DOES DELEGATION INCREASE WORKER TRAINING?
Bilanakos, Christos ; Heywood, John S. ; Sessions, John G. 等
DOES DELEGATION INCREASE WORKER TRAINING?
I. INTRODUCTION
Delegation of decision-making allows employers to capture the
superior knowledge and information of workers. The objective functions
of workers, however, may differ sharply from those of their employers.
This tradeoff between enhanced information and misaligned incentives
lies at the heart of a growing literature claiming that delegation and
incentives are complementary--the more authority delegated to workers,
the stronger must be the incentives for workers to align their
objectives with those of their employer.
We uniquely provide a theoretical illustration and supporting
empirical evidence showing that delegation increases employee training.
We extend Aghion and Tirole's (1997) seminal model to recognize
that a worker's effort cost of acquiring superior information can
be reduced through the provision of training. We compare the
profit-maximizing training intensity under two alternative
organizational structures: integration, where the firm retains the
ability to overrule the investment project recommended by the worker;
and delegation, where the firm cannot overrule the worker's
preferred investment project.
Our model predicts that the firm provides more training under
delegation than under integration if the preferences of the firm and the
worker are sufficiently congruent. This reflects the key trade-off from
delegation. The worker is induced to supply more effort in acquiring
information but the firm loses control and risks its preferred
investment project not being implemented. When the firm benefits
similarly from its own preferred project and from that of the worker,
the effort-enhancing effect of delegation dominates the loss-of-control
effect and this makes additional training a profitable investment.
In an Appendix, we add training to a richer model in which
incentive pay is also an option to foster the agent's effort. We
again show that training can be a profitable investment under delegation
and partially characterize the optimal intensity of training, wage
incentives, and effort under each authority structure (integration and
delegation), while leaving the full treatment of the relationship
between these variables for future research.
We impose delegation or integration exogenously in our model. This
seems appropriate in situations where, for example, more senior
management may require certain types of delegation, or where an external
hire imposes delegation practices in a way that is unrelated to the
firms' training practices. There may also be structural differences
across industries that make delegation more likely--something we exploit
in our instrumental variable (IV) strategy in Section V.C. Within this
framework, our nonequilibrium approach of looking at integration and
delegation separately provides a sensible and testable association
between delegation and training.
We test the prediction of our theoretical model using large
British-matched employer-employee cross sections and an associated
panel. We find that those establishments which delegate more, with
workers greatly influencing their own tasks, offer more training. This
persists across a variety of specifications, alternative sample
restrictions, and using alternative definitions of both delegation and
training. It persists in establishment fixed-effect estimates, under
alternative functional forms and in reasonable IV estimates that also
control for workplace fixed effects. (1) The empirical relationship
between delegation and training appears remarkably durable.
Our investigation remains pertinent as delegation and
decentralization of decision-making within firms have become
increasingly common since the late 1970s, being most evident in
Scandinavian and Anglo-Saxon countries (Aghion, Bloom, and Van Reenan
2014). (2) This growth in actual delegation has been matched by a recent
literature on the relationship between incentives and delegation, which
we summarize in the next section.
In what follows, Section II sets our study in the context of
related literature. Section III provides the theoretical framework and
identifies the conditions under which delegation increases training.
Section IV discusses the data and empirical methodology. Section V
presents empirical results and provides robustness checks. Section VI
concludes and offers suggestions for further research.
II. RELATED LITERATURE
Our research fits a strand of literature that models pairs of human
resource practices searching for complements or substitutes--see, for
example, Allgulin and Ellingsen (2002). This allows more tractable
theoretical models and cleaner empirical predictions. (3) We examine the
pair of delegation and training. The study of these practices has
generated two extensive but distinct literatures. To our knowledge, we
are the first to examine delegation of authority as a determinant of
training.
In this section, we briefly review the research on delegation and
incentives that we extend to include the training decision. Theoretical
work views the choice of delegation in terms of information and control.
Delegating authority can be beneficial because workers know more about
their day-to-day tasks than their employer. Yet, workers'
objectives can differ from those of their employer. This tradeoff
between information and objectives exists because the agent does not
fully communicate private information when the principal retains
authority. This arises because of bounded rationality (Jensen and
Meckling 1992) or because of the agent's strategic use of
information (Holmstrom 1984).
Grossman and Hart (1986) observe that authority might follow from a
contract that allocates decision rights within the organization (see
also Hart and Moore 1990). Such formal authority, however, does not
always coincide with real authority, the effective control over
decision-making. Aghion and Tirole (1997)--hereafter AT--explore this
tension between real and formal authority and its implications for
delegation. They recognize that information provision can be a critical
yet noncontractible relation-specific investment.
AT model a principal-firm and an agentworker who together implement
a single investment project. The firm tasks the worker with assembling
information regarding the expected payoffs across an array of potential
projects. The firm selects from two alternative organizational
structures: integration, whereby the firm maintains formal authority
over investment decisions and can ignore the worker's
recommendation as to the "best" investment project; and
delegation, whereby the worker selects a particular project and cannot
be overruled. AT show that delegating authority encourages the worker to
work harder in ascertaining which project should be implemented.
