You are on page 1of 28

British Journal of Industrial Relations doi: 10.1111/j.1467-8543.2008.00699.

x
46:4 December 2008 0007–1080 pp. 674–701

Is There a Public Sector Training


Advantage? Evidence from the
Workplace Employment Relations Survey
Philip Murphy, Paul L. Latreille, Melanie Jones
and David Blackaby

Abstract

Using matched employer–employee data from the Workplace Employment


Relations Survey (2004), we find a significant training ‘advantage’ exists for
public sector workers over private sector workers even after accounting for
differences in the composition of the two workforces. This finding is robust to all
but one change in specification, designed to account for worker sorting effects
which can lead to unobserved workplace-based effects being correlated with
individual worker characteristics. Using the average characteristics of workers
within an establishment as a control for these sorting effects all but eliminates
the estimated public sector training advantage, which has otherwise been an
empirical regularity of many individual-based training models.

1. Introduction

The provision of world-class public services remains at the heart of the


current government’s political strategy. It has been reflected in recent spend-
ing reviews, which have substantially increased expenditure on public services
and reversed what had previously been a long and steady decline in public
sector employment. Almost inevitably, as investment in public services has
increased, it has reawakened interest in how labour outcomes in the public
sector compare with those in the private sector. Although various reforms
introduced during the last 25 years have sought to make the public sector
more competitive and more responsive to market forces,1 important differ-
ences still remain between the two sectors that affect the relative terms and
conditions under which public sector workers are employed.

Philip Murphy, Paul L. Latreille, Melanie Jones and David Blackaby are at School of Business
and Economics, Swansea University.

© Blackwell Publishing Ltd/London School of Economics 2008. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.
Public Sector Training Advantage 675
To date, most of the research examining differences between the public and
private sectors has tended to focus on pay. In contrast, far less attention has
been paid to other aspects of employment that vary between the sectors, and
in particular, to differences in the incidence of employer-provided training in
the public and private sectors (an exception is Latreille et al. 2005). This is
surprising since, although not universally reported, it is now something of an
empirical regularity that irrespective of whether the unit of analysis is the
individual or the workplace/firm, the incidence of employer-provided train-
ing in the UK is higher in the public sector than in the private sector (see
Arulampalam and Booth 1997, 1998; Green et al. 1999). This phenomenon
is also evident in the Workplace Employment Relations Survey 2004
(WERS2004), where some 77 per cent of public sector employees received
non-Health and Safety (H&S) employer-provided training in the previous
12 months, compared with just 61 per cent of private sector employees. The
present article looks more closely at this issue, examining in particular
whether differences in training incidence between the two sectors are the
result of differences in the composition of the workforces and the establish-
ments in which they work, or whether public sector employees enjoy a
training ‘advantage’ relative to otherwise comparable private sector workers.
The analysis in this article is based on WERS2004, and as such, we are able
to control for a wide range of workplace-based variables, which, in addition
to the usual suspects, include measures of financial performance, man-
agement relations and whether training is included in the scope of union
bargaining. We are also able to include a measure of whether workplace
monitors equal opportunities, which allows us to examine the effect that such
policies have on employees’ access to employer-provided training. An impor-
tant feature of WERS2004 is that it is one of the few datasets currently
available that matches information on individual employees to their place of
work. This enables the analysis to be undertaken at the individual employee
level while allowing both observed and unobserved establishment heteroge-
neity to be taken into account. To our knowledge, this is the first time that
such an analysis has been undertaken for Britain.
The outline for the rest of the article is as follows. Section 2 provides some
background on why training incidences might be higher in the public sector
than in the private sector. Section 3 describes the methodology and data used
in the empirical analysis. The main results are discussed in Section 4 and
some conclusions given in Section 5.

2. Background

Table 1 shows the incidence of training in WERS2004 by duration band for


public and private sector workers separately. The table distinguishes between
workers who have not received any employer-provided (non-H&S) training
in the last 12 months and the incidence of those receiving ‘less than one day’,
‘one to less than two days’, ‘two to less than five days’, ‘five days to less than
© Blackwell Publishing Ltd/London School of Economics 2008.
676 British Journal of Industrial Relations
TABLE 1
Incidence of Employer-Provided Training in the Last 12
Months by Duration and Sector

Public Sector Private Sector

No training received 23.1 39.1


Less than 1 day 7.9 10.0
1 day to less than 2 days 17.2 14.5
2 days to less than 5 days 27.5 19.6
5 days to less than 10 days 13.7 8.8
10 days or more 10.1 7.9
Unweighted observations 5,587 12,544

Source: WERS2004.

10 days’, and ‘10 days or more’. It shows that both the incidence and duration
of training are higher in the public sector than in the private sector.2 While the
distribution of training across these durations is of interest, this is not the
focus of the present article, which instead concentrates on differences in
training incidences between the public and private sectors.
There are several reasons why the incidence of training might be higher
in the public sector. Human capital theory, for example, suggests that less
investment in (specific) training will occur where the period over which the
returns to training can be amortized is shorter, and typically job tenure tends
to be shorter in the private than in the public sector. Alternatively, a higher
rate of technological or organizational innovation in the public sector could
require more training in the sector to allow workers to familiarize themselves
with new procedures and processes.
Although these and other workforce composition effects may account for
part of the observed public–private sector training gap, it would be surprising
if such differences provided a complete explanation. An intriguing possibility
emerges, therefore, that fundamental differences in practice in the two sectors
accounts for part of the difference, allowing workers in the public sector to
enjoy training ‘advantage’ over otherwise comparable workers in the private
sector.
Differences between the public and private sectors that affect the willing-
ness of employers in the public sector to train more employees can arise from
a number of different sources. First, if there are social externalities to train-
ing, and it is difficult to guarantee the quality of publicly subsidized training
in the private sector, the government may intervene to optimize social welfare
by increasing training in the public sector. Under-provision of training can
also occur where training is of a general type and workers face capital
constraints which restrict their ability to fund their own training and private
sector employers are unwilling to finance training because of poaching fears
(Becker 1975). In this situation, the government may step in to fill the training
deficit, much as it currently does in providing general training for both health
professionals and teachers.
Second, the private sector may train less than the public sector because it is
more likely to be constrained by (short-run) profit considerations (Booth
© Blackwell Publishing Ltd/London School of Economics 2008.
Public Sector Training Advantage 677
TABLE 2
Measure of Financial Performance by Sector: Percentage of
Sampled Employees Covered by Each Measure

Public Sector Private Sector

Profit 5.2 56.8


Value added 19.1 6.9
Sales 0.4 10.6
Fees 0.4 1.9
Budget 53.7 16.4
Costs 6.3 4.7
Expenditure 5.8 0.9
Other 9.1 1.8
Unweighted observations 5,587 12,554

Source: WERS2004.

