Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
CORPORATE DEMOGRAPHY AND WAGE INEQUALITY:
VERTICAL AND HORIZONTAL SORTING AS SOURCES OF REGIONAL WAGE DISPERSION*
Jesper B. Sørensen MIT Sloan School of Management
50 Memorial Drive, E52-581 Cambridge, MA 02142-1347
Olav Sorenson UCLA Anderson Graduate School of Management
110 Westwood Plaza, Box 951481 Los Angeles, CA 90095-1481 [email protected]
September 2004
* We are indebted to Niels Westergaard-Nielsen and Tor Eriksson of the Aarhus School of Business for
allowing us to use the Pay and Performance data. Søren Leth-Sørensen, Jørn Hansen Schmidt, and Paul Bingeley were extremely helpful in facilitating access and providing time, advice and resources. We also thank Roberto Fernandez, Michael Hannan, Boyan Jovanovic, Leslie McCall, Glenn McDonald, Damon Phillips and Ezra Zuckerman for useful comments on earlier drafts of this paper. The usual disclaimer applies.
CORPORATE DEMOGRAPHY AND WAGE INEQUALITY: VERTICAL AND HORIZONTAL SORTING AS SOURCES OF REGIONAL WAGE DISPERSION
ABSTRACT
This study examines how wage inequality depends on corporate demography, or the number
and diversity of employers in a labor market. Drawing on insights from economics and
sociology, we identify two distinct theoretical mechanisms linking corporate demography to
wage inequality. One mechanism implies that wage dispersion should increase with the
number of firms competing for the same labor pool, as labor market competition drives
compensation to reflect worker heterogeneity. The second mechanism suggests that
organizational diversity should reduce inequality because it increases the odds that an
individual finds employment with a firm that can take advantage of his or her unique skills.
Our analysis of Danish census data provides support for each thesis: Within an industry, the
number of, and absence of concentration among, firms operating in a labor market increases
wage dispersion. The variety of industries offering employment within a region, however,
both reduces inequality and mitigates the effect of within-industry labor market competition.
1
How much money could a gifted shooting guard earn in a world with no
professional basketball teams? What would a gourmet chef earn in the absence of
restaurants? For both the basketball player and the chef, one can easily imagine that they
would earn less in such situations than they would if employers existed that desired their
distinctive traits. Though other organizations might value the shooting guard’s agility,
eye-hand coordination and physical stature, few, if any, would attach as much value to
the guard’s unique constellation of abilities and attributes as a basketball team. Similarly,
family and friends may appreciate the gourmet chef’s culinary creativity, senses of smell
and taste, and aesthetic talents, but firms that do not serve meals place less value on these
special talents than those that do. Denied access to their optimal lines of employment,
both the chef and the shooting guard would have to work in jobs that may not fit their
abilities particularly well. Moreover, they would potentially find themselves in
competition with others whose unique skills and characteristics do match those desired by
employers. The shooting guard, for example, might find a job with a baseball team, but
then will find himself in competition with others whose physical gifts make them
excellent baseball players. As a result, the shooting guard would likely earn less than he
would with a basketball team. (Michael Jordan’s inability to convert his basketball
success into a professional baseball career comes to mind.)
These hypothetical examples highlight the potential importance for income
inequality of the interaction between the distribution of individual abilities and attributes
in a labor market and the prevalence of employers with a need for such characteristics.
2
Most sociologists would consider uncontroversial the claim that wage differentials, to
some extent, reflect differences in individual characteristics. (Debate centers instead on
the extent to which wage differentials reflect productivity differences, as opposed to
echoing socially constructed beliefs about ability.) Other things being equal, basketball
teams compensate better the truly gifted basketball player than the less talented one, and
pay more for the taller player than the shorter one. The returns to possessing particular
abilities and attributes, however, depend on the match between individuals with those
qualities, on the one hand, and employers that can – or at least believe they can – make
effective use of them on the other (Sørensen, 1996: 1357). Michael Jordan’s fortunes
would decline in a world with baseball but no basketball teams; the chef without
restaurants as potential employers must settle for a second best outcome. Both have
specialized characteristics that they cannot capitalize through the labor market in the
absence of the right types of employers.
Some of these valued individual characteristics admittedly arise through
investments in training for particular roles. Human capital theories (Becker 1964), for
example, contend that rational individuals would not invest heavily in acquiring skills
that no employer wants. As the case of industrial restructuring suggests, however,
changes in the nature of employer demand can render past human capital investments
more or less valuable (Morris and Western 1999). Moreover, some differences in ability
arise from heterogeneity in traits that one might consider innate (e.g., height in the
context of basketball), in the sense that they do not stem from past investments in the
acquisition of particular skills. If one considers the distribution of individual abilities and
attributes at least partially exogenous to the labor market at a point in time, an important
3
sociological issue concerns the interaction between the distributions of these abilities and
attributes, and the social structure of the labor market, which determines their value.
Though some of these differences in demand from employers undoubtedly reflect real
individual productivity differences in a particular type of activity, others likely stem from
socially constructed beliefs about this relationship, as in the case of race and sex
discrimination. Regardless of the particular source of this variation in demand, however,
the allocation of rewards in society associated with a given distribution of characteristics
depends on the structure of the labor market, as individuals’ abilities and attributes meet
employers in this arena.
But what features of the labor market determine the value of these individual
characteristics? On the demand side, the labor market consists of formal organizations
that differ in a wide variety of ways. The intensity and nature of interactions between
these organizations determines the opportunities available to employees. Human and
organizational ecologists, in particular, have highlighted the degree of diversity embodied
in these formal organizations as an important determinant of community and industry
dynamics (e.g., Hawley, 1950; Hannan and Freeman, 1977). Recent organizational
research has begun to examine the linkages between labor market outcomes and the
corporate demography of labor markets – or the number and diversity of organizations
(employers) active in a particular labor market (Carroll and Hannan, 2000).1 For
example, a small number of studies have attempted to link the diversity and dynamics of
employers to individual career events, in terms of either promotion chances (Phillips,
2001), or mobility between employers (Greve, 1994; Haveman and Cohen, 1994).
1 We use “corporate demography” and “industrial demography” interchangeably.
4
Corporate demography may also influence other types of employment related
outcomes as well. In this paper, we draw on existing theory fragments in economics and
sociology to develop hypotheses regarding the influence of corporate demography on
wage dispersion. Our review of past theoretical work suggests two (seemingly
contradictory) expectations concerning the association between industry structure and
wage dispersion. The first account, drawing on economic models of the labor market,
maintains that (for fixed labor demand) a greater number of employers in a region should
increase wage inequality as firms increasingly diverge in their ability to employ human
capital effectively, thereby forcing wages to reflect underlying individual differences in
productivity (Lydall, 1959; Rosen, 1981; Wheeler, 2001). Returning to our example, the
larger the number of basketball teams, the greater the extent to which player salaries
should depend on differences in basketball ability. The second perspective, drawing on
arguments from organizational ecology, contends that wage dispersion should decline
with the number of organizations in the labor market, because workers have more
opportunities to match their characteristics to an employer in need of them (Hannan,
1988; see also Roy, 1951). In a world with only baseball teams, the athlete whose skills
match best the game of basketball will earn less than the player with skills ideally suited
to baseball, but in a world with both types of teams, inequality declines as the skills
related to basketball are more amply rewarded.
The differing predictions arise from a divergence in assumptions regarding the
nature of the labor market. Whereas most economic models of labor markets assume a
single dimension of worker quality, the ecological approach presupposes a multi-
dimensional world in which employees have diverse types of skills and organizations
5
have varying demands for them. If one assumes that organizations involved in similar
operations prefer employees with similar skills, income inequality should increase as the
number of such firms rises. On the other hand, organizations operating in different
industries frequently draw on disparate skill sets; hence, in the presence of heterogeneous
abilities and attributes, wage dispersion should decline with increasing diversity in the
industries represented in a local labor market. Moreover, the availability of these
opportunities should also allow employees to avoid working in industries in which they
could not contribute effectively (or in which they might face more severe discrimination),
thereby mitigating to some degree the salience of the number of firms within an industry
for income inequality.
We examine the relationship between corporate demography and income
inequality by analyzing regional variation in wage inequality as a function of regional
differences in corporate demography. Both within and across countries, income
inequality varies tremendously from one area to the next (Rauch, 1993; Blau and Kahn,
1996; McCall, 2000). We propose that differences in the corporate demography of the
local population of employers can help to explain this heterogeneity in inequality across
regions. We analyze comprehensive, yearly census data on the Danish population that
contains information on the geographic location of workplaces. Our results indicate that
local wage dispersion within an industry: (1) increases with the number of firms in the
industry, and (2) declines with the variety of employers available outside the industry.
