Content uploaded by Vassilis Monastiriotis
Author content
All content in this area was uploaded by Vassilis Monastiriotis
Content may be subject to copyright.
Union Retreat and Regional Economic
Performance: the UK in the 1990s
Vassilis Monastiriotis
Abstract
This paper uses a panel of regional data to investigate the impact that the well-
documented decline in trade unionism in the UK had on the economic performance
of its regions. The analysis employed here departs from the traditional firm-level
and cross-sectional analyses and looks at the economy-wide effects of unionism.
Our findings provide evidence in line with theory that predicts unions to increase
wages and reduce labour demand, leading to higher unemployment, but they also
indicate that unionism is positively related to productivity and incomes, although in
all cases the effects are non-linear. We conclude that unionism is not necessarily a
burden for the economy, so long that the beneficial wage/productivity effects
overbalance the negative effects on employment.
JEL classification: J51, R11, R23
Keywords: Trade unions, regional economic performance, panel data
Department of Geography and Environment, London School of Economics, Houghton Street, London
WC2A 2AE, UK Tel.: ++44 (0) 20 79556776; Fax: ++44 (0) 20 79557412; E-mail:
v.monastiriotis@lse.ac.uk. I wish to thank Melanie Dorson and Vivian Copeland (Reward Group
Ltd), Sean Townsend (BLPES), Ian Murray (TUC), Travor Mendez (Certification Office) and
Stephen Woodland (DTI) for their assistance with data collection. I have benefited from comments
by Gilles Duranton, Ian Gordon, Henry Overman, Jacques Poot and conference participants. All
errors and omissions, of course, remain with the author. Part of this research has been conducted
under financial support by the ESRC (Ref.: T026271203).
1
1. Introduction
Trade union membership in the UK has followed a remarkably stable
declining path over the last two decades, a trend that was only recently halted
(Metcalf, 2001). This decline was quite robust across different individual and
workplace characteristics, including education, gender, industry and location (Bland,
1999). This so-called union retreat (Martin et al., 1996) has largely been welcomed
by both policy-makers and academia. As unions are expected to impact adversely on
labour effort and productivity, increase labour costs, and generate wage-inflation
and unemployment, the decline in union membership has been taken to create a
potential for enhanced flexibility and improved economic performance.
However, the empirics of the economic impact of unionism are not fully
consistent with such a view. The voluminous empirical literature on the economic
and labour market effects of trade unionism provides robust evidence for the wage-
increasing role of unions, but reaches less clear conclusions regarding their impact
on productivity, employment and investment.1 Some studies have found unionism to
have a positive impact on investment (Machin and Wadhwani, 1989) and
productivity (Brown and Medoff, 1978; Nickell et al., 1989, 1991). Other studies,
though, have reported a significant negative impact on output (DeFina, 1983),
output growth (Nickell and Layard, 1998), employment growth (Blanchflower and
Oswald, 1988), productivity (Clark, 1984; Lovell et al., 1988) and profitability
1 A number of firm- and industry-level studies have shown that unions are associated with higher
wages, especially for the lower tail of the wage distribution (Blanchflower, 1986; Freeman, 1980,
1991; Card, 1991; Blackaby and Murphy, 1991; Gosling and Manning, 1993). Also very robust are
the findings about the effect of unions on wage inequalities. However, as the focus of this paper is on
regional macroeconomic performance, we do not discuss this issue further.
2
(Freeman, 1983; Machin, 1991). The results obtained for the employment effects of
unionism are equally diverse (Minford, 1982; Sinclair, 1987; Nickell and
Wadhwani, 1988; Blanchflower and Millward, 1988), while more aggregate studies
have also obtained inconclusive results (OECD, 1997; Nickell and Layard, 1998).
A plausible explanation for the existence of conflicting findings in the
empirical literature can be found in the selection of the unit of analysis. More often
than not cross-sectional firm-level studies find non-unionised firms to grow faster
than their unionised counterparts (e.g., Blanchflower and Millward, 1988). In
contrast, time-series and panel-data studies frequently observe insignificant or
positive employment effects for unions (e.g., Nickel et al., 1989). Naturally, an
explanation that could reconcile these findings would probably acknowledge that it
is possible for a union to suppress employment in unionised firms but for unionism
to have differentiated effects at the economy-wide level and over a period of time,
plausibly by increasing economic dynamism.2 Related to that, is the theoretical
assertion that although in a perfect competition framework unions are a burden to
the economy, once monopoly power and internal labour markets are allowed for,
unions can be shown to improve economic performance thus constituting a second-
best response to an imperfect world (Mayhew, 1983; Booth, 1991; Green et al.,
1996; Booth and Chatterji, 1998; Amable and Gatti, 2001; Bassanini and Ernst,
2002). Thus, firm-level studies focusing on competitive sectors of the economy can
plausibly observe different mechanics than those identified at the larger scale where
market imperfections are more evident.
2 This line of argument is similar to that regarding the employment effects of capital formation. At the
firm level, physical investment is often found to induce some capital-labour substitution. However, at
the economy-wide level investment is the main determinant of output growth and thus also
employment growth.
3
In fact, the idea that unionism can have both positive and adverse effects in
the economy has long been recognised in the relevant literature. In their influential
work, Freeman and Medoff (1984) identified two faces for unionism, one related to
wage increases and labour input rigidities, and another related to higher
organisational efficiency, lower labour turnover (see also Miller and Mulvey, 1993)
and more investment in physical and human capital.
