this article we are
looking at private nonfarm business MFP growth, which only includes labor and capital as inputs, BLS also publishes other measures
of MFP growth, one of which is referred to as KLEMS (K-capital, L-labor, E-energy, M-materials, and S-purchased services), which
includes additional inputs as well.
8 Although figure 2 shows a slight shift in the contribution of labor composition (from 0.2 percent to 0.3 percent) from the speedup
period to the slowdown period, looking at these data with more precision reveals a trivial shift in the contribution of labor composition
during the slowdown, with an upward shift of just 0.01 percentage point, from 0.24 percent in the speedup period to 0.25 percent in
the slowdown period.
9 Alistair Dieppe, Neville Francis, and Gene Kindberg-Hanlon observe this phenomenon in both advanced economies and emerging
market economies, in the period from 1980 to 2018: “The volatility of labor productivity is largely accounted for by [MFP] across both
advanced economies and EMDEs, where [MFP] accounts for 75–80 percent of labor productivity variance. The high proportion of
volatility present in [MFP] reflects its role as a residual, explaining all productivity variation not driven by slower-moving developments
in the capital stock and human capital.” Alistair Dieppe, Neville Francis, and Gene Kindberg-Hanlon, “A tale of two dynamics:
exploring productivity and business-cycle drivers of developed and emerging market economies,” working paper, July 24, 2020, p. 6.
10 For this article, we are looking at private nonfarm business MFP growth, which only includes labor and capital as inputs. However,
BLS also publishes other measures of MFP growth that include other inputs, referred to as KLEMS (capital, labor, energy, materials,
and services).
11 See Brian Resnick, “Dark matter, humility, and coming to grips with the unknown,” Vox, November 25, 2020.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

12 From 1929 to 1933, GDP fell by a cumulative 26.3 percent, whereas from 2007 to 2009, GDP fell by a cumulative 2.7 percent. For
more information, see National Data, National Income and Product Accounts (NIPA), Table 1.1.6. Real gross domestic product,
chained dollars (U.S. Bureau of Economic Analysis).
13 From 1933 to 1940, GDP grew at an average annual rate of 7.2 percent, and from 2009 second quarter to 2018 fourth quarter,
GDP grew at an average annual rate of 2.3 percent. NIPA, Table 1.1.6.
14 The U.S. population grew at an average annual rate of 0.7 percent from 1929 to 1940 and from 2007 to 2018. NIPA, Table 7.1
Selected per capita product and income series in current and chained dollars, line 18 (U.S. Bureau of Economic Analysis).
15 Valerie Cerra and Sweta Chaman Saxena, “Growth dynamics: the myth of economic recovery,” American Economic Review, vol.
98, no. 1, March 2008, pp. 439–57; Carmen M. Reinhart and Kenneth S. Rogoff, This Time is Different: Eight Centuries of Financial
Folly (Princeton, NJ: Princeton University Press, 2009); and Carmen M. Reinhart and Vincent R. Reinhart, “After the fall,” Working
Paper 16334 (Cambridge, MA: National Bureau of Economic Research, September 2010).
16 International Monetary Fund, World Economic Outlook, October 2009: Sustaining the Recovery (Washington: International
Monetary Fund, 2009), p. 125.
17 Daisuke Ikeda and Takushi Kurozumi, “Slow post-financial crisis recovery and monetary policy,” Working Paper 347 (Federal
Reserve Bank of Dallas Globalization Institute, October 2018), p. 4. As noted in endnote 7, in this article, I use MFP to refer to
multifactor productivity, although others in the literature refer to the measure as TFP or total factor productivity. These acronyms and
names refer to the same measure. For more information, see the BLS Division of Productivity Research and Program Development
FAQ page at https://www.bls.gov/dpr/faqs.htm#Q01.
18 These changes in demographics “include, most prominently, the demographic shifts of the surge of women into the labor force in
the 1970s–1990s and, more recently, the baby boom[ers] beginning to retire.” James H. Stock and Mark W. Watson, “Why has GDP
growth been so slow to recover?” (draft paper presented at the Boston Federal Reserve’s conference, “The elusive ‘great recovery’:
causes and implications for future business cycle dynamics,” October 2016), p. 1.
19 Ray C. Fair, “Explaining the slow U.S. recovery: 2010–2017,” Business Economics, vol. 53, no. 4, pp. 184–194, October 2018.
20 Robert E. Hall, “The routes into and out of the zero lower bound,” in Proceedings—Economic Policy Symposium—Jackson Hole
(Federal Reserve Bank of Kansas City, 2013), p. 3.
21 Robert J. Barro, “The job-filled non-recovery” (Cambridge, MA: Harvard University, September 2016), p. 6.
22 Ikeda and Kurozumi, “Slow post-financial crisis recovery and monetary policy,” p. 4.
23 David M. Byrne, Stephen D. Oliner, and Daniel E. Sichel, “Is the information technology revolution over?” International Productivity
Monitor, vol. 25, Spring 2013, p. 21.
24 Romain Duval, Gee Hee Hong, and Yannick Timmer, “Financial frictions and the great productivity slowdown,” Working Paper
17/129 (International Monetary Fund Working Papers, May 2017), p. 1.
25 Duval et al., “Financial frictions and the great productivity slowdown,” p. 17. Intangible assets are nonphysical assets such as
intellectual property, including software, research and development, patents, trademarks, and goodwill. Germán Gutiérrez and
Thomas Philippon, in “Investment-less growth: an empirical investigation,” abstract, Working Paper 22897 (Cambridge, MA: National
Bureau of Economic Research, December 2016), also cite decreased investment in intangibles; see section (in this article) “The
slowdown in the contribution of capital intensity.” In addition to having an indirect effect on the MFP growth measure, as noted by
Duval et al., a slowdown in intangibles would also have a direct effect on the contribution of capital intensity measure, by explicitly
reducing the numerator of the change in the capital-labor ratio. In addition, the fallout from the Great Recession may have also
hampered rates of reallocation—the process of moving more resources toward high-productivity firms and away from low productivity
firms—according to Lucia Foster, Cheryl Grim, and John Haltiwanger, “Reallocation in the Great Recession: cleansing or not?”
abstract, Working Paper 20427 (Cambridge, MA: National Bureau of Economic Research, August 2014). These authors observe
faltering rates of reallocation during the Great Recession, which is the inverse of the typical case for recessions, perhaps indicating
that the magnitude of this recession adversely affected prospects for productivity growth. Even worse, they state that “the reallocation

