The Effect of Union Density on Productivity and
Wages
Erling Barth, Institute for Social Research and NBER
Alex Bryson, UCL, IZA and NIESR
Harald Dale-Olsen, Institute for Social Research
Abstract
We exploit exogenous variance in the price of union membership to identify the effects of
changes in workplace union density on workplace productivity and wages in the population of
Norwegian manufacturing firms over the period 2001 to 2012. Changes in the tax treatment
of union membership create variation in the price of membership. Increases in union density
lead to substantial increases in labour productivity having accounted for the potential
endogeneity of unionization. Unions are able capitalise on their increased bargaining power
clawing back part of that additional productivity through a higher union wage premium.
Key-words: trade unions; collective bargaining; union density; productivity; wages
JEL-codes: J01, J08, J50, J51
Acknowledgement: We thank the Norwegian Research Council for funding (grant number 236786).
Corresponding author: [email protected].
1.
INTRODUCTION
Do unions obstruct or promote productivity growth? Theoretically there are several good
reasons to support either view. The rents obtained by local unions may impede investments
(Grout, 1984) and conflict generated by union bargaining may be detrimental to
management-employee relations (Bryson, Forth and Laroche, 2011). On the other hand
unions may provide a “voice” for workers and improve information flows (Freeman and
Medoff, 1984), and local union bargaining may promote efficient provision of effort (Barth,
Moene and Willumsen, 2014).
Empirically, it has proven difficult to assess the effect of unions on productivity. The
drawbacks to the observational studies assessing union effects on firm performance are
discussed in detail in Section Two, but the chief one is the absence of exogenous variance in
unionization required to draw causal inferences. Workplaces become organised for a number
of reasons that are linked with their performance. First, union formation and membership
may be highly dependent on the potential rents to be reaped. On the one hand, it pays more to
invest in unionization and membership in more productive firms. On the other hand, in firms
facing the risk of downsizing or closure, the value of membership may also be high since
unions tend to offer legal services and help with conflict resolution. Second, union members
may be highly selected. Again the direction of selection is not clear: on the one hand more
productive workers are more likely to obtain jobs in well-paying unionized firms, on the
other hand, highly productive workers with good outside options may be better off with
individual bargaining rather than under collective coverage. Regardless of the direction of the
selection, it has proven very difficult to come up with a research design that convincingly
deals with this problem.
To our knowledge, only DiNardo and Lee (2004), and the follow up studies by Lee and Mas
(2012) and Frandsen (2012) represent attempts at identifying unions' causal effects on
workplace performance. They use a regression discontinuity design related to union
recognition in the United States. We discuss their contributions in more detail below. We
contribute to the literature using exogenous variance in the price of union membership to
identify the effects of changes in workplace union density on workplace productivity and
wages in the population of Norwegian manufacturing firms over the period 2001 to 2012.
2
Exogenous shifts in the net price of union membership arise due to changes in the tax
treatment of union membership. As with most normal goods, the demand for union
membership (or the service this membership provides) is negatively related to its price, and
thus the demand for union membership fluctuates with the size of the tax subsidies. We know
of no other studies using this source of exogenous price variance as a means of instrumenting
the demand for union membership.
We calculate the potential subsidy for each individual worker in the economy, and predict a
union density for each establishment based on the tax system. The predicted union density is
then used as an instrument in our productivity and earnings regressions. Since the effective
tax subsidy varies with marginal tax brackets based on the income bracket for individuals we
take special care to control for the share of workers in each tax bracket as well as of average
worker quality in all our IV regressions. To ensure that we control for selection of workers
into workplaces we also control for worker fixed effects from an earnings regression by
averaging the individual fixed effects within the workplace.
We find increases in union density lead to substantial increases in labour productivity having
accounted for the potential endogeneity of unionization. We also show unions are able to
claw back part of that additional productivity through a higher union wage premium.
The remainder of the paper is organised as follows. Section Two briefly reviews the
theoretical and empirical literature and elaborates on the role of union density and union
institutions in helping to understand heterogeneity in union effects. Section Three describes
our data and outlines the empirical approach. Results are presented in Section Four before
concluding in Section Five with a discussion about the implications of the results for our
understanding of union effects more generally.
2.
THEORY AND PREVIOUS EMPIRICAL LITERATURE
Unions are in secular decline. Membership has been falling for decades in much of the
developed world (Schnabel, 2012), and collective bargaining is under threat, even in
countries like Germany where sectoral bargaining was previously regarded as a fixed feature
of the economic landscape (Addison et al., 2011). Two salient facts go largely unnoticed in
discussions of the economic implications of these changes. The first is that unions continue
to procure a wage premium for covered employees both in Anglo-Saxon countries
3
(Blanchflower and Bryson, 2007) and in Continental European countries like France (Breda,
2015). Second, the negative correlation between unionization and workplace or firm
performance, apparent in the 1970s and 1980s (Hirsch, 2007; Metcalf, 1989), had largely
disappeared by the 1990s, at least in Britain where much of the research was conducted
(Blanchflower and Bryson, 2009).
This has led to speculation as to why. Some maintain that declining union density, together
with a changed economic environment - notably increased global competition - began to
undermine unions' ability to monopolise the supply of labour (Brown et al., 2009). Certainly,
it is the case that where negative associations persist, they are confined to workplaces with
strong bargaining power, either by virtue of high union density or the presence of multiple
bargaining units (Bryson et al., 2011; Pencavel, 2004). Some point to a reorientation of
union strategies resulting in partnerships with employers born of union weakness (Frege and
Kelly, 2003).
In France, the negative association between unionization and workplace
performance is confined to a small number of militant unions (Bryson et al., 2011). Others
point to differential union survival among firms and industries with higher rents (Brown et
al., 2009) permitting unions to extract rents without obvious detrimental impacts on the
workplace.
The literature exploring union effects on economic outcomes is one of the oldest and most
extensive in economics. It goes back at least as far as Adam Smith's The Wealth of Nations
which he wrote in 1776. The bulk of the literature treats unions as labour cartels, intent on
strengthening the bargaining power of their members by threatening the supply of labour to
firms if employers prove unwilling to accede to their wage demands. As such, they have the
potential to extract rents from employers resulting in the payment of above-market wages.
As Adam Smith pointed out in The Wealth of Nations employers are also liable to form
cartels, not only to limit price competition, but also in an attempt to offset union bargaining
power. The wage outcome of union bargaining will depend on a number of factors. These
include the relative bargaining power of the two parties which, in turn, is related to the price
elasticity of demand for labour, the elasticity of demand for labour with respect to capital, the
substitutability of non-union for union labour and worker support for the union, usually
captured by the percentage of workers who are union members. Inter alia, the economic
implications of a bargained outcome for the firm depend on the intensity of market
competition faced by the firm, the rents available to the firm and its ability to attract and
4
retain labour.
Nevertheless, on the assumption that worker bargaining power rises, on
average, in the presence of trade unions, it seems reasonable to assume union bargaining will
raise wages above the counterfactual market wage set at the intersection between labour
supply and demand.
The implications of a union bargained wage for employment outcomes will depend, in part,
on whether unions bargain solely over wages - as in the right-to-manage model in which
employers set employment conditional on the union bargained wage - or over wages and
employment simultaneously (efficient bargaining) leading to potentially Pareto efficient
outcomes. Employment outcomes will also depend on what utility the union is seeking to
maximise. If the representative or median union member values continued employment as
well as wages, this utility may be captured by the value of the total wage bill, in which case
the union will be cognisant of potential negative employment consequences where bargained
wages are set "too high".
There are multiple channels by which trade unions may affect labour productivity, and these
effects may cut in different directions. More able workers may queue for union jobs where
they pay above market wages, a worker selection effect that may raise labour productivity in
the union sector. If selected from the queue by a unionised employer an employee may be
less likely to quit compared to a non-union scenario given the wage wedge between the union
job and the employee's outside options, in turn affecting employers' propensity to invest in
human capital. If unionised labour is more expensive than non-unionised labour this may
induce employers to substitute capital for labour, leading to capital intensification that is
productivity enhancing. A separate channel is the union "voice" effect, first identified by
Freeman and Medoff (1984), whereby unions aggregate and convey the preferences and
knowledge of workers to management in a manner that can be more efficient than eliciting
individual workers' voices, or failing to engage with workers at all. 1 Unions may also serve to
alleviate agency problems in a similar way as performance pay schemes (Vroman 1990;
Barth et al., 2012), or provide efficient effort levels within a framework of local bargaining
(Barth et al 2015).