1991). Bureaucrats and politicians in the public sector, for example, are more
likely to adopt budget maximization rather than profit maximization as an
objective, and this allows them the flexibility they need to follow non-profit
related objectives (see Dixit 1997; Niskanen 1971, 1975; Reder 1975; Tirole
1994).
WERS data show that the majority of workers in the public sector are
employed in workplaces where managers identified ‘budget’ as being the most
appropriate measure of financial performance, whereas in the private sector
it was ‘profit’ (Table 2).3 Although it continues to be difficult to assign
particular behavioural objectives to the workplaces in which individuals
work, these data nevertheless provide some support for the argument that
behavioural objectives differ systematically between the public and private
sectors, and therefore have the potential to affect training decisions.
Third, even where profit considerations determine employer behaviour, the
public sector may be more likely to undertake and pay for general training if
the government is able to appropriate a share of the productivity gains
produced by training from income tax, which accrue irrespective of whether an
individual remains in public sector employment.4 Mehaut and Perez (2004)
have also suggested that increased job security in the public sector reduces the
risk of poaching and so increases the incentive within the sector to train
employees. Similarly, the higher job security enjoyed by public sector workers
may increase their demand for job adaptation and career progression training.
In fact, the more rigid hierarchical nature of wages and job tasks found in the
public sector actually lend themselves to employers designing incremental
wage scales for well-defined job tasks, which provide workers with an incentive
to acquire firm-specific human capital (see Prendergast 1993).
Fourth, Acemoglu and Pischke (1999) have shown that firms are willing to
invest in general training where they are able to exploit transaction costs
associated with workers’ mobility, which are related to job search imperfec-
tions. Since these costs are likely to increase with the level of training, rents to
the firm increase with training, and as a result wages tend to increase with
training less than productivity. This leads to what Acemoglu and Pischke
© Blackwell Publishing Ltd/London School of Economics 2008.
678 British Journal of Industrial Relations
refer to as a compressed wage structure. As the public sector wage distribu-
tion is more compressed than the private sector, we might therefore expect to
see higher training incidences in this sector, as employers are able to keep a
greater share of the rents generated from training.
Fifth, if training is seen as just another form of rent-sharing between
employers and employees, training could be seen as a more attractive option
in the public sector where wages may be capped either as a result of
nationally-bargained pay scales or because of the imposition of cash limits.
The use of training as a form of rent sharing naturally raises the issue of
employee bargaining power and whether training is included in the scope of
bargaining at the workplace. To capture this latter effect, a measure indicat-
ing whether management negotiates with unions about training is included in
the analysis. However, the ability of workplace collective bargaining mea-
sures to capture rent-sharing effects in the public sector is somewhat limited
because of the importance of nationally negotiated agreements in the sector.
Finally, higher training could be observed in the public sector as a result
of worker sorting, where workers sort themselves into public and private
sector establishments according to unobserved characteristics that predis-
pose certain workers towards receiving employer-based training. Elsewhere
Abowd et al. (1999) have shown that worker sorting according to unobserved
individual heterogeneity explains about ‘90 per cent of inter-industry wage
differentials and about 75 per cent of the firm-size wage effect’ in a longitu-
dinal sample of 1 million French workers.

3. Methodology and data

Methodology
The training model used is similar to that outlined by Booth (1991) and
Green (1993a). Specifically, an individual receives training whenever the net
benefit to training perceived by the provider is greater than zero. As the net
benefits to training are not normally directly observed by the provider, the
model is given by:

Z*pub ij α
= X pub + ηijpub
pub
ij
(1)
Z*prv ij α
= X prv + ηijprv
prv
ij

⎧I ij = 1 if Z ij ≥ 0 ⎫ ⎧I ij = 1 if Z ij ≥ 0 ⎫
pub *pub prv *prv

⎨ pub ⎬ and ⎨ prv ⎬ (2)


⎩I ij = 0 if Z ij < 0 ⎭ ⎩I ij = 0 if Z ij < 0 ⎭
*pub *prv

where Z* is a latent variable measuring the net benefit to training perceived


by the provider, X is a vector of individual and workplace level variables
which affect the perceived returns to training, a is a vector of conformable
coefficients, I is an observed binary variable taking the value 1 if employer
© Blackwell Publishing Ltd/London School of Economics 2008.
Public Sector Training Advantage 679
training has been received by an employee in the last 12 months and 0
otherwise,5 superscripts pub and prv refer to employment in the public and
private sectors, respectively, and the subscript ij refer to the ith individual in
a sector employed in the jth establishment. The panel element of the data is
captured by including an unobserved workplace-level heterogeneity term in
the error term defined in (1). Specifically, the errors are assumed to have the
following structure:

ηij = μ j + ε ij (3)

where eij is a standard error term and mj is an unobserved random variable that
is constant for each establishment.6 The nature of the data, therefore, allow us
to control for unobserved workplace heterogeneity and estimates of a in
expressions (1) and (2) are obtained from a standard random effects logit
model.
Following Gomulka and Stern (1990) and Fairlie (1999, 2005), an estimate
of the public–private sector training gap is found from comparing the inci-
dence of training in the two sectors. This is represented by:

I pub − I prv = P ( X pub


i α
ˆ pub ) − P ( X prv
i α
ˆ prv ) (4)
where
1 1
I pub = ∑ P ( X pub
n1 i
i α
ˆ pub ) = P ( X pub
i α
ˆ pub ) I prv = ∑ P ( X prv
n2 i
i α
ˆ prv ) = P ( X prv
i α̂
prv
);
¯
I is the incidence of employer provided training in the public (pub) or private
sectors (prv); P(.) is an estimate of the probability that the ith individual has
received employer provided training; ¯ P(.) is an average of P(.) for each sector;
and n1 and n2 are the number of individuals in the public and private sectors,
respectively.7 The incidence of training in each sector, therefore, is influenced
by the composition of the workforce (X) and the way in which particular
characteristics in a sector influence the chance of a worker receiving training
(a).
It follows that any training gap between the two sectors can be shown to be
equal to the sum of two components: a characteristic component which
shows the effect of composition differences between public and private sector
workers and their effect on training propensities; and a coefficient component
which measures the extent of the public sector training advantage or dis-
advantage, and shows how the propensity to receive training for otherwise
comparable workers differs between the two sectors. Standard procedures
are available for identifying these two separate components. Formally the
public–private sector training gap can be shown to be equal to:8

I pub − I prv = [P ( X pub


i α
ˆ prv ) − P ( X prv
i α
ˆ prv )] + [P ( X pub
i α
ˆ pub ) − P ( X pi ubα̂ prv )] (5)

where α̂ prv and α̂ pub are estimates of the structure determining training
provision in the private and public sectors, respectively.

© Blackwell Publishing Ltd/London School of Economics 2008.


680 British Journal of Industrial Relations
The first term [P ( X pub ˆ prv ) − P ( X prv
i α i α
ˆ prv )] represents the difference in
training incidence attributable to differences in the characteristics and
working conditions of individuals employed in the public and private sectors
— the characteristic component is based on evaluating differences in training
propensities according to the determinants of training in the private sector
(α̂ prv ). Consequently, if public sector workers are endowed with characteris-
tics, or work in workplaces, which are more conducive to the receipt of
training than private sector workers, then the characteristic effect will reflect
this and be positive. The second term [P ( X pub i α
ˆ pub ) − P ( X pub
i α
ˆ prv )] is the part
of the public–private sector training gap, which is explained by differences in
the determinants of training in the two sectors — the coefficient component.
This shows the difference in training that would arise if otherwise equivalent
public sector workers were employed in the private sector, and therefore
measures the training advantage or disadvantage that public sector workers
enjoy over otherwise comparable private sector workers.
The decomposition shown above is typically referred to as being based
on the private sector because it uses coefficients estimated for the private
sector (α̂ prv ) to evaluate the size of the characteristic component in the
decomposition. Of course, it is also possible to write the decomposition in a
slightly different way and evaluate the size of the characteristic component
using public sector coefficients (α̂ pub ) instead.9 As many of the reforms
recently introduced into the public sector were designed to make the public
sector behave more like the private sector, particularly in terms of how the
sector responds to market conditions, we decided to base the decomposi-
tions reported in the article on the private sector training structure.
However, because the results of the decomposition can differ depending on
whether private or public sector training coefficients are used as the base, we
tested the sensitivity of the results to this choice by reporting decompositions
based on both public and private sector training structures in Section 4. We
also test the sensitivity of the results to several changes in specification,
including, among other things, an attempt to control for worker sorting
effects.
Worker sorting effects have the potential to introduce a bias in the standard
random effects model, particularly where worker sorting results in worker
characteristics being associated with unobserved workplace heterogeneity
(see Abowd et al. 1999). To address this issue, the model is re-estimated
using a Chamberlain-Mundlak transformation (Chamberlain 1980; Mundlak
1978). The transformation specifies a distribution function for mj, which
allows for dependency between unobserved workplace heterogeneity (mj) and
individual worker characteristics (Xij) to be accounted for in the estimation
procedure. Specifically, mj = gX̄j + nj, where nj ~ N(0,sn) and X̄j is a vector of
mean individual characteristics for workers in the jth establishment.10
Without wishing to pre-empt the discussion of this sensitivity analysis, we
find the public sector training advantage to be robust to all but one change in
specification. The exception was that designed to capture the effects of
worker sorting.
© Blackwell Publishing Ltd/London School of Economics 2008.
Public Sector Training Advantage 681
Data

