Wage inequality in workers' cooperatives and conventional firms

Abstract

The author evaluates the effects of democratic worker participation on the income distribution within firms. Wage inequality in French workers' cooperatives (called SCOPs) versus traditional firms is measured using the 2001-2012 panel DADS dataset which includes all French firms. The author finds significantly lower inequality in SCOPs, in line with the previous empirical literature. Going into more detail, it appears that inequality is reduced at the top of the distribution and specifically regarding qualification-based inequalities; the gender gap and the advantage of senior workers are not lower in SCOPs. These findings contribute to the literature on Labor-Managed Firms, as well as to the broader debate on rising wage inequality in developed countries.

Full text

The European Journal of Comparative Economics
Vol. 14, n. 2, pp. 303-329
ISSN 1824-2979
http://dx.doi.org/10.25428/1824-2979/201702-303-329
Wage inequality in workers’ cooperatives and
conventional firms
1
Nathalie Magne
2
Abstract
The author evaluates the effects of democratic worker participation on the income distribution within
firms. Wage inequality in French workers’ cooperatives (called SCOPs) versus traditional firms is
measured using the 2001-2012 panel DADS dataset which includes all French firms. The author finds
significantly lower inequality in SCOPs, in line with the previous empirical literature. Going into more
detail, it appears that inequality is reduced at the top of the distribution and specifically regarding
qualification-based inequalities; the gender gap and the advantage of senior workers are not lower in
SCOPs. These findings contribute to the literature on Labor-Managed Firms, as well as to the broader
debate on rising wage inequality in developed countries.
JEL: J54, D21, J31, P13
Key words: Worker cooperatives, inequality, wage equation.
1. Introduction
Explaining the increase of wage inequality has been a challenge for economists
since the 1980s. At the level of the firm, there is no consensus to explain the deviations
from marginal productivity remuneration. At a macro level, large inequalities are
recognized as having a detrimental effect on growth and economic stability, specifically
since the 2008 crisis (Dabla-Norris et al. 2015). A micro approach sheds light on the
dynamics of wage inequality and cooperatives are a very good natural laboratory as
democracy's effect on income distribution can be observed. Furthermore, the
consequences of workers’ participation on wage distribution can be studied with relation
to effort incentive, worker selection and turn-over. There are many reasons to think that
worker participation in decision making in firms should lead to lower wage inequality.
Theoretically the median voter theory leads to the conclusion that, in cooperatives
where workers vote democratically, there will be a redistribution of wealth whenever the
median income is lower than the mean (Kremer 1997). Another point of view considers
ex ante selection: agents who choose to work in a cooperative are likely to have a strong
aversion to inequality. On the other hand, if cooperatives are operating side by side with
conventional firms in a competitive market, they might not be able to have a
significantly different wage structure in the long term. An empirical answer is thus
required to the question of whether or not the wage structure actually differs between
cooperatives and conventional firms, and more specifically whether inequality is lower
in cooperatives. If it is, a second question will be raised: how much of the observed gap
1
I am indebted to Virginie Pérotin and Gabriel Burdin as well as to conference participants at IAFEP and
EACES for their helpful comments. I am also grateful to two anonymous referees for very useful
comments and suggestions. This work is supported by a public grant overseen by the French National
Research Agency (ANR) as part of the “Investissements d’Avenir” program (reference: ANR-10-
EQPX-17 - Centre d’accès sécurisé aux données – CASD).
2
University of Montpellier, Nathalie.magne@univ-montp3.fra

EJCE, vol.14, n.2 (2017)
Available online at http://eaces.liuc.it
304
is due to the workers’ characteristics and how much is due to a different remuneration
of these characteristics?
In empirical literature there is a consensus that inequalities are much lower in
cooperatives. Pencavel (2001) proves this in the case of American plywood cooperatives
and highlights the fact it can be partially explained by the small number of supervisors
3
.
Based on an extensive dataset of Uruguayan firms, Burdin (2016) shows strong evidence
of redistribution in favor of low wage workers in cooperatives and of a higher exit rate
for high-ability members. The paper offers precise measures of these two phenomena.
The case of Mondragon cooperatives has also been investigated but shows a very
different pattern from French cooperatives since statutory regulations exist regarding
wage differentials (Dow 2003). The fact that there are no such rules in French
cooperatives
4
leads to a large diversity of individual firms. It is therefore entirely relevant
to measure the impact of workers’ democratic participation on wage determiners. For
Northern Italy, Bartlett et al. (1992) show a much lower wage differential in
cooperatives since the ratio of managerial to unskilled manual workers’ wages is 75%
lower, mainly because of lower managerial salaries. More recently, Abramitzky (2008)
explores the equality schemes and distribution patterns of Israeli Kibbutzim. He shows
that the level of equality has diminished since the 1980s but the kibbutzim which have
remained egalitarian are the richest. Finally, Clemente et al. (2012) measure lower
inequality in Spanish cooperatives, with industry variations. With the exception of
Burdin (2016), all these papers have databases concentrated on one industry or one
region: Pencavel (2001) focuses on Northwest American plywood cooperatives and
Bartlett et al. (1992) use a matched sample of 85 firms in Tuscany and Emilia-Romagna,
all in light manufacturing with around 100 workers. While offering a unique field for
systematic comparison since the chosen cooperatives and conventional firms develop in
the same environment, these databases do not allow for variations in industry and
region. The DADS
5
dataset allows us to take into account all French cooperatives which
are distributed among all industries and French regions. It is also unprecedented in its
size since it includes 23 million jobs, 45000 of which are in cooperatives. Panel data is
available for years 2001 to 2012 for 1/12th of all French jobs. French SCOPs also
present the double advantage of a long history and a recent dynamism: they have been
numerous and active in a wide range of industries since the end of the 19th century and
they have created an estimated 15,000 net jobs between 2000 and 2015. France is the
third European country in terms of workers cooperatives after Spain and Italy.
Some of the mentioned articles analyze the causes of reduced inequality (size of
the firm, statutory rules, median voter theory or political convictions) while others focus
on the consequences (brain drain, lower productivity). They do not go into detail about
the distribution of wages according to categories of workers beyond the simple
distinction between high-ability and low-ability workers. Furthermore, there is no
measurement of the yield of workers’ characteristics, with the notable exception of
3
Greenberg (1986) counts 1 or 2 managers in cooperatives where 6 or 7 are present in an equivalent
conventional firm.
4
There is a label entitling firms to tax advantages if the mean of the 5 best paid workers does not exceed
5 times the minimum wage and if the firm is not listed on the stock market. However, this label is
completely independent from the cooperative status.
5
“Déclaration annuelle des données sociales” collected by the fiscal administration and made available for
researchers by the INSEE (Institut national de la statistique et des études économiques)

N. Magne, Wage inequality in workers’ cooperatives and conventional firms
Available online at http://eaces.liuc.it
305
Clemente et al. (2012). Using the Oaxaca-Blinder method, the latter pinpoints workers’
characteristics as the reason for their lower wages in cooperatives and finds similar
returns for these characteristics in both types of firms. At the end of this brief review of
the existing empirical literature, we can point out that wage inequality has not been well
documented for French SCOPs
6
despite the rich and influential history of French
workers’ cooperatives, and in all other cases, the wage distribution was not analyzed in
detail. This paper will attempt to answer the following: how different is the distribution
of wages in French SCOPs compared to conventional firms (CFs) and which categories
of workers benefit from it?
Section 2 reviews the theoretical predictions and assesses their relevance in the
context of French worker cooperatives. Section 3 presents the empirical strategy to
compare levels of inequality and returns to workers’ characteristics in both types of
organizations. Section 4 describes the extensive dataset we use in our analysis. Section 5
displays the results and relates them to the hypotheses made in section 2. Section 6
presents concluding remarks.
2. Theoretical issues and institutional background
2.1. Wage inequality theory
Although SCOPs still account for a negligible proportion of French firms, they
currently have the wind in their sails as they have been found more resistant to
economic shocks
7
and are regarded as a popular alternative model of firm.
8
At the end
of 2016, there were 2298 SCOPs in France employing 48750 workers. SCOPs are
worker cooperatives characterized by a few important statutory rules:
1. Capitalization. Workers own at least 51% of the capital. Up to 49% can
be owned by outside shareholders but it cannot be listed on the stock
market, nor can the shares fluctuate from their nominal price.
2. One person = one vote. Members vote democratically in the general
assembly
9
. Not all workers are members of the cooperative and the
proportion varies strongly between firms. Some SCOPs have clauses in
their status making it compulsory for workers to become members within
a few years, some strongly encourage it in an informal manner and others
do not exert any pressure. On average, according to the CGSCOP
10
, 69%
of workers with at least 2 years seniority are members.
6
Defourny, Estrin and Jones (1985) and Fakhfakh, Pérotin and Gago (2012) focus on productivity and
capital issues.
7
See Roelants, Eum, Terrasi (2014), p. 36.
8
The 2014 social economy law includes a clause that gives priority to workers to form a cooperative when
it comes to firm transmission from a former owner.
9
Each SCOP has specific rules and traditions but the annual vote at the general assembly includes at least
important strategic decisions, as well as the election of the manager.
10
The General Confederation of SCOPs is an organization that gathers almost all French SCOPs,
providing financial and managerial services to its members and has a function of communication and
lobbying. The Confederation publishes a few key statistics each year: http://www.les-
scop.coop/sites/fr/les-chiffres-cles/