However, this higher effort comes with a loss of control and an
increased likelihood that the chosen project fails to maximize the
firm's payoff. Which effect dominates depends upon how congruent
are the objectives of the firm and worker. The less their objectives
coincide, the more likely will the loss of control effect dominate and
the more likely will the firm retain formal authority.
We expand AT by imagining that the firm can invest in training.
This training lowers the cost of information acquisition by the worker
and makes delegation particularly effective in providing superior
information. We show that this is a profitable investment when
objectives are sufficiently congruent.
Several papers have, like us, developed extensions to AT's
framework. Baker, Gibbons, and Murphy (1999) argue that even though
delegation can only ever be informal, it might remain in equilibrium due
to reputational concerns. Hart and Holmstrom (2010) and Bolton and
Dewatripont (2013) also stress that delegation may persist despite the
typical ability of the principal to reverse a delegation decision.
Zabojnik (2002) argues that it is less costly to motivate an agent to
work on their own project rather than on the principal's project,
while De Paola and Scoppa (2006) investigate the costs and benefits of
delegation within a framework where the principal cannot observe the
agent's effort. In both cases, delegation persists as a feasible
outcome when the loss of control implied by delegation proves less
costly for the principal than the loss of information under
centralization. (4)
Estimation by Aghion, Bloom, and Van Reenan (2014) confirms that
the congruence of preferences (as proxied by trust) helps determine
delegation. Itoh, Kikutani, and Hayashida (2008) show that delegation
from core to affiliated Japanese firms is associated with incentives for
accountability. Nagar (2002), Colombo and Delmastro (2004), Foss and
Laursen (2005), and De Varo and Kurtulus (2010) demonstrate that
incentive payments for managers and for workers are associated with
delegation. Yet, De Varo and Prasad (2015) argue that noisy incentive
pay may induce risk-averse agents to select suboptimal tasks. For
instance, surgeons may not operate on high-risk patients. They confirm
that delegation and incentives are positively correlated for simple jobs
but negatively correlated for complex jobs (where task selection is
valuable). Lo et al. (2016) show that sales employees with higher tenure
and skills are delegated more pricing authority even as uncertain
product markets make delegation less likely.
While the evidence is not monolithic, we incorporate the issues of
these studies into our testing. In robustness checks, we examine whether
the influence of delegation on training varies with the presence of
incentive schemes. If incentive schemes facilitate delegation by
increasing the congruence of interests, they may also make training more
valuable to the firm. Put differently, training becomes more profitable
when it is more likely to be used in the interest of the firm. We will
also experiment with the role of an uncertain market environment.
As a last note, delegation may simply be critical for firm success.
Bloom, Sadun, and Van Reenan (2012b) show that failure to delegate
authority (often resulting from lack of trust) impedes firm growth.
Boedker et al. (2011) find that of 32 practices, delegation correlates
most closely with their "High Performing Workplace Index."
More generally, researchers emphasize that appropriate delegation
reflects successful management (Garicano and Rayo 2016) and generates
productivity differences across firms and countries (Bloom et al.
2012a). As a consequence, we adopt an empirical strategy that tries to
rule out threats to the independent role that we argue delegation can
play. We want to avoid presenting correlations that simply reflect that
superior management more likely both delegates and trains workers.
We use matched employer-employee data to estimate the determinants
of firm-sponsored training. Such estimates frequently focus on the role
of competition in labor and product markets (Acemoglu and Pischke 1998,
1999; Manning 2003). (5) While Brunello and Gambarotto (2007) confirm
that employers provide less training in more competitive labor markets,
Bilanakos et al. (2017) find that product market dominance strengthens
investment in training. We will account for such determinants when
focusing on the role of delegation. We are aware of no papers that
relate training to delegation and only one paper that relates
(exogenous) human capital to the degree of delegation within a firm. (6)
While building from these previous works, we test whether the delegation
of decision-making authority plays an independent role in an
establishment's choice of training intensity.
III. THEORETICAL MODEL
A. Setup
We consider a principal-owner, P, and an agent-employee, A, who
either implement a single investment project or choose to do nothing. P
tasks A with collecting information about the payoffs of n > 3
potential and a priori ostensibly identical projects. The
principal's gross profit associated with each project k [member of]
{1,2, ... , n} is [B.sub.k] and the agent's corresponding private
benefit (which may include on-the-job perks or the possibility of
signaling his ability) is [b.sub.k]. These payoffs do not take into
account any wage payments from P to A. The case where P and A do nothing
is summarized in a "project zero" yielding payoffs [B.sub.0] =
[b.sub.0] = 0. (7) The principal reaps B>0 from her preferred project
while the agent reaps b > 0 from his own preferred project. The
agent's benefit from P's preferred project is fib and the
principal's profit from A's preferred project is aB, where
a,[beta][member of](0,1] are exogenous congruence parameters.