WERS2004 is a stratified random sample of just under 2,300 establishments


taken from the Inter-Departmental Business Register (IDBR) maintained by
the Office for National Statistics (ONS) and includes workplaces of all sizes
except those with fewer than five employees.
Information is taken from two (matched) components of WERS. First,
firm/workplace characteristics are derived from the management question-
naire, which is completed by a senior manager dealing with personnel or
employment relations at the establishment. Second, information relating to
individual employees is taken from an employee questionnaire, which was
completed by a random sample of employees from each establishment. For
establishments with fewer than 25 employees, all employees were asked to
complete the employee questionnaire.
The covariates included in the analysis reflect the richness of the data and
include measures of an individual’s human capital (e.g. age, education,
workplace-tenure and occupation); household background (e.g. marital
status and the presence of dependent children aged under four); employ-
ment contract (e.g. permanent contract, and full-time/part-time working);
union membership; and workplace environment (e.g. establishment size,
employment downsizing, the use of computers, whether formal bargaining
arrangements include training issues, management–employee relations, rela-
tive financial performance, worker involvement as measured by the exist-
ence of quality circles and joint consultative committees, and whether the
establishment monitored the effects of any equal opportunities policy). The
means of the major variables used in the main analysis are reported in
Table 3. These are based on individual employee information on 5,587 and
12,554 public and private workers respectively (information on 4,144
employees was lost because of missing data or other data restrictions), and
are reported separately for both public and private sector workers as well as
within sector according to whether employees had received employer-
provided training in the last 12 months (Train = 1) or not (Train = 0). The
public sector is defined to include workplaces in which the formal status of
the establishment (or the organization of which it was part) were described
by the management respondent as being either a government-owned limited
company, nationalized industry or trading public corporation; a public
service agency; other non-trading public corporations; a quasi autonomous
national government organization; or local or central government, includ-
ing the National Health Service and Local Education Authorities. This defi-
nition, therefore, is similar to that used in the Labour Force Survey, which
assigns individuals to the public sector according to whether the activities of
the workplace in which they are employed are owned, funded or run by
central or local government.
Table 3 shows that the incidence of training in the public sector is higher
than that in the private sector — 77 per cent compared to 61 per cent.
Table 3 also shows that not only are the average characteristics of workers
© Blackwell Publishing Ltd/London School of Economics 2008.
682

TABLE 3
Public and Private Sector: Means of Variables by Sector and Training Status

Public Sector Private Sector

All Train = 1 Train = 0 All Train = 1 Train = 0

Individual employee variables constructed from the employee


questionnaire
Training 0.765 0.609
Age 16–21 0.021 0.021 0.022 0.077 0.080 0.073
Age 22–29 0.123 0.139 0.072 0.179 0.205 0.137
Age 30–39 0.243 0.244 0.242 0.264 0.271 0.254
Age 40–49 0.319 0.318 0.322 0.245 0.244 0.248
Age 50 or more 0.294 0.278 0.342 0.235 0.200 0.288
Use computer at work 0.813 0.876 0.608 0.728 0.795 0.623
Part-time 0.300 0.261 0.426 0.219 0.197 0.253
Permanent 0.904 0.906 0.897 0.931 0.935 0.925
Female 0.653 0.659 0.631 0.471 0.477 0.462
Children under 4 years old 0.113 0.112 0.119 0.126 0.122 0.132
British Journal of Industrial Relations

Female ¥ children under 4 years old 0.062 0.061 0.065 0.047 0.044 0.052
Single 0.170 0.180 0.139 0.253 0.268 0.228
Married 0.719 0.710 0.748 0.657 0.647 0.673
Widowed/separated/divorced 0.111 0.110 0.113 0.090 0.085 0.099
Degree or higher 0.343 0.386 0.205 0.251 0.302 0.171
A-level 0.152 0.160 0.125 0.144 0.158 0.123
O-level 0.324 0.317 0.344 0.367 0.358 0.382
Other qualifications 0.065 0.056 0.094 0.066 0.062 0.073
No qualifications 0.116 0.081 0.232 0.172 0.120 0.251
Union member 0.636 0.656 0.570 0.237 0.244 0.228
Workplace tenure less than 1 year 0.121 0.127 0.102 0.174 0.188 0.152
Workplace tenure 1 to less than 2 years 0.127 0.140 0.087 0.130 0.146 0.106
Workplace tenure 2 to less than 5 years 0.257 0.255 0.265 0.279 0.284 0.272
Workplace tenure 5 to less than 10 years 0.180 0.177 0.191 0.187 0.180 0.198

© Blackwell Publishing Ltd/London School of Economics 2008.


Workplace tenure 10 years or more 0.315 0.301 0.355 0.230 0.202 0.272
Manager or senior official occupations 0.069 0.076 0.042 0.132 0.159 0.091
Professional occupations 0.199 0.228 0.105 0.086 0.099 0.067
Associate professional and technical occupations 0.228 0.256 0.136 0.146 0.168 0.112
Administrative and secretarial occupations 0.211 0.206 0.227 0.179 0.177 0.182
Skilled trade occupations 0.025 0.019 0.043 0.087 0.067 0.118
Personal service occupations 0.138 0.135 0.148 0.058 0.071 0.040
Sales and customer service occupations 0.008 0.008 0.008 0.096 0.100 0.091
Process plant and machine operative occupations 0.018 0.013 0.035 0.101 0.072 0.146
Elementary occupations 0.104 0.059 0.256 0.115 0.087 0.153
Workplace variables constructed from the management
questionnaire
Downsized in the last year 0.283 0.283 0.283 0.413 0.398 0.436
Establishment size less than 25 0.116 0.116 0.115 0.194 0.176 0.222
Establishment size 25–50 0.179 0.178 0.183 0.148 0.137 0.165
Establishment size 51–100 0.162 0.167 0.148 0.153 0.161 0.140
Establishment size 101–200 0.168 0.165 0.177 0.159 0.159 0.159
Establishment size 201–500 0.150 0.146 0.161 0.168 0.170 0.165
Establishment size 501–1000 0.077 0.082 0.063 0.093 0.099 0.086

© Blackwell Publishing Ltd/London School of Economics 2008.


Establishment size 1001 or more 0.148 0.146 0.153 0.085 0.098 0.063
Joint consultative committee 0.549 0.554 0.535 0.368 0.391 0.333
Quality circles 0.451 0.455 0.438 0.375 0.397 0.342
Scope of bargaining includes training 0.088 0.085 0.095 0.018 0.021 0.013
Management relations good 0.906 0.910 0.890 0.895 0.905 0.879
Management relations poor 0.013 0.012 0.017 0.013 0.012 0.015
Financial position better than average in sector 0.418 0.416 0.422 0.530 0.557 0.487
Financial position worse than average in sector 0.050 0.049 0.056 0.084 0.078 0.092
Monitors effects of equal opportunities policy 0.383 0.393 0.351 0.164 0.194 0.118
Technical change 0.873 0.873 0.872 0.852 0.872 0.820
Organizational change 0.834 0.832 0.838 0.679 0.708 0.633
Unweighted observations 5,587 4,272 1,315 12,554 7,644 4,910
Public Sector Training Advantage

Notes: Train = 1, sample restricted to individuals who have received employer-provided training in the last 12 months; Train = 0, otherwise.
683
684 British Journal of Industrial Relations
in the public and private sectors different, but also that the incidence of
training in the two sectors varies according to workplace and worker char-
acteristics. For example, public sector workers tend to be older and better
qualified, are more likely to work with computers, and are more likely to
have been employed in the same workplace for longer than private sector
workers. Similarly, the incidence of female employment, non-manual
employment, and union membership are all higher in the public sector, as
is the proportion of employees working in establishments with a monitored
equal opportunities policy, quality circles and joint consultative commit-
tees. Many of these features are also associated with a higher incidence of
training, and while the impact that these variables have on training often
seem more pronounced in the private sector (particularly for many of the
establishment-based characteristics included in the analysis), the data does
suggest that there is a link between training and employee characteristics
which is consistent with the human capital model. Thus, the relative inci-
dence of training tends to be higher among full-time and more highly edu-
cated workers. Similarly, older workers seem less likely to have received
training, as are workers with 10 years or more of job tenure. The use of
computers also appears to be an important driver of training in both the
public and private sectors. However, while an examination of these means
is informative, because they cannot control for the influence of other
factors they cannot on their own provide an accurate picture of the deter-
minants of training in the two sectors; for this we need to use multivariate
analysis, and it is to these estimates that we now turn.