EJCE, vol.14, n.2 (2017)
Available online at http://eaces.liuc.it
306
3. Profit distribution. The profits are shared in three parts: participation (at
least 25%) is shared between all workers (including non-members) either
on an equal basis or in proportion to wages or hours worked. Dividends
(maximum 33%) are distributed to members in proportion to the owned
shares. Reserves (minimum 16%) are reinvested in the company and
constitute collectively owned capital that may never be paid to the
members, even in the case of shutdown.
These characteristics bring SCOPs close to what is referred to in the literature as
labor-managed firms (LMF) or employee-owned firms since Ward (1958), although
some institutional rules are not always well accounted for (particularly profit-sharing and
integration of new members) as highlighted by Kamshad (1997). Theoretically, there are
many good reasons to think wage inequality would be lower in such firms. The literature
on the subject can be divided into two parts: the theory of labor-managed firms is
founded on the median voter model whereas the non-profit literature relies mostly on
the intrinsic motivation hypothesis.
According to Kremer’s median voter theory (1997), wages are compressed in
LMF adopting the principle of one vote per worker, because whenever the median wage
is below the mean a majority of workers votes for redistribution. This has two possible
consequences. Firstly, minorities or non-members can be oppressed or arbitrarily
expropriated. Secondly, workers whose abilities are above the mean are incited to flee to
jobs where they have reasons to think they would be paid higher wages
11
. This raises the
question of the durability of such a significant difference in the wage structure on a
competitive market. Kremer (1997) argues that mobility barriers exist in the form of a
non-refundable investment all workers made in the cooperative before they had access
to any information on productivity. Empirically, Burdin (2016) shows a higher turn-over
for managers in Uruguayan labor-managed firms than in comparable conventional
firms, without it calling into question the significantly different wage structure. In the
case of SCOPs, the principle of one vote per member-worker applies, if not directly to
decide on wages, at least to elect a manager who then decides on wages
12
. There is an
initial investment which can be seen as partially non-refundable since the legal status of
SCOPs proscribes capital gains workers could benefit from if they invested their capital
in CFs’ shares. We can therefore expect a certain level of redistribution within firms
compared to a situation with no vote, redistribution which will probably not favor
minorities.
Another argument in favor of lower inequality in cooperatives comes from
Hansmann (1996). He argues that LMF will be most efficient when workers’
preferences are homogeneous and therefore the cost of collective decisions is not too
11
In a perfectly competitive labor market, if all workers outside cooperatives are paid their marginal
product, any worker with a marginal productivity above the mean of the cooperative is incited to leave.
In a more realistic labor market, we can still assume that high productivity workers would be pai d more
in an organization where no decision is submitted to vote.
12
Considering the decision processes in which wages are determined in SCOPs, incidental evidence from
interviews with SCOP managers show that they are specific to each firm. In most cases, there is a
general desire to minimize the wage differential, but some firms have a philosophy of perfect equality
whereas in others wages are at the manager’s discretion. In the latter case, since the manager is elected
by the cooperative members during the annual general assembly, members still have an indirect impact
on wage inequality. In some firms, strict rules regarding pay rises and wage range have been voted for in
the general assembly and included in the statutes.

N. Magne, Wage inequality in workers’ cooperatives and conventional firms
Available online at http://eaces.liuc.it
307
high. In that case, we should observe lower wage inequality in French cooperatives, not
because of a redistribution process, rather because of a similarity in the workers’
characteristics.
The literature on non-profit organizations insists on another hypothesis that could
lead to lower inequality in cooperatives: the intrinsic motivation of highly qualified
workers. Workers are said to be intrinsically motivated if they gain utility from their
work beside monetary compensation. Following Preston (1989), workers employed in
the non-profit sector are ready to “donate” labor, which is to say work for lower wages
in exchange for the morality of their job or other non-monetary compensation. Lower
average wages (not accompanied by lower satisfaction) should therefore be found in the
non-profit sector. The opposite argument is presented by François (2003) who
distinguishes two effort extraction technologies: one preferred by CFs that relies on
intensive and costly supervision and leads to lower wages and another preferred by non-
profit organizations that relies on intrinsic motivation and leads to higher wages. Just
like the theory, empirical evidence is mixed: in the US, Leete (2001) concludes that the
wage differential is in favor of the non-profit workers only for certain industries and
Mocan and Tekin (2000) find a non-profit wage premium in the child care industry
whereas Ruhm and Borkoski (2003) find no significant differences. In Europe, Mosca et
al. (2007) for Italy and Narcy (2011) for France find lower non-profit wages. Yet the
wage gap need not be the same for low and high wages. Intrinsic motivation is not
equally distributed among workers and more specifically it is likely to weigh more for
high-ability workers. We can suspect the existence of a wealth effect, meaning workers
are more likely to consider monetary compensation as secondary once they have secured
a certain level of wage. Among others, Narcy (2011) highlights the intrinsic motivation
of executives in the French non-profit sector. Regarding SCOPs, this leads us to the
hypothesis of lower average wage and most importantly lower levels of inequality due to
lower wages at the top of the distribution. This could explain long term differences as
intrinsically motivated workers have no reason to quit even though they anticipate
higher wages in other firms. Incidental evidence from the field seems to favor the
intrinsic motivation hypothesis
13
. In addition to the legal rules of SCOPs, some
historical context can also be helpful to formulate hypotheses about wage distribution.
2.2. Some context about French workers cooperatives
Historically, the current cooperative movement in France took off on the
initiative of workers in manufacturing and construction industries at the end of the 19th
century. Throughout the 20th century more diversification occurred and the general
confederation of SCOPs insists that cooperatives are now widely spread across all
industries. The 2010 change of name from “production workers’ cooperative company”
to “participative and cooperative company” was meant to reflect the diversity of
SCOPs. According to the SCOP confederation, an average of 220 SCOPs a year were
13
Field evidence comes from a survey that consisted in 40 interviews with SCOP managers in the Rhône-
Alpes region. The wage system and its fairness is at the center of much discussion and managers tend to
insist on the wage sacrifice they accepted in exchange for a more fulfilling job and the benefits of
participative management or out of aversion to inequality. More details can be found in Charmettant et
al. (2016).

EJCE, vol.14, n.2 (2017)
Available online at http://eaces.liuc.it
308
created between 2007 and 2012
14
. Many of these newly created firms are in the service
industry.
However, SCOPs are still over-represented in the manufacturing and construction
industries as shown in table 1.
14
http://www.les-scop.coop/sites/fr/les-chiffres-cles/

N. Magne, Wage inequality in workers’ cooperatives and conventional firms
Available online at http://eaces.liuc.it
309
Table1: Industry distribution
Proportion of firms.
2009-2012
Proportion of jobs.
2001-2012
Industries
CF
SCOP
CF
SCOP
Manufacturing industry
9.1%
17.0%
19.0%
26.9%
Construction industry
11.3%
23.3%
6.8%
31.9%
Trade, transport, accommodation
and catering
39.7%
13.9%
27.4%
10.1%
Other services
33.7%
35.8%
34.6%
24.8%
Education and health
6.2%
10.1%
11.8%
6.3%
Total
100%
100%
100%
100%
Regarding SCOP workers’ characteristics, they are also different from CF
workers, as shown in table 2. There are notably fewer women working in SCOPs, more
blue collars and fewer newly hired workers, which is obviously linked to the industry
distribution described in table 1. If minorities are disadvantaged as the median voter
theory predicts, we should observe lower wages for women and newly hired workers in
SCOPs
15
. Regarding women, there may be a self-sustaining cycle at work: it women
anticipate lower wages in SCOPs, their scarce participation will be exacerbated. This will
have to be kept in mind as a potential endogeneity source when interpreting the results.
Table 2: Workers’ characteristics
Workers’ characteristics
CF
SCOP
Women
45%
28%
Age mean
36.7 years
39.5 years
Permanent contract
71.2%
75.9%
Full-time jobs
71.0%
76.9%
Incidental jobs16
14.9%
10.8%
Seniority mean
3.6 years
5.2 years
Occupations
Executive
15%
12%
Intermediate occupation
17%
17%
Semi-skilled white collar
11%
9%
Unskilled white collar
21%
6%
Semi-skilled blue collar
20%
36%
Unskilled blue collar
12%
17%
Intern
4%
2%
Total
100%
100%
15
Regarding newly hired workers, the fact that they are less likely to be voting members must also be
taken into account.
16
Incidental jobs or “postes annexes” are defined by the INSEE as jobs including fewer than 120 hours
or fewer than 30 days or a ratio number of hour/duration inferior to 1.5 and the wage is less than three
times the minimum wage. The objective is to eliminate summer jobs or very temporary jobs that are
numerous but not representative. Those jobs constitute 14% of all jobs for a given year.