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

that did occur [during the Great Recession] was less productivity enhancing than in prior recessions.” Nicholas Bloom, Max Floetotto,
Nir Jaimovich, Itay Saporta-Eksten, and Stephen J. Terry also cite weak rates of reallocation, which they claim is undergirded by an
increase in uncertainty during recessions, contending that “increased uncertainty also reduces productivity growth because it reduces
the degree of reallocation in the economy. Higher uncertainty leads productive plants to pause expanding and unproductive plants to
pause contracting, which in the . . . U.S. economy drives much of aggregate productivity growth.” Nicholas Bloom, Max Floetotto, Nir
Jaimovich, Itay Saporta-Eksten, and Stephen J. Terry, “Really uncertain business cycles,” Working Paper 18245 (Cambridge, MA:
National Bureau of Economic Research, July 2012), p. 1–2.
26 John G. Fernald downplays the importance of the Great Recession in the post-2005 productivity slowdown, claiming that “the
Great Recession seem[s] less important than trends related to information technology (IT) that predated the Great Recession.” See
his paper, “Productivity and potential output before, during, and after the Great Recession,” Working Paper 2014–15 (Federal Reserve
Bank of San Francisco, June 2014). (Note that the effect of IT-intensive industries on the slowdown will be addressed in the “Industrylevel analysis of the U.S. labor productivity slowdown” section of this article.)
27 Ryan A. Decker, John C. Haltiwanger, Ron S. Jarmin, and Javier Miranda, “Changing business dynamism and productivity: shocks
vs. responsiveness,” Working Paper 24236 (Cambridge, MA: National Bureau of Economic Research, January 2018). Also, in addition
to the works cited in this section on the topic of productivity dispersion, note that data on productivity dispersion are now available
from BLS and the Census Bureau, via the Collaborative Micro-Productivity Project (CMP), which has developed and published
experimental statistics on within-industry dispersion. The public-use statistics (referred to as the Dispersion Statistics on Productivity)
developed via this project, were released in fall 2019 and cover all four-digit North American Industry Classification System industries
in the manufacturing sector. Restricted-use establishment-level data with microbased estimates of productivity as well as its
underlying components (e.g., output and input measures) are also available to qualified researchers on approved projects in secure
Federal Statistical Research Data Centers. More information on these data can be obtained at https://www.bls.gov/lpc/productivitydispersion.htm.
28 Michael Gort and Steven Klepper, “Time paths in the diffusion of product innovations,” Economic Journal, vol. 92, no. 367,
September 1982, pp. 630–653.
29 Decker et al., “Changing business dynamism and productivity.”
30 Decker et al., “Changing business dynamism and productivity,” p. 27. An in-depth discussion of high-tech industries, and their
relationship to the economy-wide productivity slowdown, is in the “Industry-level analysis of the U.S. labor productivity slowdown”
section of this article.
31 Ibid.
32 Jan De Loecker, Jan Eeckhout, and Gabriel Unger, “The rise of market power and the macroeconomic implications,” working
paper, 2018.
33 Ibid., p. 52.
34 Gustavo Grullon, Yelena Larkin, and Roni Michaely, “Are US Industries becoming more concentrated?” abstract, working paper,
2017.
35 Miguel Antón, Florian Ederer, Mireia Giné, and Martin C. Schmalz, “Common ownership, competition, and top management
incentives,” Finance Working Paper 511/2017 (Brussels: Belgium, European Corporate Governance Institute), February 2018; and
José Azar, Sahil Raina, and Martin C. Schmalz, “Ultimate ownership and bank competition,” working paper (SSRN, March 17, 2016).
36 John E. Kwoka, Jr. “Mergers, merger control, and remedies: a response to the FTC critique,” working paper (SSRN, March 2017);
Bruce A. Blonigen and Justin R. Pierce, “Evidence for the effects of mergers on market power and efficiency, ” Working Paper 22750
(Cambridge, MA: National Bureau of Economic Research, October 2016); Council of Economic Advisers, “Occupational licensing: a
framework for policymakers,” The White House: Washington, DC, July 2015; Jason Furman, “Barriers to shared growth: the case of
land use regulation and economic rents” (remarks at The Urban Institute, November 20, 2015); Jason Furman, “Market concentration
—note by Jason Furman” (paper presented at Organisation for Economic Cooperation and Development [OECD], Directorate for

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

Financial and Enterprise Affairs Competition Committee Hearing on Market Concentration, June 7, 2018); and Brookings Institution,
“The productivity puzzle: How can we speed up the growth of the economy?” panel discussion (Washington, DC, September 9, 2016).
In addition, Ernest Liu, Atif Mian, and Amir Sufi theorize that the low interest rates of the past decade have helped increase
concentration, asserting that “low interest rates encourage market concentration by raising industry leaders’ incentive to gain a
strategic advantage over followers, and this effect strengthens as the interest rate approaches zero.” For more information, see
Ernest Liu, Atif Mian, and Amir Sufi, “Low interest rates, market power, and productivity growth” abstract, Working Paper 25505
(Cambridge, MA: National Bureau of Economic Research, August 2019).
37 Dan Andrews, Chiara Criscuolo, and Peter N. Gal, “The global productivity slowdown, technology divergence and public policy: a
firm level perspective,” Hutchins Center Working Paper 24 (Washington, DC: Brookings Institution, September 2016), https://
www.brookings.edu/research/the-global-productivity-slowdown-technology-divergence/, p. 3.
38 Decker et al., “Changing business dynamism and productivity,” p. 24. Additionally, there is another widely cited firm-level theory
that has been offered to explain the increased productivity dispersion, which is from Adalet McGowan, Andrews, and Millot. Müge
Adalet McGowan, Dan Andrews, and Valentine Millot, “Insolvency regimes, zombie firms and capital reallocation,” OECD Economics
Department Working Papers No. 1399, June 28, 2017. Adalet McGowan et al. remark that the prevalence of so-called “zombie firms,”
or low-productivity firms that are unable to properly service their debts, have increased in OECD countries since the mid-2000s. The
authors hypothesize that the stubborn persistence of these low-growth firms is potentially not only keeping resources from reallocating
to high-growth firms, but it is also creating entry barriers and inhibiting the growth of new firms. However, this theory appears not to be
relevant specifically for the United States, which, unlike other OECD countries and especially European countries, has actually had a
decline in the share of zombie firms in recent years. For more information, see Dan Andrews and Giuseppe Nicoletti, “Confronting the
zombies: policies for productivity revival” slide 7, OECD Economic Policy Papers, no. 21 (presented at the Peterson Institute for
International Economics, January 23, 2018).
39 Jason Furman and Peter Orszag, “Slower productivity and higher inequality: Are they related?” Working Paper 18–4 (Peterson
Institute for International Economics, June 2018).
40 Ibid., p. 2. See also Alexander M. Bell, Raj Chetty, Xavier Jaravel, Neviana Petkova, and John Van Reenan, “Who becomes an
inventor in America? The importance of exposure to innovation,” Working Paper 24062 (Cambridge, MA: National Bureau of
Economic Research, November 2017); Frederico Cingano, “Trends in income inequality and its impact on economic growth,” OECD
Social, Employment and Migration Working Papers, no. 163 (Paris: Organisation for Economic Cooperation and Development
Publishing, December 9, 2014); and Jonathan D. Ostry, Andrew Berg, and Charalambos G. Tsangarides. “Redistribution, inequality,
and growth,” Staff Discussion Note (Washington, DC: International Monetary Fund, February 2014).
41 Furman and Orszag, “Slower productivity and higher inequality, p. 2.
42 Ruchir Agarwal and Patrick Gaulé, “Invisible geniuses: Could the knowledge frontier advance faster?” abstract, International
Monetary Fund Working Paper 18/268 (Washington, DC: International Monetary Fund, December 2018).
43 Robert J. Gordon, “Is U.S. economic growth over? Faltering innovation confronts the six headwinds,” Working Paper 18315
(Cambridge, MA: National Bureau of Economic Research, August 2012); and John G. Fernald, “Productivity and potential output
before, during, and after the Great Recession.”
44 Joseph A. Tainter, The Collapse of Complex Societies (Cambridge, MA: Cambridge University Press, 1988), p. 114.
45 Nicholas Bloom, Charles I. Jones, John Van Reenen, and Michael Webb, “Are ideas getting harder to find?” Working Paper 23782
(Cambridge, MA: National Bureau of Economic Research, September 2017), p. 8; and Jay Bhattacharya and Mikko Packalen,
“Stagnation and Scientific Incentives,” abstract, Working Paper 26752 (Cambridge, MA: National Bureau of Economic Research,
February 2020). Bhattacharya and Packalen argue that changing incentives may also be playing a role, specifically that an “emphasis
on citations in the measurement of scientific productivity shifted scientist rewards and behavior on the margin toward incremental
science and away from exploratory projects that are more likely to fail, but which are the fuel for future breakthroughs.”

