ArticlePDF Available

Workforce Diversity and Productivity: An Analysis of Employer-Employee Match Data

Authors:

Abstract

Calls for workforce diversity abound; arguments in support of diversity at the workplace suggest that we are a better society when we work together. In a recent opinion poll, 81% of respondents said that it is somewhat or very important “to have employees of different races, cultures and backgrounds in the workplace or businesses.” However, the actual economic costs and benefits to workforce diversity are unclear. The goal of this paper is to empirically assess the relationship between workforce diversity and the economic performance of an establishment. We use the New Worker-Establishment Characteristics Database (NWECD), a nationwide employer-employee matched data set, to estimate the association between productivity and workforce diversity. These data allow us to overcome many of the limitations of past studies because the NWECD is an establishment-level data set. Our main finding is that diversity is either positively associated with productivity or there is no significant relationship between diversity and productivity. We never find a significant negative relationship between establishment-level diversity and productivity. This leads us to conclude that establishments that employ a more diverse workforce are no less productive than establishments that employ a more homogeneous workforce.
Workforce Diversity and Productivity:
An Analysis of Employer-Employee Matched Data
by
Linda Barrington
Economics Program
The Conference Board
845 Third Ave.
New York, NY 10022-6679
and
Kenneth Troske
Department of Economics
University of Missouri-Columbia
118 Prof. Bldg
Columbia, MO 65211
July 2001
We would like to thank Robert McGuckin, David Neumark, and participants at the 2001 SOLE
conference for helpful suggestions, and Lucia Foster for providing some of the data. The work on
this paper was conducted while the authors were research associates at the U.S. Census Bureau.
All opinions, findings, and conclusions expressed herein are those of the authors and do not in
any way reflect the views of the Conference Board or the U.S. Census Bureau. The data used in
this article were collected under the provisions of Title 13 U.S. Code and are only available at the
Center for Economic Studies, U.S. Census Bureau. To obtain access to these data, contact the
Center for Economic Studies, U.S. Census Bureau, RM 211/WPII, Washington, D.C. 20233.
Abstract
Calls for workforce diversity abound; arguments in support of diversity at the workplace
suggest that we are a better society when we work together. In a recent opinion poll, 81% of
respondents said that it is somewhat or very important “to have employees of different races,
cultures and backgrounds in the workplace or businesses.” However, the actual economic costs
and benefits to workforce diversity are unclear. The goal of this paper is to empirically assess the
relationship between workforce diversity and the economic performance of an establishment.
We use the New Worker-Establishment Characteristics Database (NWECD), a nationwide
employer-employee matched data set, to estimate the association between productivity and
workforce diversity. These data allow us to overcome many of the limitations of past studies
because the NWECD is an establishment-level data set. Our main finding is that diversity is
either positively associated with productivity or there is no significant relationship between
diversity and productivity. We never find a significant negative relationship between
establishment-level diversity and productivity. This leads us to conclude that establishments that
employ a more diverse workforce are no less productive than establishments that employ a more
homogeneous workforce.
1
I. Introduction
Calls for workforce diversity abound; arguments in support of diversity at the workplace
suggest that we are a better society when we work together. In a recent opinion poll, 81% of
respondents said that it is somewhat or very important “to have employees of different races,
cultures and backgrounds in the workplace or businesses.”
1
Diversity appears to be a social goal
for the workplace, as well as schools and neighborhoods. Businesses are under pressure to
achieve diversity because it is the “right” thing to do and because a homogeneous workforce may
be evidence of discrimination.
While there is pressure on businesses to achieve a diverse workforce, there is very little
existing evidence on the effects of diversity on firm performance. One obvious question is
whether we should expect any relationship between diversity and firm performance. If firms were
being forced to hire diverse workers through governmental affirmative action policies, and firms
were currently discriminating against either female or minority workers, then diversity could
have a positive impact on firm performance because firms would be forced to hire workers
whose wages were less than the value of their marginal product. Alternatively, if there were no
discrimination in the labor market, if it were more costly to recruit and hire qualified minority
and female workers, or if firms must spend resources on programs, such as diversity training, to
overcome the discriminatory preferences of existing workers, then diversity may have a negative
impact on firm performance.
In a non-discriminatory market, there could also be a positive relationship between
diversity and firm performance, prompting profit-maximizing businesses to voluntarily seek
diverse workers. To the extent that the general public views a diverse workforce as a desirable
attribute for a firm, firms that have this attribute may acquire goodwill capital or brand loyalty.
2
This could be viewed as a way for firms to decrease the elasticity of demand for their product,
1
American Council on Education, “New National Poll on Diversity” Press Release, August 13, 2000 (Washington,
DC). See http://www.acenet.edu/news/press_release/2000/08august/diversity_poll.html.
2
Certainly negative public-opinion campaigns directed at businesses viewed as anti-women or anti-minority can
inflict high costs. A recent example is the Fall 2000 case of the regional company, Maurice’s Gourmet Barbecue
Sauce. Within weeks of the first public criticism of founder and chief Maurice Bessinger’s “racist” actions
(including selling propaganda advocating slavery on biblical grounds at the company’s headquarters), Maurice’s lost
the majority of its largest vendors who had been providing up to half its sales revenues (DiversityInc.Com, Sept. and
Oct. 2000, Allegiant Media Inc.).
2
thereby allowing them to charge a higher price and increase profits (Navarro, 1988). In addition,
profit maximization may require workforce diversity if a firm sells its product in many markets
and diversity helps to understand different market preferences (Osborne, 2000). In the end,
whether the net economic benefits of workforce diversity are positive or negative is an empirical
question.
The goal of this paper then is to examine the empirical evidence on the relationship
between workforce diversity and the economic performance of an establishment. In this study
we narrow the definition of economic performance to establishment-level labor productivity and
empirically assess the association between workforce diversity and labor productivity. Data
constraints have limited most previous research in this area to cross-industry comparisons using
aggregated statistics, or to case studies, making the results vulnerable to criticism. Cross-
industry analyses do not detect production-related intra-industry variation, which can be
substantial (Haltiwanger, Lane, and Spletzer, 2000). Furthermore, it is not uncommon for such
studies to omit average operational scale or other explanatory factors of production. Case
studies, on the other hand, may benefit from richer production-related data, but the degree to
which the results of such studies are applicable beyond the business or industry being studied is
limited.
We use the New Worker-Establishment Characteristics Database (NWECD), a
nationwide employer-employee matched data set, to estimate the association between
productivity and workforce diversity. These data allow us to overcome many of the limitations
of past studies because the NWECD is an establishment-level data set. The NWECD matches
workers who responded to the long form of the 1990 Decennial Census to the establishments
where they work, using the U.S. Census Bureau’s Standard Statistical Establishment List (SSEL).
The NWECD includes workers and establishments from all sectors of the economy. Thus, we
provide new estimates of the relationship between diversity and productivity using a large data
set containing establishments operating in nearly all regions and industries. Consequently, we
can examine the relationship between diversity and productivity separately for workers in
different industries. We focus here on the three largest industries in our data: manufacturing,
retail trade, and services. Our primary measure of productivity is labor productivity or log value
added per worker (measured as total sales minus total payroll divided by the number of workers
3
in the plant). We also construct an alternative labor productivity measure as log total sales per
worker (total sales divided by the number of workers in the plant). One shortcoming of these
data is that they do not contain measures of capital stock or material inputs for establishments in
sectors other than manufacturing, limiting us to using labor productivity measures. To the extent
that the characteristics of workers in an establishment are correlated with these other inputs, this
may bias our results. Using the data on manufacturing plants, we also examine the relationship
between multi-factor productivity and workforce diversity.
The richness of the NWECD allows us to construct multiple measures of workforce
diversity. We introduce a unique measure of workforce diversity and compare the results found
using this new measure with results found using more traditional measures of workforce
composition, namely, “percent white men,” “percent minority men,” etc. The unique diversity
measure we introduce has two components. The first component captures the combined race and
gender diversity of the establishment’s workforce; the second captures how that establishment’s
diversity is spread across occupations. The use of more than one measure of workforce diversity
provides an important sensitivity test and reveals that how one defines diversity is important.
Our main finding is that diversity is either positively associated with productivity or there
is no significant relationship between diversity and productivity. We never find a significant
negative relationship between establishment-level diversity and productivity. This result holds
for establishments in all three industries and is robust across our alternative measures of
productivity and diversity. This leads us to conclude that establishments that employ a more
diverse workforce are no less productive than establishments that employ a more homogeneous
workforce.
The rest of the paper is organized as follows. In the next section we discuss related
studies. Section III contains a discussion of our data and measurement issues. Our results are
presented in Section IV. Section V concludes and presents some caveats.
II. Related Literature
In considering the relationship between workforce diversity and economic performance,
the question that first arises is “Why should there be a relationship between diversity and
economic performance?” The answer to this question depends in part on whether or not
4
discrimination is present in hiring practices. Discrimination—hiring workers based on race or
gender instead of ability—results in a firm’s hiring workers who are paid higher wages than
alternative workers, but are no more productive (Becker, 1971). Forcing firms to become more
diverse through governmental affirmative action policies could then increase economic
performance because businesses would be forced to hire workers whose wage is less than or
equal to the value of their marginal product.
