Anda di halaman 1dari 25

OXFORD BULLETIN OF ECONOMICS AND STATISTICS, 72, 3 (2010) 0305-9049

doi: 10.1111/j.1468-0084.2010.00587.x

Why So Unhappy? The Effects of Unionization on


Job Satisfaction
Alex Bryson, Lorenzo Cappellari and Claudio Lucifora,
National Institute of Economic and Social Research, Smith Square, London, UK
(e-mail: a.bryson@niesr.ac.uk)
Istituto di Economia dellImpresa e del Lavoro, Universit Cattolica, Largo Gemelli,
1 1-20123, Milano, Italy (e-mail: lorenzo.cappellari@unicatt.it)
IZA (Bonn) and ERMES (Paris) (e-mail: claudio.lucifora@unicatt.it)

Abstract
Using linked employeremployee data we investigate the job satisfaction effect of
union membership in Britain. We develop a model that simultaneously controls for
the determinants of individual membership status and for the selection of employees
into occupations according to union coverage. We find a negative association between
membership and satisfaction. However, having accounted for selection effects, we find
that the negative association is confined to non-covered employees. This is consistent
with voice effects, whereby non-covered members voice dissatisfaction to achieve
union goals, and with the possibility that membership increases preferences for collective bargaining, thus lowering memberssatisfaction in non-covered environments.

I.

Introduction

Surveys of employees opinions typically reveal that union members reported job
satisfaction is lower than non-members. This empirical regularity has been found in
several data sets, in different countries and different time periods.1 This is puzzling as
The authors thank the Department for Business, Innovation and Skills the Economic and Social Research
Council, the Advisory, Conciliation and Arbitration Service and the Policy Studies Institute who co-sponsored
the Workplace Employment Relations Surveys, and acknowledge the UK Data Archive as the distributor of
the data. The authors are grateful to two anonymous referees for their helpful comments.
JEL Classification numbers: J28, J51

The empirical evidence is mostly concentrated in Anglo-Saxon countries. Selected studies reporting negative effects of membership on job satisfaction are: Freeman and Medoff (1984), Gordon and Denisi (1995) and
Borjas (1979) for the United Status; Guest and Conway (2004), Bender and Sloane (1998) and Bryson et al.
(2004) for the United Kingdom; Meng (1990) and Renaud (2002) for Canada; Miller (1990) for Australia;
Frenkel and Kuruvilla (1999) for South Korea.

357
Blackwell Publishing Ltd and the Department of Economics, University of Oxford, 2010. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.

358

Bulletin

unions should improve working conditions, this being among the reasons for joining
a union.
Confronted with the negative association between membership and job satisfaction, previous research has advanced two main alternative interpretations. A
first explanation is based on the sorting hypothesis, suggesting that dissatisfaction
is mainly driven by poorer working conditions in unionized environments or by
individual characteristics of unionized workers that are imperfectly controlled for
in the empirical analysis, generating spurious correlation between membership and
satisfaction (Schwochau, 1987; Bender and Sloane, 1998). A second explanation has
focused on the voice hypothesis, whereby unions may trigger members reported
dissatisfaction by allowing them to express their complaints instead of quitting the
job (e.g. through grievance procedures, as in the exit-voice theory of Freeman and
Medoff, 1984), or because they politicize the labour force and strategically foment
discontent to increase bargaining power (Borjas, 1979; Davis-Blake and Pfeffer, 1990;
Gordon and Denisi, 1995).
Previous studies have paid little attention to the modelling of the institutional
context for union activity and bargaining practices at the workplace. In particular,
union bargaining coverage at the workplace has been overlooked on the assumption
that only individual membership matters for job satisfaction. This might not be a
problem in countries like the United States where membership and coverage are
virtually synonymous, but it may be important in countries like the United Kingdom
where many workers are covered by some type of collective contract but are not
union members and vice versa. There are reasons to think that adding coverage to
the picture can improve our understanding of the membership satisfaction puzzle.
The need to use voice may differ in covered and non-covered jobs implying different
effects of membership on satisfaction in the two cases. In addition, membership may
be associated with individuals preferences for collective bargaining, and members
dissatisfaction in non-covered occupations may reflect such preferences. However,
the incentives to join the union can differ by coverage, which in turn may affect sorting
into membership. Finally, the sorting of workers into covered and non-covered jobs
may itself be non-random, implying that coverage status, like membership, may be
endogenous. For example, workers might queue for covered jobs when the latter are
rationed (Abowd and Farber, 1983) and failure to obtain a union-covered job may
itself generate dissatisfaction among members if lack of coverage is inconsistent with
their preferences.
This article is the first to consider the interplay between individual union
membership and bargaining coverage in explaining the link between membership
and satisfaction. We use linked employeremployee data representative of the British
workforce to analyse job satisfaction while simultaneously addressing employees
selection into both union membership and covered jobs within workplaces. We exploit
the linked nature of the data to identify the impact of unionization on satisfaction.
Our results indicate that membership effects on satisfaction depend crucially upon
bargaining coverage. We find no differences in satisfaction between members and
Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

Trade Unions and Job Satisfaction

359

non-members in the covered sector, consistent with the sorting interpretation. But
among workers not covered by collective bargaining, union membership is found to
lower satisfaction with the job. This finding can be interpreted in the framework of the
voice hypothesis, namely union members express their dissatisfaction with a view to
achieving what are typically union goals, such as improving bargaining power in the
absence of formal bargaining rights by encouraging organization among co-workers.
It may also reflect an effect of membership on preferences for collective bargaining,
which causes dissatisfaction in non-covered environments. Finally, we find that union
coverage and job satisfaction propensities are positively correlated, suggesting that
there are queues for union-covered jobs.
The article is organized as follows. Section II reviews the literature on job satisfaction and union membership and discusses the implications for union bargaining
coverage. Section III describes the data and presents a preliminary analysis of the
membership/satisfaction relationship. Our empirical strategy is outlined in section IV,
where job satisfaction, union membership and union coverage are jointly modelled.
Section V reports the main set of results, whereas the last section concludes.

II.

Job satisfaction, union membership and union bargaining coverage

Virtually all the studies that have been engaged in the empirical analysis of job
satisfaction and union membership have found the two variables to be negatively
associated. No matter what country, data set or time period was investigated, the
negative membership coefficient in the job satisfaction equation has emerged as an
empirical regularity. This recurrent finding has always been considered counterintuitive, as unions should achieve better working conditions, leading to higher
satisfaction with the job.
Attempts at explaining the empirical puzzle have been based on two competing
explanatory hypotheses. According to the sorting hypothesis, the negative association reflects the characteristics of individuals who join the union, or the attributes of
the workplaces in which membership is likely to prevail. Thus it is either the inherent
discontent of union members, or the fact that individuals tend to unionize when working conditions are poor, that lowers the estimated membership coefficient in the job
satisfaction equation, and not the fact that membership reduces satisfaction per se.
A corollary of this hypothesis is that once the sorting has been adequately dealt with
in the empirical model, the negative membership effect should become weaker or
disappear. Empirically, researchers have dealt with the issue by reducing the extent
of unobserved heterogeneity, for example, by instrumental variables or panel data
methods. Evidence in support of this interpretation has been provided by Gordon
and Denisi (1995), Bender and Sloane (1998), and Bryson, Cappellari and Lucifora
(2004) among others.
The alternative view to sorting is the voice hypothesis. According to Freeman
and Medoffs (1984) exit-voice theory, the presence of unions increases the returns
to voice, as opposed to exit, so that individuals tend not to quit jobs they dislike,
Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