4. Results

Training Estimates
Within sample estimates of the training model are reported in Table 4. Esti-
mates of a baseline specification based on workers employed in the public
(α̂ pub ) and private (α̂ prv ) sectors are reported separately, columns 2 and 3
show results for the public sector and columns 4 and 5 for the private sector.
Coefficient estimates are reported in the first column for each sector, while
marginal effects are reported in the second column. The coefficient estimates
indicate the direction in which a variable influences the probability of an
individual receiving employer-provided training, while marginal effects
quantify the response in probability terms.11
Based on a likelihood ratio test, the estimate of r, which is a measure of the
proportion of the total variance accounted for by the workplace variance
component, reported in the second column of Table 4 unambiguously rejects
the hypothesis that r = 0. The random effects model is therefore preferred to
a pooled logit model, which fails to account for unobserved workplace-level
heterogeneity. The estimate of r also indicates that the proportion of the total
error variance accounted for by unobserved workplace-level heterogeneity is
© Blackwell Publishing Ltd/London School of Economics 2008.
TABLE 4
Baseline Training Estimates: Random Effects Model

Public sector Private sector

Coefficients Marginal effect Coefficients Marginal effect

Individual employee variables


Age 16–21 -0.1641 -0.0258 0.4462*** 0.0971***
(0.273) (0.045) (0.114) (0.023)
Age 22–29 0.3148* 0.0437** 0.4323*** 0.0957***
(0.162) (0.021) (0.085) (0.018)
Age 30–39 -0.0691 -0.0105 0.2246*** 0.0512***
(0.1163) (0.018) (0.071) (0.016)
Age 40–49 -0.0647 -0.0098 0.1500** 0.0343**
(0.096) (0.015) (0.064) (0.015)
Use computer at work 0.8808*** 0.1550*** 0.4881*** 0.1157***
(0.127) (0.026) (0.064) (0.015)
Part time -0.5462*** -0.0876*** -0.2845*** -0.0671***
(0.092) (0.016) (0.062) (0.015)

© Blackwell Publishing Ltd/London School of Economics 2008.


Permanent 0.3776*** 0.0621*** 0.3725*** 0.0893***
(0.131) (0.023) (0.091) (0.022)
Female 0.1014 0.0154 -0.1467*** -0.0340***
(0.098) (0.015) (0.057) (0.013)
Children under 4 years old -0.1383 -0.0214 -0.1581* -0.0372*
(0.181) (0.029) (0.087) (0.021)
Female ¥ children under 4 years old -0.1259 -0.0195 -0.2672** -0.0636**
(0.231) (0.037) (0.132) (0.032)
Married -0.0763 -0.0113 -0.0108 -0.0025
(0.116) (0.017) (0.062) (0.014)
Widowed/separated/divorced -0.0185 -0.0028 -0.0575 -0.0134
(0.155) (0.023) (0.092) (0.022)
Public Sector Training Advantage

Degree or higher 0.6363*** 0.0897*** 0.5944*** 0.1309***


(0.147) (0.020) (0.085) (0.018)
685
Table 4 (contd)
686

Public sector Private sector

Coefficients Marginal effect Coefficients Marginal effect

A-level 0.5336*** 0.0710*** 0.4272*** 0.0942***


(0.149) (0.018) (0.086) (0.018)
O-level 0.4817*** 0.0685*** 0.2791*** 0.0639***
(0.124) (0.017) (0.070) 0.01575
Other qualifications 0.0796 0.0117 0.3564*** 0.0786***
(0.162) (0.023) (0.098) (0.020)
Union member 0.2526*** 0.0387*** 0.1603** 0.0367***
(0.084) (0.013) (0.063) (0.014)
Workplace tenure less than 1 year 0.3208** 0.0445** 0.4228*** 0.0937***
(0.143) (0.018) (0.080) (0.017)
Workplace tenure 1 to less than 2 years 0.6011*** 0.0779*** 0.5251*** 0.1139***
(0.141) (0.016) (0.084) (0.017)
Workplace tenure 2 to less than 5 years 0.0681 0.0101 0.1629** 0.0374**
(0.102) (0.015) (0.066) (0.015)
British Journal of Industrial Relations

Workplace tenure 5 to less than 10 years 0.1076 0.0158 0.0772 0.0178


(0.107) (0.015) (0.069) (0.016)
Manager or senior official occupations 1.0466*** 0.1164*** 0.7465*** 0.1567***
(0.223) (0.018) (0.103) (0.019)
Professional occupations 1.2136*** 0.1443*** 0.2220* 0.0500**
(0.193) (0.019) (0.116) (0.025)
Associate professional and technical occupations 1.1363*** 0.1402*** 0.4158*** 0.0918***
(0.176) (0.018) (0.099) (0.021)
Administrative and secretarial occupations 0.4044** 0.0563** 0.0215 0.0050
(0.175) (0.023) (0.097) (0.022)
Skilled trade occupations 0.5252** 0.0668*** -0.0402 -0.0094
(0.243) (0.026) (0.104) (0.024)
Personal service occupations 0.9980*** 0.1183*** 0.9973*** 0.1948***
(0.159) (0.015) (0.130) (0.020)

© Blackwell Publishing Ltd/London School of Economics 2008.


Sales and customer service occupations 0.4863 0.0623 0.4109*** 0.0902***
(0.428) (0.046) (0.109) (0.022)
Process plant and machine operative occupations 0.3022 0.0412 -0.1889* -0.0446*
(0.279) (0.034) (0.10) (0.024)
Workplace variables
Downsized in the last year 0.0772 0.0115 -0.1052 -0.0244
(0.121) (0.018) (0.075) (0.017)
Establishment size less than 25 0.1076 0.0157 -0.3105* -0.0735*
(0.228) (0.032) (0.174) (0.042)
Establishment size 25–50 0.1459 0.0212 -0.2488 -0.0588
(0.211) (0.030) (0.180) (0.043)
Establishment size 51–100 0.2382 0.0339 0.0312 0.0072
(0.211) (0.028) (0.176) (0.041)
Establishment size 101–200 -0.0980 -0.0150 -0.1425 -0.0334
(0.201) (0.031) (0.172) (0.041)
Establishment size 201–500 -0.0819 -0.0125 -0.0891 -0.0208
(0.207) (0.032) (0.166) (0.039)
Establishment size 501–1000 0.1606 0.0231 -0.0971 -0.0227

© Blackwell Publishing Ltd/London School of Economics 2008.


(0.245) (0.034) (0.184) (0.044)
Joint consultative committee -0.0740 -0.0111 0.1081 0.0249
(0.126) (0.019) (0.093) (0.021)
Quality circles 0.0178 0.0027 0.0496 0.0115
(0.114) (0.017) (0.081) (0.019)
Scope of bargaining includes training 0.1421 0.0205 0.7262** 0.1480***
(0.190) (0.026) (0.288) (0.050)
Management relations good -0.0159 -0.0024 0.2470* 0.0585*
(0.196) (0.029) (0.128) (0.031)
Management relations poor -0.0978 -0.0151 0.2008 0.0452
(0.487) (0.077) (0.336) (0.073)
Financial position better than average in sector 0.0187 0.0028 0.2662*** 0.0617***
Public Sector Training Advantage

(0.113) (0.017) (0.077) (0.018)


687
688

Table 4 (contd)

Public sector Private sector

Coefficients Marginal effect Coefficients Marginal effect

Financial position worse than average in sector -0.1973 -0.0313 -0.0546 -0.0127
(0.2467) (0.041) (0.135) (0.032)
Monitors effects of equal opportunities policy 0.1892 0.0280 0.4747*** 0.1043***
(0.120) (0.017) (0.106) (0.022)
Technical change -0.1885 -0.0270 0.1927* 0.0454*
(0.164) (0.022) (0.102) (0.024)
Organizational change 0.0264 0.0040 0.2160*** 0.0505***
(0.144) (0.022) (0.080) (0.019)
Constant -0.9511*** -1.6486***
(0.396) (0.257)
British Journal of Industrial Relations

c2(46) 529.73 814.06


[0.000] [0.000]
r 0.1399 0.1933
[0.000] [0.000]
Unweighted observations 5,587 12,554
Workplaces 412 1,113

Notes: * Indicates significance at 10 per cent level; ** Indicates significance at 5 per cent level; *** Indicates significance at 1 per cent level. Standard errors
reported in parentheses for both coefficient estimates and marginal effects. Figures in square brackets are p-values based on two different likelihood ratio tests:
the c2 statistic test refers to a test of the null hypothesis that all the slope parameters are equal to zero, while the p-value given for r is for a test of the null hypothesis
that r = 0. The omitted categories are aged 50 or more; single; no qualifications; workplace tenure 10 years or more; establishment size greater than 1,000;
elementary occupation; neither good nor poor management/employee relations; and neither better nor worse than average financial performance.