The magnitude of the association between the number of firms and wage dispersion
(finding 1 above) also declines with the availability of employers outside the industry.
These results remain robust to a more conservative estimation approach using a two-stage
6
hierarchical linear model (Bryk and Raudenbush, 1992; McCall, 2000), where we first
control for all observable human capital effects on wages, and then examine the
relationship between industry structure and the residuals from these wage equations. Our
results therefore indicate that industrial demography, by shaping the local structure of
employment opportunities, plays an important role in determining inequality in the
distribution of wealth through wages.
CORPORATE DEMOGRAPHY AND INEQUALITY
Our analysis focuses not on the characteristics of individual firms, but rather on
the demography of the population of employers located in a particular region. In this
respect, our approach departs from most of the existing research on the role of
organizations in the stratification process, which has typically highlighted the dynamics
of internal labor markets (e.g., Barnett, Baron and Stuart, 2001), or the adoption of
particular human resource practices (e.g., Kalleberg et al., 1996). Though important to
our understanding of career trajectories, these perspectives have not considered explicitly
how these factors contribute to income inequality across the whole of society. In
principle, one could view this overall inequality as a simple aggregation of the underlying
characteristics of the organizations operating within it. Such an approach, however, fails
to consider the potential interactions among the firms that jointly determine labor market
dynamics (Carroll and Hannan, 2000). In particular, the perspective offered here
highlights the fact that labor market inequality results, in part, from the structure of
competition in the labor market, and that the diversity of organizations in a labor market
affects the dynamics of competition on the demand side of the labor market.
7
Although the vast majority of research on corporate demography has focused on
understanding how the characteristics of organizational populations influence both the
fates of individual firms and the evolution of industries, several studies suggest an
important link between corporate demography and labor market outcomes. In particular,
ecological dynamics appear to have a large effect on employee mobility. Examining the
movement of employees across firms, for example, Haveman and Cohen (1994)
demonstrate that workers most commonly leave their existing jobs either to join a newly
established firm or as a result of the dissolution of their employer; Carroll and Hannan
(2000: 430-431) estimate that such dynamics-related job shifts may account for as much
as 25% to 55% of all inter-firm employee mobility. Greve (1994; Fujikawa-Greve and
Greve, 2000) has further demonstrated that the rate of such shifts depends on the
demographic characteristics of a population – finding a positive relationship between
employee mobility and heterogeneity in the size distribution of firms. And recent studies
suggest that ecological processes influence not just the movement of employees across
firms, but also their upward mobility in internal labor markets; Phillips (2001; Phillips
and Sørensen, 2003) finds that employee promotion rates rise with an employer’s
susceptibility to ecological competition.
Our study builds on this ecological perspective, but considers the effects of
industrial demography with respect to a different type of labor market outcome, wage
inequality. Drawing on insights from the economics and sociological literatures, we
identify two seemingly inconsistent predictions concerning the relationship between the
diversity of the population of employers in a region and the degree of wage inequality.
The first, growing out of economic models of labor market competition, argues that more
8
intense competition should increase the dispersion in wage levels across workers (Lydall,
1959; Rosen, 1981; Wheeler, 2001). The second, building on the logic of human and
organizational ecology and focusing more on the opportunity structure available to
employees, argues that a wider variety of opportunities should reduce wage dispersion.
We discuss each approach in turn.
Competition, sorting and inequality
The traditional neoclassical approach to labor markets in economics contends that
firms pay workers their marginal product – in other words, each employee’s wage should
match the degree to which that individual contributes to the output of the firm. The logic
for this expectation rests on the assumption that firms will hire more employees if doing
so would increase their profits, and on the belief that employees choose to work for the
firm that offers them the highest wages. To the extent that wages reflect marginal
productivity, it then follows that wage inequality emerges from differences across
individuals in productivity.
Economic models of inequality offer a variety of accounts – differing in the extent
to which they emphasize heterogeneity among workers or heterogeneity among firms – of
the sources of this differential productivity. Traditional accounts focus on heterogeneity
in human capital, either as a consequence of variation in innate abilities or as the outcome
of differential investment in the acquisition of training and skills (Becker 1964). Since
such models typically assume homogeneity among firms, they do not imply a link
between industrial demography and inequality; firms with the same production
technology, regardless of their number, should benefit equally from hiring more
productive labor. An alternative approach assumes that firms differ in their ability to
9
capitalize on the talents of employees. Analytic models suggest that wage inequality can
arise even in the absence of heterogeneity across workers because equally skilled
employees might nonetheless differ in their productivity when employed in different
settings (e.g., using different technologies). Explanations along these lines have been
forwarded, for example, to explain the positive correlation observed between wages and
firm size (see Idson and Oi, 1999, for a review).
The more interesting approaches from our perspective, however, see inequality as
the result of a labor market matching process between heterogeneous employees and
heterogeneous employers (i.e., cases where both individuals and firms differ in
productivity or ability). One potential type of matching highlights the importance of the
scale or scope for the application of individual talents. A small difference in the abilities
of two individuals would normally only justify a small difference in their wages, though
production technologies might amplify this differential in absolute terms. But imagine
that markets differ in their sizes. In that case, one would expect a sorting of individuals to
markets, with the most talented individuals working in the largest markets and securing
the largest returns (Lydall, 1959; Rosen, 1981, 1982). To see the intuition behind this
effect, return to professional basketball. A more talented basketball player might increase
the probability that a potential fan buys a ticket to attend a game. In a small market, such
as Sioux Falls, South Dakota, with a population of about 100,000, the player might
increase expected ticket sales by 500, but the same small difference applied to the
Chicago market (roughly 60 times larger) could result in 30,000 additional tickets sold.
Hence, NBA basketball players differ only slightly in talent from those in the Continental
10
Basketball Association (CBA), but the two leagues offer dramatically different salary
scales.2
Though our presentation of this dynamic appears predicated on an assumption
that wages reflect underlying differences in worker ability, this type of sorting process
really only requires that firms have consistent beliefs – whether accurate or not – about
which employees they consider most valuable. Regardless of whether these differences
reflect actual heterogeneity in skill or simply perceived differences, this sorting dynamic
has been used to explain the highly skewed income distributions among artists and
athletes (Rosen, 1981), and among managers (Lydall, 1959; Rosen, 1982) – moving up
the corporate hierarchy or to larger firms allows the leveraging of managerial talents
across a broader range of operations.3 Greater heterogeneity in the scale or scope of
operations across firms would therefore lead one to expect more pronounced wage
inequality.
Another type of sorting argument depends not on the scale of operations (for a
particular type of activity) but on synergy between worker skill and the ability of an
employer to make effective use of that skill. In developing an analytical model to
understand the relationship between the number of employers, productivity and wage
inequality, Wheeler (2001) begins by assuming that one can array both employees and
employers along a single ‘quality’ dimension.4 Higher ability employees can produce
2 Television differentiates these two leagues even further by allowing the NBA to reach audiences
beyond the populations of the cities in which the games take place. 3 See Sattinger (1983) for a review. Since economists would typically expect competition to
remove inefficient allocations from the market, the stability of this equilibrium in economic models typically depends either on search costs (e.g., Albrecht, Axell and Lang, 1986) or on market power (e.g., Bhaskar, Manning and To, 2002).
4 Wheeler’s (2001) model has a nearly identical structure to that originally used by Becker (1973) to describe the matching of individuals in “marriage markets” and later by Kremer (1993) to explain the importance of the complementarities that result from highly skilled employees working with one another.
11
more, but the quality of the employer limits their effectiveness. For example, a highly
skilled chef might contribute positively to the profitability of a McDonalds restaurant, but
she could add far more value working for an employer that could utilize her skills more
effectively, such as a gourmet restaurant. In order of their quality (from highest to
lowest), firms in Wheeler’s model then search for employees.5 If they can search without
cost, the market generates a perfect quality match between employees and employers –
that is, firms match with employees of similar rank on the quality distribution (Becker,
1973). This perfect matching between employees and firms maximizes the expected level
of wage dispersion due to the higher productivity resulting from good matches. Search
costs introduce a friction that reduces the expected extent of this dispersion, but in either
case inequality increases with heterogeneity in the quality distribution of employers
(Kremer, 1993; Wheeler, 2001).
Organizational diversity and inequality
A very different perspective on the social implications of the number of
employers available arises from the literature on human and organizational ecology.