360

Bulletin

but, rather, to express their complaints through voice mechanisms provided by the
union. Unionized workers voice dissatisfaction to improve their bargaining power,
implying that reported dissatisfaction is strategic and not necessarily indicative of
members true attitudes.2 Borjas (1979) stresses that the union politicizes the labour
force creating an incentive to voice strategic dissatisfaction, but the dissatisfaction
is not genuine. Unions may behave strategically to foment dissatisfaction among
members by making the unsatisfactory elements of their jobs more salient, so that
unionization lowers members job satisfaction vis-`a-vis non-members. Some suggest
that there are limited circumstances, such as organizing campaigns, decertification
elections and contract negotiations, where unions may do this (Davis-Blake and
Pfeffer, 1990; Gordon and Denisi, 1995). Empirically, all these versions of the hypothesis imply that the negative membership effect should still be observable even after
having accounted for endogenous sorting. Evidence supporting this interpretation of
the membership satisfaction puzzle can be found in Borjas (1979), Hersch and Stone
(1990) and Heywood, Seibert and Wei (2002) among others.
Existing research has investigated several aspects of the membership satisfaction
relationship, but so far little attention has been devoted to its interaction with union
bargaining coverage. There are several reasons why the two competing explanations
of the membership satisfaction puzzle can interact with union bargaining coverage.
First, the incentives to be a union member may be stronger in a covered environment compared with a non-covered one, because union effects on pay and working
conditions are stronger where density is higher (Stewart, 1987). The sorting of workers
into union membership can therefore be reinforced in the presence of union coverage.
Everything else equal, one should be able to find evidence supportive of the sorting
hypothesis more in covered environments than in non-covered ones.
Second, the reasons for using voice may differ in covered and non-covered jobs.
In principle, it is in covered occupations that voice should be most effective, because
only those cases are characterized by the institutional features needed for voice to be
heard in the bargaining process (Bender and Sloane, 1998).3 But if we consider alternative versions of the voice explanation, then it is plausible that voice effects could be
at work also in non-covered occupations. For example, membership could increase
workersawareness about unsatisfactory aspects of their jobs (Borjas, 1979), inducing
members in non-covered occupations to report dissatisfaction vis-`a-vis non-members,
as for those occupations there are no union representatives to voice discontent.
Furthermore, unions may have the greatest interest in fomenting dissatisfaction when
they are trying to obtain bargaining rights. Insofar as membership in the non-covered
sector creates the potential for union organizing, non-covered members voice may
be seen by employers as presenting a threat of workplace unionization which can reinforce workersbargaining power: in this sense, non-covered members are substitutes
2
Hammer and Avgar (2005) find that some workers mitigate their dissatisfaction through shirking or absenteeism because the psychological costs of quitting are too high.
3
There is however evidence that the gains to voice in a bargaining context are relatively modest (Guest
and Conway, 2004).

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

Trade Unions and Job Satisfaction

361

for the union. Finding evidence of membership effects in non-covered environments


would therefore be supportive of these versions of the voice hypothesis, although it
would not help distinguishing among them.
Third, preferences for collective bargaining can be associated with union membership, either because workers with a preference for collective bargaining are more
likely to become union members, because membership itself induces or strengthens
such preferences, or both. When covered jobs are rationed, some members will end up
in non-covered occupations, so that members lower job satisfaction in non-covered
environments may reflect their frustrated preferences for collective bargaining.
Empirically, if such preferences determine the sorting into membership, then the
observed membership/satisfaction effect in non-covered occupations should disappear once workers allocation along both dimensions of unionization is adequately
controlled for. However, if it is membership that increases workers preferences for
coverage, then the effect should still be observable even after having allowed for the
endogeneity of membership and coverage.
Ignoring coverage when studying the job satisfaction of members might not be
a problem in countries like the United States where membership and coverage are
virtually synonymous, but it may be a more serious omission in a country like Britain
where almost 40% of workers covered by bargaining are non-members and 26% of
non-covered workers are union members (see the first row of Table A1). The former
are free-riders in that they may benefit from collective bargaining coverage without
becoming union members. This free-riding rate is considerably higher than that in the
United States (Bryson and Freeman, 2006), in spite of the fact that in both countries
most collective bargaining occurs at workplace-level (Kersley et al., 2006). One
explanation is that the union membership decision in Britain is genuinely free as the
closed shop enforcing membership in covered workplaces was outlawed in the early
1990s (Millward, Bryson and Forth, 2000). The incentive to free-ride is bolstered
by the fact that British unions are unable to levy an agency fee on non-members to
cover the union agency costs in bargaining on their behalf. Table A1 also shows that
there is a high percentage of non-covered members,4 which can be explained by the
existence of union benefits unconnected with bargaining coverage which are confined
to members, such as access to union representation in grievance cases, or other
services provided by the union to its members, for example, professional indemnity
insurance.5 For these workers, membership may also be part of their occupational
identification, as is the case for health professionals whose pay is nevertheless set by
4

Until 2000, British workers had a right to union membership but no statutory right to bargaining coverage.
Coverage was at the discretion of the employer and was usually conceded in the face of strong union organizing
campaigns. Since 2000, a new law provides a right to bargaining coverage where the majority of workers desire
it. However, there is evidence that the law has had only very limited impact in practice (Gall, 2004), suggesting
that the interaction of coverage and membership has been substantively unaffected by the law.
5
For an example of how substantial this latter benefit might be see http://www.unison.org.uk
/healthcare/pages view.asp?did=1183, which outlines insurance offered by UNISON, the largest public sector
union in Britain.
Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

362

Bulletin

public sector Pay Review Bodies rather than through collective bargaining (Bach and
Givan, 2004).
The discussion in this section has shown how theories linking union membership to job satisfaction specialize when union coverage is added to the picture, and
that an explicit assessment of the role of bargaining coverage is missing from the
literature. To fill this gap in knowledge, we now use linked employeremployee data
and develop a model that estimates employees job satisfaction while controlling for
endogenous selection along the two relevant dimensions of unionization, coverage
and membership.

III.

Data and preliminary analysis of the membership/satisfaction link

Our data are the linked employeremployee British Workplace Employee Relations
Survey 1998 (WERS). With appropriate weighting, it is nationally representative
of British employees working in workplaces with 10 or more employees covering
all sectors of the economy except agriculture (Airey et al., 1999). The survey
covers a wide range of issues and contains controls for a large set of individual- and
workplace-level attributes. We use two elements of the survey. The first is the management interview, conducted face-to-face with the most senior workplace manager
responsible for employee relations. The second element is the survey of employees
where a management interview was obtained.6
Two aspects of the data make them particularly suited for analysing the interplay
between union membership and union coverage in affecting employees satisfaction.
First, the management interview contains a bargaining coverage indicator for each
occupational group in the workplace, providing a precise measure of coverage that
is difficult to obtain from surveys exclusively based on interviews with employees.7
Second, the available information on workplace attributes is very detailed, providing
variability even within cells defined by industry, size and region, that is, the set of
workplace controls typically deemed to affect job satisfaction in the literature. To the
extent that such variability has an impact on unionization and not on satisfaction,
then linked employeremployee data offer a rich source of variation that is useful
for identifying the impact of unionization on satisfaction. In this respect, linked data
may be seen as a complement to longitudinal surveys that use changes in individual
union status over time for estimating the impact of membership on satisfaction (see,
e.g. Heywood et al., 2002).
The survey asked each employee to provide a rating, on a five-point scale from
very satisfied to very dissatisfied, concerning how satisfied they were on four
6

Response rates were 80 and 64% for the management and employee questionnaires, respectively.
The manager responsible for employment relations at the workplace is asked: Which of the following
statements most closely characterises the way that pay is set for [each occupational group present at the
workplace] ... collective bargaining for more than one employer (e.g. industry-wide agreement); collective
bargaining at an organisation level; collective bargaining at this workplace; set by management at a higher
level in this organisation; set by management at this workplace; negotiation with individual employees; some
other way (e.g. Pay Review Body)? The first three codes constitute coverage by collective bargaining.
7