This paper investigates empirically the impact that declining union
membership in the UK and its regions had on regional economic performance in the
1990s.3 We adopt a macroeconomic perspective in order to examine the economy-
wide union effects. The use of the region as the unit of analysis allows us to
combine this macroeconomic perspective with the examination of a relatively
homogenous set of (regional) economies for a short -and thus relatively
homogenous- period (the 1990s). The question we ask is: to what extent has trade
union density had an impact on regional economic performance during the period of
sharp decline in unionisation rates? Further, we investigate whether this impact was
regionally uneven and whether it helped reduce regional disparities. For the first
question we examine the possibility of both linear and non-linear (concentration)
effects.4 For the second, the focus is on the possible presence of spatial
autocorrelation, which would indicate that a mechanism of spatial heterogeneity or
of spatial dependence is in operation (Anselin, 1988). The next section reviews the
3 The empirical analysis uses a panel of regional UK data for the period 1989-1998. Data availability
was the factor determining the time-span of the study, as regional data on union densities are only
available for these years, with the exception of the estimates for 1980 and 1984 based on the
Workplace Industrial Relations Surveys which are not, however, regionally representative. Three
other potential sources of information (Certification Office, TUC and DTI) were also unable to help,
despite efforts from some of their staff, for which I am thankful.
4 We also examined the role of changes in unionism for regional economic performance, but this was
always insignificant. Thus, our empirical analysis focuses solely on the levels of union density.
4
regional evolutions that characterised the UK during this period. Section 3 makes
some brief theoretical considerations, while section 4 presents the empirical
investigation and discusses the obtained results. The last section concludes.
2. The UK regions in the 1990s
The 1990s saw a substantial decline in cross regional differences in
unemployment rates, which can only partly be explained by the national trend of
declining unemployment after 1993. Despite the well-documented regional
persistence in unemployment in the UK (Martin, 1997), regional dispersions nearly
halved between 1989 and 1998. In the same period, regional differences in
employment opportunities (as measured by the employment-to-population ratio)
declined by around 25%. Employment growth rates were more or less the same at
the start and the end of the period of study, but between 1991 and 1994 all British
regions experienced net job losses as a result of the economic recession of the early
1990s. On the other hand, real wages increased throughout the period in all regions
and so did labour productivity. However, in contrast with the evolution of regional
unemployment differentials, regional wages did not exhibit strong signs of
convergence, while regional dispersions in productivity widened. In nominal terms,
regional wage differentials increased substantially (Cabinet Office, 1999).
Nevertheless, in real terms (i.e., when deflated by a regional price index5), regional
real wages in the British regions have been rather stable, with the exception of the
5 We use a two-dimensional regional price index, derived from survey data collected by the Reward
Group Ltd. The same index has been used in Duranton and Monastiriotis (2002), while the same data
source has been used previously by Borooah et al. (1996).
5
period 1989-1993.6 However, when looking at UK data, this pattern is masked by
the fact that wages in N. Ireland have been catching up with the rest of the country
throughout the period (Figure 1).
Figure 1: The evolution of regional wage inequalities in Britain and the UK
Regional dispersion in real wages, GB and UK
5
10
15
20
25
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998
Standard Deviation - UK Standard Deviation - GB
As with wages, real regional household incomes (per head) and gross per
capita product increased substantially, by an average annual rate of 3.9% and 3.1%,
respectively. Nevertheless, regional average annual growth rates of gross per capita
products varied from 1.8% in Scotland to 4.7% in the South East (excluding Greater
London). The corresponding figures for regional growth in household incomes were
1.7% (Scotland) and 6.2% (Greater London). Consequently, regional inequalities
increased remarkably throughout the 1990s. The standard deviation for the two
measures of regional incomes was increasing by an average annual rate of more than
5% in the 10 years of our study (Figure 2).
6 There is a strong reduction in regional wage dispersions in 1989-1991 -which is mainly due to the
recession that hit particularly the service sector in the southern regions (Martin, 1993)- and then a
sharp increase in 1991-1993.
6
Figure 2: The evolution of regional income inequalities in the UK
Regional dispersions in incomes, UK
400
450
500
550
600
650
700
750
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998
18000
20000
22000
24000
26000
28000
30000
32000
Real GDP p.c. (sdev) Real Household Incomes (sdev)
Trade union densities kept decreasing throughout the period of our study in
all regions. This decrease was even more dramatic than in the 1980s, as national
trade union density decreased by more than 10 percentage points, representing a
remarkable absolute decline in union membership. Nevertheless, the regional
pictures of union retreat are not identical (for a more detailed analysis on that, see
Martin et al., 1996). Regional union density rates of decline vary from a “low” 1.8%
p.a. in Wales and East Anglia to a “high” 3% p.a. in the Midlands. Moreover, the
regional patterns of union decline show some -unexpected- randomness. Union
retreat is not faster in high- (or low-) union density regions, is not related to the
regional level of economic development and is not specific to declining (or growing)
regions.7 Table 1 presents correlation coefficients and significance levels for the
estimated correlations between regional union density decline and (i) the 1989
7 However, two of the poorest regions in the UK (N. Ireland and Wales) had the higher union
densities and the slowest rates of union decline.
7
regional union densities, (ii) the 1989 level of real regional GDP per capita, and (iii)
the 1989-1998 average annual growth rates of real regional GDP per capita.
Table 1: Union density decline and regional characteristics (UK)
Correlation with rate of decline in union density
Regional Characteristics Correlation Significance
Union density in 1989 -0.214 0.505
GDP per capita in 1989 -0.310 0.317
GDP growth 1989-1998 0.170 0.597
To summarise, the regional and national economic performance of the UK in
the 1990s (effectively, after the recession at the beginning of this period) was rather
satisfactory. Unemployment rates declined everywhere, while wages, incomes and
productivity increased fast. However, this good economic performance was not
evenly distributed across the regions. Although regional differences in employment
and unemployment rates declined substantially, regional dispersions in incomes and
productivity widened. These movements of convergence in measures of
employment and divergence in measures of incomes reveal a picture of regional
heterogeneity in economic performance. As the regional evolutions of union decline
were similarly heterogeneous, it is interesting to ask to what extent the latter can
contribute to the explanation of the identified regional heterogeneity in economic
performance. Our empirical analysis ofthis issue follows after a few theoretical
considerations, presented in the next section.