Ecologists assume that organizations in a community occupy more or less well-
differentiated niches, so that they survive by drawing on their own combinations of
resources from the environment (Hawley, 1950; Hannan and Freeman, 1977). In the labor
market, for example, some firms rely more heavily on employees with physical strength
and manual dexterity, while others rely on workers with strong analytic reasoning and
communication skills. Employees likewise have diverse sets of abilities and attributes. If
5 Though sequential search might seem unreasonable, one could think of it as being equivalent to a
situation in which workers choose to work for the highest quality firm when they receive more than one offer for employment.
12
these characteristics match well with the demands of their employer, workers receive
rewards in the form of bonuses, raises and internal promotion. A poor fit, on the other
hand, relegates them to lower pay and limited opportunities for advancement. In a world
with only one employer (or in which all employers look for the same types of
employees), one would expect substantial inequality as some individuals fit well to the
requirements of this organization and receive ample compensation, while others do not
(Hannan 1988). Moreover, those without the ‘right’ skills have little opportunity to
change employers; as Glaeser (1994: 19, quoted in Wheeler 2001) observes, “in a one-
company town, individuals who are imperfectly matched to that company have nowhere
else to go. These workers will stay at the company (or leave the town), being
underproductive.”
As the diversity of employers – and consequently the variety of the employee
abilities and attributes they desire – increases, inequality should decline because a larger
proportion of the population can find employers that desire their characteristics, leaving
few un- or under-employed. In a conceptual piece discussing the implications of this idea,
Hannan (1988) highlights the likely importance of three dimensions of the industrial
population: the number of firms, their variation in ages and their differences in size. To
the extent that firms occupy distinct niches, increasing numbers of firms should reduce
inequality because each firm prefers employees with somewhat different abilities and
attributes. Similarly, conditioning on the number of firms, populations with more diverse
firm age and size distributions likely exhibit even greater heterogeneity in the
characteristics they desire. Firms often imprint on a set of operating routines and a
production technology when they begin (Baron, Burton and Hannan 1996); hence, those
13
founded at different times often operate with varying production and organizational
models, potentially drawing on divergent employee abilities and attributes. Likewise,
firm size typically correlates to both market position and the method of production,
meaning that firms of varying sizes also likely prefer different employees (cf. Greve,
1994; Fujikawa-Greve & Greve, 2000).
Though less prominent than the vertical sorting models described above, some
theoretical work in economics building on the so-called “Roy model” follows a similar
line of reasoning. Roy (1951) presented a model of an economy with two sectors, hunters
and fishers (for formalizations, see: Sattinger, 1975; Heckman and Sedlacek, 1985;
Heckman and Honore, 1990). Each type of work requires its own set of skills. Employees
choose the sector in which they work on the basis of comparative advantage – in other
words, they select the job in which they can expect to be most productive, and
consequently to earn the highest wages. As a result of this type of self-selection, the two-
sector economy yields higher average and lower variance in wages as long as the
different skills do not correlate perfectly. Elaborations of this model have revealed that
the basic result holds even if workers do not know their types and must search
sequentially for the right employer for their skills (MacDonald, 1982). Empirical
verification of this model, however, has been limited to demonstrating that two- and
three-sector models of the economy explain the distribution of wages better than a one-
sector model (Heckman and Sedlacek, 1985; Gould, 2002). The Roy model, therefore,
while pointing to a similar logic for the link between organizational diversity and income
inequality, has only considered much coarser-grained differences across employers.
14
Vertical and horizontal sorting
Both perspectives we have described rely on a matching process between
employers and employees, but differ on a key assumption. Whereas the first set of
models, such as those proposed by Rosen (1981, 1982) and Wheeler (2000), assumes a
one-dimensional distribution of worker characteristics and employer interests, the
ecological perspective assumes that these distributions exist in a multi-dimensional space.
This subtle difference in assumptions, far from being simply an issue of theoretical
convenience, is the source of their opposing predictions regarding the effects of increases
in the number of firms in a labor market. The difference parallels the distinction between
vertical and horizontal differentiation in product markets in the marketing literature (cf.
Gabszewicz and Thisse, 1979; Shaked and Sutton, 1982). Under vertical differentiation,
one assumes that all buyers would prefer the same products (workers) at a fixed level of
price (e.g., virtually everyone would prefer a business class air ticket to coach if it did not
cost extra). Horizontal differentiation, on the other hand, allows for the possibility that
consumers (employers) prefer different product (employee) attributes.
We have little reason to believe, however, that the empirical reality of corporate
demography and labor markets corresponds neatly to either of these stylized assumptions;
changes in industrial demography can generate sorting along both the vertical and
horizontal dimensions. Individuals have qualitatively different abilities and attributes,
such that their desirability depends on the employer. If organizations differ sufficiently in
their needs, the competition for labor therefore may not generate the determinate rank
ordering envisioned in the vertical sorting model. Similarly, while individual
organizations may in principle occupy distinct niches, many organizations overlap highly
with each other in terms of their demand for labor (Sørensen, 1999, 2004), suggesting
15
that the impact of an increase in the number of firms depends on their diversity. Thus, it
does not seem inconsistent with the ecological argument to expect that increases in the
number of firms of a particular type could produce sorting and increased inequality.
In this sense, one can view the vertical sorting and ecological (or horizontal
sorting) arguments as complements to each other. Among organizations drawing on a
specific set of human resources, such as those operating within the same industry using
similar production technologies and hiring logics, the vertical sorting argument likely has
the greatest force in explaining the effects of increases in competition. By pointing to the
importance of vertical differentiation within a niche, it provides a richer
conceptualization of the ecological argument of the labor market consequences of
changes in industrial demography. The ecological insights, by contrast, highlight an
important role for organizational diversity neglected in the vertical sorting model’s
(implicit) assumption of horizontally undifferentiated firms. As firms within a labor
market draw on a wider range of human resources, vertical sorting – and with it wage
inequality – declines as employees more likely find firms that fit well to their abilities
and attributes.
EMPIRICAL STRATEGY
Our argument suggests that two dimensions of corporate demography contribute
importantly to the generation of inequality: the number and the diversity of employers.
We examined the impact of these factors using a unique dataset characterizing the Danish
labor market from 1980 to 1998. This dataset allows us to relate regional variation in
wage inequality among employed workers to regional differences in corporate
demography. We begin by simply comparing the gross income inequality in an industry
16
within a particular region to the distribution of firms in the region, both within and
outside the focal industry. This analysis offers a picture of the overall relationship
between corporate demography and wage dispersion. This gross relationship may result
from differences both in the demand for observable and unobservable individual
characteristics, as well as from endogenous processes influencing these distributions.
Some employers, for example, might prefer more experienced workers, while others
might desire greater levels of education. On the other hand, this analysis cannot account
for differences in inequality arising from exogenous differences in the characteristics of
workers – in other words, variation across regions in wage dispersion might reflect
heterogeneity in the distribution of employees rather than of employers. For example,
highly educated workers may concentrate in particular regions of the country. Hence, we
also considered a substantially more conservative approach, where we estimated the
degree of wage inequality, net of individual observable characteristics, to assess the
effects of corporate demography on income inequality.
Wage setting in Denmark
One would only expect the predictions of the two theories we have outlined to
hold in labor markets that meet certain scope conditions. Two conditions are particularly
important. First, since the matching of employees to employers likely occurs primarily
through the movement of employees between jobs and firms – at least within regions –
both theories presuppose that a fair degree of flexibility characterizes the labor market, in
the sense that both workers and firms can terminate employment relationships on short
notice. Second, for wages to reflect the quality of the match between the worker and the
firm, employers must also have some discretion over the wages that they pay to
17
individual workers so that they have the freedom to adjust pay to (perceived)
productivity.6
With respect to the first scope condition, observers generally consider the Danish
labor market quite flexible, particularly by European standards. Employers in Denmark
incur very low firing costs, leading it to have the lowest level of employment protection
among European OECD countries (OECD, 1994), at a level on par with the United States
(Bingley and Westergaard-Nielsen, 2003). Perhaps as a result, the Danish labor market
has both high rates of mobility between firms and relatively short lengths of firm tenure.
Albæk and Sørensen (1998), for example, estimated an average annual rate of job
creation of 12% and an average annual level of job destruction of 11.5%, between 1980
and 1991. By comparison, Davis and Haltiwanger (1992) estimated annual job creation
and destruction rates in U.S. manufacturing between 1980 and 1988 of 8.4% and 11%.
Furthermore, Denmark has a low mean firm tenure, at the bottom of the range for OECD
countries along with the United States, Australia and the United Kingdom (OECD, 1997).
Bingley and Westergaard-Nielsen (2003) argue that this fact both reflects the low degree
of employment protection and results from the high degree of wage compression within
firms, which encourages productive workers seeking pay increases to seek out higher-
paying employers.7
As for the second condition, wage setting in Denmark historically has been quite
centralized, thanks in part to the high rates of unionization (over 70% [Iversen, 1996]).