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

Trade Unions and Job Satisfaction

363

aspects of their job: (i) the amount of influence they had over their job; (ii) the pay
they received; (iii) the sense of achievement they got from their work; and (iv) the
respect they got from supervisors and line managers. We conduct our initial analysis
on each of the four facets. Later we concentrate on the three non-pecuniary job
satisfaction items, where we find the puzzling negative association with union
membership to be more evident. Information on individual membership is derived
from a question in the employee questionnaire. We match the information on union
coverage from the manager questionnaire to individuals using their occupational
category recorded in the employee questionnaire.
Table A1 provides some descriptive statistics for the estimation sample, which is
derived from the original WERS sample after deletion of cases with missing information on variables required for the econometric analysis (approximately 4,000 cases
were deleted in this way). Employees exhibit greater dissatisfaction with pay than
with non-pecuniary facets of their jobs. There are differences in non-pecuniary job
satisfaction between employees in covered and non-covered occupations with those
in non-covered occupations expressing greater satisfaction. Breaking the data down
by union membership and coverage reveals notable differences in personal attributes.
In particular, the profile of union members differs across covered and non-covered
occupations. For example, the proportion of highly educated individuals is larger
among non-covered members compared with covered members, by approximately
50%. Conversely, the distribution of educational attainment does not differ much if
one compares non-members in covered and non-covered occupations. Non-covered
members also tend to be concentrated in professional and technical occupations,
whereas the occupational distribution of covered members is skewed towards manual
jobs. Finally, non-covered members are nearly twice as likely to be found in Health
and Education than covered members (32% vs. 17%), while covered union members
are more heavily concentrated in manufacturing (32% vs. 16%).
We now describe the relationship between job satisfaction and membership by
means of an ordered probit regression of each satisfaction indicator on a membership
dummy and a set of controls that include personal characteristics, job and workplace
attributes. If unions improve overall job quality, one should expect members to be
better off on average compared with non-members whereas, conditional on working
conditions, no job satisfaction differential related to union membership should be
observed. Such an exercise has therefore the aim of assessing whether, and to what
extent, the membership satisfaction puzzle emerges from nave specifications that
control for working conditions and personal characteristics but ignore the endogeneity
of membership and coverage. Contrasting its results with those from later sections
which model selection into membership and coverage will therefore be informative
about the relevance of the sorting interpretation. Regressions are estimated on the
whole sample and the sub-samples of covered and non-covered employees.8 Table 1
8
The regressions are weighted to account for sampling probabilities and a robust variance estimator corrects
for the presence of repeated observations within the same establishment.

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

364

Bulletin

TABLE 1
The effect of union membership on job satisfaction: estimates from models that treat
both membership and coverage as exogenous

Whole sample
Union coverage
No union coverage

Satisfaction
with pay

Satisfaction
with influence
over work

Satisfaction
with respect
from managers

Satisfaction
with sense of
of achievement

0.014 (0.012)
0.002 (0.014)
0.010 (0.024)

0.069 (0.009)
0.069 (0.013)
0.074 (0.014)

0.073 (0.010)
0.079 (0.013)
0.074 (0.014)

0.059 (0.009)
0.066 (0.013)
0.049 (0.013)

Notes: Reported are marginal effects associated with the union member dummy variable in ordered probit
regressions of job satisfaction. The effect refers to the shift in the probability of being very satisfied or satisfied associated with a change in the membership indicator from 0 to 1. The number of observations is 23,601
in the whole sample, of which 9,208 is in the sample of occupations covered by union bargaining and 14,393
in other occupations. Robust standard errors in parentheses. Regression uses survey weights. For each of the
models estimated, the hypothesis that regressors have no explanatory power is rejected with a Pvalue of
0.0000. The variables used as regressors are: gender, age dummies, education dummy, ethnicity, marital and
parental status, disability status, occupational dummies, indicators for training, gender segregation in the job,
whether the employment relationship is fixed term, whether overtime hours are always paid, the availability
of family friendly policies and the possibility to take days off when needed, weekly pay and weekly hours,
industry, establishment size, region, workforce composition, whether establishment is publicly owned and the
unemployment rate in the relevant travel to work area.

focuses on the marginal effects associated with the membership coefficient, using
the probability of being satisfied or very satisfied as the outcome of interest.9 (The
full set of ordered probit coefficients is reported in a table available from the authors
upon request, whereas the choice of the regressors entering the satisfaction equation
is discussed in the next section.) Considering pay satisfaction first, Table 1 shows that
there is no significant association between union membership and job satisfaction:
the puzzling negative effect apparent in much of the previous literature does not seem
to apply to pecuniary facets of satisfaction. This may reflect the union membership
pay premium documented, among others, by Blanchflower and Bryson (2004) using
WERS data, which may compensate for the member/non-member satisfaction differential. The puzzling satisfaction penalty associated with membership is statistically
significant for the other satisfaction items, and ranges between 5 and 8 percentage
points on the probability of being very satisfied or satisfied. As the sample
frequencies of workers who were very satisfied or satisfied reported in Table A1
range between 57 and 63% in the overall estimation sample, union membership shifts
the satisfaction probability by roughly one-tenth of the aggregate probability in the
sample. Table 1 also shows that quantitatively these differentials are very similar in
sub-samples defined by union coverage status.
Table 2 reports union membership satisfaction differentials using propensity score
matching (PSM). As in the case of the regression estimates shown before, interpreta9
Marginal effects are computed from ordered probit coefficients by considering the shift in the probability
of the outcome of interest (i.e. being very satisfied or satisfied on a given job satisfaction facet) induced
by a switch in union membership status, and are evaluated at the sample averages of the controls entering the
regression model.

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

Trade Unions and Job Satisfaction

365

tion of the member/non-member differential as the effect of union membership relies


on the assumption that selection into union membership is captured by observables.
However, unlike regression, PSM is a semi-parametric technique and does not require
assumptions to be made about the functional form of the satisfaction equation. Rather
it compares the satisfaction differential between members and non-members who are
similar with respect to observable attributes. The maintained assumption in regression
and PSM when estimating treatment-on-the-treated is that counterfactuals for
members can be found among non-members who are observationally equivalent as
indicated by their propensity for union membership. In our analysis, this propensity
is estimated as the probability of membership based on a probit estimator. The independent variables entering the probit are identical to those used in the regression
analyses, see section IV. Where members are adjudged to be too far from their nonmember counterparts they have no counterfactuals against which to estimate the union
membership effect and are therefore dropped from the analysis.10
The figures in Table 2 are the mean differences in satisfaction across members
and their matched non-member counterparts where satisfaction is measured as being
very satisfied or satisfied. As in the case of the regression analyses, results are
presented separately for the whole economy, covered and non-covered employees.
The results are qualitatively similar to those obtained from regressions presented
before. The membership/satisfaction differential is smallest with respect to pay.
TABLE 2
Propensity score estimates of the membership/satisfaction differential

Whole sample
Union coverage
No union coverage

Satisfaction
with pay

Satisfaction
with influence
over work

Satisfaction
with respect
from managers

Satisfaction
with sense of
of achievement

0.034 (0.041)
0.006 (0.817)
0.041 (0.144)

0.083 (0.000)
0.036 (0.156)
0.080 (0.001)

0.081 (0.000)
0.056 (0.029)
0.068 (0.005)

0.078 (0.000)
0.085 (0.001)
0.050 (0.034)

Notes: Mean differences in satisfaction across members and their matched non-member counterparts where
satisfaction is measured as being very satisfied or satisfied. Asymptotically robust P-values in parentheses.
Matching uses survey weights. Whole sample N = 13,473 (9,410 members and 4,063 non-members). Covered employees, N = 7,413 (5,549 members and 1,864 non-members). Non-covered employees, N = 5,577
(3,747 members and 1,830 non-members). The reason why the sums of non-covered employees and covered
employees do not equal the base for the whole economy is that matching is undertaken for each of the three
samples whole economy, covered employees and non-covered employees producing different matched
samples.

10
The matching method deployed is nearest neighbour matching. Matches for members were those nonmembers whose estimated probability of membership was up to 0.002 above or below the estimated probability
for the member. Matching is undertaken separately for all employees, covered employees and non-covered
employees. The number of members lost through enforcement of this support requirement was 48 in the
case of the whole economy estimates, 106 among covered employees and 56 among non-covered employees.
Full details of the probit estimation and diagnostic tests for the matching are available from the authors on
request.