The principal chooses the level of training, I, to provide the
agent. The training cost incurred by P is c(I) with c'(0) = 0, c
(I) > 0 and c"(I) > 0 for I > 0. Both the principal and
the agent are initially unaware of the payoffs from the various
projects. P acquires perfect information regarding the payoffs of all
projects with exogenous probability E but remains ignorant with
probability 1 - E. A chooses effort e devoted to acquiring information
about the projects' payoffs and becomes perfectly informed with
probability e but learns nothing with probability 1 - e. We assume that
training reduces the agent's marginal effort cost as captured in
the effort cost function g(e, I) with [mathematical expression not
reproducible]. (8) Since our formulation primarily intends to motivate
the empirical analysis rather than suggest a general and thorough
extension of AT's model, we ease analytical exposition by assuming
the specific functional forms c(I) = [theta][I.sup.2]/2 and g(e, I) =
[rho][e.sup.2]/2I (with [theta] > 0 and [rho] > 0) throughout this
section. This illustrative example sheds light on the main tradeoffs
associated with delegation and allows a closed-form solution.
The principal pays a wage w [greater than or equal to] 0 to the
agent, who faces a fixed outside option represented by his reservation
utility [bar.u]. (Appendix A shows how the model can be extended to
incorporate incentive pay.) We follow AT by considering integration (n)
and delegation (d). Under integration, P can overrule A's
recommendation and, if informed, adopt her preferred project. Under
delegation, P cannot overrule A's recommendation and optimally
accepts it since [alpha] > 0. Of course, an uninformed agent will
accept P's proposal (if any) given that [beta] > 0. Since the
projects cannot be contracted upon ex ante, the model follows Grossman
and Hart's (1986) incomplete contracting approach. Specifically,
the initial contract allocates formal authority to either P or A and the
overall sequence of actions is described in Figure 1.
Under integration, the payoffs of P and A, [u.sup.n.sub.p] and
[u.sup.n.sub.A] respectively, are given by:
(1) [u.sup.n.sub.p] = E x B + (1 -E)e x aB-c(I)-w
(2) [u.sup.n.sub.A] =E x [beta]b + (1 - E)e x b + w-g(e,I).
The payoffs associated with delegation are:
(3) [u.sup.d.sub.P] = e x aB + (1 - e)E x B-c(I)-w
(4) [u.sup.d.sub.A] = e x b + (1 -e)E x [beta]b + w-g(e,I).
B. Equilibrium
The model is solved recursively under each authority structure
(integration and delegation). In both cases, we first characterize
A's optimal effort given the wage and training level. Then, we move
back to identify the profit-maximizing training intensity and wage
anticipating the worker's optimal effort and taking into account
the latter's participation constraint. Finally, we establish the
conditions under which P optimally selects to delegate formal authority
to A.
CASE 1. Integration
Under integration, the agent chooses e [member of] [0,1 ] so as to
maximize [u.sup.n.sub.A] and the associated first order condition is:
(5) [partial derivative][u.sup.n.sub.A]/[partial derivative]e = (1
- E)b - ([partial derivative]g/[partial derivative]e) = 0.
Given our assumed functional forms, this condition implies the
solution:
(6) e* (I) = min {(1 - E)bI/[rho], 1}.
The agent contributes more effort the higher is his private benefit
([partial derivative]e*/[partial derivative]b > 0) and the lower is
the probability that P becomes informed ([partial derivative]e*/[partial
derivative]E < 0). Importantly, an increase in training induces A to
work harder ([partial derivative]e*/[partial derivative]I > 0) by
reducing his marginal cost of effort--that is, training is complementary
to effort in this setting.
The principal anticipates e*(I) and chooses the level of training,
[I.sup.n], and the wage, [w.sup.n], that maximize her payoff subject to
A's participation constraint:
[mathematical expression not reproducible]
Letting [[mu].sub.1] denote the multiplier of PC in the associated
Lagrangian ([L.sub.1]) and assuming an interior solution (w > 0 and 0
< e* < 1), the first-order conditions with respect to I and w are:
(7) [mathematical expression not reproducible]
(8) [partial derivative]L/[partial derivative]w = -1+[[mu].sub.1]
=0.
Since [[mu].sub.1] = 1 >0, the participation constraint will be
binding at the optimal solution implying that P can use the wage to
extract any variation in the surplus due to training. In Equation (7),
the first term is P's marginal benefit associated with the positive
impact of training on A's effort incentives while the second term
is simply the marginal cost of training. Since [partial
derivative][u.sup.n.sub.A]/ [partial derivative]I = [(1 - E).sup.2]
[b.sup.2]/2[rho] > 0, training raises A's utility and the third
term in Equation (7) shows that P has stronger training incentives when
taking A's participation considerations into account. The solution
of the above problem yields [I.sup.n] and [w.sup.n] which can then be
substituted into e*(l) to derive the level of effort, [e.sup.n], as
summarized in Equation (9):
(9) ([I.sup.n], [w.sup.n], [e.sup.n]) = ([(1 - E).sup.2] b (2aB +
b) /2[theta][rho], [bar.u] - E[beta]b-([b.sup.2] [(1 -E).sup.2]
[I.sup.n] )/2[rho], ((1 - E)b[I.sup.n])/[rho]).