© Blackwell Publishing Ltd/London School of Economics 2008.


Public Sector Training Advantage 689
larger in the private sector than in the public sector: 14 per cent in the latter
compared to 19 per cent in the former. This suggests a greater diversity of
workplaces found in the private sector when compared to those in the public
sector.
While factors that affect the likelihood of an employee receiving training in
the public and private sectors are often similar, some differ, particularly in
terms of the statistical significance attached to different variables. As
expected, and in line with results reported elsewhere by Arulampalam and
Booth (1998), training propensities in both the private and public sectors
depend upon the nature of an individual’s employment contact. Thus,
workers on permanent contracts are more likely to receive employer-
provided training, while those working part-time are less likely to receive
training in both sectors. The marginal effects reported in Table 4 for
example, indicate that a permanent worker in the public (private) sector is
6 per cent (9 per cent) more likely to have received training than a non-
permanent worker, while a part-time worker is 9 per cent (7 per cent) less
likely to have received training than a full-time worker in the public (private)
sector. Both of these results can be explained by standard investment theory,
either because the period over which a return to training can be amortized is
shorter, or because the benefits of training are discounted at a higher rate to
reflect the increased uncertainty attached to employment among these groups
of workers.
Workplace tenure is also shown to have a significant effect on training
propensities in both sectors; specifically, training is less likely to be received,
the higher is workplace tenure. Again this result is consistent with the pre-
diction made by theory; namely that employers or workers are more likely to
invest in training as the time over which they can realize a gain from the
investment increases.12 One might expect that age effects should also fit into
this standard story. In fact, in the private sector training propensities do fall
with age, exactly as predicted by theory. However, in the public sector
(excepting workers aged between 22 and 29) age is not a statistically signifi-
cant determinant of training, and as a result the age/training profile is much
flatter in this sector.13 Indeed, the relationship between tenure and training,
and age and training, tends to be much flatter in the public sector, which
suggests that either the returns to training in this sector accrue more quickly
or, alternatively, that less weight is given to the investment aspect of training
in this sector.
The effect that trade unions are likely to have on the willingness of indi-
viduals to undertake training and of employers to provide training is unclear
from a purely theoretical perspective. Mincer (1983), for example, has sug-
gested that training may be lower in unionized firms because union-imposed
seniority rules governing promotion reduce workers’ incentives to invest and
undertake training. Alternatively, it is argued that by increasing earnings and
providing a ‘voice’ for employees within the firm (see Freeman and Medoff
1984), trade unions reduce turnover and increase the incentive to employers
to invest in training.14 Generally, the UK evidence supports a net positive
© Blackwell Publishing Ltd/London School of Economics 2008.
690 British Journal of Industrial Relations
effect of trade unions on training (e.g., Green 1993b; Green et al. 1999;
Greenhalgh and Mavrotas 1994; and Harris 1999). This result is confirmed in
Table 4, which shows that trade union membership increases the likelihood
of an individual receiving training in both the public and private sectors by
around 4 per cent. Interestingly, however, while the variable indicating
whether the scope of bargaining includes training increases the likelihood of
an individual receiving training in both sectors, it is only statistically signifi-
cant in the private sector. From this perspective, therefore, it would seem that
while rent sharing may have an important effect on training decisions in the
private sector, in the public sector, where national agreements are more
common, such motives are less easy to detect.
Other results reported in Table 4 can be briefly summarized as follows.
First, training propensities are higher for better qualified workers. This result
is consistent with education and training being complements in the produc-
tion process, which Rosen (1976) attributes to the effect that education has
on the efficiency of learning by reducing the costs of acquiring new skills.
Cost considerations could also be invoked to account for the fact that train-
ing propensities tend to be higher for workers in non-manual occupations
than for workers in manual and unskilled occupations. However, differences
in manual and non-manual training propensities are also likely to be related
to frequency with which these group’s skills need to be updated as a result of
changes in the nature of the jobs that they do.
Although few household status effects are statistically significant, an
important exception is the lower training propensity found for workers with
young children (under four) in the private sector. The effect is most pro-
nounced for women with young children, and implies that women with young
children are just over 6 per cent less likely to receive training in the private
sector than otherwise comparable men. This result is consistent with women
taking responsibility in the household for childcare duties. However, the
3 per cent training penalty found for men with young children in the private
sector was not expected, even though the estimated effect was only significant
at the 10 per cent level. Women without young children in the private sector
are also 3 per cent less likely to receive training in the private sector than
otherwise comparable male workers; however, there is no comparable train-
ing penalty for women in the public sector.15
More generally, training propensities in the private sector tend to be
more sensitive to the range of workplace characteristics included in the
analysis than they are in the private sector. Hence, while managers’ per-
ceptions of manager/employee relations, relative financial performance,16
monitored equal opportunities policies, and whether the workplace had
undergone technological or organizational change in the last two years all
increased the likelihood of workers receiving employer-provided training in
the private sector, in the public sector these factors were not statistically
significant.

© Blackwell Publishing Ltd/London School of Economics 2008.


Public Sector Training Advantage 691
Decomposition Analysis
Decomposition results based on the model reported in Table 4 are shown in
Table 5.17 The overall public–private sector training gap is 16 percentage
points, and this is split into a coefficient and characteristic effect depending
on whether the public or private sector is used as the basis of the decompo-
sition. Using private (public) sector coefficients as the appropriate base,
differences in worker and workplace characteristics account for 6.5 percent-
age points (5.2 percentage points) of the public–private sector training gap,
or 40 per cent (32 per cent) of the overall gap. Compared to private sector
workers, therefore, employees in the public sector either have human capital
endowments or are employed in workplace environments that predispose
them to the receipt of employer-provided training.
Irrespective of the base chosen, the size of the coefficient effect reported in
Table 5 suggests that public sector workers enjoy a significant training
advantage over otherwise comparable private sector workers. For example,
using private (public) sector characteristics, the estimated training advantage
was 9.6 percentage points (10.9 percentage points), which is significant and
accounts for around 60 per cent (68 per cent) of the observed training gap
between public and private sector workers. Although this estimate of the
public sector training advantage is smaller than the 16 per cent advantage
estimated by Arulampalam and Booth (1998), nonetheless the size and sig-
nificance of the effect identified here suggests that public sector workers
continue to enjoy a considerable training advantage over otherwise compa-
rable workers in the private sector. The estimates also suggest that the pattern
of results is broadly similar irrespective of whether public or private sector
coefficients are used to evaluate the characteristic component.18 For this
reason, and in line with the preference reported earlier, decompositions based
on the private sector are reported in the remainder of the article.

Sensitivity Analysis
This section considers the sensitivity of the estimated public sector training
advantage to a number of changes in the specification of the underlying
training model. The results of this analysis can be summarized as follows.
First, it could be argued that the financial performance and employer–
employee relations measures included in the analysis could be endogenous.
For example, high or low training propensities may also affect an establish-
ment’s financial performance and/or its employer–employee relations. To test
the sensitivity of the results to the inclusion of these variables therefore, the
model was re-estimated with these potentially endogenous variables
excluded. The results indicate that the estimated public sector training
advantage is not sensitive to the inclusion/exclusion of these variables. For
example, the estimated public sector training advantage from the restricted
specification was 9 per cent (0.09, Standard Error (SE) 0.014), which is not
significantly different from that reported for the full specification in Table 5.
© Blackwell Publishing Ltd/London School of Economics 2008.
692

TABLE 5
Decomposition of Employer-Provided Pubic/Private Training Gap

Public versus private sector Baseline specification

Private sector base Public sector base

Difference in predicted training incidence


i α
I pub − I prv = P ( X pub i α
ˆ pub ) − P ( X prv ˆ prv ) 0.161 0.161
Characteristics Component
pub prv prv prv
i
P ( X αˆ i
) − P ( X αˆ ) i α
P ( X pub i α
ˆ pub ) − P ( X prv ˆ pub )
0.065 [40] 0.052 [32]
British Journal of Industrial Relations

(0.009) (0.014)
Coefficient Component
i α
P ( X pub i α i α i α
ˆ pub ) − P ( X pub ˆ prv ) P ( X prv ˆ pub ) − P ( X prv ˆ prv )
0.096 [60] 0.109 [68]
(0.014) (0.018)

Notes: Standard errors for the characteristic and coefficient components are shown in parentheses. Figures in square brackets indicate the percentage of the overall
training gap accounted for by the coefficient and characteristic components.