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

366

Bulletin

Indeed, in the covered and non-covered sub-samples the differential is not statistically significant. On all three non-pecuniary aspects of satisfaction members were
significantly less satisfied than their non-member counterparts, the differential being
in the range of 510%. The effects did not differ greatly across covered and noncovered employees nor across the different aspects of non-pecuniary satisfaction. The
exception is satisfaction with influence over work among covered workers where the
membership differential was not statistically significant.
The results presented thus far indicate that the membership/satisfaction differential persists having controlled for differences in the characteristics of members
and non-members, except in the case of pay satisfaction (and, in the matched estimates, with respect to influence over work among covered workers). Furthermore,
these effects are apparent in the covered and non-covered sectors. The discussion
in section II indicates that the reason for members satisfaction may be different in
covered and non-covered occupations and that both membership and coverage may
be endogenously determined along dimensions unobservable to the analyst. Moreover, descriptive statistics show that members profile is different in covered and
non-covered occupations. We therefore pursue further analyses that account for the
interplay between union membership and union coverage in shaping members job
satisfaction.

IV. An econometric model of job satisfaction, union membership and


union coverage
To estimate the effect of membership on satisfaction, we extend the instrumental
variables framework used in several previous studies to encompass selection into jobs
covered by collective agreements and those that are not covered. We do so by allowing
for the possibility that unobserved heterogeneity in job satisfaction and membership
may be correlated with the process assigning individuals to covered and non-covered
jobs. In tackling this issue we exploit the linked employeremployee structure of
the WERS data which permits us to match information on union coverage for each
occupational group in the workplace, derived from the manager questionnaire, with
individual-level information on employees.
Let c*i denote the propensity of being employed in a covered job for individual
i, i = 1. . .n. This propensity depends upon two components: the net benefit derived
from covered employment and the employers hiring decision. Both components are
functions of personal and workplace characteristics, observed (xi ) and unobserved
(i ). If there are queues for covered jobs, individuals observable and unobservable
(to the analyst) personal attributes can play a role in affecting employers hiring
decisions.11 The vector of observables xi includes personal attributes (gender, age,
education, disability status, marital and parental status, ethnicity) and workplace
11
Evidence of job queues for union jobs in the United States is discussed by Abowd and Farber (1983),
Farber (2001) and Freeman and Rogers (1999).

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

Trade Unions and Job Satisfaction

367

attributes (industry, size, region, workforce composition, age, whether the workplace
is a standalone workplace or part of a larger organization, and whether it is publicly
or privately owned) plus an indicator of local labour market conditions (the travelto-work area unemployment rate). Individual wages (a job satisfaction shifter in most
of the literature) are not included in xi as they are one of the bargaining outcomes
and therefore endogenous to coverage. We specify c*i as a linear function of its
determinants:
c *i =  xi + i ,

(1)

where  is a coefficient to be estimated. We do not observe c*i ; rather, we observe


whether individual i is covered by union bargaining, an event that signals that
c*i exceeds some latent threshold, which can be set to zero without loss of
generality. Let Ci = I (c *i > 0) indicate the event, where I ( ) is an indicator
function.
Next, let the latent propensity of being a union member, m*i , be a function of the
same set of personal and workplace attributes used in the coverage equation, plus an
unobserved component:
m *i =  xi + vi ,

(2)

where symbols have a meaning analogous to that in equation (1). Although the individual membership decision is possibly less influenced by employer behaviour than
coverage is, the set of workplace characteristics included in xi may capture factors that
are relevant to individual membership, such as industry-specific attitudes towards the
union, and therefore we use in equation (2) the control factors used in the coverage
equation. When the latent membership propensity is positive, we observe individual
i to be a union member; let Mi = I (m *i > 0) index that event. Taken together, equations (1) and (2) fully represent the choice set resulting from the combination of
the membership and coverage choices. Note that the two processes are unconditional one upon the other. As we are not interested in the effect of coverage on
membership we model the two processes in a reduced form fashion, and control
for their interrelationship by estimating the cross-process correlation, as discussed
next.
To estimate the effect of membership on satisfaction while accounting for employees union coverage status, we adopt an endogenous switching framework and allow
the impact of (observed and unobserved) job satisfaction determinants to be different
in the two bargaining regimes:
s *i = (C zi + C Mi + uCi )Ci +(NC zi + NC Mi + uNCi )(1 Ci ),

(3)

where s*i is the individual propensity to be satisfied with the job and the C and
NC subscripts refer to quantities relevant for workers in jobs that are covered or
non-covered by union bargaining. Following Clark and Oswald (1996), s*i may be
Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

368

Bulletin

thought of as the utility derived from working, itself an argument for an overall
individual welfare function. The data do not allow direct observation of s*i , but
provide information on the satisfaction rank reported by employees (Si ). We let the
observed satisfaction rank depend on the underlying satisfaction propensity through
the mapping , Si = (s*i ), which is a step function that takes on a set of ordered
values (from very dissatisfied to very satisfied) depending upon s*i crossing a
set of threshold levels. The coefficients C and NC index the effect of union membership on job satisfaction after taking account of endogenous selection into union
coverage. As this model controls for endogenous sorting, negative estimates of these
coefficients would lend support to the voice hypothesis in covered and/or noncovered occupations. The discussion of section II highlighted that voice effects may
be different in covered and non-covered environments, therefore the model allows
for a differential impact of membership on satisfaction in the two regimes. The
vector of observables zi includes all the variables that affect unionization (but with
some exceptions, see next), plus pecuniary (weekly earnings, hours of work and
whether overtime hours are always paid) and non-pecuniary (training, gender segregation in the job, whether the employment relationship is fixed term, the availability
of family friendly policies and the possibility to take days off when needed) job
attributes, which may influence job satisfaction and may to some extent be correlated with unionization.
Estimating equations (2) and (3), ignoring equation (1), is subject to an endogenous sample selection issue, as long as the unobserved individual determinants of
union coverage are correlated with unobservables in the membership and satisfaction
equations. As discussed in section II there may be queues for job positions covered
by union bargaining, and if the individual attitude towards working (the innate
propensity to be satisfied with the job) is valued and somehow observed by employers,
then we should expect the unobserved (by the researcher) determinants of coverage
and satisfaction to positively co-vary. Moreover, members lower satisfaction in
non-covered jobs may actually reflect their frustration with lack of coverage, and
simultaneously controlling for the processes generating membership and coverage
may help reducing any such spurious correlation.
Besides endogenous selection into coverage, the other source of spurious correlation is the endogeneity of membership, the only one that has been addressed by the
literature thus far. As discussed there, such a correlation is the driver of the sorting
interpretation of the membership satisfaction puzzle. We allow the sorting of individuals into membership to differ by bargaining regime, an aspect that in the light of
our theoretical discussion can be relevant, as the incentives to unionize can differ
between covered and non-covered employees.
We tackle endogeneity by allowing the unobserved individual components of
equations (1)(3) to be jointly distributed according to a quadri-variate normal
distribution with zero means, unit variances and free correlations:
(i , vi , uCi , uNCi ) N4 (0, ).
Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

(4)

Trade Unions and Job Satisfaction

369

By specifying the extra-diagonal elements of the correlation matrix  we introduce unobserved heterogeneity into the model, thereby accounting for the endogeneity
issues outlined before.12
To aid identification of the effects of interest we formulate a set of exclusion
restrictions. In particular, we need to make assumptions about variables that affect
coverage and membership but, conditional on these, have no residual impact on job
satisfaction. To this end, we assume that after controlling for factors such as the establishments industry, size, region and workforce composition throughout the models
equations, some of the workplace characteristics included in xi have no independent
effect on job satisfaction. In doing so, we exploit our linked employeremployee data
to provide identification of the model. We use two instruments, measuring factors
that affect both membership and coverage. The first instrument is establishment age.
We hypothesize that in older (more than 20 years) workplaces the chance of union
bargaining coverage is larger than in otherwise comparable establishments. There
is considerable empirical support for this hypothesis in Britain, as earlier cohorts
of workplaces were more likely to recognize unions for pay bargaining than those
born after 1980 (Machin, 2000; Millward et al., 2000). Older workplaces thus represent an environment that is more favourable to unions compared with younger ones,
which may also affect the individual membership decision, for example, through the
social custom effects discussed by Booth (1995). We assume that after controlling
for establishment characteristics such as industry and size there is no residual impact
of workplace age on job satisfaction. Second, we use as an instrument an indicator of
whether a workplace belongs to a multi-establishment firm or is a standalone workplace. Union coverage has fixed costs for employers (such as the transaction costs in
engaging in negotiation and consultation with worker representatives) which can be
accommodated better in the presence of multiple establishments, whereas employees satisfaction is more likely related to characteristics of the establishment where
they are located. Individual membership propensities are also likely to be affected by
the multi-site structure of the firm, because in single-site firms workers have a much
higher chance of being informed about job-related issues by senior management, so
they are less likely to seek alternative sources of information such as unions (Forth,
Bewley and Bryson, 2006).