From the 1930s through the 1970s, centralized bargaining between associations with
6 In principle, the scope conditions of the theories could be met even if individual employers had to pay all employees performing the same task the same wage, regardless of individual differences in productivity – as long as heterogeneity existed among firms in the wage paid and the labor market allowed a sorting of workers to firms.
7 High rates of turnover also lower the value of firm-specific training, and therefore may cause employers to focus more strongly on employees whose skills match well to the job.
18
broad memberships representing organized labor on the one side and employers on the
other primarily determined wage rates, leading to extremely low levels of income
inequality relative to other countries (Iversen, 1996). During the period we study,
however, the centralized wage bargaining system began to break down. The first element
of this decentralization, during the early 1980s, involved a shift in the role of the
nationwide labor and employer associations from reaching enforceable agreements to
setting non-enforceable targets, allowing industry-level associations to negotiate
enforceable agreements (Iversen, 1996; Madsen, Andersen and Due, 2001).
Even greater employer discretion over wage setting came later in the 1980s as this
centralized system gave way to the so-called “normalløn” system. Under this regime,
employers could pay wage supplements to all employees in the firm, supplements to all
occupants of particular jobs, and performance bonuses, though these supplements still
required agreements between the firms and unions at the local level. Iversen (1996)
argues that much of the pressure for decentralization came from employers, who, in the
face of increased international competition, felt the need for greater control over their
incentive structures. Another important shift in the level of the centralization came in
1991 with the replacement of the nomalløn system with the “minimalløn” system.
Whereas the prior system primarily allowed for differentiation through bonuses or
distinct job titles, the minimalløn system relaxed restrictions on the normal wage rates
(without bonuses) for particular jobs. The centralized bargaining process largely limited
itself to setting a minimum wage for a job. Employers could then set the wage for a
particular worker according to the worker’s productivity, experience and effort.
19
The minimalløn system has spread rapidly in the Danish labor force. By 1997 (the
end of the period we study), wage systems that included decentralized wage setting
covered 84% of the private sector labor force (Dansk Arbejdsgiverforening, 2000). Thus,
although individual firms in the 1960s and 1970s played a relatively small role in setting
individual wage rates, their influence has increased dramatically in the 1980s and
particularly in the 1990s. Consistent with this increased decentralization of wage setting,
Bingley and Westergaard-Nielsen (2003) observe an increase in the returns to firm
tenure, which they interpret as evidence of the increased ability of firms to use incentives
to retain valuable workers.
Data and measures
We analyzed data from government registers maintained by Statistics Denmark
and collected in the Integrated Database for Labor Market Research. Referred to by its
Danish acronym, IDA, this database contains comprehensive, annual information on
individuals and work establishments in the Danish market. The data amount to an annual
census: IDA includes all individuals legally residing in Denmark in a given year.
Moreover, IDA contains a wealth of individual characteristics, including information on
family structure, educational attainment, income, and work experience. Most importantly
for our purposes, the dataset links individual employees to their work establishments and
employers. Though IDA contains only limited information about each of these
establishments, it does identify both their industry and geographic location. It categorizes
workplaces into one of 111 industries using a standard (ISIC, rev. 2) classification
scheme. Our analysis, however, did not extend to all of these industries. Notably, we
excluded industries in the primary (agricultural and extractive) sector, as well as
20
industries in which public sector employment accounted for more than 15% of the
workforce.8 As a result, the sample analyzed includes 81 industries.
With respect to geography, the database associates each workplace with a
township (“kommune” in Danish). Townships, the smallest administrative units in
Denmark, divide the country into 275 mutually exclusive and exhaustive regions. In
considering the analysis of regional variation in wage inequality, the appropriate level of
aggregation is a salient concern. In particular, the sorting and diversity arguments appear
potentially sensitive to the level of aggregation at which one computes the corporate
demography measures. For example, the presence of alternative industries in nearby
townships may affect workers in a township with limited industrial diversity. We see no
easy solution. Some scholars argue that “journey-to-work” boundaries should define
regional labor markets (Beggs and Villemez, 2001), but such boundaries become at least
partially endogenous to the attractiveness of the employment opportunities in a given
area. To assess the potential impact of measuring corporate demography at the lower
level of aggregation, we computed an alternative set of demographic measures using a
79-category aggregation of townships. This aggregation combines townships based on
commuting patterns in 1980 (Andersen, 2000), and thus corresponds to the definition of a
regional labor market in terms of journey-to-work boundaries. These alternative analyses
proved reassuring, as they paralleled the conclusions reached with the more
disaggregated measures that we report here.9
8 The distribution of public sector employment is highly skewed: 62 of the 111 industries have no
public sector employment, and in 87 industries (including the primary sector), less than 5% of employees work in the public sector.
9 We also estimated models that included dummy variables for the 16 counties in Denmark (the administrative units above townships), to account for potential spatial autocorrelation in the determinants of wage dispersion. McCall (2000) similarly uses state-level dummies to test for unobserved heterogeneity
21
Our analyses drew on a subset of the IDA database constructed and maintained by
the Center for Labor Market and Social Research in Denmark. This “Pay and
Performance” dataset covers all private-sector employees in a given year; it excludes
employees of the (sizable) public sector, the self-employed, and individuals not in the
labor force. For the purposes of our investigation, these exclusions make sense: the wage-
setting dynamics in the public sector likely depend on a host of factors not considered in
the theories outlined above, and the other two groups do not receive wages.
We analyzed data from 1980 to 1998. When estimating the models, we split the
analyses into two time periods: one running from 1980 to 1991, the other from 1992 to
1998. Two factors lead us to split the data along these lines. On the theoretical side, the
collective bargaining agreements reached in 1991 marked the earliest adoptions of the
minimalløn system for wages paid in subsequent years. As noted above, this system
substantially increased the level of flexibility available to employers in the wage setting
process, so we expected that the sorting processes might operate more strongly under this
system. From a more practical perspective, Statistics Denmark also shifted to a new
industry classification scheme in 1992. Though a mapping of old to new categories
exists, splitting the analysis offered a conservative approach to ensuring that this change
in coding would not influence our results.
To assess the influence of corporate demography on wage inequality, we
regressed two measures of wage dispersion on measures of the local industry
demography: gross wage dispersion and within-group wage dispersion.
across cities in the United States. Though the incorporation of these effects increases the standard errors, the results reported below remained robust to their inclusion.
22
Gross wage dispersion: We measured gross wage dispersion as the logged
standard deviation of wages across all workers employed within a particular industry
within a township.10 The models of gross wage dispersion included the mean wage for
workers in an industry in a township since the magnitude of the standard deviation
depends in part on the mean.
As noted above, this gross measures captures inequality from several sources: (i)
differential returns to observed characteristics across regions; (ii) differential returns to
unobserved characteristics across regions; (iii) endogenous investments in abilities and
attributes in response to these differential returns; and (iv) exogenous variation in the
distribution of individual characteristics. Although one might reasonably attribute the
first three to sorting processes, the fourth source belongs in a separate category. Hence,
the estimates using gross wage dispersion measure provide an upper bound on the
importance of corporate demography to inequality. An ideal measure would eliminate the
last, exogenous source alone. Doing so, however, is impossible with the data at hand, so
instead we turned to an even more conservative approach (a lower bound): We estimated
the effects of sorting stemming only from the second source – differential returns to
unobserved characteristics – by using the within-group wage dispersion as a dependent
variable.
Within-group wage dispersion: We estimated the effects of corporate demography
on within-group wage dispersion using a two-step hierarchical linear modeling strategy.
Our approach is similar to that adopted by McCall (2000) in her analysis of regional
10 Logging the standard deviation accounts for the likely distribution of the dependent variable.
The distribution of a sample of standard deviations from a lognormal distribution (which the raw wage data match almost perfectly) is log-normally distributed. Despite this fact, a set of models using the raw standard deviation as the dependent variable yielded substantively similar results.
23
wage inequality in the United States (see also Blau and Kahn, 1996; Rauch, 1993).11 In
the first stage we estimated, for each year, a linear regression model (using OLS) of an
employee’s log wage from their primary job as a function of a variety of individual
characteristics: occupation,12 labor force experience, labor force experience squared,
years of education, firm tenure, sex, marital status, an interaction between sex and marital
status, and the number of children under the age of two.13 These regressions also included
fixed effects for each industry and township. These fixed effects purged from the model
the mean wage in each industry as well as the mean wage in each township, ensuring that
the remaining variation in incomes does not reflect differences in levels across industries
and townships.