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

46 Chad Syverson, “Will history repeat itself? Comments on ‘Is the information technology revolution over?’” International Productivity
Monitor, no. 25, Spring 2013, pp. 37–40.
47 Ana Paula Cusolito and William F. Maloney, Productivity Revisited: Shifting Paradigms in Analysis and Policy (Washington, DC:
The World Bank Group, 2018), p. 8.
48 Though figure 2 shows that both the 2005–18 slowdown period and the 1981–97 period posted the same low 0.7-percent rate, at
slightly more precision, the 2005–18 rate (0.66 percent) was nearly a percentage point lower than the 1981–97 period rate (0.74
percent).
49 Byrne et al., “Is the information technology revolution over?” p. 21.
50 Robert E. Hall, “Quantifying the lasting harm to the U.S. economy from the financial crisis,” Working Paper 20183 (Cambridge, MA:
National Bureau of Economic Research, May 2014), p. 13.
51 Ravi Bansal, Mariano Max Croce, Wenxi Liao, and Samuel Rosen, “Uncertainty-induced reallocations and growth,” Working Paper
26248 (Cambridge, MA: National Bureau of Economic Research, September 2019), p. 1.
52 Gutiérrez and Philippon, “Investment-less growth,” abstract. Tobin’s Q was first introduced by Nicholas Kaldor in 1966. For more
information, see Nicholas Kaldor, “Marginal productivity and the macro-economic theories of distribution: comment on Samuelson and
Modigliani,” Review of Economic Studies, vol. 33, no. 4, October 1966, pp. 309–319. It was popularized a decade later, however, by
James Tobin, who describes its two quantities: “One, the numerator, is the market valuation: the going price in the market for
exchanging existing assets. The other, the denominator, is the replacement or reproduction cost: the price in the market for newly
produced commodities. We believe that this ratio has considerable macroeconomic significance and usefulness, as the nexus
between financial markets and markets for goods and services.” James Tobin and William C. Brainard, “Asset markets and the cost of
capital,” in Economic Progress, Private Values, and Public Policy (Amsterdam: North-Holland Publishing Company, 1977).
53 Germán Gutiérrez and Thomas Philippon, “Investment-less growth: an empirical investigation,” Brookings Papers on Economic
Activity (Washington, DC: Brookings Institution, September 7, 2017), p. 90.
54 Ibid., p. 93.
55 Ibid., p. 136.
56 Regarding quasi-indexers, Gutiérrez and Philippon state, “Quasi-indexers have diversified holdings and low portfolio turnover—
consistent with a passive, buy-and-hold strategy of investing portfolio funds in a broad set of firms.” Gutiérrez and Philippon,
“Investment-less growth,” December 2016, p. 3.
57 Ibid., abstract.
58 Ibid., 2016, p. 19. See also Michael Jensen, “Agency costs of free cash flow, corporate finance, and takeovers,” American
Economic Review, vol. 76, no. 2, 1986, p. 323. Gutiérrez and Philippon, “Investment-less growth,” September 7, 2017, p. 109. See
also Christine Jolls, “Stock repurchases and incentive compensation,” abstract, Working Paper 6467 (Cambridge, MA: National
Bureau of Economic Research, March 1998).
59 Fernald, “Productivity and potential output before, during, and after the Great Recession,” p. 9.
60 Bloom et al., “Really uncertain business cycles,” p. 1.
61 Foster et al., “Reallocation in the Great Recession: cleansing or not?” abstract.
62 Fernald, “Productivity and potential output before, during, and after the Great Recession.”
63 Sprague, “Below trend,” https://www.bls.gov/opub/btn/volume-6/below-trend-the-us-productivity-slowdown-since-the-greatrecession.htm.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

64 In 2010, labor hours jumped up by more than 7 percentage points in a single year and transitioned from a 7.2-percent decline in
2009 to a 0.1-percent decline in 2010. This much smaller decline in labor hours growth for 2010 helped shrink the contribution of
capital intensity and expand MFP growth compared with what they had been in 2009. Specifically, labor hours growth lowered the
contribution of capital intensity in 2010 relative to that of 2009 because its low rate in 2010 (–0.1 percent) was then more similar to the
rate for capital services (0.8 percent), thus shrinking the capital-to-labor ratio relative to the ratio of the prior year. In addition, MFP
growth was dramatically accelerated as the below-average gains in both labor and capital in that year were paired with rapidly
recovering output growth (3.3 percent).
65 Also, note that the low productivity growth of the 2005–18 period itself reduced the long-term rate since 1947 from 2.3 percent—
what it had been as of 2005—to 2.1 percent as of 2018.
66 The data underlying the labor composition series bear this phenomenon out during the Great Recession: the labor cost share of
workers under 45 and workers without a college degree plunged much more than it did for workers over 45 and workers with a college
degree, during the Great Recession.
67 In this case, I use the long-term historical trend rate from 1947 to 2005, which is 2.3 percent, rather than the rate from 1947 to
2018, which is 2.1 percent. This is because, in this exercise, we are attempting to determine what growth would have been after 2005
if it had continued at the rate observed prior to 2005.
68 This per-worker loss in output technically should be viewed in terms of productivity analysis, with the output loss being represented
relative to this input to production. Furthermore, the loss in output per worker does not equate to the loss in compensation per worker,
which would have been a lesser amount, as only a portion of output accrues to workers as compensation for their labor. Giandrea and
Sprague, “Estimating the U.S. labor share,” https://doi.org/10.21916/mlr.2017.7.
69 Note that, in this hypothetical case, some of this additional leisure or nonwork time could have been the result of layoffs (which
could have resulted in unemployment, voluntary retirement from work, or, for some workers, another job), to the extent that the
reduction in overall hours was distributed inequitably among workers and reduced employment rather than average weekly hours. To
the extent that average weekly hours were reduced, this could have come in the form of increased vacation or sick leave offered and
taken, a reduction in the workweek, or a transition from full-time to part-time work for some workers.
70 For this computation of hypothetical average weekly hours, we are assuming that all the reduction in overall hours went to a
reduction in average weekly hours and not to employment. This approach was taken to show the additional leisure or nonwork time
that would have been hypothetically available for this group of workers who were employed during 2005–18.
71 Evsey D. Domar, “On the measurement of technological change,” Economic Journal, vol. 71, no. 284, December 1961, pp. 709–
729. In his article, Domar developed a method for estimating industry contributions to overall MFP growth for an aggregate sector. I
use this approach for breaking out industry contributions to private nonfarm business MFP growth. Domar showed that a given
industry’s contribution to an aggregate MFP growth rate is equal to the MFP growth rate for that industry, multiplied by a two-period
average of the ratio of output in the industry to value-added in the sector. The sum of the Domar-weighted industry MFP growth rates
approximates the private nonfarm business MFP growth rate. Also, to calculate the industry contributions to the private nonfarm
business contribution of capital intensity, I use an approach that uses Domar’s general approach and allows for the breakout of this
component. For more information regarding this approach, see Robert Inklaar, Mary O’Mahony, and Marcel Timmer, “ICT and
Europe’s productivity performance industry-level growth account comparisons with the United States,” Groningen Growth and
Development Centre (GGDC) Research Memorandum GD-68 (Netherlands: University of Groningen, GGDC, December 2003), pp.
12–13.
72 Farm industries are excluded from our analysis in this article, as we are looking at the contributions to the private nonfarm
business sector.
73 The summed MFP contributions, using the Domar method, add to 1.11 percentage points, which is slightly different from the topline private nonfarm business slowdown of 1.27 percentage points. For this section of the article, I use the 1.11 figure as the overall
value because it is consistent with the summed contributions.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

74 Byrne et al. report that “since 2004 IT has continued to make a significant contribution to U.S. labour productivity growth, though it
is no longer providing the boost that it did during the productivity resurgence from 1995 to 2004.” See Byrne et al., “Is the information
technology revolution over?” abstract.
75 Stephen D. Oliner and Daniel E. Sichel, “The resurgence of growth in the late 1990s: Is information technology the story?” Journal
of Economic Perspectives, vol. 14, no. 4, Autumn 2000, p. 16.
76 David M. Byrne and Carol Corrado, “ICT prices and ICT services: What do they tell us about productivity and technology?”
Working Paper 2017-015 (Board of Governors of the Federal Reserve System Finance and Economics Discussion Series, February
10, 2017), p. 21.
77 According to Byrne et al., “the official price indexes for semiconductors developed by BLS show that quality-adjusted
semiconductor prices are not falling nearly as rapidly as they did prior to the mid-2000s.” See Byrne et al., “Is the information
technology revolution over?” p. 23.
78 Byrne and Corrado, “ICT prices and ICT services,” p. 1.
79 Byrne et al., “Is the information technology revolution over?” p. 23.
80 David M. Byrne, John G. Fernald, and Marshall B. Reinsdorf, “Does the United States have a productivity slowdown or a
measurement problem?” Brookings Papers on Economic Activity (draft paper presented at BPEA conference, March 10–11, 2016), p.
1, https://www.brookings.edu/wp-content/uploads/2016/03/ByrneEtAl_ProductivityMeasurement_ConferenceDraft.pdf.
81 Ibid.
82 Ibid., abstract. The authors also qualify that “the effect on [MFP of their adjustment] is more muted.”
83 Chad Syverson, “Challenges to mismeasurement explanations for the U.S. productivity slowdown,” abstract, Working Paper 21974
(Cambridge, MA: National Bureau of Economic Research, February 2016).
84 Decker et al., “Changing business dynamism and productivity,” p. 27.
85 Mordecai Kurz, “On the formation of capital and wealth,” Discussion Paper 17-016 (Stanford Institute for Economic Policy
Research, June 25, 2017).
86 Ibid., abstract.
87 Lucia Foster, Cheryl Grim, John Haltiwanger, and Zoltan Wolf, “Innovation, Productivity Dispersion, and Productivity Growth,” U.S.
Census Bureau Center for Economic Studies Working Paper 18-08, February 2018, p. 4.
88 Titan Alon, David Berger, Robert Dent, Benjamin Pugsley, “Older and slower: the startup deficit’s lasting effects on aggregate
productivity growth,” Working Paper 23875 (Cambridge, MA: National Bureau of Economic Research, September 2017), p. 2.
89 Ibid., p. 2.
90 Grullon et al., “Are US industries becoming more concentrated?” abstract.
91 David Autor, David Dorn, Gordon H. Hanson, Gary Pisano, and Pian Shu, “Foreign competition and domestic innovation: evidence
from U.S. patents,” working paper, September 2018.
92 Ibid., p. 2.
93 Lucia Foster, John Haltiwanger and C. J. Krizan, “Market selection, reallocation, and restructuring in the U.S. retail trade sector in
the 1990s,” Review of Economics and Statistics, vol. 88, no. 4, November 2006, p. 748.
94 On this topic, also see Ronald S. Jarmin, Shawn D. Klimek, and Javier Miranda, “The role of retail chains,” in Timothy Dunne, J.
Bradford Jensen, and Mark J. Roberts, eds., Producer Dynamics: New Evidence from Micro Data (Chicago: University of Chicago
Press, January 2009), chap. 6.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