V.

Results

Tables 1 and 2 showed that the membership/satisfaction puzzle does not emerge
when considering satisfaction with pay. Therefore, in this section we concentrate
on satisfaction with non-pecuniary job facets. To gauge the distinctive contribution
of accounting for endogenous coverage in addition to endogenous membership, we
12
We estimate the model by maximum simulated likelihood, using the GewekeHajivassiliouKeane (GHK)
simulator and 50 Halton draws. As in the previous section, we use sample weights and a robust variance
estimator. The correlation coefficient between uCi and uNCi is not identified, as it would require individuals
simultaneously observed in covered and non-covered jobs, and does not enter the likelihood function.

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

370

Bulletin

start by considering results from restricted versions of the model of interest in which
only membership is treated as endogenous, that is, as in the existing literature on
union membership and job satisfaction. We then move to the results of the model presented in the previous section, which also controls for the (observed and unobserved)
determinants of coverage.
Table 3 presents the estimates of the effect of union membership on the nonpecuniary facets of job satisfaction and of the correlations of unobservables between
the membership and satisfaction equations (the full set of estimated coefficients is
available from the authors upon request).
Estimates are presented for the whole estimation sample and by (exogenously)
splitting observations according to individual coverage status. Considering the whole
sample first, the negative satisfaction differential distinguishing members from nonmembers is magnified by endogenization of membership status, its absolute value
doubling (approximately from 6 to 13 percentage points) in comparison with the estimates of Table 1. The estimated membership differential is similar across satisfaction
facets. The increase in the absolute value of the effects is suggestive of a positive
selection effect of intrinsically more satisfied individuals into union membership,
which is confirmed by the positive signs on the estimated correlation of error terms
between the membership and satisfaction equations. In other words, it seems that it
TABLE 3
The effect of union membership on job satisfaction: estimates from models with endogenous
union membership and exogenous union coverage
Whole
sample

Union
coverage

Marginal effect of union membership dummy in equation for job satisfaction with
Influence over work
0.133 (0.039)
0.126 (0.147)
Respect from supervisors
0.138 (0.046)
0.093 (0.153)
Sense of achievement derived from work
0.122 (0.041)
0.045 (0.242)

No union
coverage
0.200 (0.047)
0.192 (0.061)
0.160 (0.047)

Correlation of unobservables across equations for union membership and job satisfaction with
Influence over work
0.110 (0.059)
0.273 (0.238)
0.192 (0.068)
Respect from supervisors
0.113 (0.070)
0.057 (0.250)
0.178 (0.088)
Sense of achievement derived from work
0.106 (0.064)
0.195 (0.325)
0.171 (0.066)
Notes: Results are derived from a simultaneous equations model for job satisfaction and union membership. Reported are marginal effects associated with the union member dummy variable in job satisfaction
equations, and the cross-equation correlations of the errors. The effect refers to the shift in the probability
of being very satisfied or satisfied associated with a change in the membership indicator from 0 to 1. The
number of observations is 23,601 in the whole sample, of which 9,208 is in the sample of occupations covered
by union bargaining and 14,393 in other occupations. Robust standard errors in parentheses. Regression uses
survey weights. For each of the models estimated, the hypothesis that regressors have no explanatory power
is rejected with a P-value of 0.0000. The variables used as regressors for job satisfaction are the ones indicated in Table 1. Regressors for the union membership equation are: gender, age dummies, education dummy,
ethnicity, marital and parental status, disability status, occupational dummies, industry, establishment size,
region, workforce composition and the unemployment rate in the relevant travel to work area, a dummy for
establishment age larger that 20 years and indicators for whether the establishment is standalone or publicly
owned.
Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

Trade Unions and Job Satisfaction

371

is not inherent dissatisfaction per se that motivates workers to be part of the union.
Rather, the opposite seems to be true, that is, membership lowers job satisfaction,
consistent with the voice hypothesis.
Splitting the sample by bargaining coverage reveals differences in members
satisfaction relative to non-members. The membership/satisfaction differential
persists among non-covered employees, and is apparent for all three non-pecuniary
job facets. Among covered employees the differential is instead no longer significant.
Moreover, the positive selection effect of intrinsically satisfied individuals into membership is not found among covered employees, the correlations of the unobservables
dwindling to insignificance for all three facets of satisfaction, whereas the effect
remains positive and significant among non-covered employees. It seems, therefore,
that the nature of members satisfaction differs depending upon whether their occupation is covered by collective bargaining. In occupations in which there is no union
representation in bargaining, the fact of being a member increases dissatisfaction.
This result is consistent with the voice hypothesis: in the absence of formal union
representation members voice discontent as a way to increase bargaining power or
to motivate other workers to organize in the hope of obtaining union coverage. One
could also think of an alternative explanation of the negative membership effect in
non-covered occupations, namely that it reflects members frustration at the absence
of bargaining coverage. Conversely, for covered workers, members dissatisfaction is
spurious, not causal, and disappears once the determinants of membership are taken
into account.13 This is in line with the sorting hypothesis and the fact, discussed in
section II, that individual incentives to sort into membership may actually be larger
in covered than in non-covered jobs.
The evidence discussed so far is not conclusive about the nature of non-covered
members dissatisfaction. The negative effect of membership is consistent with the
voice interpretation, but could also reflect unobserved heterogeneity in the process
allocating individuals into membership and (non-)coverage. For example, these may
be individuals who became union members because they have a preference for collective bargaining, but could not find a job in the covered sector, so that the gap between
their preferences and reality lowers their job satisfaction. This may be especially true
in the presence of queues for covered jobs, which grant employers some degree of
freedom in selecting employees from those queues. A way to assess the relevance of
this alternative interpretation is to account for sorting not only into membership but
also into coverage, using the model introduced in section IV. The model allows for
heterogeneity in the propensity to be covered by union bargaining by accounting for
factors that are relevant for sorting into coverage, and influence employers hiring
13
Using an overall job satisfaction indicator and pooling covered and non-covered employees from
WERS98, Bryson et al. (2004) also found that the negative satisfaction/membership link was the result
of self-selection into membership. Their analysis was conditional on a battery of indicators of individuals
opinions on the climate of industrial relations. The fact that these opinions may be correlated with union
coverage can explain why we are able to find results consistent with theirs after splitting the sample according
to coverage status.

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

372

Bulletin

decisions in the presence of queues. If the negative membership effect in the satisfaction equation of non-covered employees reported in Tables 13 reflects endogenous
sorting of members out of covered occupations, then allowing for the simultaneous
endogeneity of membership and coverage should get rid of it.
The main coefficients of interest are reported in Table 4. This table presents
marginal effects of union membership on job satisfaction and the cross-equation
correlations of the error terms in the coverage, membership and satisfaction equations.
Although the whole set of estimates is not reported for lack of space (it is available
upon request), it is worth briefly discussing the estimates of the coefficients associated with the variables excluded from the satisfaction equations, see Table A2. The
instruments (i.e. workplace age and single-establishment organization) are individually statistically significant in shifting membership and coverage propensities at
the 1% confidence level in all cases, but for establishment age in the coverage
equation, for which the confidence level is 10%. Tests of joint significance of the
instruments in each equation and in both equations simultaneously strongly support
the use of these variables as shifters for the two unionization indicators. The two
variables also bear the expected signs, indicating that union members and occupations covered by collective agreements are more likely to prevail in workplaces that
are older and belong to multi-plant organizations. Using normality as the identifying
restriction, we also tested that the instruments were not significant in the satisfaction
equations at conventional levels of confidence, and in all but one of the six satisfaction equations estimated (satisfaction with influence in non-covered occupations)
we found support for their exclusion, as the null hypothesis of joint non-significance
could never be rejected, with P-values in the order of 0.345 or larger. Tests of joint
significance of these variables in the two satisfaction equations for each satisfaction
facet do not reject the null at conventional levels of confidence in the cases of satisfaction with respect from managers and satisfaction with sense of achievement; for
satisfaction with influence over work, the null hypothesis can be rejected at the 5%
level of confidence.
In Table 4, the effect of membership on satisfaction is never statistically significant for employees in covered occupations. This result confirms our previous finding
that in covered occupations the negative impact of membership is not causal but spurious. In contrast, when non-covered occupations are considered, we find a negative
effect of membership on satisfaction. When workers are not covered by a collective
agreement being a union member lowers reported satisfaction vis-`a-vis non-members.
Compared with the corresponding effects in Table 3, figures in Table 4 are between 7
and 12 percentage point lower in absolute values, implying a proportional reduction
between 35 and 65%. Allowing for endogeneity of coverage thus reduces the negative
effect of membership, but does not remove it. Thus, the sorting of members out of
covered occupations can only partly explain the negative membership effect found in
Tables 13, but cannot entirely account for it, because otherwise we should observe no
remaining membership effect after having endogenized coverage. Results in Table 4
are instead consistent with the existence of voice effects in non-covered occupations.
Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