We used the first stage regressions to generate the measure of within-group
income inequality used in the second stage. The dependent variable for the second stage
is the natural log of the standard deviation of the residuals for workers in a particular
industry in a particular township.14 As McCall (2000: 419) notes, one can interpret the
residual standard deviation as “a measure of variation in the earnings of workers with the
same observed characteristics, some unknown portion of which is due to differences in
the distribution of, and returns to, unobserved characteristics.” We computed this
measure for each observed combination of industry i, and township j, at year t. The
11 Our estimation procedure differed from McCall’s (2000) in one respect. Where she estimated
separate regressions in the first stage for each region in her dataset, we estimated a pooled regression across all regions in the first stage. Since we split the sample into much smaller regions, separate regressions would in many cases entail very small sample sizes. Substantively, this difference implies that we assume that the income returns to the observable individual level characteristics (such as work experience) do not vary across regions.
12 Occupation here refers to seven broad categories ranging from top management to unskilled worker, entered as dummy variables.
13 Since the sample consists of all private sector employees with valid data in a given year, the sample sizes for the first stage regressions range from over 1.3 million to more than 1.6 million cases.
14 We adjust for sampling error by adding 1/(2*n) where n denotes the number of individuals in the cell (Bryk and Raudenbush, 1992).
24
logged residual standard deviation for cell i,j,t therefore becomes our measure of income
inequality among workers in industry i, and township j, for year t. Higher values indicate
greater dispersion in the incomes of these workers, net of their observed characteristics.
Measuring corporate demography: We generated three measures to assess the
salience of the demographic characteristics of the local industry population on wage
dispersion. First, we used the simplest representation of the population, the count of the
number of firms in an industry, to capture the level of within-industry labor market
competition. We experimented with different functional forms for the effect of the
number of firms. A comparison of the curve implied by the logged number of firms to a
seven piece spline revealed almost no difference between the two. In fact, the splined
estimation improved the variance explained by only 0.3%. Given this small difference,
we retained the more parsimonious models using the logged firm count. Second, since
markets with only a single employer in an industry may differ qualitatively from those
with two or more competing firms, we also included a dummy variable for those
industry-township cells with only one firm. Finally, using employment shares, we
calculated the standard deviation of firm size in the industry; to the extent that production
systems and/or hiring practices differ systematically with scale, this measure should also
capture horizontal sorting.
To measure the importance of the diversity of firm types in a local labor market,
we created two alternative measures. The number of industries simply counts the number
of industries represented in the local population of firms.15 This number does not,
however, account for the distribution of activity across these sectors, which also likely
15 Up to 111 industries may be present in any given township, since we include primary sector
industries and industries dominated by the public sector in this measure.
25
affects the diversity of opportunities available to any particular individual. Hence, we
calculated an entropy measure to capture both the number of industries represented in the
local population as well as the concentration of employment shares across them.16
In addition to these measures of interest, the models include measures of three
other characteristics of the local labor market: the total number of employees in the
township across all industries, the total number of employees in the township in the focal
industry, and the total number of employers in the township across all industries (see
Table 1 for descriptive statistics). To the extent that industries have common demands for
employee skills, one would expect wage inequality to rise with the total number of
employers in a township as a result of vertical sorting. Also, wage dispersion likely has a
negative relationship with both the total number of employees in a township and the
number of employees within a particular industry in a township. When controlling for the
number of employers, these measures essentially capture average organization size. If
wages vary less within than across organizations, one might therefore expect regions with
larger organizations to exhibit lower inequality.17
RESULTS
Table 2 reports OLS estimates of models of the effects of corporate demography
on gross income inequality for the period 1980 to 1991, along with standard errors
adjusted for the clustering of observations at the township level. Table 3 presents the
same models estimated for data from 1992 to 1998. All models included dummy
16 The entropy measure we use is -Σpilog(pi), where pi is industry i’s share of private employment
in a township. 17 Wages might vary less within organizations for a variety of reasons: Job classifications may
vary less within than across firms. Similarly wage-setting regimes typically differ more greatly across firms. Organizations may also compress wages in the interest of improving worker perceptions of fairness.
26
variables for year and industry, so the estimates reflect the impact of regional differences
in corporate demography on regional wage inequality after purging the data of common
industry and temporal components. The dependent variable in Tables 2 and 3 is the
logged standard deviation of wages, multiplied by 1,000 for ease of presentation.
The estimates in these two tables generally support the expectations derived from
the sorting processes. Consistent with the expectations of the vertical sorting argument,
the estimates in the first column of both Tables 2 and 3 indicate that the dispersion of
wages in an industry increases with the number of employers in that industry in a
township. The dummy variable for a single employer, moreover, reveals a pronounced
increase in inequality as a township moves from having one to two employers within a
particular industry. Total industry employment meanwhile tends to depress wage
dispersion. As noted above, when included in a model with the number of firms, this
variable provides information on average firm size. In essence, it suggests an inverse
correlation between wage inequality and average firm size, which one might expect if
wages vary less within than across firms.
The second models in Tables 2 and 3 introduced one measure of the diversity of
employment opportunities, the count of industries in the township. Consistent with the
expectations of the ecological diversity argument, this variable has negative coefficient
estimates in both time periods, although the effect is not statistically significant in the
second period (Table 3). The estimated effects of the alternative entropy-based measure,
included in the third column of the tables, also have the expected valence, but do not
prove statistically significant in either time period. The final two models in both Table 2
and Table 3, however, which include interaction effects between the diversity measures
27
and the logged number of employers, provide support for the ecological argument. In
both time periods, the level of local wage inequality in an industry declines as the
diversity of employers increases. Moreover, the statistically significant interaction effect
suggests that the benefits of organizational diversity increase as the pressure for vertical
sorting, driven by the number of employers in an industry, rises.
Although the results in Tables 2 and 3 provide evidence of the important role that
industrial demography plays in inequality, they cannot escape the criticism that the
distribution of workers with different human capital characteristics across regions may
not vary randomly with respect to the distribution of firms and industries. For this reason,
Tables 4 and 5 present estimates of the same models, except that we used our within
group (i.e. residual) measure of wage inequality, purged of observed human capital
characteristics (once again, multiplied by 1,000 for ease of presentation). The results
parallel those using the gross inequality measure as the dependent variable. Competition
among employers within an industry continues to have a substantial positive effect on
residual wage inequality. As before, the diversity of employers generally lowers wage
inequality, and the magnitude of the diversity benefit increases with the number of
employers in an industry (the interaction term).
In Tables 4 and 5, we also note that wage dispersion appears to decline with the
diversity of employer firm sizes. One could interpret this effect of employment
concentration in at least two ways. On the one hand, a large number of small firms may
generate more intense labor market competition (Guadalupe, 2003), thereby magnifying
the sorting effects and wage dispersion. On the other hand, diversity in firm size may
reflect differences in organizational models and production processes, and concomitantly
28
the degree of diversity in the opportunity structure (Hannan, 1988; Greve, 1994). Small
firms more likely engage in craft-like production, while larger firms tend toward more
automated, mass market models. Small firms also typically employ more informal hiring
procedures, which may favor different individuals from the more bureaucratic human
resource policies typical of larger firms.
DISCUSSION
Our statistical analyses of regional variations in wage dispersion in Denmark
indicate that the degree of inequality in a local labor market depends on the demography
of employers in that labor market. In particular, our analyses suggest that the dispersion
in wages among workers in an industry depends both on the number of co-located
employers in the same industry, and on the availability of employers with different
production processes and technologies. More pronounced competition within the industry
engenders greater inequality through vertical sorting, while increased organizational
diversity dampens it.
Though both vertical sorting and ecological diversity have significant effects, a
comparison of the relative magnitudes of these forces proves instructive. The
standardized coefficient estimates, reported in Table 6, point to a substantial vertical
sorting effect. For example, columns one through four indicate that a one standard-
deviation increase in the (log) number of firms leads to a 16%-19% increase in gross
inequality in both time periods. As noted earlier, however, this overall effect could stem
from multiple sources: differences in the matching process on both (i) observable and (ii)
unobservable characteristics between employees and employers; (iii) endogenous
heterogeneity across regions in human capital as a result of differences in the returns to
29
investments in education and experience; and (iv) exogenous variation in human capital
across regions.