The above expression evidently shows that training and wages are
treated as substitutes by P, since the provision of more training
enables the principal to lower the wage while keeping A's
participation constraint satisfied. (9)
CASE 2. Delegation
When formal authority is delegated to A, the latter chooses e
[member of] [0,1 ] so as to maximize [u.sup.d.sub.A] and the first-order
condition is written as:
(10) [partial derivative][u.sup.d.sub.A]/[partial derivative]e = (1
- [beta]E)b- ([partial derivative]g/[partial derivative]e) = 0.
The solution yields the optimal effort function:
(11) [??](I) = min{((1-[beta]E)bI/[rho]),1}
where [partial derivative][??]/[partial derivative]b > 0,
[partial derivative][??]/ [partial derivative]E < 0 (as before) and
[partial derivative][??]/[partial derivative][beta] < 0 (since a
higher [beta] increases A's payoff from implementing P's
preferred project and so dampens A's incentive to become informed
himself). The impact of training on effort is again positive but
stronger than under integration ([partial derivative][??]/[partial
derivative]I > [partial derivative]e*/[partial derivative]I).
Comparing the optimal effort choices also reveals that [??](I) > e*
(I). Given the training level, A faces the same marginal effort cost
under either authority structure but reaps a higher marginal benefit
under delegation (since (1 - [beta]E)b > (1 - E)b) and thus has
stronger incentives to become informed in this case.
Anticipating the new optimal choice [??](I), P now selects the
training intensity, [I.sup.d], and the wage, [w.sup.d], which solve the
following problem:
[mathematical expression not reproducible]
Denoting by [[mu].sub.2] the multiplier of PC in the associated
Lagrangian ([L.sub.2]) and assuming again an interior solution, the
first-order conditions with respect to I and w become:
(12) [mathematical expression not reproducible]
(13) [partial derivative][L.sub.2]/[partial derivative]w = -1
+[[mu].sub.2] =0.
As before, [[mu].sub.2] = 1 > 0 and the participation constraint
binds at the optimal solution. In Equation (12), the term ([partial
derivative][??]/ [partial derivative]I) x aB represents P's
marginal benefit from training due to fostering A's effort
incentives. Yet, the term ([partial derivative][??]/ [partial
derivative]I) x EB represents a marginal cost associated with the
reduced likelihood that P receives B from her own preferred project and
c'(I) is the marginal cost of training. Since [partial
derivative][u.sup.d.sub.A]/[partial derivative]I = [b.sup.2] [(1 -
[beta]E).sup.2]/2[rho] > [partial derivative][u.sup.n.sub.A]/[partial
derivative]I, training increases A's utility relatively more under
delegation and P takes this stronger effect into account when choosing
the level of human capital investment. Solving for [I.sup.d] and
[w.sup.d] and substituting back into [??](I) to derive the effort level,
[e.sup.d], we finally obtain (10):
(14) [mathematical expression not reproducible]
C. The Impact of Delegation on Training Intensity
The outcomes derived in Equations (9) and
(14) can be compared to state the following Proposition.
PROPOSITION 1. When formal authority is delegated, the equilibrium
training and effort intensity as well as the wage level can be either
higher or lower than under integration. In particular:
i. [I.sup.d] < [I.sup.n] for a [member of] (0,[??]) and
[I.sup.d] > [I.sup.n] for a [member of] ([??],1]
ii. [e.sup.d] < [e.sup.n] for a [member of] (0,[??]) and
[e.sup.d] > [e.sup.n] for a [member of] ([??],1]
iii. [w.sup.d] > [w.sup.n] for a [member of] (0,[bar.a]) and
[w.sup.d] > [w.sup.n] for a [member of] ([bar.a],1]
where [mathematical expression not reproducible]
The intuition underpinning Proposition 1 can be understood by also
investigating the conditions under which the principal optimally chooses
to delegate authority. For this purpose, we write P's payoffs under
integration and delegation as:
(15) [u.sup.n.sub.p] = E x B + (1-E)[e.sup.n] x
aB-c([I.sup.n])-[w.sup.n] = E(B + [beta]b)-[bar.u] +
[theta][([I.sup.n]).sup.2]/2
(16) [u.sup.n.sub.p] = [e.sup.d] x aB + (1 -[e.sup.d]) x
EB-c([I.sup.d])-[w.sup.d] = E(B + [beta]b)-[bar.u] +
[theta][([I.sup.n]).sup.2]/2
This reformulation, combined with part (i) of Proposition 1,
immediately leads to the following result.
PROPOSITION 2. The principal prefers delegation if and only if the
congruence parameter a is sufficiently high: [u.sup.d.sub.p] <
[u.sup.n.sub.p] for a [member of] (0, [??]) and [u.sup.d.sub.p] >
[u.sup.n.sub.p] for a [member of] (2, l].