3
If discrimination were not present in hiring
decisions however, forcing businesses to hire women and minorities could result in the hiring of
less productive workers, leading to lower productivity for the firm.
However, even if forcing firms to hire diverse workers did result in increased
productivity, there could be costs associated with increasing workforce diversity that could offset
any gain. Such potential costs include increased costs associated with recruiting and hiring
women and minorities (Holzer and Neumark, 2000b), and costs associated with getting a
heterogeneous workforce to work together (Osborne, 2000). If workers self-select into
employment based on the race, gender, or religion of co-workers, a firm attempting to create a
diverse workforce may lower the utility of existing workers who may in turn shirk or take other
actions costly to the firm. The firm may therefore have to incur other costs, such as providing
diversity-training courses, in an effort to change the preferences of current workers. A diverse
workforce could also negatively impact productivity if diversity introduces communication and
cooperation challenges not present among homogeneous workers (Lang, 1986).
Holzer and Neumark (2000a) and Altonji and Blank (1999) provide recent and extensive
reviews of the evidence on discrimination in the labor market. Holzer and Neumark conclude,
“that the overall evidence points to some continuing discrimination against women and blacks”
(p 493), although the evidence is by no means conclusive. Continuing discrimination against
women and blacks would suggest performance gains could follow from increased workforce
diversity.
3
Firms with more than 50 employees who do business with the federal government, and all firms with more than 100
employees, are required to file yearly reports with the Equal Employment Opportunity Commission (EEOC) on the
percent of women and minority workers in the firm. These reports can then be used by the EEOC to file
discrimination lawsuits (Bloch, 1994). In the opinion of Bloch (1994, p. 105), “noncontractors required to file EEO-
1 reports that are monitored by the EEOC … would not be acting rationally if they were to avoid hiring minorities
and women.”
5
The most significant difference between affirmative action, which is the focus of Holzer
and Neumark’s (2000a) review, and our focus on workplace diversity is that affirmative action is
designed to produce social gains, whereas from the firm’s perspective, only private gains are
relevant. To the degree that the performance gains from increased diversity in a discriminatory
environment can be internalized by the employer, and not off-set by accompanying higher costs,
a positive relationship should be found between workforce diversity and economic performance.
However, there may be other private gains from increasing workforce diversity. Navarro
(1988) presents a theory in which firms use corporate charitable giving as a way to promote the
firm’s image. This in turn provides goodwill capital to the firm that acts to either shift out or to
decrease the elasticity of the firm’s demand curve. In either case, charitable giving is a profit
maximizing act on the part of the firm. To the degree that workforce diversity is seen as a
desirable social goal, the firm could view having a diverse workforce as another way to invest in
goodwill capital.
Alternatively, workforce diversity may actually contribute to the production of a higher-
quality or unique product.
4
Osborne (2000) argues that profit maximization may lead to
workforce diversity if a firm sells its product in many markets, and that product has multiple
characteristics that are valued differently in the different markets.
5
The firm will optimize its
workforce diversity with respect to “the product’s characteristics, the extent to which different
markets value them, and the extent to which [demographically diverse] groups inherently differ
in their capacity to provide” these characteristics (p 473).
Data constraints have limited most research attempting to empirically examine the
relationship between diversity and productivity to cross-industry comparisons using aggregated
statistics or to case studies. For example, Heckman and Payner (1995) study the textile industry
4
A diverse workforce producing a better product than that produced by a homogeneous workforce is analogous to
the argument that diverse student bodies in colleges and universities produce better educational experiences.
Students, being an input into the educational end product, as well as consumers thereof, the argument goes, will
create a richer (better) educational product if they are more diverse (see Holzer and Neumark, 2000a, p 528). In the
case of busing-style integration, however, it is also argued that positive peer effects from privileged students will
benefit the less privileged, without net harm to the education of the more privileged. This education-focused
argument does not have a clear business analogy.
5
One example of this may be General Motors’ recently announced initiative to increase the number of women in
charge of car dealers so that the characteristics of their dealers better matches the characteristics of their customers,
50 percent of whom are women (St. Louis Post-Dispatch, March 21, 2001, p. C8.).
6
in South Carolina and show that affirmative action programs increased black employment during
a period of increasing productivity in textiles. In a 1986 study of medical school graduates, Penn,
et al. (1996) find that minority-recruitment graduates saw more patients per day and were more
likely to serve rural and inner city communities and low-income patients, both possible measures
of productivity. And, in a 1987 study of police departments, Steel and Lovrich (1987) find that
departments with active affirmative programs showed no statistical difference in crime rate,
clearance rate (crimes solved), and operation costs. However, in a study of the effect of
affirmative action policies on police departments, Lott (2000) finds that increasing police force
diversity negatively affects productivity.
Studies using aggregate data, such as state level manufacturing data or inter-industry data,
have generally found no negative repercussions from increased workforce diversity. In analysis
of state level manufacturing data over the period 1966 to 1977, Leonard (1984) finds that
changes in the female or minority employment percentages had no significant impact on industry
productivity after controlling for region, industry, capital stock, and the percentage of the
industry employment that was blue collar. A study of 57 industries revealed changes in the
female or black employment percentages between 1984 and 1988 had no significant impact on
industry productivity (Conrad, 1995).
Research using employer-employee matched data to study the relationship between
diversity and productivity, however, is almost non-existent. One exception is the Holzer and
Neumark (2000b) study which examines data from a 1992-1994 survey of 3200 employers in
Atlanta, Boston, Detroit, and Los Angeles on the establishments’ affirmative action policies and
the last employee hired. These data revealed that employers who have active affirmative action
policies in recruitment seem to locate more-qualified candidates who are women and minorities.
Employers that have active affirmative action policies in hiring do hire women and minorities
with somewhat weaker credentials, but the performance of these employees is not weaker than
that of their peers. Performance is measured by the supervisor’s rating of the employee, not by
any measure of total establishment output. Additional training and evaluation programs for
affirmative action hires may have been responsible for the satisfactory productivity of those hired
with weaker credentials (Holzer and Neumark, 2000b).
7
The New Worker-Establishment Characteristics Database (NWECD) that we use for this
study can greatly improve upon past analysis of the relationship between workforce diversity and
productivity by allowing us to simultaneously control for establishment and worker
characteristics and eliminating problems introduced by using aggregated data. Work using
employer-employee matched data have already proven a rich source for studies related to worker
characteristics. Hellerstein, Neumark and Troske (1999) investigate the role that age, race, and
gender play in accounting for wage and productivity differences across plants. Among other
results, they find that “women’s wages fall short of men’s by considerably more than can be
explained by their lower marginal productivity. This is consistent with the standard wage
discrimination hypothesis.” (p 433). With regard to African American manufacturing workers,
however, they find no evidence of wage discrimination.
A different empirical test of the relationship between economic performance and
workforce diversity is to let the market “speak for itself.” If a company’s stock price signals
economic performance—present or future—then the movement of a company’s stock price vis-à-
vis its workforce diversity should reveal if a relationship exists between workforce diversity and
economic performance. Wright, et al. (1995) and Welbourne (2000) both find that the stock
market rewards owners of firms for increased diversity. Wright, et al. report that companies
recognized by the U.S. Department of Labor for having an exemplary affirmative action program
experience an increase in stock price immediately after the announcement. Welbourne reports
that having a diverse management team at the time of the initial public offering of a company’s
stock results in a higher stock price than that received by similar companies with more
homogeneous management teams. The results from both of these studies suggest that the stock
market does see net financial benefits accruing to companies with greater diversity.
It is worth noting that the positive relationship between workforce diversity and stock
price does not show exactly what the market is rewarding. It may be that the market interprets
diversity as providing a revenue edge through increased goodwill capital. Alternatively, it may
be that diversity is viewed simply as a proxy for good management—a signal that this must be a
well-managed company because well-managed companies are more diverse, not that diversity
8
will make the company more profitable.
6
The lack of causal evidence on the relationship
between diversity and economic performance is a point we pick up later when discussing our
own results.
III. The Data and the Measurement of Variables
The New Worker Establishment Characteristic Database
The NWECD is a cross-sectional data set linking workers’ responses to the 1990
Decennial Census long form to establishment data drawn from various economic censuses.
Since these data are constructed in the same way as the original Worker Establishment
Characteristics Database (Troske, 1998), and are documented extensively elsewhere (Bayard,
Hellerstein, Neumark, and Troske, 1998), we only briefly describe these data. The NWECD was
constructed by matching worker records from the 1990 Sample Edited Detail File (SEDF) to
establishment records from the 1990 Standard Statistical Establishment List (SSEL). The 1990
SEDF consists of all household responses to the 1990 Decennial Census long form. Individuals
receiving the long form were asked to identify each employed household member’s occupation,
employer location, and employer’s industry. The Census Bureau then assigned occupational,
industrial, and geographic codes to these long-form responses. Thus, the SEDF contains
standard demographic information for all respondents as well as detailed location and industry
information for each respondent’s place of work.
The SSEL is a complete list of all establishments in the U.S. in a given year and is used
by the Census Bureau to administer various economic censuses and surveys. The SSEL contains
detailed location information for all establishments along with a four-digit SIC code and a unique
establishment identifier that is common to other Census Bureau economic data.