Trade Unions and Job Satisfaction

373

TABLE 4
The effect of union membership on job satisfaction: estimates from models that treat
both membership and coverage as endogenous
Model for job satisfaction with influence over work
Effect of union membership dummy on job satisfaction
(a) If union coverage
(b) If no union coverage
Difference (a) (b)
Correlations of unobservables in model of satisfaction with influence
over work across equations for
Membership and coverage
Coverage and satisfaction if union coverage
Membership and satisfaction if union coverage
Coverage and satisfaction if no union coverage
Membership and satisfaction if no union coverage
Model for job satisfaction with respect from supervisors
Effect of union membership dummy on job satisfaction
(a) If union coverage
(b) If no union coverage
Difference (a) (b)
Correlations of unobservables in model of satisfaction with respect from
supervisors across equations for
Membership and coverage
Coverage and satisfaction if union coverage
Membership and satisfaction if union coverage
Coverage and satisfaction if no union coverage
Membership and satisfaction if no union coverage
Model for job satisfaction with sense of achievement from work
Effect of union membership dummy on job satisfaction
(a) If union coverage
(b) If no union coverage
Difference (a) (b)
Correlations of unobservables in model of satisfaction across equations for
Membership and coverage
Coverage and satisfaction if union coverage
Membership and satisfaction if union coverage
Coverage and satisfaction if no union coverage
Membership and satisfaction if no union coverage

0.052 (0.083)
0.132 (0.038)
0.183 (0.092)

0.343 (0.027)
0.174 (0.174)
0.158 (0.140)
0.178 (0.072)
0.158 (0.056)

0.034 (0.084)
0.073 (0.041)
0.108 (0.095)

0.343 (0.027)
0.128 (0.124)
0.133 (0.124)
0.280 (0.077)
0.090 (0.066)

0.003 (0.083)
0.085 (0.037)
0.082 (0.092)
0.343 (0.027)
0.119 (0.132)
0.072 (0.124)
0.155 (0.076)
0.100 (0.053)

Notes: Results are derived from the simultaneous equations model for job satisfaction, union membership
and union coverage. Models estimated by maximum simulated likelihood, using a GHK simulator with 50
Halton draws. Reported are marginal effects associated with the union member dummy variable in job satisfaction equations that switch according to coverage, and the cross-equation correlations of the errors. The
effect refers to the shift in the probability of being very satisfied or satisfied associated with a change in the
membership indicator from 0 to 1. The number of observations is 23,601. Robust standard errors in parentheses. Regression uses survey weights. For each of the models estimated, the hypothesis that regressors have no
explanatory power is rejected with a P-value of 0.0000. The variables used as regressors for job satisfaction
are the ones indicated in Table 1, those for the membership equation are indicated in Table 3 and the coverage
equation uses the same set of regressors as the membership equation.

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

374

Bulletin

Trade unions may increase members awareness about unsatisfactory aspects of the
job, inducing them to express dissatisfaction when there are no union representatives. Moreover, in the absence of formal representation, members may strategically
foment dissatisfaction so as to encourage organization among co-workers, what may
be seen as a threat of workplace unionization by employers and strengthen workers
bargaining power.14 As discussed in section II, the negative membership effect in noncovered occupation may also reflect a causal effect of membership on preferences for
collective bargaining, which get frustrated in the absence of union coverage.
Table 4 also reports tests of significance for the difference in the marginal effects
of membership on satisfaction between covered and non-covered occupations. The
difference is statistically significant at the 5% confidence level only in the case of
satisfaction with influence over work. However, for each satisfaction indicator, the
95% confidence interval (not reported) of the marginal effect for non-covered occupations does not contain the marginal effect for covered ones, suggesting that failure
to reject the null hypothesis is driven by imprecision of the estimates for the case of
covered occupations.
Further support for the fact that membership effects in covered occupations are
spurious comes from the estimated correlations of the errors for the satisfaction and
membership equations. Although not statistically significant at conventional levels,
their negative signs suggest that inherently dissatisfied workers select union membership when the occupation is covered by union bargaining. This is consistent with the
fact that the incentives to unionize may be larger when there is coverage, amplifying
the sorting effect in covered occupations.
The correlations between unobservables in the membership and satisfaction equations for non-covered employees indicate positive selection of intrinsically satisfied
individuals into membership. That is, when there is no union recognized for bargaining, it is those who are intrinsically more satisfied who are found among union
members. This effect may be interpreted in terms of employees underlying motivation towards their job. For example, in the absence of pecuniary rewards through
bargaining coverage, the returns to membership may come in the form of things like
greater information flows from management, which are prized more highly by those
who are particularly attached to their jobs. The positive coefficient could also be
capturing some underlying features of specific jobs where membership is an important part of employees occupational identification, or where particular job attributes
make it advantageous for employees to join the union and pool professional risks.
In Britain, these professions include teachers and medical professionals whose pay
is set by government Pay Review Bodies rather than through collective bargaining.

14
In Britain, as in the United States, organizing campaigns tend to take place at the workplace level. Unlike
the United States, where organizing campaigns usually entail secret ballots, in Britain organizing campaigns
are usually led by rank-and-file union members at the workplace. This is one reason why non-members are
more likely to report that they have been asked to join a union in workplaces where there are union members
(Kersley et al., 2006).

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

Trade Unions and Job Satisfaction

375

The evidence provided in Table A1 about the characteristics of non-covered members


supports these interpretations.
Estimation of the model with endogenous coverage provides insights into the
correlations of unobservables affecting membership and coverage, on one hand, and
satisfaction and coverage on the other. The correlation of the unobserved determinants
of union membership and union coverage is relatively large and precisely estimated.
The correlations between the error terms in the propensities to be satisfied and to
be employed in covered occupations are positive, and in some cases namely noncovered employees statistically significant at conventional confidence levels. The
positive signs indicate that unobservable determinants of coverage are associated with
high levels of job satisfaction. As mentioned in section IV, a possible interpretation
for this finding is that covered jobs are rationed and prospective employers are able
to select from the queue of job applicants selecting individuals according to some
traits that are likely to be correlated with productivity, like job satisfaction.15

VI.