95 Foster et al., “Market selection, reallocation, and restructuring in the U.S. retail trade sector in the 1990s,” p. 748, as cited in Emek
Basker, “Job creation or destruction? Labor-market effects of Wal-Mart expansion,” Review of Economics and Statistics, vol. 87, no. 1,
February 2005, pp. 174–183.
96 Foster et al., “Market selection, reallocation, and restructuring in the U.S. retail trade sector in the 1990s,” p. 749.
97 David Popp, Jacquelyn Pless, Ivan Haščič, and Nick Johnstone, “Innovation and entrepreneurship in the energy sector,” abstract,
Working Paper 27145 (Cambridge, MA: National Bureau of Economic Research, May 2020).

RELATED CONTENT

Related Articles
Multifactor productivity slowdown in U.S. manufacturing, Monthly Labor Review, July 2018.
Below trend: the U.S. productivity slowdown since the Great Recession, Beyond the Numbers, January 2017.
What can labor productivity tell us about the U.S. economy? Beyond the Numbers, May 2014.

Related Subjects
Productivity

Hours of work

Multifactor productivity

Employment

Economic development and growth

5/3/2021

Are temporary layoffs becoming permanent during COVID-19? : Monthly Labor Review: U.S. Bureau of Labor Statistics

U.S. BUREAU OF LABOR STATISTICS
Bureau of Labor Statistics

Publications

Monthly Labor Review

HOME

RELATED CONTENT

Related Articles
Employment recovery in the wake of the COVID-19 pandemic, Monthly Labor Review, December 2020.
Changes in consumer behaviors and financial well-being during the coronavirus pandemic: results from the U.S. Household Pulse
Survey, Monthly Labor Review, December 2020.
From the barrel to the pump: the impact of the COVID-19 pandemic on prices for petroleum products, Monthly Labor Review, October
2020.

Related Subjects
Social issues

Leave benefits

Worker safety and health

COVID-19

Industry studies

April 2021

Estimating state and local employment in recent
disasters—from Hurricane Harvey to the COVID-19
pandemic
Natural disasters, including hurricanes, floods, wildfires,
and the coronavirus disease 2019 (COVID-19) pandemic,
have challenged the standard practices used to produce
state and area employment estimates. In some cases,
these challenges have led to modifications to the handling
of reported business closures, assumptions regarding
nonresponse, and the techniques used for modeling
employment in domains with small samples for state and
metropolitan areas. This article examines how a series of
major hurricanes in 2017 and 2018 affected the estimation
of state and metropolitan area payroll employment and how
lessons learned from these disasters provided a playbook
for producing estimates during the COVID-19 pandemic.
The U.S. Bureau of Labor Statistics (BLS) Current

Steven M. Mance
mance.steven@bls.gov

Employment Statistics (CES) program produces some of
the timeliest economic indicators each month, both for the
nation as a whole and for varying levels of geographic
detail. The CES program conducts a monthly survey of

Steven M. Mance is a supervisory economist in
the Office of Employment and Unemployment
Statistics, U.S. Bureau of Labor Statistics.

about 144,000 businesses and government agencies
representing about 697,000 individual worksites.
Respondents to the CES survey provide information on the
number of employees, total payroll, and hours paid for the pay period that includes the 12th day of the month. The
CES program uses these data to produce estimates of industry employment, hours, and earnings. BLS typically
publishes national data within a week of the end of the reference month; data for the 50 states; Washington,
DC; Puerto Rico; the U.S. Virgin Islands; and about 450 metropolitan areas are released approximately 2 weeks
later.
Natural disasters challenge assumptions built into the employment estimation process regarding nonresponse and
the relationship between business openings and closures, and the CES program must address this issue in order
to produce accurate data. This article provides an overview of three aspects of the CES methodology that natural
disasters confront. It then provides case studies of how the CES program dealt with these questions in developing
state and metropolitan area estimates following five recent hurricanes—Harvey, Irma, and Maria in 2017, and

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

Florence and Michael in 2018—and how the lessons learned from the earlier experience were applied to producing
estimates in the early months of the coronavirus disease 2019 (COVID-19) pandemic. (See table 1.)
Table 1. Selected natural disasters that affected state and local employment data
Disaster
Hurricane Harvey
Hurricane Irma
Hurricane Maria
Hurricane Florence
Hurricane Michael
COVID-19 pandemic

Analyzed effect in Current Employment Statistics estimates
September 2017, Texas
September 2017, Florida
October 2017, Puerto Rico
September 2018, North Carolina and South Carolina
October 2018, Florida and Georgia
March and April 2020, 50 states, plus Washington, DC, and Puerto Rico

Source: U.S. Bureau of Labor Statistics.
Note: COVID-19 = coronavirus disease 2019.

CES methodology and natural disasters
The CES program surveys businesses each month and uses the data and a consistent, objective methodology to
produce estimates of industry employment, hours, and earnings. The methodology relies on assumptions that are
continually evaluated and generally hold in normal times. However, natural disasters are extreme events, and this
section explores three ways that such disasters raise questions about the usual CES methodological
underpinnings.

Are survey respondents different from nonrespondents?
The CES survey is designed as a probability-based survey.[1] The BLS Quarterly Census of Employment and
Wages (QCEW), which contains information on all employers required to participate in the unemployment
insurance (UI) system and covers about 97 percent of all nonfarm payroll employment, provides both a sampling
frame and the main benchmark source for the CES program.[2] In developing the QCEW as a sample frame, BLS
stratifies the data by state, industry, and establishment size. The CES survey randomly selects businesses within
these strata, with the inverse of the probability of selection used as the sample weight in estimation. The survey
estimates are benchmarked to the administrative QCEW data annually. The difference between the benchmark
data and the survey-based estimates—the benchmark revision—is regularly used to assess the accuracy of the
CES estimates. After setting benchmark employment levels each year, the CES program estimates monthly
employment growth rates with a weighted link-relative estimator. The estimator is essentially the ratio of the current
month’s weighted employment to that of the prior month in the set of matched respondents that reported nonzero
employment for the pay period containing the 12th of both months. The generally small number of active survey
units reporting zero employment in either the current or prior month is not used because of assumptions in the
business birth–death model (discussed in the next subsection).
Equation 1 provides the weighted link-relative estimate (
) for time t, where
is reported employment for a
matched respondent i in n matched sample establishments, w is the sampling weight, and d is a nonresponse
weight adjustment (calculated for industries identified to respond at substantially higher or lower rates)[3]:

U.S. BUREAU OF LABOR STATISTICS

(1)

MONTHLY LABOR REVIEW

Within each estimating domain (an area or industry combination in which the estimate calculation is performed—
other published cells are summaries of estimating cells), the estimator implicitly imputes data for nonrespondents
as well as businesses reporting zero employment in the current or prior month by using the growth rates of the
matched units. State-level estimates are produced for “estimation supersectors,” high-level industries (such as
construction or leisure and hospitality) that are estimated directly and control the sum of estimates of more detailed
industries. The sample size in these estimation supersectors is large enough to avoid using small-area modeling
techniques in most cases, but domain estimates use respondents in businesses from quite different industries (for
example, clothing stores, gas stations, and grocery stores mixed together in retail trade).
The implicit imputation follows from the form of the estimator, which does not require explicit imputation, and a
missing-at-random assumption within each estimating cell (conditioned on the industries in di). The survey
respondents stand in for the nonrespondents, and this assumption works well when they are similar, on
average.[4]
Natural disasters challenge the missing-at-random assumption. Certain types of businesses—for example, those
in federally designated disaster areas—might have more job losses than others in their state because of business
disruptions, and they might respond to the CES survey at a lower rate, as other matters take a higher priority.
Similarly, certain industries within an estimation supersector may fare differently from one another during a disaster
in a way that correlates with response rates, and the same may be true for differences in employment trends and
response rates between large and small establishments.