3. The impact of unionism on economic outcomes
In our analysis we view unionism as a socio-economic phenomenon which,
as such, can influence any economic and socio-economic aggregate. Specifically,
8
unionism can affect or reflect attitudes towards work and employment participation.
It can affect labour productivity and effort. Also, it can impact on labour demand
and investment decisions from the side of the firm. Thus, unionism can be expected
to be statistically associated with a large range of economic variables, like
productivity, wages, household incomes, employment participation, employment
and output growth, and investment.
The literature on the economics of unions does not provide a general
equilibrium model that could capture the multiplicity of the effects that unionism
can have on the economy. Rather, a number of micro-economic models exist that
describe different aspects of these effects, mainly those related to productivity,
wages and employment. Moreover, aggregate studies examining the union effects on
these and other economic indicators (i.e., growth rates, investment, education levels
and unemployment) rarely employ structural models, usually because of their cross-
country focus and problems related to sample size (see for example Koedijk and
Kremers, 1996; OECD, 1997; Nickell and Layard, 1998).
The most popular attempt to formally model the relationship between
unionism and output or productivity originates from the work of Brown and Medoff
(1978). In their production-function approach, unionism enters as a control variable
for the productivity differences among unionised and non-unionised workers (see
Monastiriotis, 2002, for a discussion of alternative production-function models of
unionism). Although such a model can be used to derive a union-augmented labour
demand, it is rather standard in the literature to investigate the employment effects
of unionism by assuming that labour demand depends on workers’ effort, which is
in turn a function of union membership (see Machin and Wadhwani, 1991). Finally,
9
the union effects on wages are typically modelled by including union status as an
exogenous variable in a Mincer (1974) type wage equation.
In this study we look at a large number of variables, in order to obtain as
wide a picture as possibleof the economic impact of unionism. We focus on
productivity and unemployment, which are the two major areas where unionism is
expected to have an impact, but we also examine the empirical relationship between
unionism and wages, household incomes, output, employment participation,
investment, and employment and output growth. By looking at such a large number
of variables it is possible to make some inferences regarding the socio-economic
mechanisms that underlie identified effects. For example, if unions are found to
increase wages, one has to look further and investigate the impact that unions have
on labour productivity, as well as on employment. If productivity and employment
decline with unionism, the wage-increasing effect must be due to the bargaining
power of unions. On the other hand, if employment and productivity expand
together with unionism, then one would have to conclude that the productivity
increases are more likely to be the direct union effect and that the wage increase is
simply a (positive) side-effect.8
In a study similar to ours (which is to our knowledge the only aggregate
panel-data study conducted at a regional level to date), Freeman (1988) examines
the impact of unionism on incomes, output and employment (for US states), in order
to estimate its wider economic effects. His analysis does not employ a structural
model of the economy. Rather, simple regressions are used to associate unionism
8 An alternative explanation would be that unions generate some type of efficiency wages from which
productivity increases follow. However, this would conflict with a possible positive employment
effect.
10
with the three economic indicators, only controlling for fixed regional effects and
the level of wages. Freeman (1988) suggests that the latter proxy for general
economic conditions potentially corrects any omitted variables problems.9
Our analysis follows that of Freeman (1988) and examines the impact that
the declining unionisation rates in the UK had on its regional economic
performance. We avoid specifying a structural model for the economy. Instead, we
select two empirical models for productivity and unemployment, controlling for
fixed effects as dictated by the appropriate specification tests and the fact that we
use a panel of cross-regional and time-series data. As the early 1990s were
characterised by a small recession, while the late 1990s was a period of relative
expansion, we also include time dummies (time-specific effects) where appropriate.
To produce some results directly comparable to Freeman (1988), but also to check
the robustness of our results, in section 4.3 we regress each economic variable on
unionism and real wages, without adding any further controls.
Before closing this section, some econometric specification issues must also
be considered. Often in panel data analyses, where one needs to control for spatial
and time fixed effects, a two-way demeaning is applied to the estimating models.10
9 He also notes that the inclusion of real wages can additionally give information about “possible
causal relations, as unions can be expected to affect unemployment, per capita income and
employment largely through earnings” (p.710).
10 The procedure is as follows. Assume that unemployment (U) is made a function of unionism (T),
wages (W), a vector of country-specific (constant-across-space) variables (X; e.g., national
unemployment and inflation rates, or national measures of international competitiveness) and a vector
of region-specific (constant-over-time) variables (Z; e.g., distance from the Capital, average
temperatures, road and rail networks). Assuming a log-linear specification, then, yields
ititititit eWdZcXbTaU
ε
+++++= (1)
Aggregating over time (Ui), then across space (Ut) and finally across the whole sample (U: global
mean) and then calculating Uit-Ui-Ut+U yields )()()()(
εεεε
+−−++−−++−−=+−− tiittiittiittiit WWWWeTTTTbUUUU (2)
As long as one is only interested in the value of b in the original relationship, estimating (2) instead
of (1) saves a significant number of degrees of freedom and also technically eliminates other possible
econometric problems.
11
This transformation is crucial, especially when one of the two dimensions of the
panel is particularly small, as it economises significantly on degrees of freedom and
helps avoid possible specification problems (Baltagi, 1995). It is moreover
necessary when constant-over-time or constant-across-space variables are included
in the model, as such variables will be perfectly collinear with the fixed effects.
In our empirical analysis none of these considerations raise any problems.
Our panel is close-to-square (N=12, T=10) and our estimating models do not
include national or fixed-over-time regressors. More importantly, rather than
imposing a two-way error component specification (spatial and temporal fixed
effects) as would be the case in using the de-meaning transformation, we wanted to
test for the presence of (fixed or random) temporal and regional effects. To do this,
consistently with standard panel data analysis techniques, we employed the Breusch-
Pagan and Hausman specification tests (testing for the existence of random region-
specific effects versus no effects at all, and versus fixed regional effects,
respectively). To test for the existence of time-specific effects as well as for the
simultaneous existence of regional and time-specific effects, we conducted a
number of F-tests for omitted variables.11 Following this exploratory analysis, our
final estimating models were specified as two-way error component models, that we
estimated with dummy variables least squares (DVLS). This is equivalent to the
two-way de-meaned specification, but allows direct estimation of the fixed effects
and assessment of their significance.