1
Freeman and Medoff (1984) adapt Hirschman's (1970) exit-voice-loyalty model, originally used by Hirschman
primarily to understand consumer preferences, to an employment relations setting, emphasising its productivityenhancing potential, as well as increasing employer pay-offs to human capital investments as employees resort
to voice over exit when confronting workplace problems.
5
Unions may also be detrimental to labour productivity. Wage-effort bargaining may result in
the sub-optimal deployment of labour through "restrictive practices" (Metcalf, 1989). Where
union bargaining breaks down resultant strike action or actions short of strikes, such as goslows, may adversely affect productivity. Unions' ability to insure workers against arbitrary
employer actions, whilst potentially conducive to job security and thus improvements in
productivity, may also lead to workers taking unauthorised absences, or "shirking" in other
ways. Unions' ability to extract rents from new investments may lead to a "hold up" problem
whereby investors, aware of the issue, may invest less than they might otherwise have done,
leading to sub-optimal capital investments (Grout, 1984).
In the worst case scenario,
investors may react adversely to the threat of unionization, taking evasive action by investing
in the non-union sector.
The empirical literature has, until quite recently, been dominated by Anglo-US studies where
sectoral bargaining is uncommon in the private sector and unions organise on a workplaceby-workplace or firm-by-firm basis.
Consequently, the focus has been establishing the
economic effects of unions obtaining bargaining rights at workplace level, and the bargaining
strength of unions at workplace level, often proxied by the proportion of employees in
membership.
There are four limitations to this literature. First, it is an empirical literature dominated by
studies that identify the partial correlation between unionization and economic outcomes, the
assumption being that selection into union status is captured by observable features of the
worker or, if panel data are available, by time-varying observable traits and time-invariant
unobservable traits. It has proven difficult to account for potentially endogenous selection
into union status due to a lack of credible instruments. Second, the majority of studies have
relied on data collected from individual workers in household surveys such as the Current
Population Survey (CPS) for the United States. Necessarily, these studies omit important
features of the firm employing the workers, so that analysts have found it difficult to tackle
biases associated with omitted variables influencing union status and the economic outcomes
of interest. Studies using linked employer-employee data tend to find that the omission of
these variables upwardly biases estimates of union effects on wages, for example (Bryson,
2002; Blanchflower and Bryson, 2004). Third, limited availability of firm-level data has
prevented analysts from undertaking workplace-level or firm-level analyses, thus limiting
what analysts have been able to say about outcomes that are best investigated at this level,
6
such as productivity and profitability.2 Fourth, the particularities of the institutional setting
characterising the liberal economies of the USA, UK, Canada, Australia and other Anglo-US
economies mean it is difficult to know whether findings from those countries generalise to
other settings characterised by more centralised and co-ordinated bargaining regimes.
Indeed, as discussed below, our prior is that they are unlikely to read over directly since
sectoral and national bargaining arrangements are likely to affect the costs and benefits of
unionization for specific firms.
The empirical regularities regarding the union wage premium stem from a literature that is
dominated by observational studies capturing the partial correlation between union status and
wages is captured in cross-sectional data or, in some cases, the association between changes
in union status and wages with panel data.3 The union wage premium - or what might, more
appropriately be termed the union wage "gap" to use Lewis's (1963; 1986) terminology varies across groups of workers, over time, and is counter-cyclical (Lewis, op. cit.). Since
union bargained wages apply to all covered workers union bargained wages tend to be a
public good rather than a private incentive good payable only to union members. Even so,
studies often find a union wage premium among members in covered workplaces, which may
partly reflect an upward bias associated with omitted variables affecting selection into
membership status and wages, or else the effects of heterogeneous union bargaining power
(Booth and Bryan, 2004). The latter arises where membership simply proxies higher union
density, something that is not observed in studies which cannot link employees to the
workplaces that employ them.
Unionization also slows the rate of employment growth in workplaces. This finding, which
Addison and Belfield (2004) termed the "one constant" in the empirical union literature,
when set alongside the persistence of a union wage premium, is consistent with right-tomanage models in which employers set employment levels conditional on the bargained
wage. However, union effects are rarely sufficient to affect workplace survival (Bryson,
2004), suggesting either than unions seek to maximise the wage bill (some weighted function
2
In principle one can aggregate workers from worker-level data to construct firms where unique firm identifiers
are available, but data are rarely available for the full population of workers in a firm and, in any case, such data
rarely contain firm-level economic metrics other than wages.
3
The latter have rarely considered the endogeneity of union switching but for an examination of the
implications of union endogenous switching in relation to pay satisfaction see Bryson and White (2016).
7
of wages and employment), that they successfully organize firms with surplus rents, or that
wage effects are partially offset by productivity improvements.
Recently analysts in the United States have sought to identify the causal impact of union
bargaining on workplace performance using a regression discontinuity design comparing
economic outcomes in workplaces where the vote just exceeded the majority threshold
required for representation with workplaces where the vote felt just short of the required
majority. Using this method DiNardo and Lee (2004) find little impact of new unionization
on business survival, employment, output, productivity or wages over the period 1984-2001.
When interpreting this result one must bear three points in mind. First, the vote for
representation captures an "intention to treat" through union representation that does not
always materialise in practice. This is because, under the US system, the majority vote
requires the employer to negotiate with the newly formed union in good faith to arrive at new
contractual terms and conditions. However, unions never get to "first contract" in a high
percentage of cases (Ferguson, 2008), suggesting the regression discontinuity captures a
lower bound estimate. Second, if union bargaining power is increasing in the demand for
unionization, as the literature on union density effects suggests, the margin just-beingunionised is likely to capture effects associated with weaker trade unions. This is precisely
what Lee and Mas (2012) find in a follow up study which shows that, using an event study
approach, the equity value of newly unionised firms drops markedly after 15-18 months,
something that is not apparent using a regression discontinuity design. They reconcile results
in Lee and Mas (2012) with those in DiNardo and Lee (2004) by showing that the negative
relationship between cumulative abnormal returns and unionization rises with the vote share
in support of the union. The implication is that firms' owners have a strong expectation that
new unionization will have an impact on firms' economic performance, especially when
union bargaining power is great. Third, unions are known to focus their attention on raising
the wages of low earners, providing the rationale for Frandsen's (2012) quantile regression
investigation. He uses the same regression discontinuity as DiNardo and Lee and Lee and
Mas and finds large counterveiling effects of new unionization on wages in different parts of
the wage distribution, with unions using their bargaining power to compress wages by
increasing the wages of the lower paid and reducing the returns to skill at the top of the
distribution. A recent paper using the same identification strategy found negative effects of
unionization on staffing levels in nursing homes but no effects on care quality, suggesting
positive labour productivity effects (Sojourner et al., 2015).
8
In a number of European countries the vast majority of workers and firms are covered by
collective bargaining. In Austria and France, for example, over 95 per cent of workers have
their pay set directly through collective bargaining - often at national or sectoral level - or
else collectively bargained rates are extended to them under statutory procedures (OECD,
2016).
Setting wages and terms and conditions at sectoral or national level necessarily involves the
aggregation of firm and worker preferences above firm level. It is unclear, a priori, whether a
bargained outcome set beyond the firm will operate to the benefit or disadvantage of a
specific firm. It depends, in part, on where the firm sits in the firm wage hierarchy and on the
firm's ability to withstand wage hikes. The bargained rate may be particularly beneficial to a
firm where its competitors struggle to pay the new rate. At the macro-level sectoral and
national bargaining is liable to compress wage dispersion since the uncovered sector is small,
thus taking wages out of competition - at least at the lower end of the labour market where
the bargained rates bite - potentially minimising any adverse effects of bargained rates on
firm performance.
The situation is more complicated in those countries where firms may be subject to national
or sector bargained rates and local bargaining, either at firm or workplace level. Often local
bargaining builds on national or sector bargained rates. How they do so depends on the
degree of coordination across bargaining levels, as studies have shown, but also on the
bargaining strength of local unions and thus their ability to bid up wages beyond the centrally
set wage.4 Studies confirm the importance of union density at workplace or firm level in
these circumstances. For example, Breda (2015) shows the union wage premium in France
rises with workplace union density where the workplace has a high market share, consistent
with workers extracting surplus rents via their local bargaining power.
The setting for our empirical investigation is Norway, a country where firms may be covered
by collective bargaining at local level (workplace of firm), sector level, national level, or a
combination of local and sector/national bargaining. Eighty-seven per cent of all employees
are covered by collective agreements. There is a high degree of coordination in bargaining,
4
For a review of this literature see Bryson (2007).