Are business births and deaths properly captured?
The ability to use the QCEW as a frame with a close match to the target population of all nonfarm payroll jobs is
one of the CES program’s greatest strengths. Two of the program’s biggest challenges are accounting for
businesses that have opened and closed since the frame was established. It is impossible to sample, enroll, and
collect data on business births in real time because the QCEW lags the CES survey by several months.
Establishment deaths—defined as worksites that no longer have any payroll employees, which may be temporary
or permanent—present a challenge because businesses generally stop reporting when there are no employees. A
small number of businesses report establishment deaths; however, they are mainly firms with multiple worksites.
These two components of job growth generally offset one another, and the degree to which CES procedures do
not capture the residual growth when applied to historical QCEW population data is known as the “net birth–death
residual.” This value is generally small and stable, and therefore the CES program forecasts historical values with
a time-series model for use in estimation. Equation 2 provides the formula for the employment estimate (

(2)
The CES program multiplies the prior month’s employment level (
then adds the current month’s forecast value (
).[5]

) by the weighted link relative (

) and

Natural disasters strain the usual assumptions about business births and deaths. Disaster conditions may halt the
formation of new businesses, while forcing many existing businesses to cease operations. But despite disruptions
—and sometimes even severe damage to physical locations—many businesses continue to pay employees
3

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

through the hardship, mitigating the number of business deaths by CES definitions. Still, the net birth–death
residual calculated from the QCEW is often noticeably lower in domains affected by natural disasters, and the
forecast error tends to be positive in these cases.

Are small-area models adequate?
The CES program produces estimates with a version of the Fay-Herriot model for estimation supersectors and for
many metropolitan areas with small sample sizes.[6] The model estimate is a weighted combination of the direct
weighted link-relative estimate and a synthetic component consisting of a historical trend value corrected by a
coefficient derived from the regression model in equation 3:
(3)
Y1i,j represents the statewide weighted link-relative estimates for a given industrial supersector (i) and state (j), Y2
represents the corresponding historical trend values,

is an estimated coefficient, and

is a model residual.

The regression modeling is performed at a statewide supersector level, and the correction factor (
from the
state-level model is linked to detailed metropolitan areas. The Fay-Herriot estimate for metropolitan area k is
displayed in equation 4:
(4)

The weight (Wi,k) assigned to the direct estimate (Y1i,k) is a function of the variance of the direct and trend (Y2i,k)
components as well as the variance from the state-level model. When the relationship between the direct
estimates and the historic trend is looser, the direct estimates receive a higher weight. The correction factor
adjusts the historic trend values to the direct estimates roughly on average across the country, but it does not
account for any clusters of heterogeneity among states or areas.
This model relies on assumptions about the similarity of industry behavior between areas in order to “borrow
strength” across those areas; the model also relies on assumptions about the relationship between statewide and
metropolitan area trends.[7] Hurricanes bring these assumptions into stark relief, as some affected states may be
dissimilar to the nation as a whole, and even within those states some areas face devastation while others remain
unscathed. Similarly, the COVID-19 pandemic affected jobs in states and cities to widely different degrees, despite
broad-based shutdowns and job losses.

Five major hurricanes
Although the CES survey began in 1915, and the program has faced many natural disasters since then, the survey
has evolved substantially over the years. The CES program fully implemented a probability-based design for the
50 states and Washington, DC, in 2003, BLS assumed responsibility for producing the state and local estimates in
2011, and estimation procedures and models have continued to evolve since then.[8] As a result, the 2017–18
hurricanes provided the best context in which to address estimation issues related to the COVID-19 pandemic.[9]
Several major hurricanes made landfall on the Gulf of Mexico and Atlantic coasts, as well as over U.S. territories in
the Caribbean, during the 2017 and 2018 hurricane seasons.[10] The timing of these storms matters because
anyone who worked or received pay for any portion of the pay period including the 12th day of the month was
counted as employed. Hurricane Harvey made landfall at Port Aransas, Texas, as a category-4 storm, on Friday,

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

August 25, 2017, and produced heavy rainfall and severe flooding in coastal Texas and other parts of the Gulf
Coast, most notably in Louisiana. The system that became Hurricane Irma also formed in August 2017 and made
landfall over several Caribbean islands as a category-5 storm in early September; it struck the Florida Keys as a
category-4 storm on Sunday, September 10, 2017, and reached the mainland later that day, continuing north along
the Florida peninsula. Hurricane Maria formed in September 2017 and caused severe devastation as it made its
way through the Caribbean that month. The center of the then-category-5 storm passed just south of St. Croix in
the U.S. Virgin Islands on September 19, 2017, before moving diagonally across Puerto Rico the following day.[11]
The next year, Hurricane Florence reached category-4 strength and weakened to a category-1 storm, before
making landfall on Friday, September 14, 2018, near Wrightsville Beach, North Carolina; it then traveled into South
Carolina, causing severe flooding and storm surges in both states. The following month, Hurricane Michael struck
the Florida panhandle on Wednesday, October 10, 2018, as a category-5 hurricane.
Hurricanes Harvey and Irma both caused tragic loss of life and rank among the most severe U.S. natural disasters,
in terms of property damage. Harvey caused 68 direct deaths in Texas and an estimated $125 billion in property
damages in the region. Irma caused 10 direct deaths across three states—7 in Florida, 2 in Georgia, and 1 in
South Carolina—and 3 direct deaths in the U.S. Virgin Islands, as well as $50 billion in property damages. The
CES survey measured far fewer job losses associated with Harvey than with Irma, however. Texas employment
grew in September 2017—adding 14,600 jobs, not far off the prior 12-month average growth of 16,900 jobs—with
net job losses in coastal metropolitan areas such as Houston-The Woodlands-Sugar Land, TX (–8,800). The state
added 39,300 jobs the following month as businesses bounced back. Florida, by contrast, lost 172,700 jobs in
September 2017, before recovering 194,700 jobs the following month. The difference resulted from the timing of
the storms. Irma struck Florida at the beginning of the week that included September 12, so workers on a weekly
payroll who lost their shifts for 7 days showed up in the estimates as a job loss. By contrast, Harvey hit Texas more
than 2 weeks earlier, so only prolonged layoffs resulted in a drop in measured employment in that state.[12]
Hurricane Maria, despite hitting Puerto Rico and the Virgin Islands well before the October 2017 reference period,
followed closely behind Irma’s strike on the territories and caused lasting devastation that was evident in
employment data for a prolonged period.

The birth–death model
In normal months, a small number of businesses report zero employment in the CES survey. The rate of reported
establishment “deaths” generally follows a stable, seasonal pattern, and some of the deaths are temporary.
Following Hurricanes Irma and Maria, the number of establishments reporting zero employment increased sharply.
Table 2 shows the proportion of respondents that reported zero employment and compares the forecasted and
actual birth–death residuals (later derived from the QCEW) in disaster months. In order to provide frames of
reference, table 2 also shows the average proportion of business deaths in the sample over the 2-year period
before the disaster and the average absolute monthly forecast error, in terms of employment. The proportion of
reported establishment deaths was double the usual rate in Florida following Hurricane Irma and 7 times the usual
rate in Puerto Rico following Hurricane Maria. This indicated that the usual birth–death relationship did not hold,
and actual birth–death values from the QCEW—calculated several months after the initial release of the CES
estimates—were substantially more negative than the forecast values. The forecast error of employment in Florida
following Irma was nearly 4 times the average absolute monthly forecast error, while the forecast error for Puerto
Rico following Maria was more than 8 times larger than average. Following other major hurricanes, the proportion

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

of reported establishment deaths was relatively normal, and the birth–death forecasts were closer to their mark
and well within normal ranges. (See table 2.)
Table 2. Reported zeros and birth–death forecast errors

Event

Proportion of businesses reporting zero

Birth–death forecast error (mean absolute

employment

error, 2014–19)

Month
Harvey (Texas)
Irma (Florida)
Maria (Puerto Rico)
Florence (North Carolina)
Florence (South Carolina)
Michael (Florida)
Michael (Georgia)
COVID-19 (50 states, plus
Washington, DC, and Puerto
Rico)

Percent

September 2017
Average, September 2014, 2015, and 2016
September 2017
Average, September 2014, 2015, and 2016
October 2017
Average, October 2014, 2015, and 2016
September 2018
Average, September 2015, 2016, and 2017
September 2018
Average, September 2 015, 2016, and 2017
October 2018
Average, October 2015, 2016, and 2017
October 2018
Average, October 2015, 2016, and 2017
March 2020
Average, March 2017, 2018, and 2019
April 2020
Average, April 2017, 2018, and 2019

0.23
0.15
0.40
0.19
2.20
0.32
0.14
0.16
0.21
0.20
0.22
0.18
0.13
0.18
0.46
0.18
3.40
0.22

Month

Value (in thousands)

September 2017

–0.3 (7.9)

September 2017

28.4 (7.5)

October 2017

13.8 (1.7)

September 2018

4.9 (3.6)

September 2018

–0.5 (2.1)

October 2018

4.5 (7.5)

October 2018

2.3 (5.0)

March 2020

Unavailable

April 2020

Unavailable

Source: U.S. Bureau of Labor Statistics.