© Blackwell Publishing Ltd/London School of Economics 2008.


Public Sector Training Advantage 693
Second, following Alemeida-Santos and Mumford (2005) we introduce a
measure of wage dispersion into the model. This measure is based on the
gross wages of full-time workers in WERS and is constructed as the log of the
ratio of the 90th and 10th percentiles of the wage distribution for an occu-
pation (nine occupations), within an education band (five education bands)
and within an industry sector (12 industry sectors). This wage dispersion
measure was then matched to individuals according to their occupation,
educational level and industry sector. The effect of increased wage dispersion
on training was as expected—that is, the more dispersed the wage structure
the lower the level of training. For example, in the private sector the marginal
effect of an increase in wage dispersion reduced the likelihood of an indi-
vidual receiving training by 1.3 per cent and in the public sector by 2.5 per
cent; however, neither individual effect was significant, nor did the inclusion
of this variable have any discernable effect on the estimated public sector
training advantage. From this limited perspective, therefore, the wage dis-
persion variable did not appear to be a major factor in determining the size
of the public sector training advantage.
Third, as shown in Table 1, other than for training lasting for ‘less than
one day’, the incidence of training in the public sector is higher in every
duration band. While modelling training durations is beyond the scope of
the present article, we did consider the effect that excluding induction train-
ing might have on the estimated public sector training advantage. As induc-
tion training is typically undertaken by an employee when they first arrive
at a workplace and is of relatively short duration, the effect of induction
training is most easily removed by eliminating workers from the sample
who either had a training episode that lasted ‘less than two days’ or who
had worked at the workplace for less than a year. As might be expected
from the data reported in Table 1, restricting the sample in this way
increases the difference in training incidences between the public and
private sectors to 23 percentage points, and on the back of this increase in
the raw differential, the estimated public sector advantage also rises to 13
per cent (0.13, SE 0.019). The difference between this estimate of the public
sector training advantage and that reported in Table 5, however, was not
statistically significant.
Fourth, to examine the effect that financial motives might have on the size
of the public sector training advantage, information provided by senior
managers on what they felt most closely described their interpretation of
financial performance was included in the analysis. The data, which are
shown in summary form in Table 2, was used in two different ways in the
analysis. First, a set of variables measuring the different financial motives
were included in the estimation. Second, the sample was restricted to employ-
ees in the public and private sectors who worked in workplaces that had the
same financial motive. As suggested in Table 2, however, to generate reason-
able sample sizes for this second approach, we had to limit the analysis to
workplaces that identified ‘budget’ considerations as being the most appro-
priate measure of financial performance.
© Blackwell Publishing Ltd/London School of Economics 2008.
694 British Journal of Industrial Relations
Including a full range of financial motive variables in the specification
reduced the estimated public sector training advantage to 6 per cent (0.060,
SE 0.016).19 While this estimate is almost 4 percentage points lower than the
equivalent estimate reported in Table 5, the difference is only statistically
significant at the 10 per cent level. However, restricting the sample to workers
employed in workplaces which have the same financial motive produced a
stronger result: in this case, not only does the raw difference in training
incidence between the public and private sectors fall to 9 per cent, but the
estimated public sector training advantage also falls to 4 per cent (0.044, SE
0.022), which is significantly different from that reported in Table 5 for the
full sample.20
There is also other evidence that suggests that the workplaces contained in
the restricted sample are different from those in the full sample. For example,
while unobserved workplace heterogeneity remains an important feature of
the data, the gap between the proportion of the error variance accounted for
by unobserved heterogeneity in public and private sector workplaces is
reduced, for example, r is 12 per cent in the public sector and 14 per cent in
the private sector, compared to equivalent values in the full sample of 14 and
19 per cent, respectively. Focusing attention on a single financial motive like
budget, therefore, not only reduces the size of the public sector training
advantage but also the degree of workplace diversity between the public and
private sectors.
Taken together these estimates provide some support for the idea that
differences in financial motives play a part in determining the size of the
public sector training advantage. In fact, as shown in Table 2, financial
motives can be used effectively to classify the majority of public and private
sector workplaces. However, even when controls for financial motive are
included in the research design, a significant training advantage is still found
in the public sector, which is also shown to be robust to a number of
important changes in specification. On the basis of this analysis, there seems
to be some support for the idea that the well-documented advantage public
sector workers enjoy in pay and pension rights might also extend to other
areas of the employment contract, including employer-provided training. But
what about the potential effects that worker sorting might have? It is to this
important issue that we now turn.
Fifth, and finally, we considered the sensitivity of the results to an alter-
native estimation method, one that is designed to address the effect that
worker sorting effects might have on the analysis. As explained in the meth-
odology section, worker sorting effects may have implications for the con-
sistency of the estimates produced by the random effects model, particularly
where they induce an association between employee characteristics and
unobserved workplace heterogeneity.
Fortunately, Chamberlain (1980) has shown that consistent estimates of
the random effects model under these circumstances can be obtained using a
Chamberlain-Mundlak transformation.21 Moreover, as estimates of the stan-
dard random effects model are consistent when worker characteristics and
© Blackwell Publishing Ltd/London School of Economics 2008.
Public Sector Training Advantage 695
workplace heterogeneity are uncorrelated, a simple test of the likely relevance
of worker sorting effects suggests itself; namely a Hausman test based on the
coefficient estimates common to both the original and transformed models.22
The calculated Hausman statistic for the private sector and public sector
estimates are c2(47) = 196 and c2(47) = 7, respectively. As the Hausman test
rejects the hypothesis that the estimates from the two models are the same in
the private sector, this suggests that the assumption of independence between
establishment heterogeneity and individual workforce characteristics is
rejected in that sector. It follows that worker sorting effects, which result in
the concentration of particular workers in particular establishments, have the
potential to explain part of the difference in training incidences between
public and private sector workers. In fact, the decomposition analysis based
on the enhanced Chamberlain-Mundlak specification indicates that the
public sector training advantage is all but eliminated once variables mea-
suring the average characteristics of workers within an establishment are
included in the model. Specifically, the estimated public sector training
advantage is reduced to only 0.6 per cent (0.006, SE 0.015), while the com-
ponent of the training difference accounted for by the characteristic effect
increases to just over 96 per cent of the overall public/private sector dif-
ference. Controlling for the average characteristics of employees within
individual establishments, therefore, seems to play an important part in
accounting for difference in training incidences between the two sectors. We
interpret this finding as being due to a worker sorting effect, which can arise
either as a result of unobserved individual heterogeneity being correlated
with employee characteristics and training, or alternatively, as a result of
training complementarities that occur as a result of the concentration of
particular groups of workers within establishments. Whatever the mecha-
nism, worker sorting effects seem to be important and capable of accounting
for much of the residual training advantage identified by the earlier analysis.