9
but 79 per cent of all employees work in workplaces with local bargaining following the
national or sectoral level bargaining rounds (Barth et al 2015). Around 70 per cent of private
sector workplaces and 77 per cent of private sector employees are covered by some form of
collective bargaining. Four-in-ten workplaces have some local collective bargaining which
covers over half (54 per cent) of employees. In contrast to France where union membership
is well below 10 per cent, but in common with other Scandinavian countries, Norway has
high levels of union membership. Half of all private sector employees are union members,
while mean union density is 40 per cent in private sector workplaces (Bryson et al., 2015).
There is no union membership wage premium for men or women in Norway, but wages rise
for all workers where workplace union density is higher (Bryson et al., 2016). No evidence
on the causal impact of union density on productivity and wages exists in Norway, and even
studies of correlations are scarce. Barth, Raaum and Naylor (2000) and Balsvik and Sæthre
(2014) provide evidence on the relationship between union density and wages. Both studies
estimate a union wage premium of around 7 per cent, i.e., when union density increases by
10 percentage points then wages increase by 0.7 per cent. Barth, Raaum and Naylor (2000)
point out that any effect of individual union membership disappears when adding controls for
union density, which implies that the bargained wage at the workplace is a public good.
3.
THE NORWEGIAN TAX LEGISLATION AND THE UNION MEMBERSHIP
FEE
The Norwegian income tax system is proto-typical of a modern welfare economy, with a
strongly graduated or progressive element, thus making the relationship between the marginal
tax rate and labour income positive but highly non-monotonic. Table A1 and Figure A1 show
the development of the tax schedule over time in Norway. Over the years, minor changes in
income tax rates and tax brackets are determined politically and made public by the Ministry
of Finance at the end of the previous year).5 In the Norwegian system taxable income is
determined by adding different kinds of income (labour income, employer-provided fringe
benefits) and subtracting certain basic deductions (major deductions are basic commuting
expenses and interest payment on house mortgages).
5
In 2006 a more comprehensive reform was introduced for redistributive reasons (Thoresen et al., 2011;
Thoresen et al., 2012). It reduced marginal income tax for higher income levels, while adjusting the deductable
elements for taxable income at the bottom. Wage earners' response to this tax reform was modest (Thoresen et
al, 2011), but the redistributive effects of the tax system as well as the degree of horizontal equality have
improved slightly (Thoresen et al., 2012).
10
One of these deductions is expenses related to the union membership fee. Table 1, Panel A)
shows the development of the gross deduction allowed for union membership. The deduction
increased slightly until 2005 before rising markedly.
[TABLE 1]
In 2002 the gross deduction was 900 NOK, rising to 1800 NOK in 2005 and then to 3660
NOK in 2010. No explicit pronouncements were made as to why this growth occurred, but it
may have been linked to changes in political power in Norway. Before October 2005 a
liberal-conservative coalition was in power. In the October 2005 election the Labour party
gained power, holding onto it in the election of 2009. In Figure 1 we see the development of
gross union membership deductions (left-hand side axis) and the Labour party’s and the
Conservative party’s elected number of seats in the Norwegian parliament (right-hand side
axis).
[FIGURE 1]
The net subsidy, i e., what an individual actually saves due to the gross union deduction,
depends on the individual’s tax bracket (Table 1 Panel B and Figure 2). Net deductions
increased in all tax brackets. The major tax brackets defined by the number of workers are
brackets T3-T5.
[FIGURE 2]
To establish whether these deductions are economically meaningful we looked closer at the
average union membership fee paid by individuals as reported to the tax authorities. These
averages, overall and within each marginal income tax bracket, are presented in Table A2.
Figure A2 shows the relative difference between the subsidy and the average union
membership fee for each tax bracket. We see that for the major tax groups the subsidy
amounts to 10 percent of the union fee in beginning of the period and then increased to
around 30-40 per cent. This means that through the tax system a majority of the individuals
save between one third and half of the fee. This is sizeable enough to plausibly affect the rate
of union membership take-up.
11
In spite of the fact that such subsidies exist in other countries such as the United States and
the United Kingdom no empirical evidence exists on the relationship between taxation and
the demand for union membership. However, a related literature exists linking the demand
for certain fringe benefits, such as U.S. health care, savings plans, company cars, stocks and
stock options, to the taxation of these goods and services (Gruber, 2001; Gruber and Lettau,
2004; Gutiérrez-i-Puigarnau and Van Ommeren, 2011; Choi et al., 2011; Austin et al., 1998;
Babenko and Tserlukevich, 2009). For example, Gruber and Lettau estimate that a removal of
the subsidization of employer-provided health care would reduce insurance spending by 45
per cent. Similarly, Gutiérrez-i-Puigarnau and Van Ommeren find that the subsidization of a
“company” car by the tax system leads to households demanding a more expensive car and
driving more miles privately. Beneficial tax treatment increases the employees’ demand for
stock options (Austin et al., 1998) as well as employers’ supply, since employees tend to
exercise stock options when corporate taxable income is high, shifting corporate tax
deductions to years with higher tax rates (Babenko and Tserlukevich, 2009). Our empirical
approach does not preclude the existence of multiplier or social interaction effects. Although
unions are usually unable to prevent non-members from benefiting from union bargained
terms and conditions, free-riding behaviour does not affect our identification strategy (Olson,
1965; Booth, 1985).
4.
EMPIRICAL APPROACH AND DATA
4.1
EMPIRICAL APPROACH
Consider the following simple Cobb Douglas production function:
𝑙𝑠
ℎ𝑠
𝑌𝑖𝑡 = 𝐴𝑒 𝜔𝑖𝑡+𝑢𝑖𝑡+𝛾𝑡+𝛽𝐷𝑖𝑡 𝐿𝑙𝑠 𝛽 𝐿ℎ𝑠 𝛽 𝐾𝛽
𝑘
where Y is value added, 𝜔𝑖𝑡 is a productivity shifter known to the firm when it chooses the
level of transitory inputs, but not observed by us, 𝛾𝑡 represents technological change, 𝐷𝑖𝑡 is
union density of firm i at time t, ls represents low skill and lh high skill workers respectively,
K is capital, and u is a stochastic term representing idiosyncratic shocks that are unknown to
the firm when it makes its decisions. The key coefficient for us is 𝛽 which captures the effect
of union density on productivity.
12
There are two problems to overcome. The first is the standard problem of endogenous inputs:
we use standard methods to deal with that. In our preferred specifications, we follow the
Petrin, Levinshon and Wooldridge (see Wooldridge 2009) control function approach by
including a proxy for 𝜔𝑖𝑡 using lagged values of capital and materials and their interactions
directly in the production function, and instrumenting for the two L’s using lagged values.
This proxy, derived from the firm’s first order condition in period t-1, effectively controls for
𝜔𝑖𝑡 in the equation, and thus removes the correlation between the lagged L’s and the error
term. For comparison we also provide results from OLS and fixed firm effects regressions.
The second potential problem is that union density, as discussed above, may potentially be
affected by the productivity of the workplace. In particular one may envisage that workers are
more likely to unionize, and unions more likely to invest in membership drives, when
potential rents over which the union wishes to bargain are high. On the other hand, when
firms face difficulties, union membership may provide important insurance and services
related to the risk of job loss, inducing a potential negative relationship between membership
and productivity. Furthermore, if less productive workers sort into union membership this
might induce a negative correlation between union density and productivity. 6 We deal with
this potential endogeneity issue by instrumenting D by the predicted union density, identified
by differences in tax breaks for different income brackets over time. The details of the tax
scheme are provided in Section Three above. We construct the instrument as follows. First
we estimate the following individual-level probit equation:
𝑃(𝑈𝑛𝑖𝑜𝑛 𝑚𝑒𝑚𝑏𝑒𝑟 = 1) = Φ(𝑎𝑇𝑖𝑡 + 𝛼𝜏𝑖𝑡 + 𝛾𝑡 + 𝜀𝑖𝑡 )
where T is the tax bracket the individual is in at time t, 𝜏 is the tax deduction on union
membership that year and 𝛾 is year fixed effects. The tax deduction varies with tax bracket as
6
While no direct evidence exists on less productive workers sorting into union membership, Mastekaasa (2013)
shows that workers with a higher probability of experiencing sick leave spells sort into union membership, and
arguably health, absenteeism and productivity could be related. The theoretical literature on worker sorting into
union membership focuses on workers' potential earnings. The wage standardization policies of unions result in
systematic differences in the union wage premium workers can expect: those with lower potential earnings get
the biggest returns from unionization relative to their market outside options while those with high potential
earnings see negative union returns relative to their market outside options. If one assumes potential earnings
are highly correlated with individual productivity then the implication is that it is the least productive workers
who queue for union jobs. However, as Abowd and Farber (1983) point out , being observed in the union sector
is a function of two things: a) queueing for a union job; b) being picked by the unionised employer from the
queue. Under their model, where the queue for union jobs exceeds the supply of union jobs the employer is able
to pick the best from the queue. The result is that union workers come from the middle of the ability
distribution.