The central tradeoff associated with delegation here is the same as
in AT's seminal model. Delegation induces A to work harder (given
the training intensity) and thus tends to increase P's expected
payoff (this is the incentive effect). At the same time, however, it may
lead to the selection of projects which are less preferred by P (this is
the loss-of-control effect). When the congruence parameter [alpha] is
sufficiently high--that is, when A's preferred project yields a
large enough benefit to P--the incentive effect dominates and P chooses
delegation. In our formulation, training is complementary to effort and
proportionally amplifies this tradeoff. Therefore, the conditions under
which delegation and training are positively related are identical to
those determining whether P delegates authority or not (as confirmed by
inspection of Proposition 2 and part (i) of Proposition 1). (11)
The top panel of Figure 2 graphically depicts the relationship
between [I.sup.d] and [I.sup.n], whereas the middle panel shows the
relationship between [e.sup.d] and [e.sup.n]. Since the positive impact
of [alpha] on training and effort is relatively stronger under
delegation, the curves [I.sup.d](a) and [e.sup.d](a) are steeper than
[I.sup.n](a) and [e.sup.n](a) and therefore the training and effort
intensities under delegation exceed those under integration beyond the
threshold values [??] and [??], respectively. The bottom panel of Figure
2 plots the wage levels under each authority structure. Increasing a
induces more training in both cases, thereby enhancing A's utility
and lowering the wage necessary to satisfy A's participation
constraint. The curve [w.sup.d] (a), however, falls faster than
[w.sup.n](a) implying that P pays a relatively lower wage under
delegation when [alpha] exceeds the cutoff level [bar.a]. More
generally, our analysis makes clear that the impact of delegation on
training is ambiguous and depends on the critical congruence parameter,
thus fueling our empirical estimates to identify the dominant pattern.
(12)
IV. DATA AND EMPIRICAL METHODOLOGY
In what follows we detail our data and present our methodology for
examining the influence of delegation on training. We stress the
potential difficulties introduced by using linked data and the need to
hold constant unmeasured establishment specific influences. We also
emphasize the need to account for potential endogeneity. In line with
our theoretical model, the objective is to get as close as possible to a
test of the exogenous influence of delegation.
A. Workplace Employment Relations Surveys Data
We draw data from the 2004 and 2011 Workplace Employment Relations
Surveys (WERS). The surveys randomly select UK workplaces with five or
more employees from the Interdepartmental Business Register, considered
the highest quality available sampling frame. A smaller panel exists of
establishments responding in both surveys. The sampling stratifies by
workplace size and industry with larger workplaces and some industries
overrepresented (Chaplin et al. 2005). As a consequence, all the
estimates we present use workplace weights (separate weights exist for
the cross sections and panel) to ensure that the results are nationally
representative of British workplaces. (13) The sampling weights adjust
for a number of factors influencing the probability of selection, and
the stratification by workplace size and industry (see Kersley et al.
2006). We exclude establishments not in the trading sector (government
and nonprofits) and those with missing data on the critical dependent
variable measuring training and on the main independent variable
capturing delegation.
Nearly all data, including the training measure, come from the
"Management Questionnaire," a face-to-face interview with the
most senior manager with responsibility for personnel matters. We use,
however, the linked "Employee Questionnaire" for our preferred
delegation measure as described below. The response rates for 2004 and
2011 were 64% and 46% yielding 2,295 and 2,680 establishments,
respectively. Response rates are decreasing through time reflecting
business surveys trends (Van Wanrooy et al. 2013). (14) After our
restrictions, the sample sizes are 994 in 2004, 1,012 in 2011, and 474
in the panel.
Managers indicate the proportion of employees formally trained. The
specific question asks: (COFFJOB) "What proportion of experienced
employees in the largest non-managerial occupational group have been
given time off from their normal daily work duties to undertake training
over the past 12 months?" The responses include None (0%), Just a
few (1%-19%), Some (20%-39%), Around half (40%-59%), Most (60%-79%),
Almost all (80%-99%), and All (100%). Table 1 provides the distribution
of responses showing that around 24% of the establishments trained none
of their employees in 2004. This fell to 19% in 2011 and was 17% in the
panel. About 30% of the establishments trained all employees in 2004.
This increased to 35% in 2011 and was about 30% in the panel sample.
(15) We exploit the categorical ranking by using ordered probits for
simple cross-sectional analysis but we must use a linear count variable
(ordering the categories from 1 to 7) for the fixed effect and IV
estimates.
This training measure remains broad and likely includes some
training that is not related to information acquisition. While
recognizing this, we emphasize that the notion of "investment"
from the theoretical model should not be taken too narrowly. Some of the
information acquisition resulting from training could include better
ways to organize the steps in production or how to optimize break times.
While not explicitly investments in "plant and equipment"
these investments seem both a good fit with the theory and likely to
result from a large variety of types of worker training. Also note that
we will make use of an alternative training measure and experiment with
creating a more pointed version of our current measure without
substantial changes in the results. (16)
Our preferred delegation measure (we will examine alternatives)
comes from the employee questionnaire. At each establishment up to 25
employees are randomly selected (every employee is questioned at
establishments with less than 25) and asked: "In general, how much
influence do you have about the range of tasks you do in your job?"