5. Conclusions

The government’s current strategy is to deliver world-class public services in


the UK, and this has recently renewed interest in comparative labour market
outcomes in the public and private sectors. This article looks at the neglected
issue of differences in training incidences of public and private sector
workers, which several studies have identified as being a consistent empirical
finding. Naturally, training incidences in the public sector can be higher than
those in the private sector because the characteristics of workers in the public
sector predispose them towards training. However, this article also lists a
number of reasons why training might be higher in the public sector for
otherwise comparable workers including social externalities, alternative
behavioural objectives, tax appropriation, job security, hierarchical wage
structures, differences in wage dispersion, rent sharing and worker sorting.
Although the data we have does not always allow us to be prescriptive about
© Blackwell Publishing Ltd/London School of Economics 2008.
696 British Journal of Industrial Relations
which of these potential factors/explanations is more important than others,
they do allow us to provide an estimate of the public sector wage advantage
that accounts for the influence of a large number of compositional differences
between workers and workplaces in the two sectors.
An important feature of the WERS2004 is that it includes information on
both public and private sector workplaces as well as the characteristics of
workers in these sectors. This allows us not only to control for unobserved
workplace heterogeneity, but also a uniquely wide range of workplace and
worker characteristics. One of the key findings to emerge from the analysis is
that the public sector training advantage is significant and surprisingly robust
to most changes in specification. For example, including a particularly strong
control for financial performance reduces, but does not eliminate, the public
sector advantage. However, a control for worker sorting effects all but elimi-
nates any remaining public sector advantage. Worker sorting therefore
seems to be a key factor in understanding why public sector workers enjoy a
training advantage even after controls are included for individual worker
characteristics.

Final version accepted on 30 July 2008.

Acknowledgements

Material from the WERS2004 has been made available by the sponsors the
Department of Trade and Industry; the Advisory, Conciliation and Arbitra-
tion Service; the Economic and Social Research Council; and the Policies
Studies Institute through the UK Data Archive. The authors would also like
to acknowledge the financial assistance provided by the DTI, although they
are not responsible for the views expressed in the article. Finally, the authors
would like to thank two anonymous referees and an editor of the journal for
making many useful recommendations; they can in no way be held respon-
sible for any errors that remain.

Notes

1. Elliott and Bender (1997) point to the fact that new working practices introduced
into the public sector to increase competition like contracting out and compulsory
competitive tendering are now both commonly used.
2. This conclusion is based on the five training duration bands reported in WERS
and therefore must be treated with caution, particularly if private sector
workers who receive training of 10 days or more tend to receive much longer
spells of training than public sector workers in this category. The entries in this
and all subsequent tables are based on unweighted or within sample data. The
general pattern of results, however, remains the same irrespective of whether
weighted or unweighted data are used. In the subsequent empirical analysis a
full set of workplace size controls are included to account for any tendency in

© Blackwell Publishing Ltd/London School of Economics 2008.


Public Sector Training Advantage 697
the survey design for WERS to over- or under-sample employees from particu-
lar workplaces.
3. Information on financial performance was identified from managers’ responses to
a question that asked them to identify a measure, which most closely corresponded
to their interpretation of financial performance. A number of different dimensions
of financial performance were identified, including profits, value added, establish-
ment budget, costs and expenditure, and these are reported in employee coverage
terms in Table 2 for both the public and private sectors separately.
4. This can be illustrated in the case of general training by using a simple two period
model. Assume training undertaken by the firm in the first period costs cT;
untrained workers produce q in both periods; and trained workers produce q + ai
in the second period, where ai is the individual specific training effect, which is
described by the distribution function F(ai). If competition between firms ensures
workers receive the value of their output, it can be shown that the returns to
training in the private and public sectors are given by:
Π (q + α i, T ) − Π (q, NT ) = −c T
prv prv

1
and Π (q + α i , T ) − Π (q, NT ) = −c T + α it
pub pub

1+ r
where II(.) is profit under the training (T) and non-training (NT) alternatives, t is
the marginal tax applied to wages in periods 1 and 2, r is the discount rate, and prv
and pub represent the public and private sectors, respectively. It follows that,
while the private sector will not provide any general training, the public sector will
provide training to 1 − F (
c T (1 + r )
t )
of the workforce.
5. The training question used in the analysis asks ‘Apart from health and safety
training, how much training have you had during the last 12 months, either paid
for or organized by your employer?’ and offers a choice of (banded) durations:
none, less than one day, one to less than two days, two to less than five days, five
to less than 10 days, and 10 days or more. The dependent variable used here
relates to a simple binary indicator of whether an individual has received any
employer-provided training in the last 12 months.
6. Public and private sector indicators have been removed from this relationship for
presentational reasons.
7. Ignoring public and private sector subscripts, the analysis is based on the
probability of training given mj = 0, i.e. P(I = 1|mj = 0), which on the basis of the
likelihood function used to model training propensities is equal to the logistic
distribution function F(Xija).
8. There is also a residual component in decompositions of this type. The residual
can occur either as a result of setting mj to zero when estimating the predicted
probabilities of training or, alternatively, in the case of some nonlinear estima-
tors, because the average of the estimated probabilities will not always equal the
average of the response variable. Although we will return to the issue of setting
mj = 0 later, the size of the residual component was found to be of second order of
importance.
9. The decomposition based on the public sector training structure is given by:
I pub − I prv = [P ( X pub
i α
ˆ pub ) − P ( X prv
i α ˆ pub )] + [P ( X prv
i α ˆ pub ) − P ( X pi rvαˆ prv )].
10. The variables included in X̄j include all characteristics that vary over workers in
the jth establishment.

© Blackwell Publishing Ltd/London School of Economics 2008.


698 British Journal of Industrial Relations
11. Marginal effects are evaluated at the means of independent variables but also take
account of the nature of the binary variables used in the analysis.
12. Significant workplace tenure effects are only found in the public sector for
employees with less than two years of tenure, while in the private sector such
effects are found for workers with up to five years of workplace tenure. This
suggests either that a higher proportion of training provided in the public sector
is induction training or that the duration of training provided in the public sector
is shorter. This possibility is examined in some detail later, although to anticipate
this analysis, the relative sizes of the marginal effects reported for the public and
private sectors in the first two workplace tenure categories and the descriptive
statistics included in Table 1 suggest this is not case.
13. Since workplace tenure tends to be higher in the public sector than the private this
finding could be due to tenure and age being more highly correlated in the public
sector. The flat age/training profile remains a feature of public sector employ-
ment, however, even when tenure effects are excluded from the analysis.
14. See also the more recent contributions by Booth and Chatterji (1998) and Booth
et al. (2003) which explore the implications of union-induced wage increases and
their effect on labour turnover and training more formally.
15. Several recent studies, however, have suggested that training incidences might be
higher for women (see Arulampalam et al. 2004; Jones et al. 2008).
16. The variables measuring relative financial performance were constructed from
responses given by managers to a question that asked them to compare their
workplace’s financial performance with other establishments in the same
industry. Managers could choose from a number of different categories: (a) a
lot better than average; (b) better than average; (c) about average; (d) below
average; (e) a lot below average; or (f) no comparison was possible or the rel-
evant data were not available. The variable ‘Financial Position Better than the
Sector Average’ was constructed from combining responses (a) and (b), while
the variable ‘Financial Position Worse than the Sector Average’ was found by
combining responses (d) and (e). All other possible responses to the question
were subsumed into the omitted category. However, as discussed later, the esti-
mates of the public sector training advantage reported in Table 5 were not sen-
sitive to the inclusion of this information. In the public (private) sector the
distribution of responses to this question was as follows: (a) a lot better than
average, 9.6 per cent (11.6 per cent); (b) better than average, 32.2 per cent (41.4
per cent); (c) about average, 42.7 per cent (32.7 per cent; (d) below average,
4.7 per cent (7.9 per cent); (e) a lot below average, 0.3 per cent (0.5 per cent);
and (f) no comparison is possible or the relevant data are not available, 10.4 per
cent (5.9 per cent).
17. Although a probability weights option cannot be used to estimate a random
effects model in the software used (STATA), we did examine the sensitivity of
the results to sampling weights by estimating a less complex logit model, which
does have a probability weights option. As it turned out, the weighted and
unweighted decomposition results from this exercise were not significantly dif-
ferent. For example, using the private sector as the basis of the decomposition,
the estimated public sector training premium for the unweighted data was 6.9
per cent (0.069, SE 0.009) while for the weighted data it was 7.4 per cent (0.074,
SE 0.010).
18. While the expected value of mj is zero, evaluating the predicted probabilities of
training at mj = 0 is not equivalent to the marginal probability of training for

© Blackwell Publishing Ltd/London School of Economics 2008.