13
well as over time as noted in Section Three. For each individual i in establishment j at time t,
we predict the probability of being a union member every year 𝑃𝑖𝑗𝑡 . Next we construct
predicted union density at the establishment as the average of the predicted probabilities for
the individuals employed in the establishment each year
̃𝑗𝑡 = 𝑃̅.𝑗𝑡
𝐷
̃𝑗𝑡 directly to the second stage regression, but rather use 𝐷
̃𝑗𝑡 as an
Note that we do not apply 𝐷
instrument for D.
The instrument for union density is thus defined at the establishmentXyear level, and varies
with the tax system and the number of workers in different tax brackets only. Since the tax
bracket is a function of the tax system as well as of workers’ income, we always condition on
the share of workers in different tax brackets in the second stage as well, whenever we use
this instrument. We also use the lagged wage costs per employee as controls, to ensure that
the effects are not arising from differences in average worker quality. In one of the
specifications we also include the average individual fixed effects from an individual
earnings regression for the whole economy, including year and age group fixed effects
4.2
DATA
We exploit population-wide administrative register data provided by Statistics Norway and
Statistics Norway’s The Capital Data Base (Raknerud et al., 2004). The former data,
collected by the Norwegian Tax Authorities and Social Services, provide information on
individuals and jobs, for example on income, earnings, work hours, wages and union
membership fees. These data comprise the whole Norwegian population of workers, workplaces and firms during the period 2001-2012. Data thus comprise 2500000 observations
each year. Unique identifying numbers exist for each kind of unit (individuals or workers,
workplaces, firms), thus allowing us to track these units over time. The Capital Data Base
provides information on firm production, such as value added and revenues, and inputs such
as capital, labour, intermediate goods, and costs over the same time period. Since the Capital
Data Base utilizes the same firm identifier as the public administrative register data, we are
able to link these data sources together. Although The Capital Data Base comprises firms
from all private sectors, the coverage outside the manufacturing industries is not complete (all
Manufacturing firms are included in the data base). Thus we chose to focus on the
manufacturing industries in the productivity analysis.
14
In our empirical approach described in Section 4.1 we start our analyses by conducting an
auxiliary regression of the probability of union membership as a function of the union
membership fee deductions provided by the tax legislation (the key outcome of these
regressions is our instrument for union density in the productivity analyses that follows). As
described in Section 3 the Norwegian income tax schedule is highly non-monotonic and
varies between income brackets (although it in general provides higher marginal taxes as
incomes increase). However, for very low incomes the tax schedule is very erratic. Over time
this is amplified due to tax reforms targeted at the bottom of the income distribution. Thus to
avoid this volatility, we focus on workers reporting taxable income year t and year t-1,
t∈(2000,2012) above 1G (G is the Social Service’s baseline figure). For the auxiliary
regressions this implies we restrict the analyses to roughly 2,400,000 jobs each year or
28,695,942 observations over the whole period.
For the main productivity analyses we restrict our analyses to Manufacturing industries and
then only industries with at least 200 observations over our time period. The reason for this is
that we conduct within industry analyses and thus need a certain number of observations for
robustness reasons. This leaves us with 6-6,500 firm observations each year, and when linked
to the administrative data the final regressions comprise around 7,000 firms and 40,000
observations.
Unionization
For the auxiliary regressions we exploit the information on union membership fees as
reported to the tax authorities for income deduction purposes. We simply construct a dummy
measuring 1 if a union membership fee is paid, 0 otherwise. Then we use the auxiliary
regressions to predict for each individual the predicted probability of union membership. In
our main analyses, the productivity regressions, we construct measures of union density by
simple taking the firm- and time specific average of the workers’ union membership
dummy/predicted probability.
Auxiliary regressions – the construction of our instrument for union density
We start our analyses by constructing our instrument for union density which we employ in
our subsequent productivity analyses. We estimate simple Probit regressions of a dummy
indicating union membership (union member =1,otherwise 0) on year dummies, intercept,
15
controls and net deduction (subsidy) related to union membership (in NOK). The regressions
are based on observations of workers reporting taxable income year t and year t-1,
t∈(2000,2012) above 1,000 NOK. In Table 2 we show the results of these regressions, as well
as the estimated marginal effects.
TABLE 2]
In Model 1 the regressions only comprise year dummies and an intercept as controls in
addition to the net deduction. We see that increasing the net deduction by 1 NOK
significantly increases the probability of union membership by 0.04 percentage points. This is
apparently a modest impact, but one has to remember that 1 NOK increase is also a modest
increase. The union deductions increase from 900 to 3750 NOK during our observation
period.7 In Model 2 we add dummies for tax brackets as controls. If anything this only
enforces our results. Finally, Model 3 shows that changes in the net deductions affect the
probability of union membership. For workers in the biggest tax bracket this implies that a 1
NOK increase in tax deductions yields 0.01 percentage points higher probability of being a
union member. Across all workers the effect is as before, around 0.04 percentage points.
These regressions clearly establish that the price of union membership matters for
individuals.
Finally, we create two measures for union density at the firm level for use in the productivity
analyses. The first measure is just the union density measured directly for this population of
workers, i.e., the firm- and time-specific average of union membership across workers. The
second measure is the firm- and time-specific average of the predicted union membership
probability based on Model 2 across workers.
Table 3 provides descriptive statistics at the firm level.
[TABLE 3]
5.
RESULTS
7
We have also estimated a model (not shown) comprising the log of the deductions instead of deductions in
levels. The marginal effect in this case implies that twice the deduction increases the predicted probability by 35
percentage points.
16
5.1. The impact of union density on productivity
Table 4 provides the main productivity results. The first model shows results from an OLS
regression on the whole manufacturing sector, including year and industry effects. Firms with
higher union density appear to be less productive than firms with low union density. The
negative relationship between union membership and productivity disappears with the
addition of firm fixed effects in Model 3, suggesting that workers in less productive firms are
more likely to be organized. In models 2 and 4 we include both the share of workers in each
tax bracket as well as lagged wage cost per employee in order to control for worker quality as
well. This does not change any of the previous results qualitatively. In Model 5 union density
is instrumented using the predicted union density as an instrument and the issue of
endogenous inputs is handled by including a proxy for the unobserved productivity term, and
contemporaneous inputs are instrumented by its lagged values, following the LevinsohnPetrin-Wooldridge strategy. Now we find a positive, significant effect of union density on
productivity.
[TABLE 4]
Table 5 provides results from models where the coefficients for labor and capital are allowed
to vary across industries. We run separate regressions for each industry, using the LPW
method (see appendix table A3 for the coefficients), and then regress the industry specific
residuals on union density and other covariates. Again we find a positive significant effect of
union density.
[TABLE 5]
The last two models include the average worker fixed effect from a worker-level log earnings
regression including year and age fixed effects. The estimated positive effect of union density
is even stronger when including worker fixed effects, suggesting that controlling for selection
into union membership is important.
In the last model we include an interaction term with competition in the product market. We
measure competition by the Herfindahl index. In line with the expectation of Freeman and
17
Medoff (1984) we find that the productivity effects are largest where the competitive pressure
is higher (see also Hirsch (2004: 429)). The standard deviation of the Herfindahl index is
0.0435, suggesting that a one standard deviation reduction in the Herfindahl (more
competition in the product market) increases the union productivity effect by .11 log points.
5.2. Union wage effects
We have found a positive effect of union density on the productivity of the firm. What is the
effect on wages? Table 6 reports results from log earnings regressions, estimated at the firm
level. The dependent variable is the firm level average residual earnings from log earnings
regressions including year and age dummies as well as worker fixed effects. The results from
models 1-3 show that firms with a higher union density pay better wages. The effect is
strongest when we use the instrument for union density and control for establishment fixed
effects, suggesting a negative selection into high union density firms. We also find evidence
of rent sharing since firm level tfp is positively correlated with wages.