Responses are recorded on a 4-point scale: 1 "None," 2 "A
little," 3 "Some," and 4 "A lot." Following De
Varo and Kurtulus (2010), we identify delegation as present when the
modal response across an establishment's workers is "A
lot" and absent when the modal response is "Some,"
"A little," and "None." Thus, we take the most
frequently occurring worker response to reflect the degree of delegation
in that workplace. (17)
While this measure is subjective, it has been shown to provide a
reasonable proxy for delegation to workers (see De Varo and Kurtulus
2010; De Varo and Prasad 2015). Yet, it differs in critical ways from
other measures of delegation. First, it need not reflect the decision of
actual firm owners. Thus, the delegation we observe may be from managers
to workers, a point we return to in our robustness exercises. Second, it
differs from measures on whether decisions are made centrally or at the
plant level (Meagher and Wait 2014). Despite these differences, it
remains appropriate for thinking about the provision of training.
Table 2 shows that the delegation responses display significant
variation across workplaces and over time. About one out of three
workplaces delegated in 2004 while 49% delegated in 2011 and almost 40%
delegated in the panel.
In supporting information, we show that delegation is most
widespread in manufacturing, utilities, and construction where more than
three-quarters of the workplaces have employees reporting "a
lot" of delegation. The share is intermediate in finance, health,
and education (around half), and relatively small in transport and
communication, wholesale and retail trade. There is no significant
correlation with the extent of training by industry. Training is most
widespread in utilities, transport and communication, education, and
health; intermediate in transport, construction, and manufacturing; and
low in wholesale and retail trade. The supporting information also
breaks the four incentive schemes by industry, as well as offering a
breakdown of the extent of training, delegation, and incentive schemes
by firm size.
B. Empirical Methodology
We seek to determine whether or not delegating to workers increases
a firm's incentive to provide training. We acknowledge that not
only training but also many of the other variables in the WERS are
likely to be endogenous. As made clear, our theoretical model takes
delegation as given and traces out the consequences. Thus, our empirical
strategy initially presents a series of estimations that may be
described as descriptive although we feel we do a good job of
controlling for most of the relevant confounders. Then, we move closer
to testing for an exogenous influence of delegation. This is done
through IV estimates and a series of robustness tests on those IV
estimates.
We initially estimate cross-sectional ordered probits in which the
categorical training measure depends on delegation. (18) Since our
delegation measure is built up from the employee questionnaire, we face
a typical generated regressor problem (see, e.g., Murphy and Topel 1985;
Pagan 1984). In response, we bootstrap the data using 1,000 replications
with replacement and throughout we report only bootstrapped standard
errors.
We first present the ordered probit of training against the
delegation measure and a limited set of controls. We recognize that
fixed costs in establishing training imply that larger organizations
provide training more efficiently (Barron, Berger, and Black 1997;
Black, Noel, and Wang 1999; Booth 1991; Holtmann and Idson 1991).
Moreover, such training programs may take time to develop and may
reflect the permanence and scope of the establishment. Thus, we control
for the (log) number of employees, if the workplace has been operating
for more than 5 years, whether the workplace is part of a larger
organization (i.e., multiworkplace) or is a single independent
workplace, and if the workplace is UK owned/controlled.
In the second estimate, we capture workforce characteristics known
to influence training provision. These include the percentage of
employees using computers, of female employees (Green and Zanchi 1997),
of part-time employees, and of trade union members (Boheim and Booth
2004; Dustmann and Schonberg 2007; Green, Machin, and Wilkinson 1999).
Recognizing the connection between the incentive to train and the extent
of labor mobility (Arulampalam and Booth 1998), we also control for the
percentages of employees on fixed-term contract, of temporary agency
employees, and of employees who separate and quit in the previous year.
We also add controls for the educational attainment of the workforce and
the share of the workforce in each of eight occupational groups and we
include seven dummies identifying the largest nonmanagerial occupational
group.
In a third estimate, we capture variation of training across
industries and regions by adding ten industry dummies and nine region
dummies. The fourth estimate adds variables for performance pay and
market structure (Bilanakos et al. 2017) and represents our most
complete specification. There are four indicators of performance pay
(whether or not the nonmanagerial workers receive payment by result,
merit pay, profit-related pay, or share ownership). The market structure
variables are whether there are no, few, or many competitors and whether
the product market is growing, mature, turbulent, or declining. (19)
We recognize the possibility of unmeasured establishment
characteristics that influence both the extent of training and
delegation. Thus, superior management may both train more and delegate
more. Failing to control for management quality could bias the
cross-section results. We respond by estimating workplace fixed-effect
models. The resulting within-establishment variation eliminates the
influence of unmeasured time-invariant determinants of training allowing
a potentially superior estimate. As fixed-effect ordered probits suffer
from the incidental parameter problem associated with many nonlinear
estimates (see Greene 2001), we supplement our analysis with ordinary
least squares (OLS) and Poisson fixed-effect estimates that do not
suffer from this problem (Hilbe and Green 2008). The results across the
estimates remain very similar and continue to show an important role for
delegation.
We further recognize that fixed-effect estimates need not eliminate
the possibility of endogeneity. Thus, superior management may be new to
the establishment generating a spurious correlation even in the
fixed-effect estimates. Moreover, training may determine delegation.