Public Sector Training Advantage 699
each observation, which integrates out the effects of workplace heterogeneity.
As a result, the probabilities used in the decomposition could be biased, even
though the nature of this bias is likely to be small. While an estimate of the
appropriate marginal probabilities of training is difficult to obtain in practice,
we did consider the nature of this bias using a simple simulation exercise sug-
gested to us by programmers at the STATA Corporation. Specifically on the
basis of the estimated value of sm, 100,000 random draws were taken from
mj ~ N(0,sm) for each observation and used to evaluate F̄(Xia + mj), which is the
mean of the logistic distribution found from the simulation exercise and which
numerically mimics the effect of integrating out the effects of workplace het-
erogeneity. The value of F̄(Xia + mj) found for each observation was then used
as an estimate of the marginal probability of training P(Xia) for each individual
in the sample. The decomposition based on the probabilities produced from this
simulation exercise gave a public sector training advantage of 0.100 and 0.108
when based on private and public sectors coefficients respectively. Comparing
these estimates with the public sector training advantage reported in Table 5,
therefore, suggest that any bias caused by setting mj = 0 when calculating the
probabilities used in the decompositions is likely to be small.
19. The estimates indicate that an individual employed in a workplace in the private
(public) sector which identified ‘budget’ as being the most appropriate measure of
financial performance was 9 per cent (5 per cent) more likely to have received
training in the last year than an otherwise comparable individual working in a
profit-oriented workplace. The equivalent marginal effects for the other catego-
ries were ‘value added’, 10 per cent (5 per cent); ‘sales’, -3 per cent (-11 per cent);
‘fees’, 6 per cent (12 per cent); ‘costs’, 1 per cent (7 per cent); ‘expenditure’, -4 per
cent (6 per cent); and ‘other’, 3 per cent (5 per cent).
20. Another factor that might be worth considering is the industrial distribution of
workplaces in the sample. Unfortunately, public sector employment tends to be
organized along industry lines and is concentrated in a small number of industrial
sectors, which makes it difficult to include a full range of industry codes in the
analysis. However, by restricting the sample to workers in education, health and
other community services, it was possible to make a worthwhile comparison
between training incidences in the public and private sectors. The estimated
public sector training advantage found from this analysis was 4 per cent (0.039,
SE 0.022). Thus, although the public sector training advantage is reduced in size,
it still suggests that public sector workers enjoy an advantage over otherwise
comparable workers in the private sector.
21. Chamberlain (1980) has also shown that a conditional fixed effects logit model
can be used to produce consistent estimates in this situation. In the conditional
fixed effects logit model, the incidental parameter problem, associated with a fixed
effects logit specification, is addressed by conditioning the fixed effects out of the
conditional distribution. However, a major drawback associated with the use of
this model is that it is not possible to identify either the incidental parameters or
the effect produced by any variable, which does not vary within an establishment.
Since this would make any analysis based on the decomposition framework
outlined earlier difficult to interpret, we decided to implement the alternative
Chamberlain-Mundlak procedure.
22. These estimates are available on request. The general pattern of results obtained
for the baseline specification reported in Table 4 continue to hold for character-
istics common to this more general model.

© Blackwell Publishing Ltd/London School of Economics 2008.


700 British Journal of Industrial Relations
References

Abowd, J., Kramarz, F. and Margolis, D. (1999). ‘High wage workers and high wage
firms’. Econometrica, 67: 251–333.
Acemoglu, D. and Pischke, S. (1999). ‘The structure of wages and investment in
general training’. Journal of Political Economy, 107: 539–72.
Alemeida-Santos, F. and Mumford, K. (2005). ‘Employee training and wage com-
pression in Britain’. Manchester School, 73: 321–42.
Arulampalam, W. and Booth, A. (1997). ‘Who gets over the training hurdle? A study
of the training experiences of young men and women in Britain’. Journal of Popu-
lation Economics, 10 (2): 197–217.
—— and —— (1998). ‘Training and labour market flexibility: is there a trade-off?’
British Journal of Industrial Relations, 36: 521–36.
——, Booth, A. L. and Bryan, M. L. (2004). ‘Training in Europe’. Journal of the
European Economic Association, 2: 346–60.
Becker, G. S. (1975). Human Capital, 2nd edn. New York: National Bureau of
Economic Research.
Booth, A. L. (1991). ‘Job-related formal training: who receives it and what is it
worth?’ Oxford Bulletin of Economics and Statistics, 53: 281–94.
—— and Chatterji, M. (1998). ‘Unions and efficient training’. Economic Journal, 108:
328–43.
——, Francesconi, M. and Zoega, G. (2003). ‘Unions, work-related training and
wages: Evidence for British men’. Industrial and Labor Relations Review, 57:
68–91.
Chamberlain, G. (1980). ‘An analysis of covariance with qualitative data’. Review of
Economic Studies, 47: 225–38.
Dixit, A. (1997). ‘Power of incentives in private versus public organisations’. Ameri-
can Economic Review, Papers and Proceedings, 87: 378–82.
Elliott, R. and Bender, K. (1997). ‘Decentralisation and pay reform in central gov-
ernment: A study of three countries’. British Journal of Industrial Relations, 35:
447–75.
Fairlie, R. (1999). ‘The absence of the African-American owned business: An
analysis of the dynamics of self-employment’. Journal of Labor Economics, 17:
80–108.
—— (2005). ‘An extension of the Blinder-Oaxaca decomposition technique to logit
and probit models’. Journal of Economic and Social Measurement, 30: 305–16.
Freeman, R. and Medoff, J. (1984). What Do Unions Do? New York: Basic Books Inc.
Gomulka, J. and Stern, N. (1990). ‘The employment of married women in the United
Kingdom 1970–83’. Economica, 57: 171–99.
Green, F. (1993a). ‘The determinants of training of male and female employees in
Britain’. Oxford Bulletin of Economics and Statistics, 55: 101–20.
—— (1993b). ‘The impact of trade union membership on training in Britain’. Applied
Economics, 25: 1033–43.
——, Machin, S. and Wilkinson, D. (1999). ‘Trade unions and training practices in
British workplaces’. Industrial and Labor Relations Review, 52: 179–95.
Greenhalgh, C. and Mavrotas, G. (1994). ‘The role of career aspirations and financial
constraints in individual access to vocational training’. Oxford Economic Papers,
46: 579–604.
Harris, R. (1999). ‘The determinants of work-related training in Britain in 1995 and
the implications of employer size’. Applied Economics, 31: 451–63.

© Blackwell Publishing Ltd/London School of Economics 2008.


Public Sector Training Advantage 701
Jones, M. K., Latreille, P. L. and Sloane, P. J. (2008). ‘Crossing the tracks? Trends
in the training of male and female workers in Great Britain’. British Journal of
Industrial Relations, 46 (2): 268–82.
Latreille, P. L., Blackaby, D. H., Murphy, P. D. and O’Leary, N. C. (2005). ‘The
public-private sector training gap for full-time male employees in Britain: evidence
from SCELI’. International Journal of Economic Research, 2: 125–45.
Mehaut, P. and Perez, C. (2004). ‘Further education and training in the French public
sector’. Public Management Review, 6 (3): 333–52.
Mincer, J. (1983). ‘Union effects: wages, turnover, and job training’. In J. D. Reid, Jr.
(ed.), Research in Labor Economics: New Approaches to Labor Unions, Supplement
2. Greenwich, CT: JAI Press.
Mundlak, Y. (1978). ‘On the pooling of time series and cross section data’. Econo-
metrica, 46: 69–85.
Niskanen, W. (1971). Bureaucracy and Representative Government, Chicago, IL:
Aldine.
—— (1975). ‘Bureaucrats and politicians’. Journal of Law and Economics, 18: 617–43.
Prendergast, C. (1993). ‘The role of promotion in inducing specific human capital
acquisition’. Quarterly Journal of Economics, 108: 523–34.
Reder, M. (1975). ‘The theory of employment and wages in the public sector’. In
D. Hamermesh (ed.), Labor in the Public and Non-Profit Sectors. Princeton, NJ:
Princeton University Press.
Rosen, S. (1976). ‘A theory of life earnings’. Journal of Political Economy, 84:
S45–S67.
Tirole, J. (1994). ‘The internal organisation of government’. Oxford Economic Papers,
46: 1–29.

© Blackwell Publishing Ltd/London School of Economics 2008.

You might also like