11 We do not report the process of model specification here, as this would only complicate the
presentation of the results. All the relevant tests, however, are available by the author upon request.
12
4. Empirical findings
4.1. Unionism and productivity
For productivity, the original model we specify uses a number of variables
that control for the business cycle, the gender, sectoral and occupational
composition of employment, the level of human capital, and the dynamism of each
regional economy. In particular, the original model includes the regional
unemployment rate, the female labour force participation rate, the share of
manufacturing and business services (banking and finance) to total regional
employment, the share of professionals and managers in total employment, the net-
immigration rate and a measure of within-regions wage inequalities.12
Unemployment controls for the business cycle but also for the possibility
that, in an effort/threat rationale, productivity should increase when the probability
of falling into unemployment increases. The employment composition variables (for
gender, sectors and occupations) should control for possible differences in
productivity across the different employment categories. The net immigration rate
should capture some of the effect on productivity that could be due to the economic
dynamism of each region. Finally, the level of wage inequalities could indicate the
existence of premiums in the wage structure in each regional economy. The results
from this regression are shown in the first column of Table 2. Despite the non-
theoretical structure of the model, the regression performs reasonably well. Non-
normality of the residuals and misspecification (RESET test) are both rejected at the
1% level, while heteroskedasticity is clearly not a problem. All coefficients have the
12 We also experimented with the average years of schooling of full-time employees, but this variable
was always highly insignificant and was thus excluded from the analysis. All data used have been
obtained from the Regional Tends publications (ONS).
13
expected signs, with unemployment and immigration being strongly and positively
associated with labour productivity. Manufacturing and, to a lesser extent, banking
are also significant in explaining regional productivity levels, while female labour
force participation is found to have a negative effect, albeit marginally insignificant.
Wage inequality and professional employment are not significant and for the latter
the obtained coefficient is counter-intuitive. However, many of the effects that could
be attributed to these two variables could be captured by the regional and temporal
fixed effects, or by the other employment composition variables, so that their poor
performance is not necessarily a problem.
With this reasonable specification for productivity, the next step is to
introduce the unionisation variable into the model. As stated earlier, we want to test
the impact of unionism on productivity, but also investigate the possibility of non-
linearities in this relationship, which could be due either to a concentration or to a
spatial spill-over mechanism. The inclusion of the union density rate into the model
changes little the results obtained for the control variables and the overall
performance of the model. The estimated impact of unionism is positive, but
significant only at the 10% level. The positive relationship between unionisation and
labour productivity can be taken to support the approaches suggesting an efficiency-
enhancing role for unions (see in this respect Sengenberger and Campbell, 1994),
but it could also be due to a logistic effect where productivity increases are the
artefact of the wage-increasing effect of unions (see on that Reynolds, 1986).
14
Table 2: Unionism and productivity
Model (1) (2) (3) (4) (5) (6) (7)
Unemployment
rate 0.949
(12.87) 0.901
(11.71) 0.742
(9.37) 0.194
(2.39) 0.789
(9.58) 0.724
(8.91) 0.193
(2.37)
Immigration rate 68.99
(2.80) 76.30
(3.10) 94.42
(4.13) 39.05
(2.25) 71.59
(3.03) 89.99
(3.86) 40.69
(2.33)
Manufacturing 28.11
(2.98) 23.82
(2.48) 33.63
(3.72) 9.761
(1.41) 18.37
(1.97) 30.16
(3.12) 11.54
(1.60)
Banking-finance 16.27
(2.00) 15.27
(1.90) -3.128
(-0.37) -9.877
(-1.62) 9.727
(1.23) -2.465
(-0.29) -10.50
(-1.71)
Female LF rate -27.03
(-1.63) -21.25
(-1.28) -24.33
(-1.60) -13.57
(-1.24) -24.33
(-1.53) -25.02
(-1.64) -12.82
(-1.17)
Professionals -1.113
(-0.63) -1.040
(-0.59) -1.048
(-0.65) 1.294
(1.11) 0.014
(0.01) -0.656
(-0.40) 1.084
(0.91)
Wage inequality 1.964
(1.17) 2.521
(1.50) 2.493
(1.63) 0.750
(0.68) 2.967
(1.84) 2.663
(1.73) 0.589
(0.52)
Union density 8.899
(1.87) -54.44
(-3.60) -16.32
(-1.41) 18.754
(3.36) -41.42
(-2.08) -24.30
(-1.69)
Union density
squared 76.23
(4.37) 20.57
(1.49) 64.96
(3.13) 26.91
(1.75)
Spatial lag of
unionism 159.4
(3.06) 59.15
(1.00) -40.66
(-0.93)
Real wage 0.052
(9.36) 0.053
(9.29)
Regional effects* 14.09
0.000 12.90
0.000 12.77
0.000 18.37
0.000 14.94
0.000 12.60
0.000 18.07
0.000
Temporal effects* 54.72
0.000 43.97
0.000 39.99
0.000 6.100
0.000 15.91
0.000 11.93
0.000 5.68
0.000
R-squared 0.960 0.962 0.969 0.984 0.966 0.969 0.984
Normality 5.51
0.064 5.48
0.064 4.28
0.118 1.54
0.463 3.33
0.189 3.43
0.180 2.06
0.357
Heteroskedasticity 1.21
0.270 2.00
0.158 0.15
0.701 12.57
0.000 3.46
0.063 0.47
0.493 11.72
0.001
Ramsey 3.51
0.019 3.86
0.012 0.27
0.849 1.04
0.380 2.39
0.075 0.16
0.920 1.22
0.308
Notes: t-statistics in parentheses. Figures in Italics show significance levels. *: this is an F-test for the
joint significance of the fixed effects. The test for normality is a joint chi-square test for skewness and
kurtosis. The test for heteroskedasticity is the Cook-Weisberg chi-square test. Ramsey is the RESET
F-test for omitted variables. All regressions have been estimated with DVLS. The sample consists of
120 observations, across 12 regions and over 10 years (1989-1998).