Worker records from the SEDF are matched to employer records in the SSEL using the
common industry and location information for employers available in both data sets. Briefly, the
matching process proceeds as follows. First, only establishments that are unique to an industry-
location cell are kept. Then all workers who indicate they work in the same location-industry
6
See Cappelli and Neumark (1999), and Black and Lynch (forthcoming, 2001) for studies that try to isolate
managerial practices from other causes of better economic performance.
9
cell as a retained establishment are linked to that establishment. These matched data are what are
contained in the NWECD. Because the SEDF contains only a sample of workers, and because
not all workers were matched, the NWECD contains a sample of workers in any given
establishment.
To obtain the data used for this project we impose several restrictions on the NWECD
data. First, we only keep establishments with more than 25 employees. Measures of per worker
earnings taken from the SSEL and SEDF are closer for larger than smaller establishments; thus
we believe the data are of higher quality for larger establishments. In addition, again because of
concerns about data quality, we eliminate workers in establishments where the average salary is
greater than $600,000. To obtain additional information about a worker’s employer, we match
the NWECD to establishment-level data from the 1987 Censuses of Manufacturing, Retail Trade,
and Services. We choose to focus on workers in these industries since they are the industries
containing the largest number of workers in our data.
Evidence presented in Bayard, Hellerstein, Neumark and Troske (1998) suggests that
workers are correctly matched to establishments, and that wage regressions using these matched
data are unlikely to be biased from sample selection associated with the matching process, even
though the data set is non-representative. One of the main sources of this non-representativeness
is that the NWECD over-samples workers in manufacturing and under-samples workers in retail
trade. This fact provides additional justification for why we choose to analyze workers
separately by industry.
Workforce Demographics
From the NWECD we construct various establishment-level measures of workforce
characteristics relevant to our analysis. Using the data on matched workers we construct
measures of the workforce demographics of the establishments. These include percent white
male workers, percent minority male workers, percent white female workers, percent minority
female workers, percent older workers, and percent college graduates. Each of these is measured
as the number of workers matched to the establishment in a given group divided by the total
number of workers matched to the establishment. We classify workers of all races other than
10
white, along with all Hispanic workers, as minority workers. In this paper, white workers means
non-Hispanic white workers. Older workers are defined as those over 50 years old.
Payroll and Occupational Diversity Measures
Previous studies of the effect of diversity have used measures of workforce characteristics
such as percent white male, percent minority male, percent white female and percent minority
female to capture diversity.
7
However a limitation of these measures when applied to
establishment-level data is that they do not scale the characteristics of the establishment
workforce against a relevant comparison group such as all workers in the industry. Neither do
they consider how the establishment’s diversity is spread across occupations. We therefore
construct an additional and novel measure of workforce diversity. This measure gauges the
establishment’s workforce diversity through two distinct comparisons. The first comparison
answers the question “How diverse is the establishment’s workforce, or payroll, relative to the
workforce in the establishment’s industry?” We refer to this as the payroll diversity term. The
payroll diversity term gauges the overall workforce diversity of the establishment relative to that
in its industry using a ratio of the distributions to capture the scale of the deviation between the
establishment and its baseline.
8
When constructing this measure we used as our measure of the
industry workforce the percent of white males, nonwhite males, white females, and nonwhite
females that appear in our NWECD data at the one-digit SIC industry level.
The second comparison, composing our occupational diversity term, answers the question
“How evenly is the diversity present in the establishment spread across occupations?” The
occupational diversity term captures the occupational distribution of the establishment’s
workforce diversity by measuring the difference between the establishment’s actual occupational
7
The advantage of these measures is that they are both easy to compute and easy to interpret.
8
See MacKenzie (1999) for a measure of diversity built on the difference of distributions. We have also constructed
diversity measures that compute the overall workforce diversity of an establishment relative to that in the
establishment’s region, and to that present in the NWECD (nationwide) sample. We also used as the baseline the
characteristics of workers in the entire SEDF data (which are a true random sample of the population). The
estimated relationship between these additional diversity measures and productivity are qualitatively similar to the
relationship between our primary diversity measures and productivity. We report the results using the diversity
measure that compares establishment-level diversity relative to that of its industry because the industry workforce is
a reasonable proxy for the establishment’s relevant labor pool, and because of the similarity of results using regional
and national baselines.
11
distribution of workers for each race-gender category and a hypothetical distribution where the
establishment’s workforce is spread across occupations independent of race and gender. As in
the payroll diversity term, the scale of the deviation between the establishment’s observed
distribution and baseline distribution is captured by the ratio of the observed to the baseline
distribution. This measure allows us to distinguish between employers that employ a more
diverse workforce but segregate workers by occupation within the establishment and employers
that have a diverse workforce within occupations. We use the EEOC’s nine occupational
categories when defining occupation. To repeat, the first term captures how diverse the
workforce is compared to the workforce of the industry (the payroll diversity), and the second
term captures how evenly the diversity present in the establishment is spread across occupations
(occupational diversity).
9
The arithmetic formula used to translate the ratios into a single numeral for each term was
devised to produce an index that ranged from 0 to1. A value of 1 for the payroll diversity term
indicates that the race-gender diversity of the establishment perfectly matches the baseline
distribution. A value of 1 for the occupational diversity term indicates that the establishment’s
employees of each race-gender group appear in each occupation in the same proportions as they
appear on the payroll. The exact formulation of this measure appears in the Appendix.
Productivity and Establishment Characteristics
Our two primary measures of labor productivity are constructed using the Economic
Census data. They are, 1) the log of value added (measured as total sales minus total payroll)
divided by the total number of workers in the establishment, and 2) the log of total sales divided
by the total number of workers in the establishment. One problem with these measures is that
they do not control for differences in other inputs such as capital stock and materials.
Unfortunately, information on capital stock and other materials is only collected for
manufacturing establishments. For manufacturing, then, we construct a third measure of
9
We have also constructed another measure of diversity based on the Chi-Squared test statistic. There are again two
terms to this measure, the payroll diversity term and the occupational diversity term. The payroll diversity term is
constructed as the test statistic from a test of independence in a contingency table constructed using the distribution
of workers in the establishment as the observed values and the distribution of workers in the relevant baseline as the
expected values (DeGroot, 1986). The occupational diversity term is constructed in a similar fashion. Both terms
are bounded below by zero and have no upper bound. Analysis using this Chi-Squared diversity measure leads to the
same conclusions as analysis with our diversity measure (results available upon request).
12
productivity, multi-factor productivity. This measure should allow us to judge how missing other
inputs in the services and retail trade industries affect our results. Multi-factor productivity is
measured as:
ln ln ln ln ln
iiKiLiMi
M
FP Q K L M
ααα
=− (1)
where Q
i
is real output from establishment i, K
i
is real capital stock, L
i
is total hours worked, and
M
i
is real value of materials.
10
The α’s are the factor elasticities. Factor elasticities are measured
as the share of the cost of the input relative to the total cost of all inputs in each four-digit
industry. All of the data come from the 1987 Census of Manufacturers.
Finally, we use the establishment data to construct controls for other establishment
characteristics that may affect establishment-level productivity such as location in the four
Census regions (Northeast, Midwest, South and West), four-digit SIC industry, ownership
structure (whether the establishment is part of a single-establishment firm) and size (measured as
log total employment).
Table 1 presents summary statistics for the establishments in our data set. One thing to
note is that the establishments in our data are fairly large on average. This is due to having
eliminated the smaller establishments from the data because of quality concerns, and because
larger establishments are more likely to have matched workers and therefore be included in the
NWECD (Bayard, Hellerstein, Neumark and Troske, 1998).
IV. Regression Estimates
In the analysis that follows we present results from estimating the following regression
equation:
ln
ii
Y
β
δε
=++
ii
XZ (2)
10
This is identical to the measure of multi-factor productivity used in Foster, Haltiwanger, and Krizan (1998). See
the discussion there for a more in-depth description of how this measure is constructed. We would like to thank
Lucia Foster for providing us with these data.
13
where Y
i
is our establishment-level productivity measure, either log value-added per worker or
log sales per worker (and for manufacturing establishments log of multi-factor productivity), X
i
is a matrix containing our establishment-level measures of workforce diversity, Z
i
is a matrix
containing other establishment-level controls, and e
i
is an establishment-specific error term.
Included in Z
i
are, log total employment (Log (TE)), the percent of workers in the plant that are
over 50 (Percent Older Workers), the percent of workers in the plant that are college graduates
(Percent College Grad.), a dummy variable indicating whether the establishment is part of a firm
with no other establishments (Single-Unit), dummy variables indicating which of the four census
regions the establishment is located, and a set of dummy variables indicating in which four-digit
SIC code the establishment is classified. We estimate equation (2) separately for establishments
in manufacturing, services, and retail trade.
Tables 2 through 4 contain our main results. Presented in Table 2 are the regression
results using our novel measure of workforce diversity. Focusing first on manufacturing
establishments, we see that the coefficient on payroll diversity is positively and significantly
related to productivity (at 99%) whether productivity is measured by log(value-added/TE) or
log(Sales/TE). However, there is no significant relationship between occupational diversity and
productivity in either regression.