Thus, it could be that only once an establishment has trained its
workforce will it have trust in its ability to meaningfully delegate
authority. To account for such fears of endogeneity and reverse
causation, we adopt an IV strategy based on industrial aggregation
(Fisman and Svensson 2007; Lai and Ng 2014) that we describe in detail
when presenting the results. We implement this IV strategy both for the
cross-section and workplace fixed-effect estimates in the panel.
Finally, we undertake a series of robustness exercises designed to
probe the stability and reliability of the relationship. These involve
alternative variables for the key concepts, altering specifications, and
estimating within critical subsamples. The results appear remarkably
robust and at least point strongly toward an exogenous influence of
delegation.
V. EMPIRICAL RESULTS
A. Ordered Probit Analysis
The first column of Table 3 presents the estimate for 2004. The
coefficient on employee delegation is positive and statistically
significant fitting the contention that delegation increases the extent
of training. It also shows the traditional result that larger
establishments provide more training. Column 2 adds workforce
characteristics indicating that establishments provide more training
when their employees work with computers, are female, unionized, and
employed full time. The magnitude of the estimated coefficient on
delegation increases and becomes significant at the 1% level. Column 3
shows that the magnitude of the coefficient on delegation increases
again after allowing the extent of training to differ across industries,
occupations, and regions. Finally, column 4 shows that dominant firms
train more (see Bilanakos et al. 2017) as do firms that provide
profit-related pay. Worker mobility, however, is associated with less
training. The coefficient on delegation retains size and significance.
There exists no indication that more complete specifications reduce the
role of delegation.
Column 5 provides estimates from an OLS model that treats training
as a cardinal count value from 1 to 7. The coefficient indicates that
delegation is associated with an increase of 0.422 of a training
category. While we will shortly present the full marginal effects from
the ordered probit, we report the OLS in order to make comparisons with
the other linear estimates that we report later.
In Table 4, we reproduce the series of estimates for 2011. The
pattern of the controls and the size and significance of the delegation
coefficient remain remarkably similar. Again, in column 5, we present an
OLS estimate indicating that delegation is associated with an increase
of 0.488 of a training category.
In Table 5, we report the full marginal effects of delegation from
the final ordered probit estimates. Column 1 indicates that in 2004
delegation is associated with a decrease of 0.039 in the probability of
offering no training, and an increase of 0.077 in the probability of
training all workers. Column 2 indicates that in 2011, delegation is
associated with a decrease of 0.030 in the probability of offering no
training and an increase of 0.090 in the probability of offering
training to all workers. An increase of 0.090 represents a 26% increase
on the mean probability of 0.349. The marginal effects are broadly
similar across the two surveys and suggest that the magnitudes of the
statistical relationship are economically consequential. Workplaces that
delegate offer more training, a relationship that we now probe more
deeply.
B. Panel Estimates
Despite the fact that our measure of delegation comes from the
employee questionnaire, an innovation in the WERS allows us to retain
our delegation measure in the workplace panel. Prior to the most recent
two waves (2011 and 2004), the WERS panel was a separate set of
establishments that could not be taken back to the linked employee data
available in the cross sections. For the first time, the panel is now
part of the cross sections and so linked to the employee data for 2011
and 2004. Incorporating workplace fixed effects removes time-invariant
unobserved heterogeneity such as some workplaces having routinely
superior management. Thus, it presents a relevant robustness exercise.
Column 1 of Table 6 presents a pooled ordered probit estimate
without fixed effects on the panel of workplaces. The delegation
coefficient is very close in size to those from the cross sections
suggesting that the panel is not a highly selected sample. Column 2
presents the OLS estimate which treats the ordered categories of
training as cardinal and indicates that delegation is associated with
0.379 of a category more training. (20)
Column 3 shows that the OLS fixed-effect coefficient remains highly
significant and modestly larger in magnitude than in the pooled OLS.
Thus, there appears no evidence that unmeasured time-invariant
characteristics generate a downward bias. The column 3 estimates
indicate that firms which delegate increase their training by 0.389 of a
category.
As a robustness check, we estimate alternative functional forms for
the fixed-effects estimate. The dependent variable may be better
considered a count variable allowing estimation of the fixed-effect
Poisson regression. This is one of the few nonlinear fixed-effect
estimators without incidental parameters concerns (Hilbe and Green
2008). The results are presented in column 4 and show that the estimated
coefficient retains an economically significant magnitude and
statistical significance. We also estimate conditional fixed-effect
logits (which eliminate the incidental parameter problem) by dividing
the categories of training into high and low, as well as fixed-effect
ordered probits (available as a canned routine in LIMDEP) that retain
the incidental parameter problem. We present the results in Appendix S1,
Table S1, Supporting Information, to this paper and confirm the pattern
of estimates shown in Table 6. The estimates are remarkably similar
across establishments and within establishments, thus when an
establishment moves to delegate it provides more training. (21)
C. Endogeneity and IV Estimates
While the fixed-effect estimates are reassuring, we recognize that
the positive correlation between delegation and training could still
emerge endogenously. As an illustration, superior management could
arrive between the two observations. Here, a critical determinant is not
time invariant. More dramatically, the causation could be reversed. A
more trained workforce reassures the firm that employees are
sufficiently equipped to have more influence.