The more conservative approach, reported in columns five through eight, focuses
only on the matching of employees to employers based on unobserved employee
attributes. Not surprisingly, this approach dramatically reduces the standardized effect. A
one standard deviation increase in the log number of firms in an industry only increases
the within group inequality by 5%-6%. The true total effect of vertical sorting on wages
falls somewhere between these two estimates. Though any estimate of total effects would
ideally exclude exogenous sources of productivity differences, they should include not
just the sorting on unobservable dimensions captured in the analysis of the residual wage
dependent variable, but also sorting on observable characteristics and endogenous
heterogeneity in human capital investment due to differences in employment
opportunities. Isolating the exact magnitude of the overall effect, however, will likely
prove elusive, as it requires a means of differentiating between exogenous differences in
the distribution of individual characteristics and endogenous investments in human
capital.18
By contrast, the magnitude of the diversity effects appears substantially smaller
than that of vertical sorting. The size of the diversity benefit for employees in a particular
industry, however, also depends on the number of firms in that industry, making
interpretation somewhat difficult. Consider the effects of the number of industries for
overall inequality in the first time period (the fourth model in Table 2); Figure 1
illustrates the effects graphically. Industrial diversity has little effect of its own on gross
18 The usual instrumental variables approaches to estimation do not apply in a straightforward
manner here because they depend on the relationship between the levels of variables rather than in the degree of variance produced by the sorting process.
30
inequality. It primarily acts to mitigate the intensity of vertical sorting. For example, at
the minimum level of industry diversity (a region with only one industry), a one standard
deviation increase in the number of firms in the industry corresponds to an 18% rise in
inequality. In a region with maximal diversity (including employers from all industries),
the same increase in the number of firms would result in only a 3% rise in gross wage
inequality.
Although the interaction effects make it difficult to parse the exact proportions,
our results suggest that, as measured here, vertical sorting processes account for more of
the variation in regional wage inequality than organizational diversity. At least two
factors could account for this imbalance. On the one hand, the relative importance of
vertical sorting may indicate that employers in different industries overlap to a large
extent in their labor requirements. In the vertical sorting argument, greater inequality
results as increased competition between firms magnifies individual differences in
productivity. A portion of these productivity differences stems from the fact that certain
individuals possess the specialized skills that some employers demand, while others do
not, as in our earlier example of baseball and basketball players. Yet these productivity
differences also stem from heterogeneity in general abilities and skills, transferable across
employers; the star basketball player will likely play baseball better than most other
people. The relative magnitudes of the vertical and horizontal differentiation effects
therefore depend on the extent to which employers overlap in their demands for labor; in
particular, as employers’ labor demands increasingly overlap, vertical sorting pressures
rise (cf. Gould, 2002).
31
On the other hand, this difference in the relative importance of vertical sorting
may reflect how well our measures of organizational diversity capture the extent of
horizontal differentiation among employers in their labor demands. Our measures of
diversity rely on the relatively coarse information available in an aggregated industry
classification, which renders at best a crude approximation of the true degree of
organizational diversity. Moreover, industry classification schemes have been developed
as a means of sorting organizations according to their output (products or services) and
may have limited power to distinguish across organizations in their demand for labor as
an input. Though imperfect, these industry codes nonetheless offer one of the few
accepted means for measuring organizational diversity on a large scale.
Although our estimation strategy only uses geographic variation as a means of
gaining empirical leverage on the identification of the effects of corporate demography
on inequality, our results nonetheless have interesting implications with respect to the
literature on economic geography. Industrial clusters typically refer to regions with a
large number of small firms operating in a single industry, where those firms account for
a large share of overall employment. Policymakers have shown considerable interest in
using industrial clusters as a model for economic development, and economists have
forwarded several economic models to explain why such a configuration of productive
activity might stimulate growth (for a review, see Sorenson and Audia, 2000). Little
attention, however, has been given to the potential consequences of clustering beyond
firm performance. Our results suggest that clustering may have important consequences
for the level of inequality in society. Clusters reside at the back, left corner of Figure 1
32
(many firms in a region with little industrial diversity); these regions generate the most
intense vertical sorting processes, and hence the greatest degree of income inequality.
The insight that corporate demography influences the stratification process
through its implications for the matching of individuals to employers also potentially
opens new ground for future research. For example, our current analyses do not consider
under what conditions these processes should operate most strongly. Future work may
thus consider what factors amplify and mitigate the effects of competition and sorting.
Consider, for example, search costs. These costs introduce friction into the matching
process, thereby reducing the efficiency of matches between firms and employees (i.e.
increasing the likelihood that a high quality firm employs a low quality employee) and –
because the inefficient matching leads to regression to the mean in firm productivity –
concomitantly depressing income inequality. Such search costs would have only limited
bearing on the relationship between organizations and inequality if all firms encountered
equivalent costs, but they may vary systematically across firms or industries due to labor
demands or operating routines (e.g., Gould, 2002), or across regions as a result of
differences in institutional environments (McCall, 2000).
Individuals may also vary in the degree to which these sorting processes constrain
them. The persistence of differences across regions in their local inequality depends to
some extent on the geographic immobility of individuals; otherwise, the entire country
would operate as a single labor market and regions would differ little in their degrees of
wage dispersion. Individuals nonetheless differ in the degree to which they remain rooted
in a local community. For instance, women, and particularly single mothers, typically
commute much shorter distances than men (e.g., Gordon, Kumar and Richardson, 1989).
33
If men’s greater proclivity to travel to work allows them access to regions offering higher
returns for their skills, this difference in geographic mobility could contribute to
inequality between men and women in labor market outcomes. Hence, future research
might fruitfully investigate the ways in which organizational demography contributes to
stratification across different types of individual attributes.
CONCLUSION
Few sociologists today would deny the important ways in which organizations
“link the ‘macro’ and ‘micro’ dimensions of work organization and inequality” (Baron
and Bielby, 1980: 738). Indeed, decompositions of the sources of income inequality
across individuals suggest that cross-firm differences in compensation account for as
much inequality as human capital measures (Groshen, 1991), and that the importance of
these factors in explaining inequality may have risen over the last three decades
(DiNardo, Fortin and Lemieux, 1996). At the same time, a curious disjuncture has
developed in the literature, where work on the role of organizations in the stratification
process has largely grown apart from a larger literature on the determinants of inequality
at higher levels of analysis, such at the regional level (e.g., McCall, 2000) or across
society as a whole (e.g., Nielsen and Alderson, 1997), which scarcely mentions the
importance of organizations in this process.
This state of affairs reflects the fact that research on the role of organizations in
stratification has largely adopted a “focal organization” perspective (Scott, 2002) and
thus typically does not consider how these organizational level outcomes aggregate to
produce inequality at the societal level. The simple summation of processes internal to
individual, focal organizations offers one means of aggregating these outcomes to more
34
macro levels of inequality. But such an approach could easily lead researchers astray.
Individual organizations reside within a broader ecology of organizations, and
competitive (as well as symbiotic) relationships between firms have important
implications for organizational behavior.
Though beyond the scope of our research, our results may contribute to a better
understanding of the broader trends in income inequality. For instance, one of the puzzles
in inequality research concerns the remarkable rise in wage and income inequality in
advanced economies, particularly in the last decades of the twentieth century, following
decades of decline in inequality (Nielsen and Alderson, 1997; Morris and Western,
1999). Though a wide variety of factors likely contribute to this upswing in inequality,
one commonly discussed factor is the change in the industrial structure of advanced
economies (Morris and Western, 1999). This change in structure has largely taken the
form of a reduction in industrial diversity, as employment has moved away from a
declining manufacturing sector to service sector industries. To the extent that different
sectors increasingly relying on the same types of skills, increased inequality emerges as a
consequence of this process (cf. Gould, 2002). Changes in the number and size
distribution of firms over time may also contribute to rising inequality. In the first half of
the century, the number of employers declined rapidly as industries consolidated to
realize economies of scale through mass production. More recently, however, more
flexible manufacturing technologies have engendered a boom of specialists, dramatically
expanding the number of employers in many industries and reducing average firm size
(Carroll and Hannan 2000: 20). The coincidence between these trends suggests that
35
future research might also usefully consider the linkage between the evolution of industry
structure and societal levels of inequality.
36
REFERENCES
Albæk, Karsten, and Bent E. Sørensen (1998). “Worker flows and job flows in Danish manufacturing, 1980-91.” Economic Journal, 108: 1750-1771
Albrecht, James W., Bo Axell, and Harald Lang (1986). “General equilibrium wage and price distributions.” Quarterly Journal of Economics, 101: 687-706.
Barnett, William P., James N. Baron, and Toby E. Stuart (2000). “Avenues of attainment: Occupational demography and organizational careers in the California civil service.” American Journal of Sociology, 106: 88-144.
Baron, James N., and William T. Bielby (1980). “Bringing the firms back in: Stratification, segmentation and the organization of work.” American Sociological Review, 45: 737-765
Baron, James N., M. Diane Burton, and Michael T. Hannan (1996). “The road taken: Origins and Evolution of Employment Systems in Emerging Companies.” Industrial and Corporate Change, 5: 239-276.