EJCE, vol.14, n.2 (2017)
Available online at http://eaces.liuc.it
310
The institutionalization of SCOPs took place in France at the end of the 19th
century, with the objective of promoting workers’ independence from employers and
capitalist exploitation, specifically in the construction and manufacturing industries
(Demoustier 1984, Toucas-Truyen 2005). The fight for equality targeted primarily class
equality. In the workplace, this materialized in an effort to lower inequality between
capitalists and workers and between white collars and blue collars. The strong cohesion
of the cooperative movement gives us reason to believe the original preoccupations are
still relevant today. More specifically, we can make the hypothesis that SCOPs are
primarily trying to lower inequalities due to qualification and occupations. This could
also apply to newly created SCOPs, due to the strength and stability of the cooperative
network and to the key role of the CGSCOP in most SCOP creations as a financial
support, a consultant and later on a training center for members and managers.
It is worth referring to the empirical literature on labor-unions and their impact
on wage dispersion. Using different methods, Freeman (1980), Lemieux (1998) and
Card et al. (2003) find a reduction of the differential between blue collar and white collar
workers and lower returns to skill and experience in the union sector. The literature
focusing on the impact of unions on the gender wage-gap is less consensual. Main and
Reilly (1992), Card (2003) and Koevets (2007) find that unions tend to raise women’s
wages more than men’s without entirely filling the gap. In France, Leclair and Petit
(2004) and Duguet and Petit (2009) find that unions actually increase the gender-gap.
The common history of empowerment of workers could lead to the hypothesis of
similar wage policies in both SCOPs and labor-unions. This quick overview of the union
literature allows us to reinforce our hypothesis that wage distribution in SCOPs might
be more compressed, mostly reducing the gap between qualified and unqualified
workers. However, distinct objectives and ideological disagreements between the two
forms of organization are not to be ignored, as evidenced by the history of conflicts
between SCOPs and French unions.
Finally, since profit sharing is an important part of SCOPs workers’ income,
wages in SCOPs are likely to be higher than in CFs in more profitable firms
17
. As a
result, workers in large SCOPs should be more advantaged than workers in large CFs
since large firms are shown to be more profitable on average (Josefy, Kuban, Ireland
and Hitt 2015). The expected effect of industry is more ambiguous because more
profitable industries could be different under SCOP and CF status, due to different
capital endowment (Fakhfakh, Perotin and Gago 2012).
From this theoretical and institutional analysis, we draw six hypotheses worth
testing empirically when it comes to wage dispersion in SCOPs and CFs:
H1: Wages are less dispersed in SCOPs than in CFs. Within workplace inequality is lower.
H2: Wage inequality is reduced mainly at the top of the distribution.
H3: Wage inequality is reduced mainly between qualified and unqualified positions. Return to skill is
lower in SCOPs.
17
A quick calculation from CGSCOP data allowed us to estimate average profit-sharing at 5% of total
payroll in SCOPs for 2012, whereas the results of the PIPA survey for all French CFs (available at this
address: http://dares.travail-emploi.gouv.fr/dares-etudes-et-statistiques/statistiques-de-a-a-z/article/la-
participation-l-interessement-et-l-epargne-salariale) estimates profit-sharing at 1% of total payroll.

N. Magne, Wage inequality in workers’ cooperatives and conventional firms
Available online at http://eaces.liuc.it
311
H4: The gender gap is not lower in SCOPs than in CFs.
H5: Return to seniority is higher in SCOPs than in CFs.
H6: Workplace size gives a larger wage advantage to SCOP workers compared to CF workers.
In order to test these hypotheses, we now have to establish an empirical strategy,
which draws from the union literature, as well as from other attempts to measure wage
distribution among two distinct categories of workers, namely literature on the gender
gap and on the public-private gap. Before going into more detail, we describe our data
set.
3. Database and descriptive statistics
The DADS are collected from all French firms every year by the national institute
of statistics (INSEE) and made available for researchers via a secure procedure. Its
completeness is guaranteed by the fact that it is a compulsory declaration that every
employer must make every year. We have two panel datasets that allow us to compare
SCOPs and CFs wage distributions. The first includes wage per worker for 1/12th of
national jobs for years 2001 to 2012 and the second consists of workplace observations,
for which we are able to measure average hourly wage and very detailed distribution
variables within the workplace for years 2009 to 2012. We are interested in gross wage
distribution, which includes all compensation paid to employees, including bonuses,
profit sharing, taxable fringe benefits and all employee social security contributions
18
.
We obtain hourly wage by dividing annual earnings by annual paid hours
19
. Wages are
trimmed to eliminate 0.5% of the lowest and highest values. In addition, we have a
number of wage determining variables. For individual level observations, we have:
individual characteristics such as age, sex, occupation, place of birth; job characteristics
such as full-time employment, nature of the contract, seniority; and firm characteristics
such as firm size, industry, localization, collective agreement. In the workplace dataset
we have very detailed distribution variables for wages and hours worked by each
occupation, gender and type of contract, as well as decile for gross wages.
Some observations are eliminated: industries with no SCOPs such as the
agricultural industry or the finance and insurance industry
20
, individual employers, the
public sector, incidental jobs (see table 2). Once this selection has been made, we have a
workplace dataset with 4.2 million observations (including 9500 SCOPs) in 4 years and a
job dataset with 10 million observations (including 34000 jobs in SCOPs) in 12 years.
Table 3 shows average wage by subgroup of workers for SCOPs and CFs: the
average wage for all workers is 15.6 euro per hour but qualified workers (CEO and
executives) have lower wages in SCOPs whereas others have slightly higher wages. This
is coherent with hypothesis H1, H2 and H3.
18
Net wage is also available in the DADS but because of the method used to measure it, it only includes
taxable profit-sharing. As a result, it seems less relevant for a comparison between SCOPs and CFs
since we expect profit-sharing to be significantly different between the two types of firms.
19
Positions with zero hours declared are excluded; this may concern, for example, home-workers.
20
The complete list of eliminated industries can be found in the annex.

EJCE, vol.14, n.2 (2017)
Available online at http://eaces.liuc.it
312
Table 3, Average gross hourly wage by socio-professional group and gender in SCOPs and CFs, for years
2001 to 2012. Unit: euro per hour (constant euro at its 2012 level)
CF
SCOP
Total
15.6
15.6
CEO
43.7
41.2
Executive
30.1
27.2
Intermediate Occupation
17.5
18.1
Semi-skilled white collar
13.3
13.7
Unskilled white collar
11.5
11.6
Semi-skilled blue collar
13.6
13.9
Unskilled blue collar
11.5
11.1
Intern
7.2
6.4
Men
16.8
16.2
Women
14.0
13.9
Figure 1 represents the evolution of individual hourly wage throughout the
2001-2012 period for SCOPs and CFs. SCOP wages are higher in 2001 and 2002 but
lower from 2004 on.
Figure 1
The univariate Kernel density estimation for individual hourly wages in SCOPs
and CFs for the year 2003
21
gives a first idea of how different the distribution can be
(figure 2). A higher proportion of workers are paid just above minimum wage in CF
than in SCOP
22
. The flatter density curve observed for SCOPs indicates that more
21
2003 was chosen because average wage is approximately the same in SCOPs and CFs for that year.
22
The distribution starts below the hourly minimum statutory wage which can be accounted for by a
more realistic declaration of hours worked. Moreover, minimum wage does not apply for interns.