Looking at the coefficients on the other variables in columns (1) and (2) we see results
that are fairly standard. Larger establishments are more productive than smaller establishments
and establishments that are part of single-unit firms are significantly less productive than plants
that are part of multi-unit firms. Finally, we see that the percent of older workers is negatively
related to productivity while the percent of college graduates is positively related to productivity.
The former result is consistent with the results in Hellerstein, Neumark and Troske (1999, 2001)
who use the WECD to examine the relationship between worker characteristics and
establishment-level productivity and profitability. One concern we have about the estimated
relationship between the percent of older workers and productivity is that this negative
relationship may be due to the fact that we are not controlling for capital and other materials used
in production. It may be that older workers tend to work in establishments with older, less
productive capital. As we indicated above, since we do not have any establishment-level
measures of capital or materials for establishments in services or retail trade, we are unable to
14
investigate this hypothesis in these industries.
11
However, since we do have this information for
manufacturing establishments, we examine this issue below for these establishments.
The results for service establishments mirror the results for manufacturing
establishments. Payroll diversity is positively and significantly correlated with both productivity
measures while occupational diversity is not significantly related to productivity. The
relationship between our other controls and productivity are also quite similar to the relationships
estimated using manufacturing establishments.
The estimated relationship between our diversity measures and productivity is notably
different for establishments in retail trade. Here neither payroll diversity nor occupational
diversity is significantly related to productivity.
12
Our conclusion based on the results in this
table is that plants that are more diverse, at least as measured in this fashion, are either slightly
more productive, or at least as productive, as less diverse establishments.
One issue about which we are concerned is the role of measurement error in our estimates
in Table 2.
13
Since we only have a matched sample of workers in an establishment, our diversity
measure for each establishment is only an estimate of the true diversity in the establishment.
However, in our regressions we treat these estimates as known variables. In order to try and
judge the possible effect of measurement error on our estimates we repeat our previous analysis
focusing on establishments with 500 or more employees, what we term large establishments.
Given that a larger number and a more representative sample of workers are matched in large
establishments (Bayard, Hellerstein, Neumark, and Troske, 1998), the measurement error should
be smaller among these establishments. Unfortunately, we have very few establishments in retail
trade with more than 500 workers. Therefore, we re-estimate equation (1) focusing only on large
establishments in manufacturing and services. The results from these regressions using the
payroll and occupational diversity measures are found in Table 3.
11
Information about capital stock, capital investment and other materials used in production is only asked of
establishments in the Census of Manufacturers.
12
The association between productivity and both the percent college grad. and the single-unit dummy is also
insignificant for retail trade establishments.
13
See Hellerstein, Neumark and Troske (1999) and the reference therein, for a more complete discussion of the
issue of measurement error in this context, and for an assessment of the possible effects of measurement error in a
similar setting.
15
Looking at the results for manufacturing and service establishments we see that among
large establishments the payroll diversity measure remain positively and significantly related to
both productivity measures. Now however, occupational diversity is also positively and
significantly associated with productivity in manufacturing, suggesting that our concerns
regarding measurement error were valid. Limiting our sample to establishments with more than
500 workers increased the likelihood of matching more workers in each occupational category.
The result is a stronger estimated relationship between productivity and diversity across
occupations within manufacturing establishments. These results suggest that, at least for large
manufacturing establishments, not only are more diverse establishments more productive but so
are establishments that have diverse workforces within occupation.
Among large service establishments, however, the positive relationships between
occupational diversity and our measures of productivity do not become significant when the
analysis is limited to large establishments.
In Table 4 we present regression results using more traditional measures of diversity. In
particular, we include the percent of workers in the establishments that are white male, minority
male, and white female to capture the workforce demographic mix. The percent of workers that
are minority female is our excluded group. The results using this more traditional measure of
diversity present a less clear picture of how diversity is related to productivity. Looking at the
results for manufacturing shows that establishments with a higher percentage of male workers
relative to minority female workers are more productive, suggesting that less diverse
establishments are more productive. However, the positive association between male workers
and productivity is significant for both the percent of white male workers and the percent of
minority male workers. While the expected productivity of an establishment with more nonwhite
males relative to nonwhite females may be higher, the question remains, is the establishment
more diverse?
The problem with these traditional measures is two-fold. First, as mentioned above,
measures such as percent white male do not capture how the characteristics of the workforce in
an establishment compares to the relevant workforce or labor market. If the representation of
minority males on an establishment’s payroll is greater than their representation in the relevant
comparison pool, one could reasonably argue that reducing the share of minority men on the
16
payroll should increase, not decrease, diversity. Any measure of diversity should be based on a
comparison between the characteristics of workers in some unit relative to the relevant base.
Second, in these regression estimates, such measures limit “diversity” to a series of pairwise
comparisons: minority women vis-à-vis white men, minority women vis-à-vis minority men,
minority women vis-à-vis white women.
14
More reasonable, and representative of the concerns
of employers, would be to calibrate the relative balance of all groups of workers present on the
payroll simultaneously. We view these two shortcomings of the traditional measures as being
quite serious. It is for these reasons that we devised our alternative measure of diversity, and find
the results of Table 2 more illuminating than those of Table 4.
Finally, Table 5 presents our results from estimating equation (1) using log multi-factor
productivity as our dependent variable and focusing exclusively on manufacturing
establishments. Column (1) presents results using our payroll and occupational diversity
measure while column (2) presents results using the traditional diversity measures. As with the
results reported in Table 3 for large manufacturing establishments, the estimated relationship
between productivity and workforce diversity is stronger here than in Table 2. Looking first at
column (1) we see that, after adjusting for differences in capital and materials, both the payroll
and occupational diversity measures are positively and significantly related to productivity.
Therefore, it appears to be the case that the estimated relationships between payroll diversity and
productivity in Table 2 are not due to the fact that we failed to control for cross-establishment
differences in capital or materials, at least for manufacturing establishments. The omission of
controls for differences in capital or materials does, however, weaken the association between
occupational diversity and productivity.
Controlling for cross-establishment differences in capital or materials also affects the
estimated relationship between productivity and the presence of older workers as well as the
traditional measure of workforce diversity. The coefficients on percent older workers in columns
(1) and (2) and the coefficients on the traditional measures of diversity (percent white male
workers, percent minority male workers and percent white female workers) in column (2) are not
significantly related to productivity after controlling for cross-establishment differences in capital
14
One could also have omitted white men instead of minority women, which, while more common, does not alter the
underlying dichotomous structure of the definition.
17
and materials. Apparently the significant relationship between these variables and productivity
seen in Tables 2, 3 and 4 is a function of failing to control for differences in the use of capital and
materials in manufacturing establishments.
Interestingly, after controlling for capital and materials, there remains a positive
coefficient on the percent college grad. variable. This may be attributed to the fact that we have
not controlled for differences in the cost of labor in these regressions (in equation (1) L
i
is
measured in hours, not dollars).
V. Conclusion
To examine the relationship between diversity and productivity we use a recently
constructed employer-employee matched database, the NWECD, and construct a unique measure
of workforce diversity. We find that diversity is either positively associated with establishment-
level productivity, or there is no significant relationship between diversity and establishment-
level productivity. The empirical weight, however, appears to lean more heavily on the side of a
positive association, at least for manufacturing. We never find a significantly negative
association between productivity and our measure of diversity. To the extent that governmental
affirmative action policies are forcing employers to hire a more diverse workforce, these results
suggest that affirmative action policies do not hurt establishment performance. To the extent that
firms are voluntarily hiring more diverse workers, these results suggest that it is possible to do so
without hurting establishment productivity.
However, there are a number of important caveats to consider. First, we may miss some
important costs or benefits that are associated with employing more diverse workers. Previous
work suggests that it may be more costly to recruit qualified minority and female workers
(Holzer and Neumark, 2000b). In addition, firms may have to spend resources on programs,
such as diversity training, in order to successfully incorporate minority and female workers into
their workforce. To the extent that we miss these costs, then we may overstate the relationship
between diversity and productivity. Alternatively, if one of the main benefits of workplace
diversity is an increase in brand loyalty or good will capital on the part of customers—a benefit
that is not captured in the current output price—we may understate the relationship between
diversity and firm performance. Third, given that the NWECD is a cross-sectional database, we
18
are unable to examine the causal relationship between diversity and productivity. One possible
explanation for our findings is that more talented managers achieve a diverse workplace while
simultaneously producing more output with the same inputs than less talented managers. Highly
productive firms are more diverse, but becoming diverse will not necessarily make a firm highly
productive. Finally, again due to data limitations, we are unable to examine how establishments
move from having a homogeneous workforce to having a heterogeneous workforce and whether
certain management or organizational practices achieve both a diverse and more productive
workforce. Clearly, addressing this and the previous limitation requires panel data on
establishments, along with data on management practices in the establishment—data we hope to
create in the future. For now the results in this paper should be interpreted as being descriptive
of the diversity-productivity pattern in manufacturing, retail trade, and service establishments.