[TABLE 6]
In models 4-6 we interact union density with tfp on the assumption that a strong union would
be able to reap more of the benefits from higher tfp than a weaker union, and thus that the
rent sharing parameter, which is often thought of as measuring the bargaining power of the
union, is larger with a stronger union. This is not what we find: the rent sharing parameter is
not larger with a stronger union. In the OLS specification the interaction is even significantly
negative, and in the IV specifications there is no significant interaction effect. The positive
union density effect remains, and the rent sharing parameter is at least as strong as measured
in the more parsimonious specification. One interpretation could be as follows: there is rent
sharing going on regardless of union density, however, a strong union is able to negotiate
higher wages regardless of tfp. In other words, the union ensures that even low rent firms
have to pay higher earnings.
5.
CONCLUSION
We find a positive effect of firm level union density on firm level productivity. The OLS
results show a negative relationship between union density and productivity, but the negative
relationship disappears once we control for firm fixed effects, and it turns significantly
positive after controlling for endogenous union density using variation in tax breaks over
18
time as an instrument for union membership. The OLS results were not completely
surprising: as discussed in Sections 2 and 4, if the risk of job loss increases demand for union
services (and thus membership) or less productive workers sort into union membership, this
induces such negative correlations. Our IV-approach takes this into account, thus when we
exploit the variation in union density caused by the exogenous variation in the price of union
membership, we identify a positive causal impact on productivity.
We observe that the productivity impact is greater when product market competition is
stronger. At the moment, however, we cannot disentangle whether this is due to competition
forcing employers to take productive action when facing stronger unions, or if this somehow
reflects unions influencing worker behaviour more strongly when their firm is facing stronger
competition.
We also find a positive relationship between firm level wages and union density. Perhaps
surprisingly, the effect is positive and occurs independently of a rent sharing effect, which is
also present in our data. The positive rent sharing parameter suggests that unions may have an
interest in boosting productivity in the firm, and local unions may thus contribute to local
productivity. However, since union density provides higher wages regardless of tfp, also
firms with a high union density may have a stronger incentive to improve productivity, since
a low-productivity/low-wage track is not feasible with a strong union. We are thus unable to
conclude that the boost in productivity arises from workers being more aligned with the
interests of the firm, or firms having stronger incentives to keep productivity high.
19
References
Abowd, J. and Farber, H. (1983). ‘Job queues and union status of workers’, Industrial and
Labor Relations Review, Vol. 35, pp. 354-67
Addison, J. T. and Belfield, C. R. (2004) "Unions and Employment Growth: The One
Constant?", Industrial Relations, 43, 2: 305-323
Addison, J. T., Bryson, A., Teixeira, P., and Pahnke, A. (2011) 'Slip Sliding Away: Further
Union Decline in Germany and Britain', Scottish Journal of Political Economy, 58,
4, 490-518
Balsvik, R. and M. Sæthre (2014) "Rent sharing with Footloose Production. Foreign
Ownership and Wages Revisited", NHH Discussion paper SAM 30-2014.
Barth, E. Naylor, R. and O.Raaum (2000) Union Wage Effects: Does Membership Matter?,
Manchester School , Vol. 68 (3): 259-275
Barth, E., Bratsberg, B., Hægeland, T. and Raaum, O. (2012) “Performance Pay, Union
Bargaining and Within-Firm Wage Inequality”. Oxford Bulletin of Economics and
Statistics. vol 74:327-362
Barth, E. Moene, K. and Willumsen, F. (2014) “The Scandinavian Model – an interpretation”
Journal of Public Economics Vol. 117:60-72
Barth, E. and K. Nergaard (2015), «Kollektive partsforhold: Status quo eller endring?». In
Dale-Olsen, H. (Ed.). Norsk arbeidsliv etter turbulente tider. Oslo: Gyldendal
akademisk, 83 – 104.
Blanchflower, D. and Bryson, A. (2003) "Changes over time in union relative wage effects in
the UK and the US Revisited", chapter 7 in International Handbook of Trade
Unions, John T. Addison and Claus Schnabel (eds.), Edward Elgar, Cheltenham
England and Northampton Mass., USA
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, 133-140
Blanchflower, D. and Bryson, A. (2007) "What Effect Do Unions Have on Wages Now and
Would "What Do Unions Do?" Be Surprised’ in James T. Bennett and Bruce E.
Kaufman (eds.), What Do Unions Do?: A Twenty-Year Perspective, pp. 79-113,
Transaction Publishers, New Brunswick USA and London UK
Blanchflower, D. G. and Bryson, A. (2009) ‘Trade union decline and the economics of the
workplace’, Chapter 3 in W. Brown, A. Bryson, J. Forth and K. Whitfield (eds.) The
Evolution of the Modern Workplace, pp. 48-73, Cambridge University Press
20
Booth, A. L. (1994) The Economics of the Trade Union, Cambridge University Press
Booth, A. L. and Bryan, M. L. (2004) "The union membership wage premium puzzle: is there
a free-rider problem?", Industrial and Labor Relations Review, 57, 3: 402-421
Breda, T. (2015) "Firms' Rents, Workers' Bargaining Power and the Union Wage Premium",
The Economic Journal, DOI: 10.1111/ecoj.12198
Brown, W., Bryson, A. and Forth, J. (2009) "Competition and the retreat from collective
bargaining", Chapter 2 in W. Brown, A. Bryson, J. Forth and K. Whitfield (eds.) The
Evolution of the Modern Workplace, pp. 22-47, Cambridge University Press
Bryson, A. (2002) The Union Membership Wage Premium: An Analysis Using Propensity
Score Matching, Centre for Economic Performance Discussion Paper No. 530,
London School of Economics
Bryson, A. (2004) ‘Union effects on workplace closure, 1990-1998’, British Journal of
Industrial Relations, Vol. 42, 2: 283-302
Bryson, A. (2014) "Union Wage Effects", IZA World of Labor, 35: 1-10
Bryson, A., Dale-Olsen, H. and Barth, E. (2015) "Do unions kill innovation?", UCL mimeo
Bryson, A., Dale-Olsen, H. and Nergaard, K. (2016) "Gender Differences in the Union Wage
Premium? A Comparative Study", UCL mimeo
Bryson, A., Forth, J. and Laroche, P. (2011) ‘Evolution or Revolution? The Impact of Unions
on Workplace Performance in Britain and France’, European Journal of Industrial
Relations, 17, 2: 171-187
Bryson, A. and White, M. (2016) 'Not So Dissatisfied After All? The Impact of Union
Coverage on Job Satisfaction', Oxford Economic Papers, doi: 10.1093/oep/gpw018
Card, D., Lemieux, T. and Riddell, C. W. (2007) "Unions and Wage Inequality", Chapter 5 in
J. T. Bennett and B. E. Kaufman (eds.) What Do Unions Do? A Twenty Year
Perspective, Transaction Publishers, London, pp. 114-159
DiNardo, J. and Lee, D. S. (2004) "Economic Impacts of New Unionization on Private Sector
Employers: 1984-2001", The Quarterly Journal of Economics, 119, 4: 1383-1441
Dunlop, J. (1944) Wage Determination Under Trade Unions, New York: MacMillan
Ferguson, J. P. (2008) "The Eyes of the Needles: A Sequential Model of Union Organizing
Drives, 1999-2004", Industrial and Labor Relations Review, 62, 1: 3-21
Forth, J. and Bryson, A. (2015) "Trade Union Membership and Influence", NIESR mimeo,
http://www.niesr.ac.uk/sites/default/files/publications/Forth%20and%20Bryson%20(2015)%20Trade
%20Union%20Membership%20and%20Influence%201999-2014.pdf
21
Frandsen, B. (2012) "Why Unions Still Matter: The Effects of Unionization on the
Distribution of Employee Earnings", MIT mimeo
Freeman, R. and J. Medoff (1984), What do Unions do? New York: Basic Books.
Frega, C.M. and Kelly, J. (2003) "Union revitalization strategies in comparative perspective",
European Journal of Industrial Relations, 9, 1: 7-24
Grout, P. A. (1984) “Investment and Wages in the Absence of Binding Contracts: A Nash
Bargaining Approach,” Econometrica, 52, 449 – 460.
Hicks, J. (1932) The Theory of Wages, New York: MacMillan
Hirsch, B. T. (2004), “What Do Union Do for Economic Performance?,” Journal of Labor
Research, 25, 415 – 455.