To explore this possibility, we alternatively introduce a number of additional
measures of unionism in our basic model, as shown in columns 3-7 of Table 2. The
inclusion of a squared term for unionism shows that the productivity effects of
unionism are non-linear (convex).13 The performance of the regression improves
13 The combined union effect becomes positive just below the average unionisation rate and increases
thereafter.
15
further and this time both union variables are significant even at the 1%. We then
introduce real wages as an additional explanatory variable, to test the possibility that
the estimated union effect is due to the union effect on wages (column 4).
The inclusion of real wages into the model generates some heteroskedasticity
in the residuals and, as expected, renders some of the control variables insignificant.
Further, it reduces the value of the obtained coefficients for unemployment and
migration, confirming that both of these variables are directly linked to wages.14 The
most important impact, however, is the reduction in the significance of the union
coefficients, which are now statistically not different from zero. This effect is
confirmed also when the square of unions is removed from the model (not shown).
This result leads us to conclude that the productivity effect of unionism operates
through wages and is thus a logistic artefact. However, one cannot exclude the
possibility of some direct convex union effects, as the union variables are still
significant at the 20% level in the quadratic specification. In any case, such a direct
effect cannot be taken to be dominant in any economic sense.
The next step towards the investigation of the productivity effects of
unionism is to explore the presence of possible spatial effects. Columns 5 and 6
present a pair of spatially dynamic specifications, the first based on the linear model
and the second based on the quadratic specification. As it can be seen, the linear
specification produces significant spatial effects which are positive, i.e. in the same
direction as the “local” union effect. Although this seems to suggest that unions do
produce spatial spill-overs, the results obtained from the quadratic specification
indicate that this spatial spill-over disappears when one controls for the non-linear
14 This, however, does not generate significant collinearity problems in our estimated regressions.
16
“local” effect. Since the spatial effect disappears in the quadratic specification, one
is forced to conclude that its significance in the linear model is only due to a “spatial
dependence” effect, where unionism is determined at a wider spatial scale than that
of the administrative region. This finding, of course, is as expected and reinforces
our confidence on the validity of the results we obtain.
Further, again as expected, the significance of the union effects declines
when real wages are included again in the model. This time, however, the local
union variables are marginally significant (at 10%), thus indicating that there are
marginal (convex) union effects on productivity that cannot be attributed to the
wage-increasing role of unions. In this case, the efficiency-enhancing role of unions
cannot be rejected. To further examine this possibility, we ran a number of simple
regressions for real wages, including unionism and the regional and temporal fixed
effects as the short-list of explanatory variables. The results (see Table 4) suggest
that unionism has a direct positive effect on wages, but also that part of this effect
can be captured by the inclusion of productivity into the model (not shown). Thus,
again, the presence of significant productivity effects of unionism cannot be
rejected.
4.2. Unionism and unemployment
The next relationship we focus on is the impact of unionism on
unemployment. As with productivity, we prefer an empirical specification in order
to avoid data related problems arising from the regional structure of the data. Again,
we include regional and temporal fixed effects, which control for unspecified
constant-across-space and constant-over-time factors. As was the case with the
17
productivity regressions, this specification is strongly supported by the appropriate
tests. Unemployment is made a function of the employment-to-population ratio,
which acts as a proxy for contemporaneous labour demand conditions. Additionally,
a number of variables controlling for the sectoral, occupational and gender structure
of the labour force, as well as for the economic dynamism of each region, are
introduced. These variables include the share of professionals to total employment,
the employment shares of manufacturing and business services, the female labour
force participation rate and the net immigration rate. Table 3 presents the results
from the basic regression (first column) and from the specifications where
unionisation variables are also introduced into the model.
The basic regression performs reasonably well, with no indication of mis-
specification problems at the 1% level. The labour demand and economic dynamism
variables are highly significant and have the correct signs, while the most significant
among the employment composition variables are those controlling for sectoral
employment shares. The introduction of the union density variable (second column)
changes little the results obtained previously. The overall performance of the model
is now improved and unionism is found to have a positive impact on unemployment
(significant at 10%).
Next, we examine the presence of non-linearities in the unemployment-
unionism relationship. Column 3 in Table 3 presents the results from the model
where unionism enters in a quadratic form. As it can be seen, the convex unionism
effect found earlier for the case of productivity is also found here. However, the
performance of the model is not significantly improved while the errors now
become heteroskedastic. When real wages are plugged into the model (column 4),
18
the significance of the unionisation variables drops remarkably and this is true also
in the case of the linear specification (results not shown). Wages are directly related
to unemployment and clearly they are the vehicle via which unionism impacts on
unemployment. We conclude that unionism does not affect labour demand and
employment directly, but only through its impact on wages.