N. Magne, Wage inequality in workers’ cooperatives and conventional firms
Available online at http://eaces.liuc.it
313
workers are paid between 12 and 20 euros an hour than in CFs whereas fewer workers
have high wages (above 20 euros an hour). The distributions are significantly different
for SCOPs and CFs as proved by a Kolmogorov-Smirnov test.
Figure 2. Density function of individual hourly gross wage in CF and SCOP for year
2003. The given unit is euro per hour. Statutory minimum wage is 7.19
4. Empirical strategy
Before tackling our six hypotheses H1 to H6 on the SCOP impact on the wage
distribution, we estimate the SCOP impact on the average wage level, through the
following equation:
        (1)
with  a dummy indicating if worker i works in a SCOP during year t, 
individual fixed effects and X control variables.
In order to test the six hypotheses outlined in section 2, we proceed in two parts,
using firstly the workplace dataset and secondly the job dataset. The method used for
the workplace dataset is very straightforward and provides a precise measure of the
effect of the SCOP status on wage distribution within the workplace. Quantile analyses
are made within workplaces. We run a regression per decile at the level of the
workplace, controlling for the size of the workplace, the industry, the proportion of
permanent labor contracts, the proportion of each occupation (measured in number of
workers and in hours worked), the proportion of women and the localization. The
dependent variable is the logarithm of the ith decile of full-time equivalent wage, or in
other words the full-time equivalent wage of the worker at the ith decile of the workplace
wage distribution. We also carry out the regressions with interdecile ratios as dependent
variables (successively D9/D1, D9/D5 and D5/D1). We include a dummy variable

EJCE, vol.14, n.2 (2017)
Available online at http://eaces.liuc.it
314
equal to 1 if the firm is a SCOP. We also include year dummy variables in order to
capture the variation in wages due to changes in the economic environment or any
institutional changes at a national level. This allows us to test hypotheses H1 and H2.
The equation is estimated with panel data from 2009 to 2012, with random effects as
too few firms switch from classical to SCOP status to be representative
23
. Moreover,
there is no reason to suspect unobserved heterogeneity since we control for industry,
size and skill composition of the workforce at a very fine level (see annex). In other
words, we can safely assume that there is no omitted time invariant characteristic. Any
effect of the SCOP variable on the level of wages or wage deciles is indeed what we
want to measure. We estimate the following equation:
      
  (2)
Where  includes all control variables detailed in the annex as well as year
dummies, SCOPj is the dummy equal to 1 if the firm is a SCOP and  the error term.
Next, we must go into more detail regarding the impact of the SCOP status on
wage distribution within the firm in order to test H3 and H4. To achieve this, the
dependent variable is replaced by hourly wage for men and women separately, as well as
for each aggregated occupation, and then by ratios between these. All regressions are
run with clustered standard errors to allow for intragroup correlation.
The job-level observation dataset allows us to analyze wage distribution more
precisely, taking into account individual workers’ characteristics and the impact of these
on wages. The Oaxaca-Blinder method is used to test hypotheses H3, H4, H5 and H6.
This method was first used by Oaxaca (1973) and Blinder (1973) for decomposing the
gender wage gap. It has been used more recently to compare the private and public
sectors (Melly 2005), different sexual orientation workers (Antecol et al. 2008) as well as
cooperative workers and capitalist firm workers (Clemente et al. 2012). The wage gap
between SCOPs and CFs can be broken down into two terms: the explained part (or
characteristic component) which is accounted for by the different characteristics of both
groups of workers and the unexplained part (or return component) which comes from
different remunerations of the workers’ characteristics and is identified as discrimination
by Oaxaca (1973) and Blinder (1973) in the case of gender gap
24
. We are mostly
interested in the detailed return component as we are trying to answer the following
question: which characteristics are remunerated differently in SCOPs? We estimate
equation (3):

  
  

  
  
  

 (3)
where 
 is the mean wage in log for CFs, 
 the mean wage in log for SCOPs,

 the mean wage determinant for CFs and 
 the mean wage determinant for
SCOPs. Details about the included variables can be found in the annex. We estimate
23
From 2009 to 2012, 202 firms have changed status from conventional firm to SCOP or vice versa.
24
The interpretation in terms of discrimination relies on the hypothesis that there are no unobserved
variables that could have a different impact on men’s and women’s wages.

N. Magne, Wage inequality in workers’ cooperatives and conventional firms
Available online at http://eaces.liuc.it
315

  
 and 
. 
 measures the part of the difference between average wage
in SCOP and CF that is due to different characteristics of the workers. 
  

measures the return component, or the “discrimination” assuming there is no
unobserved variable. We will be able to determine which part of wage gap between
SCOPs and CFs is due to different workers’ characteristics and which is due to different
returns to characteristics. If we find 
  
 significant and negative for the
qualification variables, this will mean that qualified workers are paid less in SCOPs than
they are in CFs, therefore proving hypothesis H3 to be true. Clustered standard errors
are used in the estimation to allow for intragroup correlation.
However, this estimation technique does not take individual fixed effects into
account. Although it was safe to assume that there were no invariant time characteristics
at the workplace level, we have every reason to believe there are individual fixed effects.
More specifically, SCOP workers’ unobserved characteristics are likely to have an
impact on their wage on average. We want to control for these characteristics in order
to isolate the SCOP effect on wage distribution. There is a technique to take into
account fixed effects in the panel data for Oaxaca-Blinder: the regression must be run in
two steps. The wage gap can be written as follows.

















  
















  

  
  
  

 (4)
       (5)
Where 
















 is the predicted mean wage in log for CFs,

















 the predicted mean wage in log for SCOPs (obtained from equation
(5) including fixed effects). All explanatory variables are centered. Standard errors are
calculated using the bootstrap option to correct for the two-step estimation procedure.
However, as noted by Heitmüller (2005), the omission of time-invariant variables in the
fixed-effect model leads to a bias in the decomposition results. This estimation
technique will allow us to distinguish between the wage gap due to characteristics on
one hand and return to characteristics on the other hand but it does not allow us to test
hypothesis H3 to H6 while controlling for individual fixed effects.
As we are primarily interested in the unexplained part of the decomposition, there
is a more straightforward method which is not affected by the same bias while taking
into account all individual heterogeneity: the use of interaction variables in a fixed-effect
regression. We run Mincerian-type wage regressions with interaction variables between
each of the independent variables and the SCOP dummy. Unlike in the workplace
regression, this variable is not time-invariant as workers can and frequently do change
from CFs to SCOPs and vice versa. This does not give us any information about the
“characteristic component” of the Oaxaca-Blinder decomposition but allows us to
compare the return component for CFs and SCOPs and therefore test hypotheses H3,
H4, H5 and H6. We estimate the following equation:

EJCE, vol.14, n.2 (2017)
Available online at http://eaces.liuc.it
316
 
        
          
        
       (6)
The equation is estimated with fixed effects and cluster standard errors. A
significantly positive β10 would mean starting to work in a SCOP is more beneficial (or
less detrimental) for men than it is for women, proving H4 to be true. The variable
 indicates the qualification and the position occupied in the firm as explained in
box 1: therefore coefficient  allows us to test hypothesis H3 (lower return to skill in
SCOPs). Finally, year dummies are included in equation (6) and interacted with the
SCOP dummy in order to allow for different effects of economic or institutional shocks
on SCOP and CF wages
25
.
Box 1. Classification of professions and socio-professional categories
It classifies the population by a combination of profession, hierarchical position and
status (salaried employee or otherwise). It has been very commonly used in France
since the 1950s for national statistics and has been updated by the INSEE in 1982
and 2003. It comprises three embedded levels of aggregation: the socio-professional
group (8 items), the socio-professional category (29 items) and the professions (486
items).
We use the second level in the workplace regressions (for precise controls on the
decomposition of the workforce) and the first level in the job level regressions. The
first level is the one used in table 3 for descriptive statistics and includes 8 categories:
CEO, executive, intermediate occupation, semi-skilled white collar, unskilled white
collar, semi-skilled blue collar, unskilled blue collar and intern. More precise
definitions of each category can be found on the INSEE website:
https://www.insee.fr/fr/information/2406153 and in Burnod and Chenu (2001) for
the distinctions between semi-skilled and unskilled white and blue collars.
5. Results and robustness checks
Before focusing on the dispersion of wages in SCOPs versus CFs, we measured
the impact of the SCOP status on the average hourly wage by estimating equation (1).
As shown in table 3, on average and without controlling for any variable, SCOP workers
have a slight wage advantage (column 1: hourly gross wage is 1% higher in SCOPs than
in CFs). However, as soon as we exclude incidental jobs and add year dummies, the
effect stops being significant (column 3 and 4), which is coherent with the statistics
reported in table 2. This could be due to the fact that SCOPs rely less on incidental jobs
or that they tend to pay incidental jobs better. Moreover, when workplace characteristics
are controlled for, the impact becomes negative (column (5): SCOP workers earn on
average 1.3% less). The inclusion of individual controls and individual fixed effects
25
In particular, the empirical literature (Burdin and Dean (2009), Pencavel et al. (2006)) shows more
flexible wages in worker cooperatives than in conventional firms.