Businesses seem to be devoting real resources to promoting (or at least to creating the
appearance of promoting) a diverse workforce. A recent survey conducted by the Society of
Human Resource Managers found that almost 50 percent of the HR managers surveyed worked
for a firm with a diversity program. This leads to two related questions. 1) Why should a firm
want to employ a diverse workforce? 2) What effect does employing a diverse workforce have
on the performance of the firm? We have provided a number of possible answers to question 1,
however, our primary emphasis in this paper has been examining question 2. Our hope is that by
examining the effect diversity has on productivity, we will begin to understand why firms may
want to employ a diverse workforce. While our results are primarily descriptive, given the lack
of existing evidence in this area, we feel that these results provide valuable new insights into the
establishment-level relationship between diversity and performance.
19
References
Altonji, Joseph and Rebecca Blank. 1999. “Race and Gender in the Labor Market,” in Handbook
of Labor Economics, Vol. 3C, Orley Ashenfelter and David Card, editors (Amsterdam:
North Holland): pp. 3143-259.
Bayard, Kimberly, Judith Hellerstein, David Neumark, and Kenneth Troske. 1998. "New
Evidence on Sex Segregation and Sex Differences in Wages from Matched Employee-
Employer Data." Mimeograph.
Becker, Gary. 1971. The Economics of Discrimination, Second Edition, (Chicago, IL: University
of Chicago Press).
Black, Sandra E. and Lisa M. Lynch. Forthcoming, 2001. “How to Compete: The Impact of
Workplace Practices and Information Technology on Productivity.” Review of
Economics and Statistics.
Bloch, Farrell. 1994. Antidiscrimination Law and Minority Employment, (Chicago, IL:
University of Chicago Press).
Cappelli, Peter and David Neumark. 1999. “Do ‘High Performance’ Work Practices Improve
Establishment-Level Outcomes?” NBER Working Paper No. 7374, October.
Conrad, Cecilia. 1995. “The Economic Cost of Affirmative Action” in Economic Perspectives on
Affirmative Action, Margaret Simms, editor (Washington, DC : Joint Center for
Political and Economic Studies).
DeGroot, Morris. 1986. Probability and Statistics, Second Edition, (Reading, MA: Addison-
Wesley).
Foster, Lucia, John Haltiwanger and C.J. Krizan. 1998. Aggregate Productivity Growth: Lessons
from Microeconomic Evidence.” NBER Working Paper No. 6803, November.
Haltiwanger, John C. Julia I. Lane and James R. Spletzer. 2000. “Wages, Productivity, and the
Dynamic Interaction of Businesses and Workers.” Mimeograph. (October 5, 2000).
Heckman, James and Brook Payner. 1989. “Determining the Impact of Federal
Antidiscrimination Policy on the Economic Status of Blacks: A Study of South Carolina.”
American Economic Review
, Vol. 79, (March):pp. 138-177.
Hellerstein, Judith, David Neumark, and Kenneth Troske. 1999. “Wages, Productivity, and
Worker Characteristics: Evidence from Plant-Level Production Functions and Wage
Equations.” Journal of Labor Economics
, Vol. 17, No. 3, (July): pp. 409-446.
Hellerstein, Judith, David Neumark, and Kenneth Troske. Forthcoming, 2001. “Market Forces
and Sex Discrimination.” Journal of Human Resources
.
Holzer, Harry and David Neumark. 2000a. “Assessing Affirmative Action” Journal of Economic
Literature, Vol. 38, (September): pp 483-568.
Holzer, Harry and David Neumark. 2000b. “What Does Affirmative Action Do?” Industrial and
Labor Relations Review, Vol. 53, No. 2, pp. 240-71.
20
Lang, Kevin. 1986. “A Language Theory of Discrimination” Quarterly Journal of Economics,
Vol. 101, No. 2 (May): pp. 363-82.
Leonard, Jonathan. 1984. “Anti-Discrimination or Reverse Discrimination: The Impact of
Changing Demographics, Title VII and Affirmative Action on Productivity.” Journal of
Human Resources, Vol. 19, No. 2, (Spring): pp. 145-174.
Lott, John. 2000. “Does a Helping Hand Put Others at Risk: Affirmative Action, Police
Departments, and Crime.” Economic Inquiry. Vol. 38, No. 2, (April): pp 239-77.
MacKenzie, Kenneth D. 1999. “The Inequality Between Two Distributions: Applications to the
Analysis of Diversity Data.” Computational and Mathematical Organization Theory,
Vol. 5, No. 1, pp. 45-65.
Navarro, Peter. 1988. “Why Do Corporations Give to Charity?” Journal of Business
, Vol. 61,
(January): pp 65-93.
Osborne, Evan 2000. “The Deceptively Simple Economics of Workplace Diversity” Journal of
Labor Research, Vol. XXI, No. 3, (Summer): pp 463-475.
Penn, Nolan, Percy Russell, and Harold Simon. 1986. “Affirmative Action at Work: A Survey of
Graduates of the University of California, San Diego, Medical School.” American
Journal of Public Health, Vol. 76, No.9, pp 1144-46.
Steel, Brent and Nicholas Lovrich. 1987. “Equality and Efficiency Tradeoffs in Affirmative
Action—Real or Imagined? The Case of Women in Policing.” Social Science Journal,
Vol. 24, No. 1, pp. 53-70.
Troske, Kenneth R., 1998. “The Worker Establishment Characteristic Database.” In John
Haltiwanger, Marilyn Manser, and Robert Topel (eds.), Labor Statistics Measurement
Issues (Chicago: NBER), pp 371-403.
Welbourne, Theresa. 2000. “Wall Street Likes Its Women: An Examination of Women in the
Top Management Teams of Initial Public Offerings.” ILR CAHRS, Cornell University,
Working Paper Series, # 99-07.
Wright, Peter, Steven Ferris, Janine Hiller, and Mark Kroll. 1995. “Competitiveness Through
Management of Diversity: Effects of Stock Price Valuation.” Academy of Management
Journal, Vol. 38, No. 1, pp. 272-87.
21
Appendix
Alternative Measurement of Diversity
The alternative measure of diversity described below captures both the multiple
dimensions of diversity and measures how that diversity is distributed across occupations within
the workplace.
To formalize the desired properties of a diversity index, begin by defining two matrixes.
First, a Payroll Diversity matrix, P, that will illustrate the diversity of the workforce as it appears
on the payroll (people hired, irrespective of occupation). Second, an Occupational Diversity
matrix, O, that will illustrate the diversity of the workforce as it is spread across occupational
categories within the establishment. (A table defining all notation appears below).
To begin, assume just two dimensions for diversity, M1 and M2, (these could be race and
gender, for example). M1 and M2 are each subdivided into some number of mutually exclusive
classifications. X1 and X2 are the respective number of sub-classifications within M1 and M2.
For example, if M1 is race, X1 could be 3—non-Hispanic white, African American, and
Hispanic, or X1 could be 2—non-Hispanic white, and minority. Then P is a matrix of X1*X2
cells. Define D to equal the number of diversity cells in P, or the number of cross-classifications.
Thus D=X1*X2. For example, if race and gender are the only two dimensions of diversity, and
race is classified as non-Hispanic white and minority, matrix P would have D=4 cells: non-
Hispanic white men, minority men, non-Hispanic white women and minority women. Each cell,
i, in P contains the proportion of the establishment’s workforce (or payroll) that falls into each
cross-classification of the dimensions of diversity.
Matrix P for M1=Race, M2=Gender,
X1=2 (non-Hispanic white, minority) and X2=2 (male, female)
M2=gender
M1=race
Man Woman
White i1= % of workforce
that are white men
i2=% of workforce that
are white women
Minority i3=% of workforce
that are minority men
i4=% of workforce that
are minority women
Now assume some number of occupational categories, say three—Occ1, Occ2, Occ3 (these could
be white, blue and pink collar jobs, for example), and define C to be equal to the number of
occupational categories. Now O, the Occupational Diversity matrix, is a matrix of L cells, where
L=D*C = (X1*X2)*C, or the number of cells in matrix P times the number of occupational
categories. As O is defined above, an establishment with just three occupational categories
(C=3, where Occ1=blue collar, Occ2=pink collar, and Occ3=white collar) would have L=12
cells, if D=4. Each cell, ij, in O contains the proportion of the establishment workforce that falls
into each race, gender, and occupational cell.
22
Matrix O for D=4 (four diversity cross-classifications from P)
and C=3 (three occupational categories)
Occ1, j=1
(blue collar)
Occ2, j=2
(pink collar)
Occ3, j=3
(white collar)
White Man,
I=1
% of workforce that are
white men in Occ 1
% of workforce that are
white men in Occ 2
% of workforce that are
white men in Occ 3
White
Woman, I=2
% of workforce that are
white women in Occ 1
% of workforce that are
white women in Occ 2
% of workforce that are
white women in Occ 3
Minority
man, I=3
% of workforce that are
minority men in Occ 1
% of workforce that are
minority men in Occ 2
% of workforce that are
minority men in Occ 3
Minority
woman, I=4
% of workforce that are
minority women in Occ 1
% of workforce that are
minority women in Occ 2
% of workforce that are
minority women in Occ 3
In constructing an alternative diversity measure, the following properties are desired:
1. Given some construction of diversity dimensions, M1, M2, etc., and the respective cross-
classifications within each dimension, the index should vary directly with the number of cross-
classifications actually represented in the workforce. The more non-zero cells in matrix P, the
higher the index.