Table 3: Unionism and unemployment
Model (1) (2) (3) (4) (5) (6) (7)
Employment-to-
population -89.01
(-11.68) -85.66
(-11.10) -76.25
(-9.94) -46.54
(-5.86) -77.65
(-10.08) -74.68
(-9.71) -46.56
(-5.82)
Immigration rate -51.43
(-2.26) -43.34
(-1.90) -25.08
(-1.14) -47.20
(-2.53) -44.20
(-2.04) -30.95
(-1.40) -47.31
(-2.50)
Manufacturing -16.54
(-1.96) -19.38
(-2.29) -9.709
(-1.16) -22.12
(-3.05) -22.02
(-2.73) -14.01
(-1.60) -22.22
(-2.94)
Banking-finance -12.71
(-1.73) -13.66
(-1.88) -25.78
(-3.41) -19.99
(-3.13) -17.94
(-2.56) -24.51
(-3.25) -19.96
(-3.10)
Female LF rate 21.18
(1.38) 25.45
(1.66) 20.34
(1.41) 19.65
(1.62) 19.92
(1.36) 18.84
(1.31) 19.61
(1.61)
Professionals 1.814
(1.12) 1.848
(1.16) 1.616
(1.08) 2.433
(1.93) 2.671
(1.74) 2.133
(1.40) 2.445
(1.89)
Union density 8.005
(1.90) -40.12
(-2.95) -3.277
(-0.26) 16.13
(3.44) -21.45
(-1.17) -2.841
(-0.18)
Union density
squared 56.67
(3.70) 4.366
(0.29) 39.89
(2.11) 4.015
(0.24)
Spatial lag of
unionism 148.97
(3.35) 81.07
(1.49) 2.227
(0.05)
Real wage 0.032
(6.34) 0.032
(6.05)
Regional effects* 13.06
0.000 13.72
0.000 16.23
0.000 19.66
0.000 6.35
0.000 5.28
0.000 6.60
0.000
Temporal effects* 37.27
0.000 16.65
0.000 19.96
0.000 33.01
0.000 9.60
0.000 10.45
0.000 11.80
0.000
R-squared 0.964 0.966 0.970 0.979 0.970 0.971 0.979
Normality 1.17
0.558 2.06
0.356 2.38
0.304 0.25
0.880 4.64
0.098 3.87
0.144 0.26
0.876
Heteroskedasticity 2.10
0.147 4.53
0.033 14.80
0.000 9.79
0.002 12.30
0.001 16.43
0.000 9.79
0.002
Ramsey 3.27
0.025 2.76
0.047 4.31
0.007 4.24
0.008 2.79
0.045 3.81
0.013 4.69
0.004
Notes: See notes in Table 2.
The last three columns examine the presence of spatial spill-over effects in
the relationship between unionism and unemployment. As it can be seen, the
19
evidence of a positive spatial spill-over effect is again strong and this time it is not
totally removed when the concentration (quadratic) effect is also included. The fact
that unionisation can increase unemployment in neighbouring regions points again,
as was the case with the productivity regressions, to a spatial dependence effect,
although this time the evidence of a spatial spill-over effect on unemployment is
stronger. It must be noted, however, that when the “local” average wage is included
in the spatially dynamic model (last column), any evidence of a unionisation effect
on unemployment disappears, in the same way it does in the spatially static model.
4.3. Robustness and further results
The empirical findings obtained so far suggest that unionism affects the
economy largely through its effects on wages. We want to investigate further this
mechanism and also check the robustness of the results obtained. For this reason, in
this subsection we widen our investigation, examining the relationship between
unionism and a wider array of economic variables and allowing only one control
variable (real wages) besides the regional and temporal fixed effects. As was the
case before, we test again for the possibility of concentration and spatial effects.
Tables 4 and 5 present the empirical findings.
As can be seen, for the unemployment and productivity regressions the
results are pretty much in line with those obtained from the more complex models.
Although for unemployment the residuals are heteroskedastic and not normally
distributed, which might suggest possible misspecification, we still find a significant
convex union effect and a marginally insignificant spatial effect. The concentration
effect dominates over the spatial spill-over effect, while all union effects disappear
20
when we control for the deflated average regional wage (Table 5). For the
productivity regressions heteroskedasticity and non-normality are much less of a
problem. Again, unionism is found to have a convex effect and this time the spatial
effect is stronger and persists even after controlling for the concentration effect.
However, again all of the union effect on productivity disappears when we include
real wages in the model.
Table 4: Unionism and the economy (concentration and spatial effects)
Model Unemployment Productivity Investment share Employment-
population ratio
Union -61.87
(-3.71) -39.75
(-1.83) -80.57
(-4.33) -37.14
(-1.57) -0.263
(-1.93) -0.324
(-1.80) 0.409
(2.62) 0.229
(1.12)
Union
squared 89.47
(4.96) 70.15
(3.22) 120.27
(5.97) 82.35
(3.48) 0.326
(2.21) 0.379
(2.11) -0.606
(-3.58) -0.449
(-2.19)
Spatial lag
of union 111.47
(1.56) 218.79
(2.82) -0.307
(-0.52) -0.905
(-1.35)
Regional
effects* 21.26
0.000 20.19
0.000 18.31
0.000 18.91
0.000 11.86
0.000 11.54
0.000 40.83
0.000 40.73
0.000
Temporal
effects* 66.36
0.000 67.76
0.000 30.80
0.000 21.11
0.000 18.66
0.000 16.36
0.000 21.57
0.000 20.62
0.000
R-squared 0.931 0.933 0.919 0.925 0.888 0.888 0.958 0.959
Normality 15.42
0.000 18.78
0.000 5.65
0.059 9.19
0.010 8.03
0.018 8.10
0.017 5.90
0.052 6.41
0.041
Heterosk. 17.06
0.000 19.31
0.000 0.62
0.432 0.93
0.334 25.57
0.000 23.66
0.000 0.01
0.910 0.63
0.429
Ramsey 3.72
0.014 3.23
0.026 4.06
0.010 10.10
0.000 1.74
0.164 1.75
0.162 2.37
0.076 2.55
0.060
Notes: See notes in Table 2.
Unionism is found to have a non-linear effect on investment (convex), wages
(convex) and the employment-to-population ratio (concave).15 At high values of
union density, the effect on investment and wages is positive, while the effect on
labour force participation (labour demand) is negative. For average union densities,
however, the effect on investment is also negative. Interestingly, the union effect on
15 The results obtained for the levels of real regional output per capita and household incomes were
very similar to those obtained for productivity and are thus not presented here.