Although reported establishment deaths generally are not used in the estimation process, the seven-fold increase
in their rate in Puerto Rico following Hurricane Maria necessitated a different approach. BLS determined the risk of
underestimating the number of business deaths to be substantial enough that reported zeros were treated the
same as other matched reports and were used in the sample link relative. This change negatively affected the
October 2017 employment estimates by 11,800 jobs, nearly offsetting the birth–death forecast error of 13,800.
Because many of the closures were temporary, the returns from zero employment were also used in the estimates.
Following Hurricane Irma, the CES program did not initially use reported business deaths in the weighted linkrelative estimator for Florida, but after benchmarking the September 2017 employment data, the program
recognized that the initial sample-based estimates undercounted the extent of job loss.[13] The QCEW data
contained a large increase in the number of establishments reporting zero employment, and had the sample-based
estimates used the reported zeros, the estimated employment level would have been 35,000 lower—more closely
tracking the population data—which would have accounted for the birth–death forecast error. (The actual value
was 28,000 lower than the forecast value.) The CES program included the returns for businesses that had
reported declines to zero following Irma in the sample-based estimates that were produced from the new
September 2017 benchmark level, increasing Florida employment estimates in October 2017 by 32,000. Including
these returns from zero and making other adjustments to account for reporting differences between the CES

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

survey and the QCEW resulted in October–December 2017 reestimates that would more closely track the eventual
benchmark data released in March 2019.

Nonresponse analysis
Following several major hurricanes, concerns arose over the possibility of businesses in severely affected areas
responding at lower rates than those in the rest of the state, which would result in a violation of the missing-atrandom assumption. The CES program has investigated nonresponse weight adjustments following hurricanes
and other disasters such as floods and wildfires dating back to Katrina, in 2005, in order to appropriately represent
disaster areas in statewide estimates. Although these adjustments can be calculated for different levels of
devastation—for instance using flood maps—the set of counties designated by the Federal Emergency
Management Agency (FEMA) as eligible for individual and public assistance has typically defined disaster areas
for purposes of nonresponse analysis. The program now regularly monitors response rates and the impact of
differential nonresponse in states affected by natural disasters. However, response rates were not significantly
lower in disaster areas following the 2017–18 hurricanes, in part because of dedicated data collection efforts.
Applying nonresponse adjustments following the 2017–18 hurricanes would have changed estimates little, with no
obvious benefit, and therefore the estimates were not adjusted for differential nonresponse.

Small-area estimation
The CES program widely uses variants of the Fay-Herriot model in producing employment estimates, particularly
at the metropolitan-area level.[14] In areas affected by recent major hurricanes, the direct employment estimates
were lower than the synthetic components 65 percent of the time. This indicates a breakdown in the assumptions
of similarity across states and among areas within each state. Chart 1 shows a comparison of the link relatives for
the direct and synthetic parts of Fay-Herriot models in metropolitan areas containing counties designated by FEMA
as eligible for individual and public assistance. (Three outlier values are not displayed.) The reference line has a
slope of 1, and values above that line indicate a direct estimate lower than the synthetic one.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

Although the direct estimates showed the job-losing effects of these hurricanes more, on average, than did the
model as a whole, the sample in these domains is small and often quite volatile. Had the CES program used direct
estimates instead of Fay-Herriot models, the root-mean-square benchmark revision to the link relatives would have
been 37 percent higher.
Experiences following the 2017 and 2018 hurricanes led to the development of other modeling approaches that
would generate better results and not violate important assumptions. Instead of modeling all statewide
supersectors together and applying the results to metropolitan areas, the domains directly affected by hurricanes
could be pooled and modeled together. Had it been used, this “hurricane variant” of the Fay-Herriot model would
have produced benchmark revisions that were, on average, 12 percent lower, and it would have improved 64
percent of estimates. The variant model would have improved accuracy the most for the hurricanes with the largest
effect on estimates, reducing benchmark revisions for Puerto Rico following Maria by 39 percent and for Florida
following Irma by 11 percent. (The hurricane variant improved estimates 71 percent of the time for Puerto Rico and
74 percent of the time for Florida.) The hurricane variant model did not perform well for Texas following Harvey:
although the correction factor better captured the drop in employment, worse model fit resulted in a much larger
weight on the direct components, which would have resulted in higher benchmark revisions. (See table 3.)

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

Table 3. Root-mean-square benchmark revision of link relatives for Fay-Herriot modeled series in
metropolitan areas during major hurricanes
Hurricane

Direct sample-based Fay-Herriot Fay-Herriot hurricane variant Percentage improved by hurricane variant

All (n = 139)
Florence (n
= 21)
Harvey (n =
22)
Irma (n =
68)
Maria (n =
14)
Michael (n =
14)

0.056

0.041

0.036

64.7

0.031

0.027

0.026

61.9

0.060

0.021

0.033

31.8

0.047

0.036

0.032

73.5

0.102

0.075

0.046

71.4

0.058

0.056

71.4

Source: U.S. Bureau of Labor Statistics.

Applying lessons learned to producing estimates during the
COVID-19 pandemic
A historically unprecedented number of jobs were lost and regained in the months after the onset of the COVID-19
pandemic.[15] Health officials identified the first case of COVID-19 in the United States in Washington State on
January 21, 2020.[16] However, it would be several weeks before community spread of the virus led to severe
business disruptions and shutdowns. Many novel data sources show declining activity throughout the month of
March 2020,[17] even before the issuance of formal shelter-in-place and stay-at-home orders.[18] The week of
March 8–14 marked a turning point in business activity. Restaurant reservation data from OpenTable illustrate the
evolving decline in business activity that week: as of Sunday, March 8, reservations were down only 2 percent over
the year, but by the following Saturday (March 14), they showed a 42-percent decline.[19] Ohio was the first state
to close all public schools on March 12, and by the end of the following week (March 15–21), nearly all public
schools had closed,[20] air passenger travel had fallen by 75 percent,[21] and restaurant reservations had dropped
by nearly 100 percent. The National Bureau of Economic Research (NBER) Business Cycle Dating Committee
determined that the longest economic expansion in U.S history ended in February 2020 and a recession began in
March.[22] Nationwide job losses in March (–1.7 million) and April (–20.7 million) marked the steepest employment
decline in history, and were followed by the two largest monthly job gains ever in May (+2.8 million) and June (+4.8
million). The steepness and suddenness of these job losses, followed by a rapid (if partial) recovery, were more
reminiscent to the losses seen after major hurricanes than those seen during a typical recession.

The birth–death model
The nationwide increases in the number of establishments reporting zero employment in March 2020 (2.6 times
normal) and April 2020 (15 times normal) could only be compared with the spikes following Irma in Florida (2.1
times normal) and Maria in Puerto Rico (6.9 times normal). Evidence from Irma and Maria indicates that using
reported zeros could reduce error. Beginning with the March final and April preliminary estimates, the CES state
and area program incorporated “excess” reported zeros in the link relative in calculating the employment
estimates. In order to avoid statistical bias, the program applied a weight-reduction adjustment value ranging
between 0 and 1 to establishments reporting zero employment. In cases in which the proportion of reported zeros
9

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

at the state estimation supersector level was at or below normal, the factor was set to 0, and in those cases in
which the proportion was many times the usual rate, the factor approached 1. The average weight adjustment
factor for a reported zero was 0.603 in March and 0.954 in April. As with Hurricanes Irma and Maria, employment
counts also needed to appropriately reflect returns from zero. Weight adjustment factors were also developed for
returns, and they were first applied in the May preliminary estimates, as many business reopened. The use of
excess reported drops-to and returns-from zero lowered the final sum of state employment estimates by 331,000
in March and 3.0 million in April, while increasing them by 1.0 million in May and 934,000 in June, as many
businesses reopened.
In addition, the CES program augmented the time-series forecast component of the net birth–death model with
information from the nonzero matched sample as an exogenous regressor in a regression model with
autoregressive integrated moving average errors (regARIMA), beginning with the April values.[23] Earlier research
indicated that this method could improve forecast accuracy during business cycle turning points, although it was
not feasible to implement following hurricanes.[24] The CES program updated the forecast models at the national
level and controlled the sum of the state subsector forecasts to the new national forecasts, incorporating sample
information on the states and industries seeing spikes in reported zero employment to allocate the distribution of
differences. This lowered the sum of state birth–death factors by 799,000 in April compared with what they would
have been without the regressor. The models in turn processed more positive sample information in May and June
2020 into higher forecast values.