Hirsch, B. T. (2007) "What Do Unions Do For Economic Performance?", Chapter 7 in J. T.
Bennett and B. E. Kaufman (eds.) What Do Unions Do? A Twenty Year Perspective,
pp. 193-237, Transaction Publishers, New Brunswick
Hirschman, A. (1970) Exit, Voice and Loyalty. Cambridge, Harvard University Press.
Holden, S. and O. Raaum (1991) "Wage moderation and union structure", Oxford Economic
Papers, 43: 409-423
Lee, D.S., and Mas, A. (2012) "Long-run impacts of unions on firms: New evidence from
financial markets, 1961–1999", The Quarterly Journal of Economics, 127: 333-378
Lewis, H. G. (1963) Unionism and Relative Wages in the United States: An Empirical
Inquiry, University of Chicago Press, Chicago, IL, USA
Lewis, H. G. (1986) Union Relative Wage Effects: A Survey, University of Chicago Press,
Chicago, IL, USA
Mastekaasa, A. (2013), “Unionization and Certified Sickness Absence: Norwegian
Evidence,” Industrial and Labor Relations Review, 66, 117 – 141.
Metcalf, D. (1989) "Water notes dry up: the impact of the Donovan reform proposals and
Thatcherism at work on labour productivity in British manufacturing industry",
British Journal of Industrial Relations, 27, 1: 1-31
Metcalf, D. (2003), “Unions and Productivity, Financial Performance and Investments,” In
Addison, J. and C. Schnabel (eds.), International Handbook of Trade Unions.
Edward Elgar.
Moene, K. and M. Wallerstein (1997), “Pay Inequality,” Journal of Labor Economics, 15,
403 – 430.
Nergaard, K., Dale-Olsen, H. and Barth, E. (2015), «Lavere organisasjonsgrad, et spørsmål
om nykommere?», Søkelys på arbeidslivet, 32, 91-110.
22
OECD (2016) Economic Policy Reforms 2016: Going for Growth, OECD, Paris
http://www.keepeek.com/Digital-Asset-Management/oecd/economics/economic-policyreforms-2016/coverage-rates-of-collective-bargaining-agreements-and-trade-uniondensity-rates_growth-2016-graph54-en#page1
Pemberton, J. (1988) "A 'Managerial' Model of the Trade Union", The Economic Journal, 98,
392: 755-771
Pencavel, J. (2004) "The surprising retreat of union Britain", Chapter 5 in D. Card, R. Blundell
and R. B. Freeman (eds.) Seeking a Premier Economy: The Economic Effects of British
Economic Reforms, 1980-2000, University of Chicago Press, 181-232
Rosenfeld, J. (2014) What Unions No Longer Do, Harvard University Press
Slichter, S., Healy, J. and Livernash, R. (1960), The Impact of Collective Bargaining on
Management, Washington DC, The Brookings Institution.
Schnabel, C. (2012) "Union Membership and Density: Some (Not So) Stylized Facts and
Challenges", IZA Discussion Paper No. 6792
Schnabel, C. and Wagner, J. (2007) " Union density and determinants of union membership in 18
EU countries: evidence from micro data, 2002/03", Industrial Relations Journal, 38, 532
Sojourner, A. J., Frandsen, B. R., Town, R. J., Grabowski, D. C. and Chen, M. M. (2015)
"Impacts of Unionization on Quality and Productivity: Regression Discontinuity
Evidence from Nursing Homes", Industrial and Labor Relations Review, 68, 4: 771-806
Thoresen, T. O., T. E. Vattø, and K. O. Aarbu (2011), Tax changes and income responses of
wage earners. Quasi-experimental evaluation and labor supply model simulation
suggest that Norwegian are less responsive. Manuscript. Statistics Norway.
Thoresen, T. O., E. E. Bøe, E. Fjærli, and E. Halvorsen (2012), “A Suggestion for Evaluating
the Redistributional Effects of Tax Changes: With and application to the 2006
Norwegian Tax Reform,” Public Finance Review, 40, 303 – 338.
Vroman, S. (1990) "The union-nonunion wage differential and monitoring costs," Economics
Letters, Elsevier, vol. 32(4), pages 405-409, April.
Woolridge, J. M. (2009), “On estimating firm-level production functions using proxy
variables to control for unobservables,” Economics Letters, 104, 112–114.
23
24
Table 1 Subsidy of union membership for workers earning more than the Social Service’s
baseline figure.
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
A) Basic union deduction
Gross
0.90
0.90
1.45
1.80
union
deduction
B) Net union membership tax deduction
UD0.07
0.07
0.11
0.14
TB1
UD0.32
0.32
0.52
0.64
TB2
UD0.27
0.26
0.43
0.52
TB3
UD0.32
0.32
0.52
0.64
TB4
UD0.44
0.44
0.71
0.89
TB5
UD0.50
0.50
0.80
1.00
TB6
1.80
1.80
2.70
3.15
3.15
3.66
3.66
3.75
0.14
0.14
0.21
-
-
-
-
-
0.64
0.64
0.97
1.13
1.13
1.31
1.31
-
0.49
0.47
0.69
0.81
0.81
0.94
0.94
0.95
0.64
0.64
0.97
1.13
1.13
1.31
1.31
1.34
0.89
0.81
1.21
1.41
1.41
1.64
1.64
1.68
1.00
0.86
1.29
1.51
1.51
1.75
1.75
1.79
Note: Panel A) reports the gross union membership tax deductions. Panel B) reports the net union membership tax
deductions. Note that in Panel D) figures are only reported for incomes above the Social Service’s baseline figure. UD-TBn
denotes net union deductions in tax bracket n. Union deductions are measured in 1000 NOK. See Table A1 in the appendix
for information on tax brackets and marginal tax rates.
25
Table 2 The impact of subsidizing union membership on the probability of union membership
Model 1
Model 2
Model 3
Net union deduction
0.0009**
0.0012**
(1.27e-6)
(3.9e-6)
Δ Net union
0.0011**
deduction
(4.5e-6)
**
Tax bracket 1
-0.7971
(0.0045)
Tax bracket 2
-0.9155**
(0.0013)
Tax bracket 3
-0.3647**
(0.0013)
Tax bracket 5
-0.2811**
(0.0011)
Tax bracket 6
-0.7441**
(0.0019)
Additional controls:
Pseudo-R2
N(observations)
Marginal effects:
Net union deduction
Δ Net union
deduction
Year dummies (and intercept) included in all models
0.0133
0.0393
0.0016
28695942
28695942
28695942
0.0004**
0.0005**
0.0004**
Note: Method: Probit regressions. Dependent variable: I(union member) taking the value 1 if union member. Population: all
workers in Norway during the period 2001-2012 reporting taxable income year t and year t-1, t∈(2000,2012) above 1G (G is
the Social Service’s baseline figure). Standard errors are reported in parentheses. ** denote significant at the 1 percent level
of significance.
26
Table 3 Descriptive statistics. Firm-level.