Becker, Gary S. (1973). “A theory of marriage: Part I.” Journal of Political Economy, 81: 813-846.
Becker, Gary S. (1964). Human Capital. New York: NBER and Columbia University Press.
Beggs, John J., and Wayne J. Villemez (2001). “Regional labor markets.” Pp. 503-530 in I. Berg and A.L. Kalleberg (Eds.), Sourcebook on Labor Markets: Evolving Structures and Processes. New York: Plenum
Bhaskar, V., Alan Manning, and Ted To (2002). “Oligopsony and monopsonistic competition in labor markets.” Journal of Economic Perspectives, 16: 155-174
Bingley, Paul, and Niels Westergaard-Nielsen (2003). “Returns to tenure, firm-specific human capital and worker heterogeneity.” International Journal of Manpower, 24: 774-788.
Blau, Francine D., and Lawrence M. Kahn (1996). “International differences in male wage inequality: Institutions versus market forces.” Journal of Political Economy, 104: 791-837
Bryk, Anthony S. and Stephen W. Raudenbusch (1992). Hierarchical Linear Models: Applications and Data Analysis Methods. Newbury Park, CA: Sage
Carroll, Glenn R., and Michael T. Hannan (2000). The Demography of Corporations and Industries. Princeton, NJ: Princeton University Press
Dansk Arbejdsgiverforening (2000).
DiNardo, John, Nicole M. Fortin, and Thomas Lemieux (1996). “Labor market institutions and the distribution of wages, 1973-1992: A semi-parametric approach.” Econometrica, 64: 1001-1044
37
Davis, Steven J., and John C. Haltiwanger (1992). “Gross job creation, gross job destruction and employment reallocation.” Quarterly Journal of Economics, 107: 819-863
Fujiwara-Greve, Takako, and Henrich Greve (2000). “Organizational ecology and job mobility.” Social Forces, 79: 547-568
Gabszewicz, Jean J., and Jacques-Francois Thisse (1979). “Price competition, quality and income disparities.” Journal of Economic Theory, 22: 340-359
Glaeser, Edward L. (1994) “Economic Growth and Urban Density: A Review Essay.” Working paper E-94-7. Stanford, CA: Hoover Institution.
Goldin, Claudia, and Lawrence F. Katz (1998). “The origins of technology-skill complementarity.” Quarterly Journal of Economics, 113: 693-732
Gordon, Peter, Ajay Kumar, and Harry Richardson (1989). “Gender differences in metropolitan travel behavior.” Regional Studies, 23: 499-510
Gould, Eric D. (2002). “Rising wage inequality, comparative advantage, and the growing importance of general skills in the United States.” Journal of Labor Economics, 20: 105-147
Greve, Henrich (1994). “Industry diversity effects on job mobility.” Acta Sociologica, 37: 119-139
Groshen, Erica (1991). “Sources of intra-industry wage dispersion: How much do employers matter?” Quarterly Journal of Economics, 106: 869-884
Guadalupe, Maria (2003). “Does product market competition increase wage inequality?” Working paper, London School of Economics
Hannan, Michael T., and John Freeman (1977). “The population ecology of organizations.” American Journal of Sociology, 82: 929-964
Hannan, Michael T. (1988). “Social change, organizational diversity and individual careers.” Pp. 161-174 in M.W. Riley (Ed.), Social Structure and Human Lives. Newbury Park, CA: Sage
Haveman, Heather A., and Lisa E. Cohen (1994). “The ecological dynamics of careers: The impact of organizational founding, dissolution, and merger on job mobility.” American Journal of Sociology, 100: 104-152
Hawley, Amos H. (1950). Human Ecology: A Theory of Community Structure. New York: Ronald Press
Heckman, James J., and Bo Honore (1990). “The empirical content of the Roy model.” Econometrica, 58: 1121-1149
Heckman, James J., and Guilherme L. Sedlacek (1985). “Heterogeneity, aggregation and market wage functions: An empirical model of self selection in labor markets.” Journal of Political Economy, 93: 1077-1125
Idson, Todd L., and Walter Y. Oi (1999). “Workers are more productive in large firms.” American Economic Review, 82: 104-108
38
Iversen, Torben (1996). “Power, Flexibility and the Breakdown of Centralized Wage Bargaining: The Cases of Denmark and Sweden in Comparative Perspective.” Comparative Politics, 28: 399-436.
Kalleberg, Arne L., David Knoke, Peter V. Marsden, and Joe L. Spaeth. 1996. Organizations in America: Analyzing their Structures and Human Resource Practices. Newbury Park, CA: Sage.
Kremer, Michael (1993). “The O-ring theory of economic development.” Quarterly Journal of Economics, 108: 551-575
Lydall, Harold F. (1959). “The distribution of employment incomes.” Econometrica, 27: 110-115
MacDonald, Glenn M. (1982). “A market equilibrium theory of job assignment and sequential accumulation of information.” American Economic Review, 72: 1038-1055.
Madsen, Jørgen Steen, Søren Kaj Andersen, and Jesper Due. (2001). “From centralised decentralisation towards multi-level regulation: Danish employment relations between continuity and change.” Working paper, Employment Relations Research Centre, University of Copenhagen.
McCall, Leslie (2000). “Explaining levels of within-group wage inequality in U.S. labor markets.” Demography, 37: 415-430.
Merton, Robert K. (1948). “The self-confirming prophesy.” Antioch Review, Summer: 193-210
Morris, Martina, and Bruce Western. (1999). “Inequality in earnings at the close of the twentieth century.” Annual Review of Sociology, 25: 623-657.
Nielsen, Francois, and Arthur S. Alderson (1997). “The Kuznets curve and the great U-turn: Income inequality in U.S. counties, 1970-1990.” American Sociological Review, 62: 12-33
OECD (1994). The OECD Jobs Study, Part II. Paris : Organisation for Economic Co-operation and Development
OECD (1997). OECD Employment Outlook 1997. Paris : Organisation for Economic Co-operation and Development
Phillips, Damon J. (2001). “The promotion paradox: The relationship between organizational mortality and employee promotion chances in Silicon Valley law firms, 1946-1996.” American Journal of Sociology, 106: 1058-1098
Phillips, Damon J., and Jesper B. Sørensen (2003). “Competitive position and promotion rates: Commercial television station top management, 1953-1988.” Social Forces, 81: 819-841
Rauch, James E. (1993). “Productivity gains from geographic concentration of human capital: Evidence from the cities.” Journal of Urban Economics, 43: 380-400
Rosen, Sherwin (1981). “The economics of superstars.” American Economic Review, 71: 845-858
39
Rosen, Sherwin (1982). “Authority, control and the distribution of earnings.” Bell Journal of Economics, 13: 311-323
Roy, Andrew D. (1951). “Some thoughts on the distribution of earnings.” Oxford Economic Papers, 3: 135-146
Sattinger, Michael (1975). “Comparative advantage and the distributions of earnings and abilities.” Econometrica, 43: 455-468
Sattinger, Michael (1983). “Assignment models of the distribution of earnings.” Journal of Economic Literature, 31: 831-880
Scott, W. Richard (2002). Organizations: Rational, Natural and Open Systems. New York: Prentice Hall
Shaked, Avner, and John Sutton (1982). “Relaxing price competition through product differentiation.” Review of Economic Studies, 49: 3-13.
Sørensen, Aage B. (1996). “The structural basis of social inequality.” American Journal of Sociology, 101: 1333-1365
Sørensen, Jesper B. (1999). “The ecology of organizational demography: Managerial tenure distributions and organizational competition.” Industrial and Corporate Change, 8: 713-744
Sørensen, Jesper B. (2004). “Recruitment-based competition between industries: A community ecology.” Industrial and Corporate Change, 13: 149-170
Sorenson, Olav, and Pino G. Audia (2000). “The social structure of entrepreneurial activity: Geographic concentration of footwear production in the United States, 1940-1989.” American Journal of Sociology, 106: 424-462
Wheeler, Christopher H. (2001). “Search, sorting and urban agglomeration.” Journal of Labor Economics, 19: 879-899.
40
Table 1: Descriptive Statistics 1980-1991 1992-1998 Variable Mean σ Mean σ
Wage inequality 3.628 0.705 3.975 0.628
Log mean industry wage 4.470 0.351 4.874 0.267
Residual wage inequality -1.074 0.413 -1.128 0.411
Industry employment in township 112.417 350.516 110.093 316.143
Total township employment (000) 7.612 20.096 8.070 19.677
Total N firms in township 747.744 1511.421 716.193 1392.012
Single industry employer 0.089 0.285 0.109 0.311
Standard dev. of employer sizes 13.857 54.975 14.295 58.355
Log N industry firms in township 1.718 1.015 1.627 1.023
N industries in township 51.866 19.422 56.142 20.411
Entropy of industry shares 3.232 0.519 3.307 0.522
Note: Wage inequality is the log standard deviation of wages in an industry-township cell. Residual wage inequality is the log standard deviation of the residuals from the first stage human capital equation. See text for details.