Nonresponse analysis
The COVID-19 recession led to nonresponse concerns about the state and area employment estimates that were
different from but related to those explored after hurricanes. Subdividing states into affected and unaffected
regions, as is done for a hurricane, was not a promising option because of the geographic pervasiveness of
business disruption. The CES program explored making differential-response-rate adjustments based on business
size (the hypothesis being that larger firms may be less disrupted and better able to continue paying employees),
but the results were unclear and any adjustments may have exacerbated the effects of confounding factors. The
program also explored conducting nonresponse analysis based on linked unemployment insurance filings, but
doing so did not uncover notable differences between respondents and nonrespondents.
In producing the April 2020 state-level estimates, a major nonresponse problem appeared as certain detailed
industries within the estimation supersectors were broadly underrepresented across states in the responding CES
sample in a way that correlated with employment trends. In retail trade, clothing and clothing accessories stores
closed almost everywhere in April 2020, and these establishments responded to the survey at relatively low rates,
while general merchandise stores and building material and garden equipment stores had comparatively resilient
employment and responded at higher rates. Within leisure and hospitality, full-service restaurants lost relatively
more jobs and responded to the CES survey at lower rates than limited-service restaurants, while in the “other
services” supersector, the laundry and personal care services industry showed steep losses and was
underrepresented in the responding sample. This nonresponse issue did not arise in the national estimates
because they are produced at a detailed industry level using samples from all states, in contrast to the approach
used in the state and local estimates, where estimation supersectors pool samples from heterogeneous industries
into a discrete geographic domain.[25]

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

The CES program took two approaches to nonresponse adjustments for these industries. If it was possible to do
so, especially in very large states, the program calculated sample-based estimates at a more detailed level and
summed them to replace the estimation supersectors.[26] In states for which this approach was infeasible, the
CES program calculated nonresponse adjustments similar to those calculated (but not applied) following
hurricanes and those obtained by using preidentified nonresponse factors in the link-relative estimator (di). Many of
the response-rate differences were longstanding, but they had not substantially affected the estimates. In total,
nonresponse adjustments served to lower the sum of April 2020 state employment estimates by 461,000 in leisure
and hospitality, 413,000 in retail trade, and 73,000 in other services. As jobs returned, the differential nonresponse
problem presented in the opposite direction, and correcting for it increased employment estimates.

Small-area estimation
The CES program explored an approach similar to the hurricane variant of the Fay-Herriot model in March 2020.
The program considered a scenario in which certain states and metropolitan areas in the Northeast and Pacific
Northwest that had early cases of COVID-19 might have substantially larger employment drops in March, in which
case it could have been beneficial to pool those areas together and model them separately. This was not the case,
however, as job losses that month presented along a continuum, with no clearly identifiable division, and were only
loosely related to early case counts. The ordinary Fay-Herriot models were also problematic. The wide range of job
losses among states resulted in very high model variance and subsequently near-total reliance on the direct
estimates. Although the direct estimates are approximately unbiased, they tend to have high variability when
sample sizes are small, and the poor performance of direct estimates following hurricanes in modeled cells
indicated the potential for low accuracy.
Instead, BLS generated and in many cases used a small-area model that generalizes the Fay-Herriot model and
relaxes many of its assumptions by jointly modeling variance and point estimates and clustering the data.[27] This
model outperformed the existing Fay-Herriot model in simulations, in terms of benchmark revisions, during both
the steep downturn of the 2007–09 recession and the subsequent recovery and expansion. When applied during
the COVID-19 recession, metropolitan area estimates were more consistent with statewide values and showed a
similar story: compared with the Fay-Herriot estimates, the new model reduced the root-mean-square difference
between metropolitan and statewide estimates in the same state or industry by 29 percent in the April 2020
preliminary estimates. The combination of lessened reliance on volatile direct estimates (because of better model
fit) and state-specific effects being captured in the synthetic component of the clustered model resulted in the
tighter relationship among metropolitan area estimates and corresponding state-level estimates.

Conclusion
Timely economic data is needed to capture rapid shifts in business conditions shortly after they occur, and payroll
employment estimates from the CES program are among the timeliest economic indicators available each month
for states and metropolitan areas. Natural disasters often cause sudden, steep job losses, but the measurement of
employment change following disasters presents challenges. The CES survey must accurately capture the
relationship between business openings and closings, account for differences between respondents and
nonrespondents, and use robust models to accurately estimate employment in small domains. Five major
hurricanes that occurred in 2017 and 2018 brought these concerns to light and resulted in the development of
potential solutions. Using reported zeros proved beneficial to the estimation process after the survey showed sharp

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

increases in their number. Modeled estimates could be improved by refining the pool of observations that are
processed together.
When the COVID-19 pandemic caused the steepest job losses in U.S. history, the CES program applied the
lessons learned from producing estimates following major hurricanes to better capture business deaths, properly
represent industries experiencing the most severe declines, and incorporate models that capture important
differences between, and commonalities among, states. Using reported zeros proved beneficial following
Hurricanes Irma and Maria, and the CES program generalized their use with the wave of business closures that
began in March 2020. Although nonresponse was not a major problem after the 2017–18 hurricanes, monitoring
processes set up in the wake of those hurricanes helped BLS identify industry-based differential nonresponse
adjustments when the steepest job losses in history occurred. Techniques pooling similar observations in FayHerriot models showed promise, and the CES program applied a model that more formally clustered the data to
produce small-area estimates during the pandemic.
Although job losses during the COVID-19 recession have been far steeper and more sudden than those occurring
in typical downturns, the associated measurement challenges highlight areas of potential improvement that could
help the CES survey better capture future business-cycle turning points and sudden changes in state and area
employment that result from disasters.

ACKNOWLEDGMENTS: Julie Gershunskaya, Greg Erkens, Chris Grieves, and Dan Zhao were essential in
solving many of the methodological problems we encountered following recent hurricanes and, in particular,
during the COVID-19 pandemic. Paige Schroeder, Brad Rhein, Greg Erkens, and Alba Báez all provided
comments that noticeably improved this article. I want to thank them, as well as the CES State and Area
program office staff more broadly, who showed a great deal of ingenuity, flexibility, and hard work in
producing the estimates.

SUGGESTED CITATION

Steven M. Mance, "Estimating state and local employment in recent disasters—from Hurricane Harvey to the
COVID-19 pandemic," Monthly Labor Review, U.S. Bureau of Labor Statistics, April 2021, https://doi.org/10.21916/
mlr.2021.9.
NOTES
1 For more information on the design of the U.S. Bureau of Labor Statistics (BLS) Current Employment Statistics (CES) survey and its
evolution, see Laura A. Kelter, “One hundred years of Current Employment Statistics—the history of CES sample design,” Monthly
Labor Review, August 2016, https://doi.org/10.21916/mlr.2016.35.
2 For more information on the process for benchmarking state and local CES employment data, see Kirk Mueller, “Benchmarking the
Current Employment Statistics state and area estimates,” Monthly Labor Review, November 2017, https://doi.org/10.21916/mlr.
2017.26.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

