Dispersion in Dispersion: Measuring Establishment‐Level Differences in Productivity

• Published date: 01 December 2023
• Author: Cindy Cunningham,Lucia Foster,Cheryl Grim,John Haltiwanger,Sabrina Wulff Pabilonia,Jay Stewart,Zoltan Wolf
• Date: 01 December 2023
• DOI: http://doi.org/10.1111/roiw.12616

bs_bs_banner

Review of Income and Wealth

Series 69, Number 4, December 2023

DOI: 10.1111/roiw.12616

DISPERSION IN DISPERSION: MEASURING ESTABLISHMENT-LEVEL

DIFFERENCES IN PRODUCTIVITY

BY CINDY CUNNINGHAM

U.S. Bureauof Labor Statistics

LUCIA FOSTER and CHERYL GRIM

Center for Economic Studies, U.S. Census Bureau

JOHN HALTIWANGER

University of Maryland, NBER, and IZA

SABRINA WULFF PABILONIA and JAY STEWART

U.S. Bureauof Labor Statistics and IZA

AND

ZOLTAN WOLF

New Light Technologies

We describe new experimental productivity dispersion statistics, Dispersion Statistics on Productivity

(DiSP), jointly produced by the Bureau of Labor Statistics (BLS) and the Census Bureau, that com-

plement the ofcial BLS industry-level productivitystatistics. The BLS has a long history of producing

industry-level productivitystatistics, which represent the averageestablishment-level productivity within

industries when appropriatelyweighted. These statistics cannot, however, tell us about the variationin

productivity levelsacross establishments within those industries. Dispersion in productivity acrossbusi-

nesses can provide information about the nature of competition and frictions within sectors and the

sources of rising wage inequality across businesses. DiSP data show enormous differences in produc-

tivity across establishmentswithin industries in the manufacturing sector. We nd substantial variation

in dispersion across industries, increasing dispersion from 1997 to 2016, and countercyclical total fac-

tor productivity dispersion. Wehope DiSP will enable further research into understanding productivity

differences across industries and establishmentsand over time.

JEL Codes: D24, E24, E32

Keywords:manufacturing, reallocation, business cycles, productivity dispersion

Note: Any opinions and conclusions expressed herein are those of the authors and do not rep-

resent the views of the U.S.Census Bureau or the Bureau of Labor Statistics. The Census Bureau has

reviewedthis data product for unauthorized disclosure of condential information and has approvedthe

disclosure avoidancepractices applied to this release (Approval IDs: DRB-FY19-393, DRB-FY19-526,

CBDRB-FY20-410). John Haltiwanger was also a Schedule A part-time employee of the U.S. Census

Bureau at the time of the writing of this paper. We thank John Earle, Kevin Fox, Bernd Görgiz, Bart

Hobijn, Mark Roberts,Jon Samuels, Chad Syverson, the Federal Economic Statistics Advisory Com-

mittee, the BLS Technical Advisory Committee, and participants at the Conference on Research on

Income and Wealthand the 2015 Federal Statistical Research Data Center Conference for their helpful

comments.

*Correspondence to: Cheryl Grim, Center for Economic Studies, U.S. Census Bureau

(cheryl.ann.grim@census.gov).

© 2022 International Association forResearch in Income and Wealth. This article has been

contributed to by U.S. Government employees and their workis in the public domain in the USA.

999

Review of Income and Wealth, Series 69, Number 4, December 2023

1. INTRODUCTION

Productivity measures are critical for understanding economic performance

in the U.S. economy. In this paper, we describe a new productivity data product,

Dispersion Statistics on Productivity (DiSP), jointly developed and published by

the U.S. Bureau of Labor Statistics (BLS) and the U.S. Census Bureau.1BLS pro-

duces the ofcial labor and total factor productivity (TFP) growth statistics for

major sectors and industries in the U.S.These statistics are constructed using aggre-

gate industry-level data and can be thought of as changes in the rst moment of

establishment-level productivity (appropriately weighted). That is, these statistics

show how productivity changes on average within sectors and industries, but they

cannot provide insight into the variationin productivity levels across establishments

within sectors or industries. Research has shown that changes in the dispersion of

productivity across establishments in the same industry is related to changes in the

growth of aggregate productivity (both economy-wide and at the industry level)

through a variety of channels.

To ll this void, BLS and theCensus Bureau initiated the Collaborative

Micro-productivity Project (CMP) to develop and publish experimental statistics

on within-industry productivity dispersion (i.e., second-moment measures of

establishment-level productivity) and to produce restricted-use research datasets.