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
lnVA t
8.11
(1.62)
lnC t
6.95
(2.21)
lnG t
8.45
(1.89)
lnLun t
1.45
(1.25)
lnLsk t
1.85
(1.33)
Uniont
0.26
(0.32)
PredUniont 0.55
(0.07)
TB1
0.01
(0.06)
TB2
0.10
(0.18)
TB3
0.07
(0.15)
TB4
0.43
(0.29)
TB5
0.38
(0.31)
TB6
0.01
(0.05)
O.margin
2.56
(%)
(10.9)
LnW
5.62
(0.55)
Worker
0.25
fixed effect (0.42)
N
6389
8.12
(1.62)
6.83
(2.60)
8.42
(1.88)
1.41
(1.24)
1.83
(1.34)
0.26
(0.31)
0.55
(0.06)
0.01
(0.04)
0.08
(0.16)
0.06
(0.14)
0.52
(0.30)
0.33
(0.30)
0.01
(0.06)
2.51
(11.0)
5.65
(0.57)
0.22
(0.40)
6461
8.15
(1.62)
6.80
(2.46)
8.42
(1.87)
1.39
(1.22)
1.86
(1.32)
0.26
(0.31)
0.54
(0.06)
0.01
(0.04)
0.07
(0.16)
0.07
(0.16)
0.55
(0.30)
0.30
(0.29)
0.01
(0.05)
2.68
(11.0)
5.66
(0.55)
0.20
(0.38)
6463
8.23
(1.60)
6.82
(2.47)
8.52
(1.87)
1.37
(1.20)
1.84
(1.32)
0.25
(0.31)
0.54
(0.06)
0.01
(0.05)
0.06
(0.15)
0.08
(0.16)
0.56
(0.30)
0.29
(0.28)
0.01
(0.06)
4.03
(10.8)
5.73
(0.58)
0.19
(0.39)
6477
8.28
(1.62)
6.85
(2.36)
8.60
(1.90)
1.38
(1.19)
1.87
(1.32)
0.25
(0.31)
0.53
(0.06)
0.01
(0.04)
0.03
(0.11)
0.08
(0.17)
0.63
(0.29)
0.23
(0.26)
0.01
(0.08)
3.94
(10.7)
5.77
(0.58)
0.19
(0.40)
6594
8.39
(1.65)
6.92
(2.31)
8.69
(1.94)
1.41
(1.19)
1.88
(1.32)
0.25
(0.31)
0.54
(0.05)
0.01
(0.06)
0.02
(0.08)
0.09
(0.18)
0.64
(0.29)
0.22
(0.26)
0.02
(0.10)
4.11
(11.3)
5.85
(0.61)
0.19
(0.41)
6691
8.50
(1.63)
7.02
(2.29)
8.80
(1.93)
1.45
(1.19)
1.91
(1.32)
0.24
(0.30)
0.54
(0.05)
0.00
(0.00)
0.02
(0.07)
0.08
(0.17)
0.58
(0.30)
0.27
(0.26)
0.05
(0.15)
4.79
(11.5)
5.95
(0.52)
0.19
(0.39)
6402
8.56
(1.63)
7.16
(2.31)
8.86
(1.94)
1.49
(1.19)
1.95
(1.32)
0.25
(0.30)
0.53
(0.05)
0.00
(0.03)
0.09
(0.18)
0.56
(0.30)
0.29
(0.27)
0.06
(0.15)
3.94
(12.1)
5.99
(0.53)
0.19
(0.42)
5993
8.46
(1.65)
7.19
(2.28)
8.75
(1.95)
1.43
(1.18)
1.93
(1.32)
0.24
(0.30)
0.54
(0.06)
0.00
(0.04)
0.08
(0.17)
0.57
(0.31)
0.29
(0.27)
0.06
(0.16)
1.74
(12.1)
5.97
(0.55)
0.18
(0.38)
5995
8.51
(1.64)
7.16
(2.29)
8.77
(1.94)
1.40
(1.15)
1.92
(1.31)
0.25
(0.30)
0.54
(0.06)
0.00
(0.02)
0.08
(0.17)
0.59
(0.31)
0.27
(0.27)
0.06
(0.16)
1.85
(12.1)
6.01
(0.55)
0.17
(0.37)
6153
8.57
(1.63)
7.17
(2.28)
8.85
(1.93)
1.39
(1.15)
1.91
(1.31)
0.24
(0.29)
0.53
(0.06)
0.00
(0.02)
0.08
(0.16)
0.59
(0.30)
0.26
(0.26)
0.06
(0.15)
2.68
(11.5)
6.06
(0.55)
0.18
(0.35)
5914
8.58
(1.70)
7.08
(2.34)
8.83
(1.99)
1.41
(1.16)
1.92
(1.34)
0.24
(0.30)
0.53
(0.06)
0.08
(0.17)
0.60
(0.31)
0.26
(0.26)
0.06
(0.16)
2.77
(11.7)
6.11
(0.55)
0.17
(0.35)
6042
Note: Population: Firms in Statistics Norway’s The Capital Data Base in industries with at least 200 observations (all within
Manufacturing) linked to individual worker information. Note that the disappearance of workers in the lowest tax brackets is
due to the imposed lower earnings limit (only jobs providing income above the Social Service’s baseline figure is included).
LnVA lnC and lnG denote log value added, log capital and log goods, respectively. lnL denotes log number of workers,
while superscript un and sk differentiate between unskilled and skilled workers. Union and PredUnion denote union density
and firm-average of the predicted union membership probability estimated from Model 2 in Table 2, respectively. TBn, n=16, denote share of workers within tax bracket n. O.margin and lnW denote firms’ operating margin and firm-average log
hourly wage, respectively. Worker fixed effect denotes the fixed worker effect from a worker-level log hourly wage
regression on year dummies(10) and age vigintile(19) dummies.
27
Table 4 The impact of union density on workplace productivity. Homogenous production
technology across industries.
Model 1
Model 2
Model 3
Model 4
Model 8
lnLsk
lnLunsk
lnC
Union
Method:
0.720**
(0.010)
0.254**
(0.008)
0.078**
(0.009)
-0.086**
(0.020)
0.636**
(0.007)
0.303**
(0.007)
0.054**
(0.007)
-0.063**
(0.014)
0.440**
(0.012)
0.226**
(0.008)
0.043**
(0.007)
0.029
(0.024)
0.446**
(0.011)
0.234**
(0.007)
0.039**
(0.007)
0.035
(0.023)
0.446**
(0.020)
0.232**
(0.009)
0.043**
(0.008)
0.653**
(0.159)
OLS
OLS
FE
FE
LPWGMMIV
First stage of Union:
PredUnion
0.461**
(0.043)
0.110**
(0.005)
0.032**
(0.005)
lnLsk t-1
lnLunsk t-1
Additional endogenous variables (instruments): lnLsk t, lnLunsk t (PredUnion, lnLsk t-1, lnLunsk t-1)
Kleibergen-Paap F:
Additional controls
Years(10)
Yes
Yes
Yes
Yes
Industry (130)
Yes
Yes
Lagged tax brackets
Yes
Yes
Fixed firm effects(6120)
Yes
Yes
Lagged ln wage cost per employee
Yes
Yes
Polynomial(lnG t-1, lnCt-1)
F
FxT
7106
42441
7106
42441
6120
41455
6120
41455
60.85
Yes
Yes
Yes
Yes
Yes
6748
40662
Note: Estimation of Cobb-Douglas production functions. Method: OLS, FE or GMM-IV (based on Woolridge’s
improvements on the method of Levinsohn and Petrin). Lagged unobserved productivity is approximated by a 3 rd order
polynomial. Robust standard errors adjusted for firm-level clustering are reported in parentheses. ** and * denote significant
at the 1 and 5 percent level of significance, respectively.
28
Table 5 The impact of union density on workplace productivity. Heterogeneous production
technology across industries.
Model 1
lnLsk
lnLunsk
lnC
A)Production technology
SIC2
0.415**(0.027)
0.158**(0.023)
0.118**(0.022)
10
**
**
0.344 (0.046)
0.304 (0.046)
0.017(0.015)
13
**
**
0.383 (0.079)
0.130 (0.062)
0.065**(0.026)
14
**
**
0.488 (0.024)
0.187 (0.023)
0.048**(0.011)
16
0.494**(0.030)
0.201**(0.021)
0.065**(0.010)
18
**
0.513 (0.096)
0.043(0.054)
0.184**(0.058)
20
0.439**(0.033)
0.149**(0.030)
0.109**(0.026)
22
**
**
0.312 (0.032)
0.182 (0.029)
0.038*(0.016)
23
**
**
0.422 (0.073)
0.169 (0.043)
0.093**(0.029)
24
0.510**(0.018)
0.194**(0.016)
0.037**(0.014)
25
**
0.721 (0.049)
-0.068(0.036)
0.055 (0.030)
26
**
0.562 (0.048)
0.050(0.038)
0.055**(0.020)
27
**
**
0.523 (0.025)
0.127 (0.021)
0.061**(0.016)
28
0.476**(0.046)
0.219**(0.046)
0.087 (0.058)
29
**
0.743 (0.032)
0.106**(0.036)
0.035 *(0.016)
30
0.423**(0.032)
0.185**(0.029)
0.043**(0.016)
31
**
**
0.501 (0.053)
0.143 (0.034)
0.030 (0.017)
32
Method:
Model 1
B) Residual IV regression
Union
0.937**(0.088)
UnionXHerfin.
Method:
IV
Years
Yes
Lagged tax brackets:
Lagged ln wage cost per employee
Skill groups
Lagged mean worker effect
Industry(3-digit)
First stage Union:
PredUnion
1.290**(0.058)
PredUnion X Herfin.
First stage UnionXHerfin:
PredUnion
PredUnion X Herfin.