Conclusion

In this article, we contribute to the literature on union membership and job satisfaction
showing that the effect of union membership on job satisfaction depends upon whether
members are covered by union bargaining or not. We used linked employeremployee
data to investigate the job satisfaction effect of unionization in Britain while also
accounting for the process that assigns workers to covered and non-covered occupations. For covered members, our results support explanations based on the sorting
argument, that is, members are inherently more dissatisfied with their jobs than their
non-member counterparts. This is what one could expect given the British context in
which covered workers do not need to be union members to benefit from union
bargaining so that individual membership in covered environments is driven by
inherent dissatisfaction with the job. In principle, membership could cause lower
job satisfaction in covered occupations because the returns to voice are higher when
there is collective bargaining, but our results rule out this possibility.
The results for employees in non-covered occupations are very different. We
find that membership reduces satisfaction. The result is consistent with a voice effect
wherein union members voice their job dissatisfaction with a view to achieving union
goals. Non-covered members may voice their dissatisfaction to encourage organization among co-workers in the hope of improving their bargaining power, or because
membership increases their awareness about unsatisfactory job aspects and there are
15
Note that this interpretation applies also to the non-covered sector, that is, also for those cases where
the correlation coefficient refers to unobservables in satisfaction and coverage propensities. To the extent that
workers expectations about the benefits of coverage do not get frustrated once in the job, the queuing interpretation also implies a positive effect of coverage on satisfaction. We did not find support for such an implication.
We used parameter estimates from the model with endogenous coverage to compute the covered/noncovered satisfaction differential. Evaluating the differential at the sample average of regressors, we estimated
the satisfaction differential to be negligible and not statistically significant for all the three satisfaction facets.

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

376

Bulletin

no union representatives to voice their discontent. The evidence is also consistent


with a causal effect of membership on preferences for collective bargaining.
Finally, we show that the individual propensity to be satisfied with the job is larger
in covered than in non-covered occupations. Where covered jobs are perceived by
employees as better jobs, and (as shown by some studies reviewed in this article)
workers demand for union representation exceeds the supply, there will be queues
for such jobs. Unionized employers will be able to pick the best workers from this
queue. To the extent that job satisfaction is positively associated with traits such as
motivation and ability that are at least partly observable by employers, we should
indeed expect to find the more satisfied among covered workers. Our results are
therefore consistent with an open-shop environment such as Britain, where workers
are free to choose whether to unionize or not, jobs covered by collective agreements
are rationed and workers queue for union jobs.
Final Manuscript Received: November 2009

References
Abowd, J. and Farber, H. (1983). Job queues and union status of workers, Industrial and Labor
Relations Review, Vol. 35, pp. 354367.
Airey, C., Hales, J., Hamilton, R., Korovessis, C., McKernan, A. and Purdon, S. (1999). The Workplace
Employee Relations Survey (WERS) 19978: Technical Report, National Centre for Social Research,
London.
Bach, S. and Givan, R. K. (2004). Public service unionism in a restructured public sector: challenges
and prospects, in Kelly J. and Willman P. (eds.) Union Organisation and Activity, Routledge, London,
pp. 89109.
Bender, K. and Sloane, P. J. (1998). Job satisfaction, trade unions, and exit-voice revisited, Industrial
and Labor Relations Review, Vol. 51, pp. 222240.
Blanchflower, D. and Bryson, A. (2004). Union relative wage effects in the United States and the United
Kingdom, Proceedings of the 56th Annual Meeting of the Industrial Relations Research Association,
Champaign, IL, pp. 133140.
Booth, A. L. (1995). The Economics of the Trade Union, Cambridge University Press, Cambridge.
Borjas, G. J. (1979). Job satisfaction, wages and unions, The Journal of Human Resources, Vol. XIV,
pp. 2140.
Bryson, A. and Freeman, R. (2006). Worker Needs and Voice in the US and the UK, NBER Working
Paper No. 12310, Cambridge, MA.
Bryson, A., Cappellari, L. and Lucifora, C. (2004). Does union membership really reduce job satisfaction? British Journal of Industrial Relations, Vol. 42, pp. 439459.
Clark, A. and Oswald, A. (1996). Satisfaction and comparison income, Journal of Public Economics,
Vol. 61, pp. 359381.
Davis-Blake, A. and Pfeffer, J. (1990). Unions and job satisfaction: an alternative view, Work and
Occupations, Vol. 17, 259283.
Farber, H. S. (2001). Notes on the Economics of Labor Unions, Industrial Relations Section Working
Paper 452, Princeton University.
Forth, J., Bewley, H. and Bryson, A. (2006). Small and Medium-Sized Enterprises: Findings from the
2004 Workplace Employment Relations Survey, Department of Trade and Industry, London.
Freeman, R. B. and Medoff, J. L. (1984). What Do Unions Do? Basic Books, New York.
Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

Trade Unions and Job Satisfaction

377

Freeman, R. B. and Rogers, J. (1999). What Workers Want, ILR Press, Cornell.
Frenkel, S. and Kuruvilla, S. (1999). Union-member relations and satisfaction with unions in South
Korea, British Journal of Industrial Relations, Vol. 37, pp. 101105.
Gall, G. (2004). Trade union recognition in Britain, 19952002: turning a corner? Industrial Relations
Journal, Vol. 35, pp. 249270.
Gordon, M. E. and Denisi,A. S. (1995). Are-examination of the relationship between union membership
and job satisfaction, Industrial and Labor Relations Review, Vol. 48, pp. 222236.
Guest, D. and Conway, N. (2004). Exploring the paradox of unionised worker dissatisfaction, Industrial
Relations Journal, Vol. 35, pp. 102121.
Hammer, T. H. and Avgar, A. (2005). The impact of unions on job satisfaction, organizational commitment, and turnover, Journal of Labor Research, Vol. 26, pp. 241266.
Hersch, J. and Stone, J. A. (1990). Is union job satisfaction real? Journal of Human Resources, Vol.
XXV, pp. 736751.
Heywood, J. S., Siebert, W. S. and Wei, X. (2002). Worker sorting and job satisfaction: the case of
union and government jobs, Industrial and Labor Relations Review, Vol. 55, pp. 595609.
Kersley, B., Alpin, C., Forth, J., Bryson, A., Bewley, H., Dix, G. and Oxenbridge, S. (2006). Inside the
Workplace: Findings from the 2004 Workplace Employment Relations Survey, Routledge, London.
Machin, S. (2000). Union decline in Britain, British Journal of Industrial Relations, Vol. 38,
pp. 631645.
Meng, R. (1990). The relationship between unions and job satisfaction, Applied Economics, Vol. 22,
pp. 16351648.
Miller, P. W. (1990). Trade unions and job satisfaction, Australian Economic Papers, Vol. 29,
pp. 226248.
Millward, N., Bryson, A. and Forth, J. (2000). All Change at Work? Routledge, London.
Renaud, S. (2002). Rethinking the union membership/job satisfaction relationship: some empirical
evidence in Canada, International Journal of Manpower, Vol. 23, pp. 137150.
Schwochau, S. (1987). Union effects on job attitudes, Industrial and Labor Relations Review, Vol. 40,
pp. 209224.
Stewart, M. B. (1987). Collective bargaining arrangements, closed shops and relative pay, Economic
Journal, Vol. 97, pp. 140155.

Appendix
TABLE A1
Descriptive statistics of the WERS98 sample
Whole Covered Covered
Non-covered Non-covered
sample members non-members members
non-members
Number of observations
23,601
Satisfaction with pay
Very dissatisfied
12.12
Dissatisfied
28.42
Neither satisfied nor dissatisfied 23.74
Satisfied
32.11
Very satisfied
3.61
Satisfaction with sense of achievement
Very dissatisfied
4.65
Dissatisfied
10.52

5,721

3,487

3,854

10,539

13.54
29.99
22.24
31
3.24

10.22
27.45
25.2
34.1
3.03

13.1
29.42
23.04
30.19
4.25

11.63
27.56
24.31
32.72
3.78

6.64
13.18

4
9.76

5.16
11.54

3.64
9.04
(continued overleaf )

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

378

Bulletin

TABLE A1
(continued)
Whole Covered Covered
Non-covered Non-covered
sample members non-members members
non-members
Neither satisfied nor dissatisfied
21.64
Satisfied
48.84
Very satisfied
14.34
Satisfaction with respect from supervisors
Very dissatisfied
8.59
Dissatisfied
12.86
Neither satisfied nor dissatisfied
20.55
Satisfied
44.16
Very satisfied
13.83
Satisfaction with influence over work
Very dissatisfied
3.23
Dissatisfied
12.12
Neither satisfied nor dissatisfied
25.97
Satisfied
47.49
Very satisfied
11.19
Female
47.99
Aged less than 20
4.74
Aged 2024
7.39
Aged 2529
12.66
Aged 3039
28.24
Aged 4049
24.78
Aged 5059
18.01
Aged 60 or more
4.20
No educational qualification
24.80
Has at most O levels
26.86
Has at least A- levels
36.34
Disabled
5.95
Non-white
3.54
Has children
53.24
Married
69.93
Manager
9.18
Professional
11.25
Associate professional and technical
8.13
Clerical and secretarial
17.80
Craft and skilled service
10.77
Personal and protective service
11.20
Sales
9.24
Operative and assembly
12.96
Other occupations
9.48
Job equally done by men and women 29.77
Availability of family friendly policies 51.63
Can take day off if needed
97.13
Overtime always paid
45.82
Open-ended contract
92.87