2. Assume there is a baseline population that the establishment’s workforce should mirror.
Define B
i
, to be the representative share in the baseline population of each diversity dimension
cross-classification. The index should vary directly with the workforce’s proportionate
representation of the baseline share for each cell, i, with no increase coming from exceeding the
baseline proportion in a given cell, i. For example, if the proportion of white women in the
establishment’s workforce grows closer to, but does not exceed the proportion of women in the
baseline population, the index should increase.
3. The index should vary directly with the occupational spread within each diversity cross-
classification. If workers in each diversity cross-classification are more evenly spread across
occupations in the establishment, the index should increase. In other words, if two
establishments have equal payroll diversity, the diversity index should register greater diversity
for the establishment where minority women are better dispersed across occupational categories,
all else equal.
4. The index should be flexible, able to incorporate broad or narrow definitions of diversity. It
should be able to handle many or few diversity dimensions as well as the number of cross-
classifications, i.e., increasing or decreasing D of matrix P.
5. It should be invariant to changes in the total number of workers or occupational categories.
Simply refining the occupational categories should not improve the index if workers are not
better spread across those jobs.
6. The index should not give a big pay-off to tokenism. Increasing the proportion in a cell of P or
O from zero to some small decimal should not cause a large increase in the index.
7. The index should be bounded by zero and one for ease of intuitive interpretation; one
indicating perfect diversity, zero indicating no-diversity.
Our alternative measure is defined with these properties in mind. The measure is composed
of two terms, one to capture the degree of payroll diversity, the matrix P; the other, the degree of
occupational diversity, stemming from matrix O. An index of 1,1 would indicate both perfect
23
payroll diversity (the establishment’s payroll perfectly mirrors the diversity of the baseline
population) and perfect occupational diversity (the establishment’s workers are evenly spread
across its occupations in perfect proportion to their representation in the workforce). Term 1
(Payroll Diversity) is defined as follows:
1
((,1))1
Term 1
1
D
i
i
i
W
Min
B
D
=
=
å
(A1)
where W
i
is the proportion of the workforce in each diversity cross-classification cell i,
B
i
is the
proportion of the baseline population in each diversity cross-classification cell i,
D is the number of cells in the Payroll Diversity matrix.
This term sums the workforce diversity proportions relative to the baseline population
proportions across the matrix P. The closer the establishment’s workforce proportions are to the
baseline population, the better the workforce reflects the baseline, (say the regional workforce
diversity mix), and the higher the index. If the workforce over represents any one diversity
classification, only one is included in the sum. In other words, the ratio is capped at one.
Subtracting 1 from this sum forces the index to equal 1 if the workforce perfectly mirrors the
baseline population and zero if there is no diversity. The D-1 term in the denominator is also
necessary to force this term to be between 0 and 1, inclusive.
Term 2 (Occupational Diversity) measures how well the diversity present in the
establishment’s workforce is spread across its occupations. Term 2 is defined as follows:
()
i=1 1
(,1)1
For 0; Term 2=
*1
DC
ij
j
ij
ij
S
Min
E
E
DC Z
=
é
ù
æö
ê
ú
ç÷
ç÷
ê
ú
èø
ë
û
éù
−−
ëû
åå
(A2)
where S
ij
is the observed proportion of the workforce that belongs to cell i,j, E
ij
= Occ
j
*W
i
= the
expected proportion of the workforce that belongs to cell i,j, D is the number of diversity cells in
matrix P, C is the number of occupational categories, and Z is the number of expected cells in
matrix O that contain zero—the number of times an occupation contains no workers of a specific
diversity cross-classification.
As with term 1, only the minimum of the ratio and 1 is included in the summation. This
time the summation is summed over i and j. The ratio is the observed proportions in cell i,j
divided by the expected proportions. The observed proportions (S
ij
) are those actually present in
the establishment. The expected proportions (E
ij
) are what you would expect to find if the
workforce were spread evenly across occupations with respect to each diversity cross-
classification. It is the product of the proportion of the workforce in occupation j (Occ
j
) and the
proportion of the workforce in diversity classification i (W
i
). This ratio is based on the notion of
statistical independence—the expected value is the value that would result if workers were
spread across the establishment’s given occupational structure independent of their diversity
cross-classification.
In words, the numerator of Term 2, calculated only for E
ij
not equal to zero, is the sum
over i (from 1 to D) and j (from 1 to C) of the minimum of 1 and the ratio of the observed to the
24
expected proportions of the workforce that belong to each i,j cell, minus 1. The denominator of
Term 2 is the number of cells in the O matrix, minus 1, and minus the number of cells in matrix
O for which the expected value is zero. The subtraction of Z (the number of cells in O for which
the expected value is zero) in the denominator is required by the restriction that the summation in
the numerator be calculated only for E
ij
not equal zero. The maximum number of terms in the
numerator is D*C- Z. Subtracting Z from the denominator assures that the numerator and
denominator are computed for the same number of cells. Also, since 1 is subtracted from the
numerator to restrict the numerator to values between 0 and 1, inclusive, 1 must also be
subtracted in the denominator.
Notation Table for Alternative Diversity Measure
B
i
= The baseline share of a diversity cross-classification, e.g., the proportional
representation of Hispanic men in the regional workforce.
C = The number of occupations.
D = X1 * X2 = The number of cell in matrix P.
E
ij
= The expected proportion of establishment’s workforce in cell ij of the matrix O.
Constructed from the proportion in Occ
j
* W
i
.
L = D * C = The number of cells in the matrix O.
M1,
M2, etc = Dimensions of diversity, e.g., M1 = race, M2 = gender.
O = Occupational Diversity Matrix.
Occ
j
= Occupations in establishment, j = 1 to C.
P = Payroll Diversity Matrix.
S
ij
= The observed proportion of establishment’s workforce in cell ij of the matrix O.
W
I
= The proportion of establishment’s workforce in diversity cross-classification cell i of
the matrix P.
X1 = Number of classifications in diversity dimension M1.
X2 = Number of classifications in diversity dimension M2.
Z = Number of cells for which E
ij
is zero.
25
Table 1: Summary Statistics for Establishments in Our Data
Mean
(1)
Std. Dev
(2)
Total Employment 216.33 467.52
Log ((Value-Added)/TE) 3.92 1.18
Log (Sales/TE) 4.28 0.93
Percent White Male Workers 0.46 0.30
Percent White Female Workers 0.42 0.30
Percent Minority Male Workers 0.06 0.11
Percent Minority Female Workers 0.06 0.12
Percent College Grad. 0.12 0.14
Percent Workers Over 50 0.19 0.15
Percent of Workers Matched 0.15 0.11
Payroll Diversity 0.39 0.25
Occupational Diversity 0.55 0.18
Number 18,509
26
Table 2: Regression of Plant-level Labor Productivity on Plant-level Diversity
Manufacturing Services Retail Sales
Log (Value-Added/TE)
(1)
Log(Sales/TE)
(2)
Log(Value-Added/TE)
(3)
Log(Sales/TE)
(4)
Log(Value-Added/TE)
(5)
Log(Sales/TE)
(6)
Payroll Diversity 0.12
(0.04)
0.11
(0.03)
0.14
(0.04)
0.12
(0.03)
-0.02
(0.05)
-0.03
(0.04)
Occupational Diversity 0.03
(0.04)
0.04
(0.03)
0.09
(0.05)
0.05
(0.04)
-0.04
(0.06)
-0.05
(0.05)
Percent Older Workers -0.08
(0.04)
-0.07
(0.03)
-0.19
(0.06)
-0.09
(0.04)
-0.23
(0.06)
-0.21
(0.05)
Percent College Grad. 0.33
(0.06)
0.34
(0.04)
0.23
(0.06)
0.26
(0.04)
-0.12
(0.08)
-0.04
(0.07)
Log (TE) 0.08
(0.01)
0.08
(0.01)
0.08
(0.01)
0.08
(0.01)
0.06
(0.02)
0.08
(0.02)
Single-Unit -0.34
(0.02)
-0.27
(0.01)
-0.13
(0.02)
-0.10
(0.01)
-0.03
(0.03)
-0.02
(0.02)
R-squared 0.58 0.61 0.53 0.62 0.70 0.73
Number 11773 11773 4545 4545 2191 2191
Note: Standard errors in parenthesis. Regressions include controls for four-digit SIC industry and for the four census regions.
27
Table 3: Regression of Plant-level Labor Productivity on Plant-level Diversity—Large
Establishments (establishments with 500 or more employees)
Manufacturing Services
Log (Value-Added/TE)
(1)
Log(Sales/TE)
(2)
Log(ValueAdded/TE)
(3)
Log(Sales/TE)
(4)
Payroll Diversity 0.45
(0.11)
0.37
(0.09)
0.17
(0.06)
0.13
(0.04)
Occupation Diversity 0.58
(0.24)
0.42
(0.19)
0.04
(0.15)
0.06
(0.09)
Percent Older Workers -0.77
(0.25)
-0.51
(0.20)
-0.97
(0.23)
-0.16
(0.14)
Percent College Grad. 0.67
(0.31)
0.70
(0.25)
0.55
(0.15)
0.41
(0.09)
Log (TE) 0.01
(0.10)
-0.05
(0.04)
-0.03
(0.03)
0.02
(0.02)
Single-Unit -0.26
(0.13)
-0.25
(0.10)
-0.04
(0.03)
-0.05
(0.02)
R-squared 0.71 0.74 0.44 0.40
Number 1032 1032 601 601
Note: Standard errors in parenthesis. Regressions include controls for four-digit SIC industry and for the four census regions.