N. Magne, Wage inequality in workers’ cooperatives and conventional firms
Available online at http://eaces.liuc.it
317
increases this negative impact (column (6) and (7)): everything else being equal, SCOP
workers earn on average 3.5% less than CF workers. On average, wages are slightly
higher in SCOPs but this is due to the firms’ characteristics (namely, SCOPs are on
average larger and more numerous in industries with higher wages) and to the workers’
characteristics (as reported in table 2, there are more men, more tenured workers and
fewer unskilled workers in SCOPs). When controls are introduced, the sign of the
coefficient becomes negative which implies that returns to characteristics are different in
CFs and SCOPs, therefore making hypotheses H3 to H6 fully relevant.
Table 4: Impact of being in a SCOP on wages. Results for estimation of coefficient β1 from equation (1).
Number of observations: 14,815,230 (34,416 are SCOP workers)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Estimated
coefficient for the
SCOP dummy
.010***
(.002)
.009***
(.002)
-.0009
(.003)
-.021***
(.003)
-.013***
(.002)
-.030***
(.002)
-.035***
(.004)
Year dummies
No
Yes
No
Yes
No
No
Yes
Workplace
controls
No
No
No
No
Yes
No
Yes
Individual
characteristics
controls
No
No
No
No
No
Yes
Yes
Individual fixed
effects
No
No
No
No
No
No
Yes
Excluding
incidental jobs
No
No
Yes
Yes
No
No
Yes
R²
0
.002
0
.002
.13
.58
.51
We now focus on wage dispersion, through hourly gross wage decile regressions.
As shown in table 5, the worker at the 10th percentile of the wage distribution in a firm
earns 2% more in a SCOP than in a CF. No significant differences are observable for
the 2nd and 3rd deciles, and all deciles above are lower in SCOPs. The negative effect
increases up to the worker at the 90th percentile of the wage distribution in a firm, who
is seen to earn 12% less in a SCOP than in a CF
26
. Inequalities appear to be lower in
SCOPs because low paid workers are slightly better off and because high paid workers
are worse off. Interdecile regressions confirm that inequality is mostly reduced at the
top of the distribution: the ratio D9/D1 is 14% lower in SCOPs than in CFs, D9/D5 is
9% lower and D5/D1 only 5% lower
27
. Wages are more concentrated in SCOPs and
inequality is reduced mostly at the top of the distribution. Hypothesis H1 and H2 are
confirmed. An additional question is the evolution of these levels of inequality during
26
As a robustness test, regressions are also run with a balanced panel, keeping only firms that were in the
panel for 4 years. The results are qualitatively the same except for the first decile which is no longer
higher in SCOPs. As a whole, all negative impacts of the SCOP variable are stronger and positive
impacts weaker with the balanced panel dataset. This could be due to differences between newly created
SCOPs and newly created CFs (for example, newly created SCOPs could have higher wages than newly
created CFs if they face higher selection from investors and bankers).
27
The interdecile coefficients are the same with the balanced panel, ruling out the possibility that only
newly created SCOPs would have lower inequality and the wage distribution would quickly converge
towards the CFs norm.

EJCE, vol.14, n.2 (2017)
Available online at http://eaces.liuc.it
318
the 2009-2012 period. To provide an answer, we add interacted year dummies with the
SCOP dummy in equation (2). The results show that the gap actually rises between 2009
and 2012: more specifically, the interdecile ratios are smaller in SCOPs during the whole
period but even more so in 2012 than in 2009. This could be due to the effect of the
crisis which increased inequalities in CFs but not in SCOPs. We now want to know
more about the characteristics that yield different wages in SCOPs and CFs.
Table 5 – Quantile and ratio regressions, workplace observations, panel 2009-2012. Estimation results of
equation (2): the same equation is estimated with the dependent variable successively equal to D1 to D9,
interdecile ratios and aggregated wages for different categories of workers at the workplace level. The
reported coefficient measures the SCOP impact on these variables (γ in equation (2)).
Dependent variable
Unbalanced
panel
Number of
observations and
R²
Decile 1
.021***
(.007)
742,232
.30
Decile 2
.003
(.007)
742,232
.38
Decile 3
-.010
(.007)
742,232
.40
Decile 4
-.022***
(.008)
742,232
.40
Decile 5
-.036***
(.008)
742,232
.41
Decile 6
-.051***
(.008)
742,232
.41
Decile 7
-.070***
(.008)
742,232
.41
Decile 8
-.089***
(.008)
742,232
.40
Decile 9
-.120***
(.009)
742,232
.38
Intedecile D9/D1
-.138***
(.008)
742,232
.20
Intedecile D9/D5
-.087***
(.006)
742,232
.14
Intedecile D5/D1
-.048***
(.006)
742,232
.17
Executive average wage
-.035***
(.008)
1,401,584
.16
Intermediate occupation average wage
.025***
(.006)
2,772,995
.20
White collar average wage
.034***
(.007)
3,061,109
.18
Blue collar average wage
.031***
(.005)
2,260,291
.17
Ratio Intermediate occupation/ executive
-.046***
(.011)
1,007,456
.06

N. Magne, Wage inequality in workers’ cooperatives and conventional firms
Available online at http://eaces.liuc.it
319
Dependent variable
Unbalanced
panel
Number of
observations and
R²
Ratio Intermediate occupation/ blue collar
.007
(.007)
1,404,090
.05
Ratio Intermediate occupation / white collar
-.004*
(.002)
2,679,014
.04
Ratio executive / blue collar
-.028***
(.010)
794,286
.11
Women average wage
.011**
(.005)
3,382,315
.38
Men average wage
.003
(.005)
3,353,863
.46
Ratio men / women
-.023***
(.006)
2,478,640
.18
We now focus on wage gaps between positions and between genders (H3 and
H4). The workplace regressions (table 5) display interesting results: average hourly wage
within the workplace is higher in SCOPs for women, blue collars, white collars and
intermediate occupations, whereas it is lower for executives and not significantly
different for men. The ratio of executive hourly wage on manual worker hourly wage
and the gender ratio are found to be lower in SCOPs. However, this could be due to
different individual characteristics of executives and manual workers, as well as male and
female workers in SCOPs since we only control for workplace variables. For example,
women could have more qualified jobs in SCOPs. The same is true regarding the
reduced gap between qualified and unqualified positions: it could be due to different
unobserved skills. The job dataset allows us to overcome these limitations since taking
into account individual fixed effects ensures that the observed return differences are in
fact due to the SCOP status: this is the purpose of the estimations reported in table 6
and 7.
The Oaxaca-Blinder estimation results show that, overall, there is no significant
difference between average hourly wage in SCOPs and CFs (this is coherent with
descriptive statistics reported in table 3). However, the characteristic component is
significant and negative and the return component is significant and positive, which
implies two conclusions. Firstly, the workers characteristics give SCOPs a wage
advantage: for example, there are more men, older workers and more tenured workers
in SCOPs which pushes SCOP wages upwards. The industries in which SCOPs are
found also have higher average wages (as the negative coefficient for industry
characteristics demonstrates). However, some characteristic components have positive
signs: for example there are more executives in CFs (table 2) which pushes CF wages
upward. Secondly, returns to characteristics give CFs a wage advantage overall and the
decomposition in table 6 shows the variables that yield a higher return in CFs: working
in a richer region, being an executive, an unskilled white collar or an intern. On the
other hand, tenured workers, semi-skilled white collars, unskilled blue collars and
workers in larger firms are better paid in SCOPs than in CFs. This is a first validation of
hypothesis H3. However, there is one problem with this decomposition, as mentioned
in section 4: it does not control for individual fixed effects. We run the two-step
regressions (equation 4 and 5) including fixed effects (table 6 column 2) and it shows a