Kleibergen-Paap F:
486.6
LPW-GMMIV
Model 2
0.729**(0.139)
IV
Yes
Yes
0.729**(0.047)
242.1
Model 3
Model 4
Model 5
0.364*(0.170)
0.858**(0.164)
IV
Yes
Yes
Yes
Yes
IV
Yes
Yes
0.910**(0.178)
-2.514**(1.241)
IV
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
0.522**(0.048)
0.609**(0.048)
0.565**(0.077)
1.888 (2.206)
113.6
161.3
-0.034x (0.017)
1.850 (0.704)
77.18
Note: Panel A): Estimation of Cobb-Douglas production functions. Method: GMM-IV (based on Wooldridge’s
improvements on the method of Levinsohn and Petrin). Dependent variable: ln(value added). Each row reports results
separately for 2-digit industries. Panel B). IV-regressions of the residual from GMM-IV-regressions of Panel A) on union
density, where union density is instrumented by the average predicted union density based on tax subsidies (see text). Herfin
denotes the Herfindahl index based on operating revenues and calculated for 2-digit industries. Lagged mean worker effect is
the firm average of the estimated fixed worker effect from a worker-level log hourly wage regression on year dummies(10)
and age vigintile(19) dummies. Robust standard errors adjusted for firm-level clustering are reported in parentheses. ** and *
denote significant at the 1 and 5 percent level of significance, respectively.
29
Table 6 The impact of union density on log hourly wages.
Model 1 Model 2 Model 3 Model 4
Union
Model 5
Model 6
0.031**
(0.004)
0.017**
(0.007)
-0.039**
(0.003)
0.031**
(0.003)
0.170**
(0.026)
-0.002
(0.042)
0.024
(0.016)
0.222*
(0.113)
-0.085
(0.054)
0.064**
(0.016)
IV
FE-IV
OLS
IV
FE-IV
2.001**
(0.139)
0.479**
(0.051)
1.964**
(0.136)
-0.662**
(0.222)
0.482**
(0.052)
0.084
(0.081)
2.001**
(0.139)
0.479**
(0.051)
0.075
(0.160)
2.392**
(0.701)
0.007
(0.121)
2.271**
(0.706)
208.76
88.35
64.75
43.10
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
6462
39058
6462
39058
6462
39058
6462
39058
6462
39058
6462
39058
0.033**
(0.004)
0.017**
(0.007)
0.000
(0.004)
0.131**
(0.032)
0.130**
(0.026)
0.129**
(0.030)
0.257**
(0.118)
0.222**
(0.113)
0.186x
(0.112)
0.015**
(0.003)
0.130**
(0.026)
0.238*
(0.018)
0.017**
(0.003)
0.023**
(0.003)
OLS
Union X TFP
TFP
Method:
First stage of Union:
PredUnion
PredUnion X TFP
First stage of Union X TFP:
PredUnion
PredUnion X TFP
Weak instrument
Kleibergen-Paap F:
Additional controls
Years(10)
Industry (130)
Lagged tax brackets
Fixed firm effects
F
FxT
Yes
Yes
Yes
Yes
Differential impact at different productivity levels
-1:
-
-
-
0:
-
-
-
1:
-
-
-
Note: Log hourly wages express the firm average of the residuals from a log hourly wage regression on year dummies(10),
age vigintile(19) dummies and fixed worker effects. Union density and TFP express deviation from mean. Method: OLS, FE,
IV and IV-FE. When analyzing differential union density impact for different productivity levels, -1 and 1 denote -1
standard deviation in the TFP-distribution from mean and +1 standard deviation in the TFP-distribution from mean,
respectively. Each observation is weighted by the number of wage observations per firm. Robust standard errors adjusted for
firm-level clustering are reported in parentheses. **, * and x denote significant at the 1, 5 and 10 percent level of significance,
respectively.
30
Table A1 Tax brackets and marginal tax rates for workers earning more than the Social
Service’s baseline figure.
2001
2002
A) Tax brackets
TB1-UL
60.6
60.6
TB2-UL
144.5 138.3
TB3-UL
183.1 179.2
TB4-UL
289.0
320
TB5-UL
793.0
830
TB6-UL
>793 >830
B) Marginal tax rates
Mtr_TB1
7.8
7.8
Mtr_TB2
35.8
35.8
Mtr_TB3
29.6
29.4
Mtr_TB4
35.8
35.8
Mtr_TB5
49.3
49.3
Mtr_TB6
55.3
55.3
2003
2004
2005
2006
2007
2008
63.4
132.5
190.4
347
872
>872
64.7
132.5
197.9
354.3
906.9
>906.9
66.0
110
185.1
381
800
>800
70
100
180
400
760
>760
66
90
76.9
78.9
80.3
179.1 186.1 195.4 202.2
400
420
441 456.4
650 682.5 716.6 741.7
>650 >682.5 >716.6 >741.7
7.8
35.8
29.4
35.8
49.3
55.3
7.8
35.8
29.1
35.8
49.3
55.3
7.8
35.8
27.1
35.8
49.3
55.3
7.8
35.8
26.3
35.8
44.8
47.8
7.8
35.8
25.7
35.8
44.8
47.8
35.8
25.7
35.8
44.8
47.8
2009
35.8
25.7
35.8
44.8
47.8
2010
35.8
25.7
35.8
44.8
47.8
2011
2012
81.7
208.8 205.6
471.2 490
765.8 796.4
>765.8 >796.4
35.8
25.7
35.7
44.8
47.8
25.2
35.8
44.8
47.8
Note: Panel A) reports the income tax brackets. Panel B) reports the marginal tax rates within tax brackets. Note that in
Panels A)-B) figures are only reported for incomes above the Social Service’s baseline figure. TBn-UL denote the upper
limit of a tax bracket n. MTR_TBn then denotes the marginal tax rate of tax bracket n. Income thresholds are measured in
1000 NOK. Marginal tax rates are measured in percent.
31
Table A2 Average yearly union membership fees for union members over time
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Average
union fee
UFTB1
UFTB2
UFTB3
UFTB4
UFTB5
UFTB6
3.43
3.58
3.74
3.86
3.99
4.06
4.24
4.36
4.51
4.64
4.82
4.98
1.40
1.40
1.39
1.39
1.46
1.47
1.59
-
-
-
-
-
2.05
1.98
1.95
1.88
1.67
1.60
1.66
1.68
1.59
1.60
1.60
-
2.61
2.60
2.64
2.62
2.47
2.33
2.31
2.29
2.30
2.32
2.37
2.34
3.42
3.61
3.80
3.90
4.06
4.10
4.15
4.27
4.43
4.55
4.74
4.88
3.90
4.08
4.24
4.42
4.54
4.64
4.92
5.03
5.17
5.31
5.52
5.57
3.38
3.34
3.45
3.69
4.07
4.50
4.57
4.50
4.62
4.84
5.02
5.14
Note: The table reports average year union membership fee reported to the tax authorities (overall and within the tax
brackets). UF-TBn denotes average union membership fee for workers within tax bracket n. Union fees are measured in
1000 NOK.
32
Figure 1 Gross union tax deduction and political governance
4
0.4
3.5
0.35
3
0.3
2.5
0.25
2
0.2
1.5
0.15
1
0.1
0.5
0
2001
0.05
2002
2003
2004
2005
Union deduction
2006
2007
2008
Labour party
2009
2010
2011
0
2012
Conservatives
Note: Gross union tax deduction (in 1000 NOK) is measured on the left-hand side axis. The labour party’s and the
conservative party’s elected number of seats in the parliament are measured (in percent of 169) on the right-hand side axis.
Before October 2005 a liberal-conservative coalition had governmental power. In the October 2005 election the Labour party
gained governmental power, which they kept in the election of 2009.
33
Figure 2 Net union tax deductions over time
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
2001
2002
2003
2004
2005
2006
2007
2008
2009
T1
T2
T3
T4
T5
T6
2010
2011
2012
Note: T1-T6 denote tax brackets. See Tables 1 and A1 for more information. Note T2 and T4 are equal except for 2012.
34
Figure A1 Marginal tax rates and taxable income
60
50
40
30
20
10
0
50
150
250
2001
350
2003
450
550
2005
650
2007
750
2009
850
950
2011
Note: The figure show the marginal tax rate for selected years for income between 50000 NOK and 1000000 NOK. The
income scale is measured in 1000 NOK.
35
Figure A2 Net union tax deductions relative to average yearly union membership fees
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
2001
2002
2003
2004
2005
2006
2007
2008
2009
T1
T2
T3
T4
T5
T6
Note: T1-T6 denote tax brackets. See Tables 1 and A1-A2 for more information.
36
2010
2011
2012