28
Table 4: Regression of Plant-level Labor Productivity on Measures of Worker Characteristics
Manufacturing Services Retail Sales
Log (Value-Added/TE)
(1)
Log(Sales/TE)
(2)
Log(Value-Added/TE)
(3)
Log(Sales/TE)
(4)
Log(Value-Added/TE)
(5)
Log(Sales/TE)
(6)
Percent Non-Hispanic
white Male
0.48
(0.07)
0.42
(0.06)
-0.06
(0.08)
0.03
(0.06)
0.21
(0.09)
0.19
(0.08)
Percent Minority Male 0.59
(0.09)
0.50
(0.07)
0.16
(0.16)
0.29
(0.11)
0.33
(0.15)
0.28
(0.13)
Percent Non-Hispanic
white Female
0.12
(0.07)
0.07
(0.06)
-0.02
(0.06)
-0.04
(0.04)
0.07
(0.08)
0.05
(0.07)
Percent Minority Female --- --- --- --- --- ---
Percent Older Workers -0.09
(0.04)
-0.08
(0.03)
-0.19
(0.06)
-0.09
(0.04)
-0.23
(0.06)
-0.20
(0.05)
Percent College Grad. 0.30
(0.06)
0.31
(0.04)
0.26
(0.06)
0.26
(0.04)
-0.15
(0.08)
0.06
(0.07)
Log (TE) 0.09
(0.01)
0.08
(0.01)
0.09
(0.01)
0.09
(0.01)
0.05
(0.02)
0.08
(0.02)
Single-Unit -0.34
(0.02)
-0.27
(0.01)
-0.13
(0.02)
-0.10
(0.01)
-0.03
(0.03)
-0.02
(0.02)
R-squared 0.59 0.62 0.53 0.62 0.70 0.74
Number 11773 11773 4545 4545 2191 2191
Note: Standard errors in parenthesis. Regressions include controls for four-digit SIC industry and for the four census regions.
29
Table 5: Regression of Total Factor Productivity (TFP) on Plant-level Diversity—Just
Manufacturing Establishments
Alternative Diversity Measures
(1)
Traditional Diversity Measures
(2)
Payroll Diversity 0.08
(0.018)
---
Occupational Diversity 0.07
(0.021)
---
Percent White Male --- -0.02
(0.035)
Percent Minority Male --- -0.01
(0.047)
Percent White Female --- -0.01
(0.036)
Percent Minority Female --- ---
Percent Older Workers -0.01
(0.021)
-0.01
(0.022)
Percent College Grad. 0.06
(0.027)
0.07
(0.027)
Log (TE) 0.02
(0.004)
0.02
(0.004)
Single-Unit -0.03
(0.008)
-0.04
(0.008)
R-squared 0.68 0.68
Number 11043 11043
Note: Standard errors in parenthesis. Regressions include controls for four-digit SIC industry and for the four census regions.
... firm-level diversity, a literature that tends to neglect the importance of the milieu in which the firm is located (Barrington & Troske, 2001;Parrotta, Pozzoli, & Pytlikova, 2014). We also differentiate across different regional contexts along the urban-rural hierarchy so that our approach enables a deeper understanding on how both firm-level and regional-level diversity relate to firm survival in different regional milieus. ...
... However, we argue that diversity is much broader and should include dimensions such as education, occupation, and demographics. Although the studies by Parrotta et al. (2014) and Barrington and Troske (2001) distinguish among different types of diversity, they focus on firm productivity. In contrast, our paper focuses on firm survival. ...
Article
Full-text available
We analyze the influence that the diversity of individuals, both within a firm and in the region where the firm is located, has on firm survival. We estimate a hazard model using micro data for Swedish firms for the years 2002 to 2013. Results show that firm‐specific diversity in education, age and gender are positively associated with firm survival. However, firm‐specific diversity by place of origin is negatively associated with firm survival. Yet the cultural diversity among the inhabitants in the region where the firm is located enhances firm survival. Magnitudes of the effects vary by region and industrial sector. This article is protected by copyright. All rights reserved.
... Another strand of the literature relies on linked employer-employee data (LEED). Barrington and Troske (2001) examined the impact of plant-level diversity with respect to age and gender on plant-level productivity (i.e., value-added and sales per worker and TFP) in the manufacturing, retail trade, and services industry. Based on cross-sectional LEED for 1999, their OLS estimates reject the hypothesis that workforce diversity would be detrimental for the productivity of U.S. plants. ...
... Many companies have welcomed the new era's diversity. Several studies have shown that older workers are just as active, engaged, and professional as younger workers (Algahtani, 2013; Barrington & Troske, 2001). Williams and O'Reilly (1998) documented that employees of a heterogeneous age group are more creative, diverse, and active than employees of a homogeneous age group. ...
Article
Full-text available
Purpose of the Study: The fundamental objective of this study was to examine the association between workforce diversity and employee performance in commercial banks of Pakistan. Methodology: Gender, age, education, and ethnic diversity contribute to the workforce's diversity. A standardized questionnaire was distributed amongst the 400 employees of commercial banks of Pakistan to collect data. However, 281 employees, constituting 70.25%, responded. The research hypotheses were tested by using a multi-regression model through Stata16 for data analysis. Main Findings: The outcomes indicated a positive and significant association between gender diversity and employee performance. Gender diversity and age diversity are strong predictors of employee performance in Pakistan. However, the findings revealed that employee performance is not affected by education or ethnic diversity. Applications of the study: This study has significant implications from both a theoretical and a realistic perspective. This study contributes by adding sound theoretical literature to the body of knowledge on workforce diversity and employee performance. From a practical perspective, this study sheds more light on how workplace diversity affects employee performance in the banking sector of an emerging economy like Pakistan. The results of this study can also be used to build strategies to improve the implementation of employee diversity in an existing market. Novelty/Originality of this Study: Studies have been conducted to analyze the impact of diversity on employee performance, but this is the first study that has investigated the association between workforce diversity and employees performance with rich facets and particularly in the context of an emerging economy like Pakistan and specifically in commercial banks.
... Furthermore, women continue to have the upper hand on the "invisible care" economy, which relates to care giving and domestic work. Discrimination on hiring workers based on gender has resulted in a firm's hiring workers who are paid higher wages than alternative workers, but are no more productive (Barrington & Troske, 2013).However, according to Emiki and Eunmi (2009), workforce diversity remains ineffective if gender issues are not first recognized and managed. ...
Article
Full-text available
mployee affective commitment is critical to the success of every E organization. However, the issues of unequal age distribution, gender discrimination, educational background, religious and unethical practice during recruitment, promotion, training and compensation have adversely affected employee commitments. There seems to be paucity of studies on the effect of workforce diversity on employee affective commitment in the banking sector in Nigeria. This study examined the effect of workforce diversity on employee affective commitment of selected deposit money banks in Lagos State, Nigeria. Survey research design was adopted for the study. The population of the study covered 4,873 management staff of the selected deposit money banks in Lagos State, Nigeria. The sample size of 464 respondents was determined through the Raosoft sample size calculator. Multi stage sampling method was utilized for sample selection. Data were collected through adapted validated questionnaire which Cronbach's alpha reliability coefficient ranged from 0.709 to 0.891. The instrument achieved a response rate of 78.01%. Data collected were analyzed using descriptive and inferential statistics. The result revealed that workforce diversity sub-variables had significant effect on 2 employee affective commitment (Adj. R = 0.287; F(5, 357) = 49.495; p<0.05); of selected deposit money banks in Lagos State, Nigeria. The study concluded that workforce diversity sub-variables affect the continuance commitment of employees of deposit money banks in Lagos State, Nigeria. The study recommended that management of selected deposit money banks should ensure that equal opportunities are given to employees in the banks irrespective of their age, gender, educational qualification, religious and ethics diversity as these will enhance employee affective commitment in the banks.
... Second, we argue that the board plays an essential role in influencing the performance and productivity of employees through a signaling effect (Spence, 1973) as well as progressive management programs advocated by the board that increase employee satisfaction and productivity (Barrington and Troske, 2001;Creek et al., 2019), which in turn contributes to firm financial performance. To our knowledge, no research has examined how board diversity impacts employee productivity or specifically whether employee productivity mediates the relationship between board diversity and firm performance. ...