EJCE, vol.14, n.2 (2017)
Available online at http://eaces.liuc.it
320
higher predicted hourly wage for SCOP workers, which is in line with the negative
characteristic component in column 1: if the observed characteristics and the individual
fixed effects (controlled for in equation 5) were paid the same in both firms, SCOP
wages would be significantly higher. The characteristic component and its
decomposition show the same signs as the one-step estimation but with smaller values,
due to the fact that we are now comparing estimated wages, as opposed to observed
wages in column (1). The return component however cannot be interpreted directly as
the explanatory power of independent variable is already embedded in the predicted
wages. In order to test hypothesis H3 to H6 while taking into account individual fixed
effects, we have to estimate a model with interaction variables (equation 6).
Table 6. Oaxaca-Blinder results (estimation of equation 3 in column 1 and estimation of equation 4 in
column 2). Return and characteristic components were computed for each explanatory variable of interest.
We control for year dummies.
(1)
Gross hourly
wage
One-step
estimation
(2)
Gross hourly
wage
Two-step
estimation
Total
CF
2.64***
(.0003)
2.64***
(.00006)
SCOP
2.63***
(.006)
2.65***
(.001)
Difference
.008
(.006)
-.0096***
(.001)
Characteristic
-.034***
(.001)
-.007***
(.001)
Return
.042***
(.004)
-.002***
(.00002)
Regions
Characteristic
.008***
(.0006)
.007***
(.0002)
Return
.004***
(.009)
-.0005***
(.0002)
Size
Characteristic
.0002
(.0003)
.00006
(.00005)
Return
-.013***
(.012)
-.0005***
(.0001)
CEO
Characteristic
-.0009
(.0008)
-.0004**
(.0002)
Return
.0002
(.0004)
-9.2e-06
(8.1e-06)
Executive
Characteristic
.014***
(.002)
.007***
(.0005)
Return
.010***
(.002)
-.001***
(.00004)
Intermediate occupation
Characteristic
(ref)
(ref)
Return
(ref)
(ref)

N. Magne, Wage inequality in workers’ cooperatives and conventional firms
Available online at http://eaces.liuc.it
321
(1)
Gross hourly
wage
One-step
estimation
(2)
Gross hourly
wage
Two-step
estimation
Semi-skilled white collar
Characteristic
-.004***
(.0007)
-.020***
(.0002)
Return
-.003***
(.001)
.002***
(.0005)
Unskilled white collar
Characteristic
-.05***
(.0007)
-.001***
(.0001)
Return
.001**
(.0006)
.0006*
(.0004)
Semi-skilled blue collar
Characteristic
.045***
(.002)
.015***
(.0003)
Return
.002
(.004)
-.0002
(.0002)
Unskilled blue collar
Characteristic
.020***
(.002)
.007***
(.0003)
Return
-.003*
(.002)
.00001
(.00006)
Intern
Characteristic
-.010***
(.001)
-.007***
(.0006)
Return
.003**
(.0007)
-.00004**
(.00002)
Industry
Characteristic
-.023***
(.0006)
-.009***
(.0001)
Return
-.005
(.035)
-.0001***
(.00002)
Male
Characteristic
-.014***
(.0005)
-.00002***
(7.7e-06)
Return
-.003
(.006)
.00005
(.0002)
Age
Characteristic
-.013***
(.001)
.0005***
(.00002)
Return
.011
(.037)
-.00002
(.00005)
Seniority
Characteristic
-.008***
(.0007)
-.006***
(.0002)
Return
-.011***
(.004)
.0007***
(.0001)
Part-time
Characteristic
.0002***
(.0003)
.003***
(.0001)
Return
-.002
(.002)
.00006
(.00005)
Number of observations
9,877,079
23,613 SCOPs
9,853,466 CFs
9,877,079
23,613 SCOPs
9,853,466 CFs

EJCE, vol.14, n.2 (2017)
Available online at http://eaces.liuc.it
322
As shown in table 7, some variables have less of an impact on wages in SCOPs
than in CFs. The most striking difference is the occupied position (see more
information about this variable in box 1), which counts much less as a wage
determinant: executives are paid less in SCOPs than in CFs. This effect holds when
individual fixed effects are controlled for (column 2). The hausman test was carried out
and the hypothesis that the individual-level effects are adequately modeled by a random-
effects model is rejected. Therefore our model should imply fixed effects, as written in
equation (6). The coefficient of the interaction variable can be interpreted as follows: its
negative sign means the effect of higher qualifications on wages is weaker in SCOPs
than it is in CFs within individuals. In other words, an executive who switches from a
CF to a SCOP will lose more in terms of remuneration than an unqualified worker. All
qualification variables are dummies and therefore should be interpreted in relation to
the intermediate occupation used as a reference. For instance, switching from a CF to a
SCOP for an executive implies a wage loss 5% higher than for an intermediate worker.
On the other hand, unskilled workers (white and blue collars) appear to be better paid in
SCOPs than CFs. This confirms hypothesis H3, although it is worth noting the
exception of interns who are paid less in SCOPs.
Table 7: Estimate of the wage equation for individual jobs, with interaction variables (equation 6) Panel
data 2001-2012, excluding incidental jobs.
(1)
Generalized least square
(2)
With individual
fixed effects
Position
CEO
.49***
(.001)
.37***
(.003)
Executive
.30***
(.0004)
.20***
(.0007)
Intermediate occupation
(Ref)
(Ref)
Semi-skilled white collar
-.13***
(.0004)
-.09***
(.0005)
Unskilled white collar
-.20***
(.0003)
-.13***
(.0005)
Semi-skilled blue collar
-.17***
(.0003)
-.09***
(.0005)
Unskilled blue collar
-.24***
(.0004)
-.13***
(.0006)
Intern
-.74***
(.0005)
-.63***
(.001)
Contract
Full-time contract
(Ref)
(Ref)
Part-time contract
.04***
(.0002)
.06***
(.0003)
Seniority
.005***
(.0002)
.004***
(.00004)
Individual characteristics
Female
(ref)

N. Magne, Wage inequality in workers’ cooperatives and conventional firms
Available online at http://eaces.liuc.it
323
(1)
Generalized least square
(2)
With individual
fixed effects
Male
.09***
(.0004)
Age
.008***
(.00001)
Interaction variables
Position
CEO
-.06***
(.02)
-.02
(.04)
Executive
-.08***
(.007)
-.05***
(.01)
Intermediate occupation
(Ref)
(Ref)
Semi-skilled white collar
.03***
(.007)
.02**
(.01)
Unskilled white collar
.04***
(.01)
.04***
(.01)
Semi-skilled blue collar
.006
(.006)
.006
(.009)
Unskilled blue collar
.02***
(.006)
.02*
(.01)
Intern
-.14***
(.01)
-.14***
(.03)
Contract
Full-time contract
(ref)
(ref)
Part-time contract
.002
(.004)
-.01
(.01)
Seniority
.002***
(.0003)
.003***
(.001)
Individual caracteristics
Female
(ref)
(ref)
Male
.01***
(.005)
.02*
(.01)
Age
-.001***
(.0002)
-.001**
(.0004)
Number of observations
10,398,099
10,398,099
R²
.51
.60
Gender inequality on the other hand is not diminished in SCOPs; it even tends to
be higher. The coefficient of the interaction variable between the male dummy and the
SCOP dummy can be interpreted as the impact of being a man on the within-individual
effect of the SCOP status on wages. The male variable is not included as it is time-
invariant: its effect is included in the fixed effect. In all our regressions, this coefficient
was found to be positive when significant, which means that women see their wages
reduced more than men (or not less) when they switch from CF to SCOP. Hypothesis
H4 is therefore validated. However, the gender variable is potentially endogenous since
the proportion of men and women working in the firm depends on the hiring strategy