The public-use statistics developed in this project, DiSP, were released for the rst

time in the fall of 2019. The rst release covered all four-digit NAICS industries

in the manufacturing sector and the years 1997–2016. The most recent release of

DiSP in the fall of 2021 extended the coverage to 1987–2018.2Moving forward,

the data will be updated annually. Restricted-use establishment-level data with

micro-based estimates of productivity and its underlying components (e.g., output

and input measures) are also available to qualied researchers on approved projects

in secure Federal Statistical Research Data Centers (FSRDCs).3

Economic theory and recent empirical evidence suggest that the second

moments of productivity are informative on several important dimensions. One

of the most important ndings in the literature on micro-level productivity

is that large productivity differences across establishments exist even within

narrowly dened industries.4For example, using data from the 1977 Cen-

sus of Manufactures (CM), Syverson (2004a) found that an establishment at

the 90th percentile of the within-four-digit-SIC labor productivity distribution

is on average about four times as productive as an establishment at the 10th

percentile.

Syverson’s ndings generated considerable interest in the causes and con-

sequences of this dispersion. Possible market explanations include the concavity

1DiSP data are available on both BLS and Census Bureau websites at: https://www.bls.gov/

productivity/tables/ and https://www.census.gov/disp. The DiSP is an experimental dataset and plans

are underway to expand the dataproduct in several ways as described in this paper.

2The additional earlier years in the 2021 release reect the ongoing efforts to enhance the DiSP

product.

3For more information on the FSRDCs: http://www.census.gov/fsrdc. An earlier version of this

dataset was analyzed in Fosteret al. (2016a).

4Syverson (2011) provides a surveyof this literature.

© 2022 International Association forResearch in Income and Wealth. This article has been

contributed to by U.S. Government employees and their workis in the public domain in the USA.

1000

Review of Income and Wealth, Series 69, Number 4, December 2023

of the prot function that prevents the most-productive business from taking

over an industry, frictions in factor adjustments (such as costs of adjusting input

factors), barriers to entry and exit, and distortions that inhibit the equalization of

marginal products across businesses (such as the regulatory environment). Drivers

of establishment-level productivity variation include differences in management

skills, the quality of production factors, innovation, and investments in R&D.

Research has shown that the dispersion inestablishment-level productivity

varies across sectors, bygeographic area, and over time. For example, Syver-

son (2004a,2004b) shows that variation in dispersion across industries and

geographic areas is related to product substitutability, market structure, and com-

petition. Hsieh and Klenow (2009) argue that both cross-country variation and

within-country variation in the dispersion in productivity are related to distortions

that inhibit productivity-enhancing reallocation. Asker et al. (2014) present evi-

dence that the patterns of dispersion reect dynamic factor adjustment frictions

within sectors. The ndings in Foster et al. (2016b) suggest that productivity dif-

ferences across establishments may be generated by differences in efciency levels,

demand shocks, frictions/distortions, or all of the above. Foster et al. (2021a)and

Cunningham et al. (2021) show that industries experiencing a surge in innovation

exhibit a burst of rm entry, followed by an increase in productivity dispersion

during an experimentation and shakeout phase, followed ultimately by an increase

in industry-level productivity.

Establishment-level productivity differences are also correlated with impor-

tant economic outcomes at the micro level, such as the survival and growth of

establishments. There is an extensive literature on the connection between produc-

tivity,reallocation, and growth (Baily et al., 1992; Griliches and Regev, 1995; Foster

et al., 2001; Diewert and Fox, 2010; Petrin et al., 2011; Foster et al., 2016a; Decker

et al., 2020; Blackwood et al., 2021). These studies show that moreproductive busi-

nesses are more likely to survive and grow. These ndings contribute to the per-

spective that reallocation—theprocess by which economic activity is allocated to its

highest valued use—is an important contributor to aggregate productivity growth.

Productivity dispersion is also important for understanding rising wage

inequality, which has been shown to be a between-rm phenomenon (Davis and

Haltiwanger, 1991; Barth et al., 2016; Song et al., 2019; Haltiwanger and Splet-

zer, 2020). In addition, several studies have found that high-wage establishments

are also more productiveand that rising between-establishment dispersion in wages

is closely associated with rising between-establishment dispersion in productivity

(e.g., Dunne et al., 2004). Economic theories of search and matching provide the

theoretical connection between productivity dispersion and wage dispersion (e.g.,

Burdett and Mortensen, 1998). Search and matching frictions create quasi-rents

for worker-rm matches that make it optimal for high-productivity rms to pay

high wages.

Our results using the DiSP experimental data conrm earlier ndings of size-

able differences in productivity across establishments within industries. To preview

our results, we nd that,on average, a manufacturing establishment at the 75th per-

centile of the within-industry labor productivity distribution is more than twice as

productive as an establishment at the 25th percentile. If we instead focus on TFP,

we nd that an establishment at the 75th percentile is almost twice as productive

© 2022 International Association forResearch in Income and Wealth. This article has been

contributed to by U.S. Government employees and their workis in the public domain in the USA.

1001