21
investment persists after controlling for wages (which are insignificant, suggesting
that investment decisions are not directly linked to labour cost considerations at the
regional level), whereas that on labour force participation disappears, as was the
case with the productivity and unemployment effects. Finally, there is no evidence
of a spatial effect for the cases of investment and labour force participation, while
for wages a significant positive spill-over effect is obtained, pointing again to a
spatial dependence mechanism (where economic outcomes that are due to unionism
are determined in areas wider than that of the administrative region) as opposed to a
possible spatial competition effect (spatial heterogeneity).16
Table 5: Unionism and the economy (wage effects)
Model Unempl. Prod/vity Inv. Share Emp/pop Real wage
Union -8.131
(-0.59) -2.607
(-0.26) -0.333
(-2.20) -0.038
(-0.27) -12.12
(-4.79) -4.12
(-1.33)
Union
squared 8.271
(0.51) 2.483
(0.21) 0.432
(2.42) 0.070
(0.42) 18.31
(6.68) 11.32
(3.65)
Spatial lag
of union 40.31
(3.97)
Real wage 0.044
(8.91) 0.064
(17.65) -0.0001
(-1.06) -0.0004
(-7.29)
Regional
effects* 33.19
0.000 70.97
0.000 9.00
0.000 64.67
0.000 11.76
0.000 16.64
0.000
Temporal
effects* 60.48
0.000 7.23
0.000 9.01
0.000 10.47
0.000 42.49
0.000 39.20
0.000
R-squared 0.962 0.981 0.889 0.973 0.924 0.935
Normality 20.15
0.000 1.36
0.505 6.72
0.035 0.68
0.711 6.43
0.040 11.88
0.003
Heterosk. 16.28
0.000 11.30
0.001 20.84
0.000 0.24
0.612 2.97
0.085 1.87
0.171
Ramsey 2.08
0.109 0.62
0.601 2.95
0.037 1.20
0.313 6.01
0.001 6.31
0.001
Notes: See notes in Table 2.
The last set of relationships we looked at was about the possible impact that
unionism may have had on regional growth in the UK, examining the relationship
16 We also experimented with non-linear terms for the spatial lags but these did not change the quality
of the results presented in Tables 4 and 5.
22
between regional union densities and output growth, productivity growth, growth of
household incomes, and employment growth. In none of the cases did unionism
prove to be significant. Because this result was very consistent across different
growth variables, we were forced to conclude that unionism did not affect the
growth rates (and, thus, the patterns of regional convergence and divergence) of the
UK regions during the 1990s.
5. Conclusions
This paper examined the economic effects of the decline in British union
densities at the regional level. Based on a panel of 120 observations, covering a 10-
year period for the Standard Statistical Regions of the UK, evidence was obtained
suggesting that trade unionism is positively associated with labour productivity,
incomes, wages and investment, although it is found to impact adversely on
unemployment and labour force participation. These findings are very consistent
with the ones derived by Freeman (1988), using a similar specification albeit for a
different country and period. The effects of unionism have been found to operate at
a scale wider than that of the administrative region and, more importantly, through
wages. Moreover, these effects are clearly non-linear. Thus, low levels of unionism
are found to be associated with lower levels of investment, productivity and wages,
while they seem to have negligible (albeit beneficial) effects on employment. It is
only at high union densities that the positive effects on productivity and investment
are observed. Finally, the absence of negative spatial spill-overs indicates the
absence of spatial competition related to unionism and, thus, that evolutions in the
latter do not by themselves generate mechanisms of regional divergence.
23
Two of these findings have important implications for policy, in light of the
recent reversal in the trends of union decline. First, to the extent that policy does not
(and should not) rely on unions to achieve improvements in economic performance,
it seems that there is an optimal level of unionisation where the economic impact
(positive and negative) of unionism is minimised. Very strong unions increase
productivity but also unemployment, whereas very weak unions have the opposite
effect. The second implication for policy refers to the main mechanism via which
unionism impacts on the economy, the wage-increasing role of unions. Although not
directly in the hands of policy, this mechanism can be to a large extent controlled by
it, mainly through arbitrage and consultations, as it is for example in the Dutch case
(see in particular van der Laan, 1993, and Ozaki, 1999). Again, this suggests that
maintaining average levels of union power (as proxied by union density) can help
avoid the adverse economic effects of both very weak and very strong unions.
A last but very significant observation related to the empirical findings
presented here relates to the differences in the effects of unions on investment and
employment. Based on these findings, it seems that in high union density regions the
need for expansion is met with higher investment in physical capital (and possibly
an intensification of the production process), rather than with increased
employment. This finding seems to support the view of unions as productivity-
enhancing organisations where the presence of a strongly unionised labour force
leads to higher investment and efficiency in production.
To summarise, there are three significant findings derived from the present
analysis. First, unions impact on the economy mainly through wages. Second, this
impact is non-linear, with significant deflating concentration effects. Finally, spatial
24
dependence does not translate into spatial competition and thus does not enforce
regional disparities. For policy, these findings imply that reasonably strong unions
are most likely not a burden to the economy. Although unionism cannot be seen as a
policy tool for regional development and convergence (especially in terms of
employment outcomes), its retreat does not necessarily promote economic
prosperity. For theory, the most relevant finding is that related to the non-linearity of
the identified effects. It seems that, at least at the economy-wide level, unionism
produces differentiated economic outcomes, depending on its degree of
concentration. Integrating this observation into the empirical and theoretical analysis
of unionism can help us gain a deeper understanding of the economic role of unions
and possibly bridge some of the contradictory findings reported in the literature.
25
REFERENCES
Amable B. and D. Gatti, 2001, Product market competition and employment
protection: the case for policy complementarity, mimeo, University of Lille II.
Anselin L., 1988, Spatial Econometrics: Methods and Models, Kluwer Academic
Publishers, Boston.
Baltagi B., 1995, Econometric Analysis of Panel Data, Wiley, Chichester, UK.
Bassanini A. and E. Ernst, 2002, Labour market institutions, product market
regulation and innovation: cross-country evidence, OECD Economics
Department Working Paper No2002/2, OECD.
Blackaby D. and P. Murphy, 1991, Industry characteristics and inter-regional wage
differences, Scottish Journal of Political Economy, 38, 2, 142-161.