EJCE, vol.14, n.2 (2017)
Available online at http://eaces.liuc.it
324
of the firm and possible self-selection by the workers (women could anticipate lower
wages and be more reluctant to work in SCOPs
28
). In case of endogeneity, the gender-
gap coefficient would be underestimated for SCOPs; in other words, if endogeneity
were controlled for, women could come out as even more disadvantaged in SCOPs.
This point is an interesting direction for future research on SCOP recruitment strategies.
Meanwhile, equation (6) was estimated for the sub-sample of the service industry (where
women are less under-represented in SCOPs: 41% for 50% in CFs) and the gender gap
was also found to be higher in SCOPs: estimated coefficient β10 was .02 (higher than in
the whole regression as shown in table 7, which feeds the endogeneity hypothesis) and
significant at the 5% level.
Seniority is found to have a stronger impact in SCOPs than in CFs as shown by
the positive interaction variables, validating hypothesis H5: one more year of seniority
causes a 0.4% wage augmentation in CFs and a 0.7% augmentation in SCOPs. Age
however has a weaker impact on wage in SCOPs than in CFs. Whereas the size of the
firm has a positive effect on all workers’ wages, this effect is stronger in SCOPs for
firms above 100 employees. Hypothesis H6 seems to be validated as well. However, we
do not control for the seniority of the firm: this effect could be due to bigger firms
being older, as we have reason to believe that older SCOPs are likely to be more
successful because of their locked assets. Finally, there is no significant difference
between SCOPs and CFs regarding the effects of industries, regions or part-time
contracts.
Estimation of equation (6) involves many interaction terms and, as such, may lead
to an increase in multicollinearity. Multicollinearity was assessed using variance inflation
factor (VIF) with simple OLS estimation of equation (6). The mean VIF for all 130
regressors is 8.7, with the age interacted variable as well as 7 interacted industry variable
showing a VIF superior to 10 (usually considered as the acceptable threshold). As a
robustness test, we run the estimations with more aggregated industries and without the
age interacted variable (collinear with seniority). The results on our variables of interest
remained unchanged. All regressions were run including and excluding incidental jobs
(for the workplace dataset, excluding workplaces with only incidental jobs at the end of
the year) as well as with balanced panels (keeping only individuals or workplaces that
were present throughout the whole period). Multicollinearity with the balanced panel
estimation led us to drop interacted year dummies. Estimations were also carried out
with daily wage instead of hourly wage (restricting the sample to full-time employees) to
account for any error in declared hours worked. We then restricted the sample to the
2006-2012 period to incorporate the two variables only available for that sub-period: the
type of contract and the existence of a collective agreement at the workplace level. Our
main results remained unchanged. Finally, potentially endogenous control variables were
removed from the workplace regressions (table 5), namely the proportion of permanent
labor contracts and the proportion of women. The impact of the SCOP dummy on
deciles and interdecile ratios remained unchanged.
Overall, we find strong evidence of lower inequality in SCOPs than in CFs. This
reduction does not affect all workers in the same way. SCOPs actually show an increase
28
It should also be mentioned that the average gender gap (reported in table 3) without any control
variable is actually lower in SCOPs (men earn 20% more than women in CFs and only 17% more in
SCOPs), which is due to the fact that a higher proportion of women are executives in SCOPs that in
CFs. This makes the issue of gender inequality in French SCOPs more complex and worthy of further
research.

N. Magne, Wage inequality in workers’ cooperatives and conventional firms
Available online at http://eaces.liuc.it
325
in seniority and gender inequality and inequality due to the size of the firm. They have
no impact on part-time contracts or regional and industry inequalities, whereas they
strongly diminish inequality due to qualifications.
6. Conclusion
Our extensive panel database allows us to compare wage distribution in French
worker cooperatives (SCOPs) versus conventional firms (CFs) while controlling for a
large set of variables. We estimate Mincerian-type wage equations to quantify the wage
impact of working in a SCOP and find a slightly negative impact of the SCOP status on
wages. We then estimate the same equations using deciles and interdecile ratios as
dependent variables and the results show much lower inequalities in SCOPs than in
CFs, both within and between firms. More specifically, inequalities are lower mostly at
the top of the wage distribution. The Oaxaca-Blinder regression allows us to distinguish
between wage differences due to workers’ or firms’ characteristics and differences due
to return to characteristics. It appears that SCOP workers’ characteristics would confer
them higher wages if they were paid the same as in CFs, but the returns of these
characteristics (particularly qualification) are lower in SCOPs. Finally, we introduce
interaction variables in order to measure any difference in the return of individuals’ or
firms’ characteristics on wages in CFs and SCOPs, while controlling for individual fixed
effects. We find that qualification-based inequalities are those most significantly lowered
in SCOPs. Inequalities between regions and industries are not significantly different,
whereas gender and seniority inequalities, as well as inequality due to the size of the firm
are raised.
These findings contribute to the empirical debate about LMF objectives, showing
a maximization of the median income more than the average wage. Our results support
Kremer’s median voter theory since the gap between high and low wages is reduced in
SCOPs and minorities (namely women and new workers) are not favored. However, it
does not sufficiently explain the durability of the reduced wage gap since there is no
strict mobility barrier preventing highly qualified workers from leaving SCOPs (although
higher return on seniority could be a indirect barrier). A precise measurement of
qualified workers’ turn-over is an interesting direction for future research. We can make
the hypothesis that intrinsic motivation and more specifically aversion to inequality plays
an important role in qualified workers’ decision to work in SCOPs. More generally, a
thorough analysis of SCOPs recruitment and dismissal strategies promises to be highly
instructive.
The larger wage advantage in SCOPs for senior workers raises the question of a
possible difference within SCOPs between members and hired workers. We are not able
to differentiate between the two in our dataset but it is an interesting direction for
further research. Higher wages in larger or older SCOPs could be linked to productivity
differentials throughout cooperatives’ life cycle, which could be explored by matching
the DADS with financial data. Finally, the remaining gender inequality sheds light on
the inequality SCOPs focus on: qualification inequality. Beyond the specific case of
workers’ cooperatives and in the political context of fast rising inequalities in developed
countries, these findings also give some insight into the consequences of workers’
democratic participation in terms of income distribution.

EJCE, vol.14, n.2 (2017)
Available online at http://eaces.liuc.it
326
Bibliography
Abramitzky R. (2008), ‘The limits of equality: insights from the Israeli Kibbutz’, Quarterly Journal
of Economics, 123(3), 1111–1159
Antecol H., Jong A., Steinberger M. (2008) ‘The Sexual Orientation Wage Gap: The Role of
Occupational Sorting and Human Capital’, Industrial and Labor Relations Review, 61, 518-43
Bartlett W. et al. (1992), ‘Labor-Managed Cooperatives and Private Firms in North Central Italy:
An Empirical Comparison.’ Industrial and Labor Relations Review, 46, 103–118
Becker S., Ichino A. (2002) ‘Estimation of average treatment effects based on propensity scores’,
The Stata Journal, 2, 358–377
Blinder A. (1973), ‘Wage discrimination: reduced form and structural estimates’, The Journal of
human resources 8, 436-55
Burdín G. (2016), ‘Equality under Threat by the Talented: Evidence from Worker-Managed
Firms’, The Economic Journal, 126, 1372-1403
Burdín G., Dean A. (2009), ‘New evidence on wages and employment in worker cooperatives
compared with capitalist firms’, Journal of Comparative Economics, 37, 517–533
Burnod G., Chenu A. (2001), ‘Employés qualifiés et non qualifies : une proposition
d’aménagement de la nomenclature des catégories socioprofessionnelles’, Travail et Emploi, 86,
87–105.
Card D., Lemieux T., Riddell C. (2003), ‘Unionization and wage inequality: a comparative study
of the US, the UK and Canada’, NBER Working Paper, 9473
Charmettant H. et al. (2016), ‘Les Scop face aux tensions entre flexibilité et sécurité de l’emploi :
sécuflexibilité’, Formation Emploi, 134
Clemente J. et al. (2012), ‘The wage gap between cooperatives and capitalist firms: evidence from
Spain’, Annals of Public and Cooperative Economics, 83, 337-56
Dabla-Norris E. et al. (2015) ‘Causes and Consequences of Income Inequality: A Global
Perspective’, Staff Discussion Notes, IMF
Defourny J., Estrin S., Jones D. C. (1985), ‘The effects of workers’ participation on enterprise
performance: Empirical evidence from French cooperatives’, International Journal of Industrial
Organization, 3, 197–217.
Demoustier D. (1984), Les Coopératives de production, La Découverte, Paris.
Dow G. K. (2003), Governing the Firm: Workers’ Control in Theory and Practice, Cambridge University
Press.
Duguet E., Petit P. (2009), ‘La présence syndicale réduit-elle la discrimination salariale à
l’encontre des femmes ? Un examen sur l’année 2002’, Revue d’économie politique, 119, 401-50
Fakhfakh F., Pérotin V., Gago M. (2012), ‘Productivity, Capital, and Labor in Labor-Managed
and Conventional Firms: An Investigation on French Data’, Industrial and Labor Relations Review,
65, 847–879
Francois P. (2003), ‘Not-For-Profit Provision Of Public Services’, The Economic Journal, 113,
C53–C61
Freeman R. B. (1980), ‘Unionism and the dispersion of wages’, Industrial and Labor Relations
Review, 34, 3-23
Greenberg E. S. (1986), Workplace democracy: the political effects of participation, Cornell University
Press
Gueslin A. (1998), L’Invention de l’économie sociale, Economica, Paris
Hansmann H. (1996), The Ownership of Enterprise, Harvard University Press
Heitmueller A. (2005), ‘A Note on Decompositions in Fixed Effects Models in the Presence of
Time-Invariant Characteristics’, IZA Discussion Paper