3 This formula is simplified to remove components that are not relevant to this article. A robust estimation technique is used to identify
“atypical” reporters, which are removed from the link and only represent themselves, and reduce the weight on other influential
reporters. Employees of religious organizations, which are in scope for CES but not on the sampling frame, are addressed by
removing their benchmark level from the prior month’s employment before applying the weighted link relative and then added back to
the current month’s level.
4 Decompositions of benchmark revisions have indicated nonresponse to not be a principal driver of total survey error in the CES
survey. See, for example, Larry L. Huff and Julie G. Gershunskaya, “Components of error analysis in the Current Employment
Statistics survey,” Proceedings of the Survey Research Methods Section, American Statistical Association, October 2009, http://
www.asasrms.org/Proceedings/y2009f.html; the paper is also available on the BLS website at https://www.bls.gov/osmr/researchpapers/2009/st090050.htm. Although the implicit imputation of the weighted link-relative estimator is generally used, there is some
explicit imputation in state and area estimates for large nonrespondents that exhibit a stable, seasonal difference from the rest of the
universe using historical QCEW data. These explicit imputations are often not made during disasters.
5 For more information about the CES net birth–death model, see Kirk Mueller, “Impact of business births and deaths in the payroll
survey,” Monthly Labor Review, May 2006, https://www.bls.gov/opub/mlr/2006/05/art4full.pdf.
6 Fay-Herriot models are commonly used in producing small-area estimates and were introduced by Robert E. Fay III and Roger A.
Herriot in “Estimates of income for small places: an application of James-Stein procedures to census data,” Journal of the American
Statistical Association, vol. 74, no. 366, June 1979, pp. 269–277, https://www.jstor.org/stable/pdf/2286322.pdf.
7 More formally, this involves distributional assumptions linking the direct estimates, historical trend projections, and actual values
from a high-level model to more detailed domains. For more detail, see Julie Gershunskaya, “Estimation for detailed publication levels
in the Current Employment Statistics survey,” Proceedings of the Survey Research Methods Section, American Statistical Association,
October 2012, https://www.bls.gov/osmr/research-papers/2012/st120190.htm.
8 Puerto Rico transitioned from a quota-based design to the current probability-based design with the January 2014 estimates.
Although the U.S. Virgin Islands were severely affected by Hurricane Maria, employment estimates for the Virgin Islands are not
examined in this article because they employ a quota-based sample with an estimation methodology that differs from that of the rest
of the CES program.
9 The handling of recent disasters was built upon past experiences, primarily Hurricane Katrina, a thorough examination of which can
be found in Molly Barth Garber, Linda Unger, James White, and John Wohlford, “Hurricane Katrina’s effects on industry employment
and wages,” Monthly Labor Review, August 2006, https://www.bls.gov/opub/mlr/2006/08/art3full.pdf.
10 Information on the track and development of these storms and estimates of the damage they caused are from the National
Hurricane Center’s Tropical Cyclone Reports; for more information, see https://www.nhc.noaa.gov/data/tcr/.
11 Devastation from Hurricanes Irma and Maria was severe enough that officials in the U.S. Virgin Islands were unable to collect
establishment data for several months, resulting in a delay in the release of preliminary September, October, and November 2017
estimates for the territory. Unlike the 50 states and the District of Columbia, the workforce agencies in the U.S. Virgin Islands and
Puerto Rico collect the majority of the microdata used in their CES estimates. More information is available at https://www.bls.gov/
sae/notices/2017/hurricanes-irma-maria-september-october-november-december-2017-payroll-data-for-puerto-rico-and-the-us-virginislands.htm.
12 An analysis of credit card data by researchers at the Federal Reserve Board indicated that retail trade spending in the areas
affected by Hurricanes Harvey and Irma dropped steeply but returned to normal within 2 weeks of the storms making landfall, which
helps explain the comparatively smaller effect of Harvey on CES estimates. See Aditya Aladangady Shifrah, Aron-Dine, Wendy Dunn,
Laura Feiveson, Paul Lengermann, and Claudia Sahm, “From transactions data to economic statistics: constructing real-time, highfrequency, geographic measures of consumer spending,” Finance and Economics Discussion Series 2019-057 (Board of Governors
of the Federal Reserve System, 2019), https://doi.org/10.17016/FEDS.2019.057.
13 Some large business deaths were accounted for in Florida following Hurricane Irma, but they solely represented themselves.
14 Fay-Herriot models are also used in some statewide estimation supersectors, mostly in smaller industries such as mining and
logging. Another type of small-domain model is also used by the CES program, primarily in detailed industries.
15 Coronavirus disease 2019 (COVID-19) is caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). For
more information, see “COVID-19 science update” (Centers for Disease Control and Prevention, January 15, 2021), https://
www.cdc.gov/library/covid19/01152021_covidupdate.html.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

16 Lena H. Sun and Lenny Bernstein, “First U.S. case of potentially deadly Chinese coronavirus confirmed in Washington state,” The
Washington Post, January 21, 2020, https://www.washingtonpost.com/health/2020/01/21/coronavirus-us-case/.
17 For example, researchers at the Federal Reserve Bank of Minneapolis used smartphone data to create indexes that measure
changes in travel and social encounters, which showed broad declines through March and into April 2020. See Victor Couture, Jessie
Handbury, Jonathan I. Dingel, Kevin R. Williams, and Allison Green, “Measuring movement and social contact with smartphone data:
a real-time application to COVID-19,” Institute Working Paper 35 (Federal Reserve Bank of Minneapolis, Opportunity and Inclusive
Growth Institute, July 2020), https://doi.org/10.21034/iwp.35.
18 There is evidence that the formal shutdown orders only explain a small proportion of the decline in business activity. See Austan
Goolsbee and Chad Syverson, “Fear, lockdown, and diversion: comparing drivers of pandemic economic decline 2020,” Working
Paper 27432 (Cambridge, MA: National Bureau of Economic Research, June 2020), https://doi.org/10.3386/w27432.
19 For more information on restaurant reservation data, see the “State of the Industry” page on the OpenTable website at https://
www.opentable.com/state-of-industry.
20 Ohio became the first state to close schools on March 12, and 27 states and territories had done so by March 16. By March 25, all
U.S. public schools were closed. See “The coronavirus spring: the historic closing of U.S. schools,” Education Week, July 1, 2020,
https://www.edweek.org/ew/section/multimedia/the-coronavirus-spring-the-historic-closing-of.html.
21 Data on passenger throughput during the COVID-19 pandemic from the Transportation Security Administration are available at
https://www.tsa.gov/coronavirus/passenger-throughput.
22 The Business Cycle Dating Committee of the National Bureau of Economic Research determined that a peak in economic activity
occurred in February 2020, marking the end of the economic expansion that began in June 2009 and the beginning of a recession.
For more information, see “Determination of the February 2020 peak in U.S. economic activity” (Cambridge, MA: National Bureau of
Economic Research, June 8, 2020), https://www.nber.org/cycles/june2020.html.
23 For more information on regression models with autoregressive integrated moving average (regARIMA), see Kathleen M.
McDonald-Johnson and Catherine C. Hood, “Outlier selection for RegARIMA models” (U.S. Census Bureau, 2001), https://
www.census.gov/ts/papers/asa2001kmm.pdf.
24 For more information on the regression model used to modify birth–death forecasts, see Victoria Battista, “Back to the future: using
current regression variables to forecast forward from historical net birth/death employment,” Proceedings of the Government Statistics
Section, American Statistical Association, October 2013, https://www.bls.gov/osmr/research-papers/2013/st130140.htm.
25 See Huff and Gershunskaya, “Components of error analysis in the Current Employment Statistics survey.”
26 In retail trade and other services, the post-stratified estimates were produced at the three-digit NAICS subsector level. In leisure
and hospitality, the estimates became the sum of arts, entertainment, and recreation; accommodation; full-service restaurants; and
the remainder of food services and drinking places. In some cases, QCEW ratios were applied to the March leisure and hospitality
estimates to create a prior-month employment level from which to apply the weighted link relative.
27 For more details on the generalized small-area model, see Julie Gershunskaya and Terrance D. Savitsky, “Bayesian
nonparametric joint model for point estimates and variances under biased domain variances,” Proceedings of the Survey Research
Methods Section, American Statistical Association, November 2019, https://www.bls.gov/osmr/research-papers/2019/st190020.htm.

RELATED CONTENT

Related Articles
Employment projections in a pandemic environment, Monthly Labor Review, February 2021.
Employment recovery in the wake of the COVID-19 pandemic, Monthly Labor Review, December 2020.
How many workers are employed in sectors directly affected by COVID-19 shutdowns, where do they work, and how much do they
earn?, Monthly Labor Review, April 2020.

U.S. BUREAU OF LABOR STATISTICS

MONTHLY LABOR REVIEW

Related Subjects
Survey methods
Benchmarking
Nonresponse adjustment
Jobs
Statistical programs and methods
Employment
Experimental methodology
Survey procedures
COVID-19
Recession

Full text of Monthly Labor Review : April 2021

FRASER