Article
Full-text available
Purpose This paper aims to reexamine the link between board racial diversity and firm performance. It focuses on the mechanism through which board racial diversity could affect performance. The paper proposes and empirically tests the role of employee productivity and R&D productivity in the relationship between board racial diversity and firm financial performance. Design/methodology/approach The paper adopts a mediation analysis framework with the bootstrapping method to test both the direct and indirect effect of board diversity on firm performance. The data used in the study come from S&P 1500 with variables composed from COMPUSTAT, Institutional Shareholder Services and Wharton Research Data Services. Findings Contrary to prior findings, the results indicate that there is neither direct effect of board racial diversity on firm performance nor is there an indirect effect through either employee productivity or R&D productivity. Research limitations/implications Because the data used in the paper are based on large public firms, the results may not generalize to small or private firms. Practical implications The findings of the paper suggest that not all diversity measures matter in the same way and firms should carefully make board appointments to reduce the perception that they select directors for any reason other than qualifications. Originality/value The paper advances the literature on board diversity by examining two previously unexplored mediating variables of employee productivity and R&D productivity. It also uses more rigorous mediation analysis with bootstrapping method and a validation sample to improve robustness of the results.
... The current study aimed to assess the influence of education diversity, language diversity, and religious diversity and work experience diversity on employee performance of teaching and non-teaching staff of government colleges of the Khairpur Sindh Pakistan. Results of the research are in line with the literature cited such as hypothesis 1 was predicted that education diversity has a significant influence on employee performance and the same has been proved from collected data and is in line with the empirical evidence (Barrington, & Troske, 2001 (Brannen, et al., 2010) all show positive significant effects on employee performance and similar has been proved from collected data in the current study. ...
Article
Full-text available
The current study attempted to examine the influence of education diversity, language diversity, work experience diversity, and religious diversity on employee performance of teaching and non-teaching staff of government colleges of the Khairpur Sindh Pakistan. Keeping in view the objectives of the study, data were collected from the sample of 309 individuals selected through stratified random sampling from government colleges of the Khairpur Sindh Pakistan. Data screening was done through SPSS23, and the model was assessed through Smart PLS3.2.8. The study found positive and significant findings of language diversity, education diversity, work experience diversity, and religious diversity with employee performance. The findings of the study are well established under the umbrella of Social identity theory and are equally valuable as a source of information for management in the education industry for keeping the momentum of remarkable performance.
... Resonating these conflicting predictions, studies have reached mixed conclusions on the productivity effect of workforce diversity (e.g. Barrington and Troske 2001;Ozgen and De Graff 2013;Parrotta et al. 2014;Trax et al. 2015). Therefore, it remains unclear how diversity affects job switches. ...
Article
Full-text available
We investigate how the demographic composition of the workforce along the sex, nationality, education, age and tenure dimensions affects job switches. Fitting duration models for workers’ job‐to‐job turnover rate that control for workplace fixed effects in a representative sample of large manufacturing plants in Germany during 1975–2016, we find that larger co‐worker similarity in all five dimensions substantially depresses job‐to‐job moves, whereas workplace diversity is of limited importance. In line with conventional wisdom, which has that birds of a feather flock together, our interpretation of the results is that workers prefer having co‐workers of their kind and place less value on diverse workplaces.
Article
Full-text available
To analyze the moderating effect of management between workforce diversity and performance of the employees and to determine the relationship between workforce diversity and employee's performance. The quantitative study was conducted in private sector firms. A sample of 90 employees of different private firms was selected in Dhaka city of Bangladesh. The data was collected through structured questionnaires. Pearson Correlation and Regression were run to analyze the data. Findings show that there is a positive and significant relationship between workforce diversity and the performance of the employees. The sample size was too short it can be replicated in future by large samples in the same sector. The relation of management between workforce diversity and employee's performance can be checked in future.
Article
Full-text available
The purpose of the study was to assess the impact of age and gender diversity on employee performance in an organisation. A descriptive research design was adopted with Zambia Compulsory Standards Agency (ZCSA) being the focus organisation which has a total of 103 employees from various units. Stratified and random sampling techniques were used to draw a sample size of 50 respondents from various hierarchical levels. The main data collection instruments were self-administered questionnaires and interview guides. A set of descriptive statistics including frequencies, percentages, the mean and standard deviation were used to generate tables, bar graphs and pie charts to present the results of the study using SPSS version18. The key findings of the study were that both age and gender diversity have a bearing on the performance of an employee and ultimately on the organisational output. It was ascertained that employees felt comfortable working with either older or younger employees. Others felt positive about being involved in teams that consist of employees with varying ages as this improved their performance. Furthermore, the study found out that gender diversity has a high predictive power on employee performance. On the other hand, organisations that create an inclusive environment for all employees irrespective of gender and create policies that eliminate discrimination from the workplace can greatly benefit from gender diversity. A significant relationship between gender diversity and work performance has shown that employees like to work with the opposite gender so long as they can carry out their job professionally. The study concluded that age diversity is a very crucial resource for firms that intend to have sustainable workforce, higher production, productivity and profitability.
Article
Full-text available
This study estimates the changes over time in the relative productivities of minorities and females. I find no significant evidence that the productivity of minorities or females decreased relative to that of white males as relative minority and female employment increased during the 1960s and 1970s. This study also compares the impact of Title VII and affirmative action and presents evidence that Title VII litigation has played a significant role in increasing black employment. This suggests that the employment of minorities and females has not entailed large efficiency costs and that Title VII litigation has had some success in fighting racial discrimination. The EEOC has sometimes been credited with opening up new pools of labor that corporations somehow contrived to ignore, and occasionally with hastening the breakdown of traditional barriers to labor mobility.... But in the context of the market's endless search for efficiency, these anomalies would have been eliminated anyway, leaving only the question of whether they were worth the expenditures compelled by law. Affirmative action is a net cost to the economy....And the true dynamic effects-the opportunity cost of all this expense and effort, the diminution of competition, inefficiencies due to the employment and promotion of marginal labor and the consequent demoralization of good workers-can only be a matter of conjecture, although they are clearly the most important of all. Senator Orrin Hatch, 1980
Article
conclusion My approach for the first time connects profit maximization to work force composition given full information. Physical productivity, which dominates so much popular dis-cussion of this issue, matters, but demand-side factors are also important. Defining a firm’s optimal work force composition is nonsensical until at least the product’s char-acteristics, the extent to which different markets value them, and the extent to which groups inherently differ in their capacity to provide them are precisely defined.
Article
This study's premise is that firms that can lower their costs and enhance their differentiation through the effective management of their human resources have a competitive advantage. Using data from 1986 through 1992, we examined the impact that announcements of U.S. Department of Labor awards for exemplary affirmative action programs had upon the stock returns of winning corporations and the effect that announcements of damage awards from the settlement of discrimination lawsuits had on the stock returns of corporations. The results suggest that announcements of awards may be associated with competitive advantage and that discrimination-related announcements may be associated with inability to achieve such advantage.
Article
This article investigates rival predictions about the consequences of affirmative action on behalf of women in one previously male-dominated public sector occupation—police service. While some predict that the enhanced use of women police officers will diminish the quality of public services, others argue precisely the opposite point of view. It is suggested here that such predictions have much more to do with the political and ideological perspectives of advocates and opponents of affirmative action than with empirical observation. It is suggested further that this question of the impact upon the quality of public service of enhanced utilization of women is much deserving of careful empirical study. Such an analysis is presented here, and the results observed indicate that little if any effect upon police agency performance can be attributed to differing levels of female police officers.
Article
This chapter summarizes recent research in economics that investigates differentials by race and gender in the labor market. We start with a statistical overview of the trends in labor market outcomes by race, gender and Hispanic origin, including some simple regressions on the determinants of wages and employment. This is followed in Section 3 by an extended review of current theories about discrimination in the labor market, including recent extensions of taste-based theories, theories of occupational exclusion, and theories of statistical discrimination. Section 4 discusses empirical research that provides direct evidence of discrimination in the labor market, beyond “unexplained gaps” in wage or employment regressions. The remainder of the chapter reviews the evidence on race and gender gaps, particularly wage gaps. Section 5 reviews research on the impact of pre-market human capital differences in education and family background that differ by race and gender. Section 6 reviews the impact of differences in both the levels and the returns to experience and seniority, with discussion of the role of training and labor market search and turnover on race and gender differentials. Section 7 reviews the role of job characteristics (particularly occupational characteristics) in the gender wage gap. Section 8 reviews the smaller literature on differences in fringe benefits by gender. Section 9 is an extensive discussion of the empirical work that accounts for changes in the trends in race and gender differentials over time. Of particular interest is the new research literature that investigates the impact of widening wage inequality on race and gender wage gaps. Section 10 reviews research that relates policy changes to race and gender differentials, including anti-discrimination policy. The chapter concludes with comments about a future research agenda.
Article
A new measure called diversity difference is proposed for the inequality of a pair of distributions. The diversity difference measure satisfies eight properties of a measure of inequality. This measure is simple to calculate and provides easily interpreted results. Existing inequality measures examine the distribution of a single variable whose data are arranged in a monotonic order. The new measure can employ multiple variables and does not require each to be monotonic but can be used if the data happen to be monotonic. The pair of distributions is useful for organizational diversity data because one of the distributions represents the actual proportions of employees in any class or set of classes and the other distribution is the benchmark or anchoring distribution. Data from the measure can be displayed in diversity difference trees for quick interpretation. The diversity difference measure can be arranged to define a Lorenz curve. An example with three classes (gender, race, and age) is employed to provide examples of the measure, the resulting Lorenz curve, and the disparity ratio.