N. Magne, Wage inequality in workers’ cooperatives and conventional firms
Available online at http://eaces.liuc.it
327
Josefy M. et al. (2015), ‘All Things Great and Small: Organizational Size, Boundaries of the Firm,
and a Changing Environment’, The Academy of Management Annals, 9, 715–802
Kamshad K. M. (1997), ‘A model of the free-entry producer cooperative’, Annals of Public and
Cooperative Economics, 68, 225–245
Kremer M. (1997), ‘Why are worker cooperatives so rare?’, NBER Working Paper, 6118
Koevoets W. (2007), ‘Union wage premiums in Great Britain: Coverage or membership?’,
Labour Economics, 14, 53-71
Leclair M., Petit P. (2004), ‘Présence syndicale dans les établissements : quel effet sur les salaires
masculins et féminins ?’ Economie et statistiques, 371, 23-47
Leete L. (2001), ‘Whither the Nonprofit Wage Differential? Estimates from the 1990 Census’,
Journal of Labor Economics, 19, 136-70
Lemieux T. (1998), ‘Estimating the Effects of Unions on Wage Inequality in a Panel Data Model
with Comparative Advantage and Nonrandom Selection’, Journal of Labor Economics, 16, 261-91
Melly B. (2005), ‘Decomposition of differences in distribution using quantile regression’, Labour
Economics, 12, 577-90
Mocan H.N., Tekin, E., (2003) ‘Nonprofit Sector and Part-Time Work: An Analysis of
Employer-Employee Matched Data on Child Care Workers’, The Review of Economics and
Statistics, 85, 38–50
Mosca M., Musella M., Pastore F. (2007), ‘Relational Goods, Monitoring and Non-Pecuniary
Compensations in the Nonprofit Sector: The Case of the Italian Social Services’, Annals of
Public and Cooperative Economics, 78, 57-86
Narcy M. (2011), ‘Would nonprofit workers accept to earn less? Evidence from France’, Applied
Economics, 43, 313–326
Main B., Reilly B. (1992), ‘Women and the Union Wage Gap’, The Economic Journal, 102, 49-66
Oaxaca R. (1973), ‘Male-female wage differentials in urban labor markets’, International Economic
Review, 14, 693-709
Ostry J.D., Berg A., Tsangarides C. (2014), ‘Redistribution, Inequality and Growth’, Staff
discussion Notes, IMF
Pencavel J. (2001), Worker Participation: Lessons from Worker Co-ops of the Pacific Northwest, Russell
Sage Foundation
Pencavel J., Pistaferri L., Schivardi F. (2006), ‘Wages, Employment, and Capital in Capitalist and
Worker-Owned Firms’, Industrial and Labor Relations Review, 60, 23–44
Preston A. E. (1989), ‘The Nonprofit Worker in a For-Profit World’, Journal of Labor Economics,
7, 438–463.
Roelants B. et al. (2012), The Resilience of the Cooperative Model, CECOP CICOPA Europe
Roelants B., Eum H., Terrasi E. (2014), Cooperatives and Employment: a global report, CECOP
CICOPA Europe
Rosenbaum P. R., Rubin D. B. (1983), ‘The central role of the propensity score in observational
studies for causal effects’, Biometrika, 70, 41–55
Ruhm, C. J., Borkoski C. (2003), ‘Compensation in the Nonprofit Sector’, Journal of Human
Resources, 38, 992-1021
Toucas-Truyen P., Dreyfus M. (2005), Les Coopérateurs : deux siècles de pratiques coopératives, Editions
de l’Atelier, Paris
Ward B., (1958), ‘The Firm in Illyria: Market Syndicalism’ The American Economic Review, 48, 566–
589

EJCE, vol.14, n.2 (2017)
Available online at http://eaces.liuc.it
328
Appendices
Table 8. Variables from individual dataset
Job characteristics
Wage
Total wage (net and gross)
Hours
Total number of hours worked
Duration of the
contract
In days
Type of contract
Permanent contract (CDI), temporary contract (CDD), “incidental job” (shorter than 30 days
and 120 hours, paid less than 3 month minimum wage) or apprenticeship. This variable is
only available for the period 2006-2012. Robustness tests were carried out for this sub-
period and did not change our main results.
Working hours
Full-time or part-time
Seniority
In years
Occupation
Head manager, executive, intermediate occupation, semi-skilled white-collar, unskilled white-
collar, semi-skilled blue-collar, unskilled blue-collar.
Individual characteristics
Age and age2
Workers’ age in years and age²
Man
Dummy variable equal to 1 if the worker is a man
Seniority
Seniority within the firm in years
Workplace characteristics
SIRET
Workplace identifier
Size
11 dummies for the number of employees in the workplace (non-incidental jobs on the 31st
of December)
1: from 1 to 4
2: from 5 to 9
3: from 10 to 19
4: from 20 to 49
5: from 50 to 99
6: from 100 to 249
7: from 250 to 499
8: from 500 to 999
9: from 1000 to 1999
10: from 2000 to 4999
11: 5000 and above
Industry
13 industries corresponding to the A17 national classification (4 industries have been
removed because they do not include any SCOPs: agriculture, finance and insurance, real
estate and public administration)
1: Manufacture of food products, beverages and tobacco products
2: Manufacture of electrical, computer and electronic equipment; Manufacture of machinery
3: Manufacture of transport equipment
4: Other manufacturing
5: Energy, water supply, sewerage, waste management and remediation activities
6: Construction
7: Wholesale and retail trade; repair of motor vehicles and motorcycles
8: Transportation and storage
9: Accommodation and food service activities
10: Information and communication
11: Professional, scientific, technical, administrative and support service activities
12: Education, human health and social work activities
13: Other service activities
Collective agreement
Dummy equal to 1 if the workplace is submitted to a collective agreement (95%). This
variable is only available for the sub-period 2006-2012
Region
23 regional dummies: Ile-de-France, Champagne-Ardenne, Picardie, Haute-Normandie,
Centre, Basse-Normandie, Bourgogne, Nord-Pas-de-Calais, Lorraine, Alsace, Franche-
Comté, Pays de la Loire, Bretagne, Poitou-Charentes, Aquitaine, Midi-Pyrénées, Limousin,
Rhône-Alpes, Auvergne, Languedoc-Roussillon, Provence-Alpes-Côte d’Azur, Corse,
Dom

N. Magne, Wage inequality in workers’ cooperatives and conventional firms
Available online at http://eaces.liuc.it
329
Table 9. Additional variables from workplace dataset
Workplace characteristics
SIRET
Workplace identifier
Wage
Total payroll for the year (net and gross)
Hours
Total number of hours worked by all employees
Industry
28 industries corresponding to the A38 national classification (10 industries have been
removed because they do not include any SCOPs: agriculture, mining and quarrying,
electricity, coke and refined petroleum products, pharmaceutical products, finance and
insurance, real estate, public administration, activities of households as employers and
extraterritorial activities)
1: Manufacture of food products, beverages and tobacco products
2: Manufacture of textile, wearing apparel, leather and related products
3: Manufacture of wood and paper products; printing and reproduction of recorded media
4: Manufacture of chemicals and chemical products
5: Manufacture of rubber and plastics products and other non-metallic mineral products
6: Manufacture of basic metals and fabricated metal products, except machinery and
equipment
7: Manufacture of computer, electronic and optical products
8: Manufacture of electrical equipment
9: Manufacture of machinery and equipment
10: Manufacture of transport equipment
11: Other manufacturing; repair and installation of machinery and equipment
12: Water supply; sewerage, waste management and remediation
13: Construction
14: Wholesale and retail trade; repair of motor vehicles and motorcycles
15: Transportation and storage
16: Accommodation and food service activities
17: Publishing, audiovisual and broadcasting activities
18: Telecommunication
19: IT and other information services
20: Legal, accounting, management, architecture, engineering, technical testing and analysis
activities
21: Scientific research and development
22: Other professional, scientific and technical activities
23: Administrative and support service activities
24: Education
25: Human health activities
26: Residential care and social work activities
27: Arts, entertainment and recreation
28: Other service activities
Distribution of the workforce
Proportion of types
of contracts
Permanent contract (CDI), temporary contract (CDD), “incidental job” (shorter than 30 days
and 120 hours, paid less than 3 months minimum wage) or apprenticeship
Proportion of
hours per gender
Proportion of hours worked by men/ women
Proportion of wage
per gender
Total payroll for men/women
Proportion of
hours per
occupation
Proportion of hours worked by head manager, executive, intermediate occupation, semi-skilled
white-collar, unskilled white-collar, semi-skilled blue-collar, unskilled blue-collar.
Proportion of wage
per occupation
Total payroll for: head manager, executive, intermediate occupation, semi-skilled white-collar,
unskilled white-collar, semi-skilled blue-collar, unskilled blue-collar.
Decile
Decile of full-time-equivalent net and gross wages (only for years 2009-2010)