Blanchflower D. and A. Oswald, 1988, The economic effects of Britain's trade
unions, Centre for Labour Economics Discussion Paper No 324, LSE.
Blanchflower D. and N. Millward, 1988, Trade unions and employment change,
European Economic Review, 32, 717-726.
Blanchflower D., 1986, What effects do unions have on relative wages in Great
Britain?, British Journal of Industrial Relations, 24, 2, 195-204.
Bland P., 1999, Trade union membership and recognition 199701998: an analysis of
data from the Certification Officer and the Labour Force Survey, Labour Market
Trends (July), pp.343-351.
Booth A. and M. Chatterji, 1998, Unions and efficient training, Economic Journal,
108, 328-343.
Booth A., 1991, Job-related formal training: who receives it and what is it worth?,
Oxford Bulletin of Economics and Statistics, 53, 3, 281-294.
Borooah V., P. McGregor, P. McKee and G. Mulholland, 1996, Cost-of-living
differences between the regions of the UK, ch.5 in J. Hills, ed., New
Inequalities, Cambridge University Press, UK.
Brown C. and J. Medoff, 1978, Trade unions in the production process, Journal of
Political Economy, 86, 355-378.
Cabinet Office, 1999, Sharing the Nation’s Prosperity: variations in economic and
social conditions across the UK, Report to the Prime Minister, London.
26
Card D., 1991, The effect of unions on the distribution of wages, IR Section
Discussion Paper No 287, Princeton University.
Clark K., 1984, Unionisation and firm performance: the impact on profits, growth
and productivity, American Economic Review 74, 893-919.
DeFina R., 1983, Unions, relative wages and economic efficiency, Journal of Labor
Economics, 1, 4, 408-429
Duranton G. and V. Monastiriotis, 2002, Mind the gaps: the evolution of regional
inequalities in the UK, 1982-1997, Journal of Regional Science, 42, 1, pp.214-
253.
Freeman R. and J. Medoff, 1984, What Do Unions Do?, Basic Books, N.Y.
Freeman R., 1980, Unionism and the dispersion of wages, Industrial and Labour
Relations Review, 34, 1, 3-23.
Freeman R., 1983, Unionism, price-cost margin and the return to capital, NBER
Working Paper No1164.
Freeman R., 1988, Union density and economic performance, European Economic
Review, 32, 707-716.
Freeman R., 1991, How much has deunionisation contributed to the rise in male
earnings inequality? NBER Discussion Paper No 3826.
Glosing A. and S. Machin, 1993, Trade unions and the dispersion of earnings in UK
establishments, 1980-1990, CEP Discussion Paper No140, LSE.
Green F., S. Machin and D. Wilkinson, 1996, Trade unions and training practices in
British workplaces, mimeo, University College London.
Koedijk K. and J. Kremers, 1996, Deregulation: a political economy analysis,
Economic Policy 26, 443-467.
Lovell C., R. Sickles and R. Warren Jr., 1988, The effects of unionisation on labour
productivity, Journal of Labour Research, 9, 55-63.
Machin S. and S. Wadhwani, 1989, The effects of unions on organisational change,
investment and employment, Centre for Labour Economics Discussion Paper
No 355, LSE.
Machin S. and S. Wadhwani, 1991, The effects of unions on organisational change
and employment, Economic Journal, 101, 835-854.
27
Machin S., 1991, Unions and the capture of economic rents, International Journal of
Industrial Organisation, 5, 327-350.
Martin R., 1993, Remapping British regional policy: the end of the North-South
divide?, Regional Studies, Vol.27, No8, pp.797-805.
Martin R., 1997, Regional unemployment disparities and their dynamics, Regional
Studies, Vol.31, No3, pp.237-252.
Martin R., P. Sunley and J. Wills, 1996, Union Retreat and the Regions, Regional
Policy and Development 8, Jessica Kingsley Publishers, UK.
Mayhew K., 1983, Trade Unions and the Labour Market, Martin Robertson & Co
Ltd., Oxford.
Metcalf D., 2001, British unions: dissolution or resurgence revisited, CEP
Discussion Paper No493, LSE.
Miller P. and C. Mulvey, 1993, What do Australian trade unions do?, Economic
Record, 69, 315-342.
Mincer J., 1974, Schooling, Experience and Earnings, National Bureau of
Economic Research, NY.
Minford P., 1982, Trade unions destroy a million jobs, Journal of Economic Affairs,
12.
Monastiriotis V., 2002, Trade unionism and productivity: empirical models and
interpretation, unpublished manuscript, Department of Geography and
Environment, LSE.
Nickell S. and R. Layard, 1998, Labour market institutions and economic
performance, CEP Discussion Paper, September, LSE.
Nickell S. and S. Wadhwani, 1988, Unions, wages and employment, European
Economic Review, 32, 727-733.
Nickell S., S. Wadhwani and S. Wall, 1989, Unions and productivity growth in
Britain, 1974-1986, Centre for Labour Economics Discussion Paper No 353,
LSE.
Nickell S., S. Wadhwani and S. Wall, 1991, Productivity growth in UK companies
1975-1986, CEP Discussion Paper No 26, LSE.
OECD, 1997, Economic performance and the structure of collective bargaining, ch.3
in Employment Outlook, OECD, Paris.
28
Ozaki M. (ed), 1999, Negotiating Flexibility: the role of the social partners and the
state, ILO, Geneva.
Reynolds M., 1986, Trade unions in the production process reconsidered, Journal of
Political Economy, 94, 2, 443-447.
Sengenberger W. and D. Campbell, eds., 1994, Creating Economic Opportunities:
the role of labour standards in industrial restructuring, International Institute
for Labour Studies, Geneva.
Sinclair P., 1987, Unemployment: economic theory and evidence, Basil Blackwell,
Oxford.
Van der Laan L., 1993, Regional differentiation in trade union density, Environment
and Planning A, 25, 255-272.