41
Table 2: Fixed effects estimates of the effects of corporate demography on wage inequality, 1980-1991
Variable (1) (2) (3) (4) (5) Log Mean industry wage 1,910.976† 1,913.112† 1,911.863† 1,911.791† 1,910.232† (15.401) (15.484) (15.430) (15.392) (15.282) Industry employment in township -0.041* -0.044* -0.042* -0.028 -0.040* (0.017) (0.018) (0.018) (0.017) (0.018) Total township employment (000) -3.207† -3.746† -3.783† -4.285† -4.875† (0.860) (0.810) (0.947) (0.733) (0.897) Total N firms in township 0.041† 0.050† 0.050† 0.067† 0.070† (0.012) (0.011) (0.013) (0.010) (0.012) Single industry employer -156.029† -153.331† -155.219† -124.276† -135.810† (11.823) (11.843) (11.908) (10.526) (11.170) Standard dev. of employer sizes -0.067 -0.055 -0.067 -0.108 -0.078 (0.059) (0.058) (0.059) (0.065) (0.063) Log N industry firms in township 150.102† 156.239† 152.367† 169.126† 160.128† (6.019) (6.244) (6.256) (4.802) (5.527) N industries in township -0.496* -0.990† (0.218) (0.211) Entropy of industry shares -8.926 -21.318† (6.624) (6.511) Log N firms * N industries -1.227† (0.151) Log N firms * Entropy -34.043† (6.001)
Note: The dependent variable is the log of the standard deviation of hourly wages for employees in a given industry in a township, multiplied by 1,000. Standard errors are adjusted for clustering at the township level. Models include dummy variables for year and industry. Interaction effects are centered. N for all models is 108,537 industry-township cells. Two-sided t-tests: * p<.05 † p<.01
42
Table 3: Fixed effects estimates of the effects of corporate demography on wage inequality, 1992-1998
Variable (1) (2) (3) (4) (5) Log Mean industry wage 1,724.802† 1,725.637† 1,725.509† 1,721.599† 1,722.756† (16.868) (16.928) (16.896) (16.913) (16.918) Industry employment in township -0.077† -0.078† -0.078† -0.056* -0.076† (0.028) (0.028) (0.029) (0.027) (0.029) Total township employment (000) -2.143 -2.355* -2.685* -2.431* -3.438* (1.191) (1.200) (1.357) (1.211) (1.421) Total N firms in township 0.029 0.032 0.037 0.043* 0.052* (0.017) (0.017) (0.019) (0.018) (0.021) Single industry employer -148.942† -147.866† -148.172† -125.764† -134.023† (13.034) (13.110) (13.138) (12.631) (13.278) Standard dev. of employer sizes 0.097 0.102 0.097 0.044 0.091 (0.055) (0.055) (0.055) (0.062) (0.058) Log N industry firms in township 136.905† 139.460† 139.164† 151.738† 146.147† (6.310) (6.675) (6.881) (5.628) (6.593) N industries in township -0.211 -0.721† (0.238) (0.226) Entropy of industry shares -8.650 -19.331* (7.858) (7.583) Log N firms * N industries -1.205† (0.149) Log N firms * Entropy -29.350† (7.374)
Note: The dependent variable is the log standard deviation of hourly wages for employees in a given industry in a township, multiplied by 1,000. Standard errors are adjusted for clustering at the township level. Models include dummy variables for year and industry. Interaction effects are centered. N for all models is 65,837 industry-township cells. Two-sided t-tests: * p<.05 † p<.01
43
Table 4: Fixed effects estimates of the effects of corporate demography on residual wage inequality, 1980-1991 Variable (1) (2) (3) (4) (5) Industry employment in township -0.044 -0.048 -0.045 -0.039 -0.044 (0.026) (0.027) (0.026) (0.027) (0.027) Total township employment (000) 0.997 0.401 0.606 0.115 -0.155 (1.422) (1.432) (1.568) (1.411) (1.511) Total N firms in township -0.009 0.000 -0.004 0.009 0.010 (0.020) (0.020) (0.022) (0.020) (0.021) Single industry employer -168.704† -165.632† -168.146† -150.574† -154.826† (10.644) (10.679) (10.665) (10.135) (10.720) Standard dev. of employer sizes -0.384† -0.370† -0.384† -0.397† -0.392† (0.114) (0.112) (0.113) (0.120) (0.118) Log N industry firms in township 38.877† 46.004† 40.465† 52.664† 45.745† (5.565) (5.657) (5.872) (4.650) (5.354) N industries in township -0.565* -0.821† (0.220) (0.211) Entropy of industry shares -6.122 -14.655* (7.027) (6.949) Log N firms * N industries -0.635† (0.124) Log N firms * Entropy -23.356† (4.642)
Note: The dependent variable is the log standard deviation of the residuals from the first-stage human capital equation regressions adjusted for sampling error, multiplied by 1,000. See text for details. Standard errors are adjusted for clustering at the township level. Models include dummy variables for year and industry. Interaction effects are centered. N for all models is 108,658 industry-township cells. Two-sided t-tests: * p<.05 † p<.01
44
Table 5: Fixed effects estimates of the effects of corporate demography on residual wage inequality, 1992-1998
Variable (1) (2) (3) (4) (5) Industry employment in township -0.051 -0.056 -0.052 -0.044 -0.050 (0.030) (0.031) (0.030) (0.031) (0.030) Total township employment (000) -0.549 -1.106 -1.162 -0.875 -1.503 (1.012) (0.980) (0.977) (1.139) (1.105) Total N firms in township 0.010 0.020 0.015 0.026 0.027 (0.015) (0.014) (0.017) (0.014) (0.016) Single industry employer -117.412† -114.534† -116.944† -102.409† -105.425† (11.682) (11.784) (11.719) (11.188) (11.516) Standard dev. of employer sizes -0.226* -0.213* -0.226* -0.244* -0.231* (0.094) (0.092) (0.094) (0.100) (0.099) Log N industry firms in township 42.425† 49.378† 43.817† 56.023† 49.402† (6.268) (6.655) (6.731) (5.739) (6.176) N industries in township -0.565† -0.847† (0.210) (0.200) Entropy of industry shares -5.241 -13.984* (7.030) (6.645) Log N firms * N industries -0.659† (0.124) Log N firms * Entropy -23.884† (5.153)
Note: The dependent variable is the log standard deviation of the residuals from the first-stage human capital equation regressions adjusted for sampling error, multiplied by 1,000. See text for details. Standard errors are adjusted for clustering at the township level. Models include dummy variables for year and industry. Interaction effects are centered. N for all models is 65,894 industry-township cells. Two-sided t-tests: * p<.05 † p<.01
45
Table 6: Effects of one standard deviation increases in independent variables on gross and within group wage dispersion
Gross dispersion Within group dispersion 1980-1991 1992-1998 1980-1991 1992-1998 (1) (2) (3) (4) (5) (6) (7) (8)
Industry employment in township -0.01 -0.01 -0.02 -0.02 -0.01 -0.02 -0.01 -0.02
Total township employment -0.08 -0.09 -0.05 -0.07 -0.00 -0.00 -0.02 -0.03
Vertical sorting
Total N firms in township 0.11 0.11 0.06 0.08 0.01 0.02 0.04 0.04
Single industry employer‡ -0.12 -0.13 -0.13 -0.14 -0.14 -0.17 -0.11 -0.11
Log N industry firms in township 0.19 0.18 0.17 0.16 0.06 0.05 0.06 0.05
Horizontal sorting
SD of employer sizes (/ 10) -0.06 -0.04 0.02 0.05 -0.22 -0.21 -0.14 -0.14
N industries in township -0.02 -0.02 -0.02 -0.02
Entropy of industry shares -0.01 -0.01 -0.01 -0.01
Note: Cell entries are the standardized coefficient estimates for the respective standard dependent variables, computed from the relevant models in Tables 2-4. The effect of the single industry employer dummy variable is reported as a change from 0 to 1.
46
Figure 1: Joint effects of the number of firms and industry diversity on gross wage inequality
5 15 25 35 45 55 65 75 85 95 1050
2
4
6
0.6
1.1
1.6
2.1
2.6
3.1
3.6
Mu
ltip
lier
of
gro
ss d
isp
ers
ion
Number of industries
Ln (firms)