24
45
11

20.06
51.53
14.64

19.28
48.29
15.73

21.75
50.03
15.53

11.9
15.94
21.65
40.17
10.34

6.04
11.08
20.03
46.45
16.41

10.7
14.41
22.19
40.34
12.36

6.94
11.3
19.6
46.81
15.36

4.91
16.03
26.13
44.45
8.48
34.97
1.17
2.93
10.17
30.96
30.58
20.77
3.42
30.59
26.67
29.02
8.40
3.39
59.90
77.02
5.50
8.50
7.77
17.70
16.03
8.99
4.02
22.08
9.41
26.64
52.71
97.40
54.76
96.88

1.95
9.98
27.52
49.94
10.6
60.57
5.03
7.66
13.40
26.55
24.50
18.30
4.56
22.31
30.37
36.90
4.50
3.64
55.08
69.04
6.93
9.79
6.02
31.21
7.22
12.15
8.88
7.15
10.65
30.70
59.59
98.24
39.59
86.00

4.58
13.81
27.24
44.53
9.83
49.49
0.85
3.66
11.35
30.77
30.60
19.33
3.43
23.28
21.19
45.67
5.98
3.86
60.24
76.91
7.14
23.49
15.10
7.90
10.63
12.78
4.63
12.59
5.75
31.53
54.15
94.07
42.08
94.11

1.95
9.98
27.52
49.94
10.6
50.50
7.78
10.84
14.15
26.50
19.94
16.05
4.74
23.01
27.75
36.96
5.10
3.48
46.92
64.23
12.44
9.14
6.69
17.04
9.14
11.55
13.58
10.06
10.37
30.55
47.85
97.65
44.24
92.45

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

Trade Unions and Job Satisfaction

379

TABLE A1
(continued)
Whole Covered Covered
Non-covered Non-covered
sample members non-members members
non-members
At least 10 days of training
in the past year
Paid less than 50 per week
Paid 5180 per week
Paid 81140 per week
Paid 141180 per week
Paid 181220 per week
Paid 221260 per week
Paid 261310 per week
Paid 311360 per week
Paid 361430 per week
Paid 431540 per week
Paid 541680 per week
Paid more than 681 per week
Total hours worked on average week
1024 employees
2549 employees
5099 employees
100199 employees
200499 employees
500 or more employees
East Anglia
East Midlands
London
North
North West
Scotland
Rest of the South East
South West
Wales
West Midlands
Yorkshire and Humberside
Manufacturing
Electricity, gas, water
Construction
Wholesale and retail
Hotels and restaurants
Transports and communication
Financial services
Other business and services
Public administration
Education
Health
Other community services

8.79

9.92

7.87

12.30

7.35

6.90
6.99
12.28
8.90
11.42
10.56
10.03
8.35
9.93
7.53
3.77
3.36
36.01
4.58
8.66
9.83
7.00
9.99
10.40
18.54
8.45
4.35
9.86
8.35
14.14
14.95
14.61
20.32
23.65
12.33
24.16
0.64
3.09
14.83
4.09
6.00
4.16
8.38
8.77
10.25
12.42
3.20

1.77
2.89
8.02
6.20
13.26
13.17
13.71
12.72
13.94
8.43
3.64
2.25
38.43
3.32
8.19
7.81
12.67
11.93
14.38
10.64
9.16
4.35
8.96
8.58
6.42
9.27
14.08
25.11
38.70
6.41
32.18
1.53
2.37
6.29
0.95
12.32
4.42
2.15
17.78
7.18
10.23
2.59

10.17
10.67
16.86
8.93
11.89
10.30
8.06
6.57
6.90
5.32
2.30
2.03
31.74
6.27
11.17
8.69
6.82
7.89
12.04
19.13
8.23
3.00
10.11
6.66
11.62
12.10
15.97
20.68
31.35
8.27
15.92
0.92
2.32
13.22
2.40
4.55
7.28
3.20
14.56
18.07
14.14
3.43

1.81
3.36
8.09
8.57
10.08
10.18
11.32
9.55
17.00
12.12
5.10
2.81
38.29
3.78
7.89
10.62
6.91
12.30
10.00
14.89
6.38
6.58
13.33
7.31
10.89
14.75
14.65
22.58
28.90
8.24
21.29
0.63
3.08
6.41
1.40
4.91
3.55
2.77
11.53
19.43
23.21
1.79

10.25
9.21
14.48
10.41
10.74
9.39
8.27
6.21
6.44
6.23
3.85
4.52
35.29
4.99
8.40
10.96
4.12
8.86
7.96
23.67
8.82
4.02
9.12
9.09
19.98
18.83
14.47
16.98
11.75
17.98
23.38
0.10
3.70
22.52
7.11
3.49
3.28
15.03
1.42
6.52
9.54
3.91
(continued overleaf )

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

380

Bulletin

TABLE A1
(continued)
Whole Covered Covered
Non-covered Non-covered
sample members non-members members
non-members
Share female employees
Share part-time employees
Share high-skilled employees
Share medium-skilled employees
Share low-skilled employees
TTWA unemployment rate > 5%
Single establishment
Publicly owned establishment
Establishment older than 20 years

48.31
25.35
8.36
31.75
59.89
50.28
23.80
30.55
51.14

37.62
17.07
6.85
31.89
61.27
57.87
10.23
45.89
65.40

55.63
29.71
7.76
31.27
60.97
44.57
17.08
52.70
57.19

52.76
26.72
6.60
43.06
50.34
57.42
16.46
52.79
61.74

50.24
27.91
9.91
28.14
61.95
45.72
35.30
8.63
38.43

Note: TTWA, travel to work area.

TABLE A2
Regression coefficients for the variables excluded from the satisfaction equations in the model
with endogenous membership and coverage and tests of significance of those variables
in the coverage, membership and satisfaction equations
Satisfaction
with influence
over work
Coefficient SE

Satisfaction
with respect from
from supervisors
Coefficient SE

Satisfaction
with sense of
achievement
Coefficient SE

Regression coefficients from union coverage equation


Single establishment
0.437
0.092
Establishment age 20
0.146
0.076

0.437
0.144

0.091
0.076

0.435
0.148

0.092
0.076

Regression coefficients from union membership equation


Single establishment
0.527
0.066
Establishment age 20
0.222
0.055

0.524
0.226

0.066
0.055

0.523
0.226

0.066
0.055

2 (d.f.)
Significance of instruments in equation for
Union coverage
26.11 (2)
Union membership
79.28 (2)
Union coverage and
83.36 (4)
union membership
Satisfaction if union coverage
0.19 (2)
Satisfaction if no union coverage 10.7 (2)
Satisfaction if union coverage and 11.5 (4)
no union coverage

P-value 2 (d.f.)

P-value 2 (d.f.)

P-value

0.000
0.000
0.000

26.77 (2)
78.82 (2)
83.51 (4)

0.000
0.000
0.000

26.22 (2)
77.97 (2)
82.27 (4)

0.000
0.000
0.000

0.911
0.005
0.022

0.58 (2)
1.66 (2)
2.14 (4)

0.749
0.437
0.710

1.77 (2)
2.13 (2)
4.11 (4)

0.412
0.345
0.392

Notes: Regression coefficients and robust standard errors are derived from the simultaneous equations
model for job satisfaction, union membership and union coverage. Regression models are estimated using a
GHK simulator with 50 Halton draws and survey weights.

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2010

Copyright of Oxford Bulletin of Economics & Statistics is the property of Wiley-Blackwell and its content may
not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written
permission. However, users may print, download, or email articles for individual use.

Anda mungkin juga menyukai