To examine these possibilities, we undertake an IV strategy based
on industrial aggregation (examples include Fisman and Svensson 2007 and
Lai and Ng 2014). The strategy posits that unmeasured characteristics of
an industry help to define the extent of delegation by workplaces within
that industry. In our case, these industry characteristics stand as
exogenous influences that make it more or less likely that firms within
the industry will delegate. They may reflect the nature of the product
and the underlying technology. The empirical implementation generates an
identifying variable that aggregates the delegation indicator. This
aggregate varies by workplace within the industry by excluding the
workplace for which it is computed. Thus, the identifying variable is
the proportion of workplaces in industry cells reporting "A
lot" of delegation after removing the given workplace from the
industry cell.
Table 7 provides estimates from the panel and again contrasts an
estimate that does not hold constant workplace fixed effects (columns 1
and 2) with one that does hold them constant (columns 3 and 4). Using
two-stage least squares (2SLS), linear probability models are estimated
in the first stages for the endogenous delegation indicators with the
cardinal value of the training variable as the ultimate second
stage-dependent variable. The second stage returns the estimated values
from the first stage along with the joint variables to estimates on
training. The first stage routinely shows a strong positive correlation
between the industry average and the excluded establishment value. There
is no evidence of a weak instrument and the other diagnostics are also
supportive. The second-stage panel estimate of delegation is very close
to the magnitudes for the two cross sections, again showing that the
panel data is not a highly selected sample. (22) Here, the standard
errors are clustered at the workplace level.
Columns 3-4 combine the IV strategy with a workplace fixed-effect
estimate. We alter the procedure slightly from the pooled estimates by
now clustering errors at the industry level as each pair of workplace
observations now contribute at most once to the workplace fixed-effect
estimate. Column 3 presents the first stage results. Estimates suggest
that establishments that are part of a larger organization, offer
payment by result, and operate in growing markets are more likely to
delegate. The latter two results are in line with Lo et al. (2016) who
find that price delegation is higher when firms offer pay for
performance and when there is less market uncertainty. Column 4 presents
the second stage results. This second-stage IV estimate with workplace
fixed effects is modestly larger. The results indicate an increase of
0.538 of a category. This is noticeably larger than the comparable
fixed-effect estimate of 0.389 of a category without the IV (see column
3 of Table 6). As the bands of training are roughly 20 percentage points
each, moving 0.538 of a band is roughly equivalent to training an
additional 11% of the establishment's workforce. The diagnostics
continue to be supportive. (23)
These estimates suggest that plausibly independent movements of
delegation are associated with the extent of training. The fact that the
IV estimate tends to be larger may reflect that instrumenting has
reduced measurement error and the associated attenuation bias.
Alternatively, it may reflect a truly larger local treatment effect. The
critical point remains that there exists no evidence that failure to
instrument generates an upward bias. When the IV is combined with the
fixed-effect estimates, they add confidence to our results and seem
sensible. Moreover, these estimates come closer to testing our
theoretical notion that exogenous changes in delegation increase
training.
Yet, we recognize that our IV does not guarantee unbiasedness when
some other potentially endogenous variables remain uninstrumented. So we
ran a variation that limited the included controls to only the region
and industry dummies. These specifications look broadly similar with no
indication of a weak instrument, and an IV indicating that delegation is
associated with a significant increase of 0.388 of a training category
(Appendix S1, Table S3).
We undertook a series of additional variations in our IV strategy.
With better data we could generate a lagged IV that would eliminate the
data of the specific firm being considered and also use data from an
earlier wave. Such lagging can help generate IVs that are less likely to
be endogenous. Unfortunately, we can trace firms for only two waves
(2004 and 2011) so we cannot fully lag our current IV. We did estimate
the 2011 training estimates using the 2004 instrument for delegation. We
use the panel data from 2004 to generate industry averages that exclude
the specific firm. This is used as an exogenous variable to generate the
2011 establishment-specific IV. This is presented in the first two
columns of Table 8 and looks broadly similar both to the cross-sectional
IV estimates in Appendix S1, Table S1 and to the results in Table 7
suggesting that delegation is associated with 0.728 of a category
increase in training. Note that by using the panel, we eliminate
concerns that firms observed in 2011 might have entered in response to
2004 conditions. The identifying assumption is that past decisions by
firms in this industry are exogenous to today's delegation decision
by the firm in question, even though the firm in question was present in
the industry in the past.
We varied this strategy by using the larger cross sections in 2004
and 2011 (which are not linked). We use the earlier cross section to
aggregate industry averages that we use as excluded variables to
generate a new 2011 IV. There is no individual establishment variation
in these averages as the firms are not linked (the variation is by
industry). We limit attention to only those 2011 establishments more
than 7 years old. This eliminates firms that may have entered the
industry in response to critical endogenous variables. The familiar
result of delegation being associated with about a half a training
category increase emerges and remains weakly significant in otherwise
supportive IV estimates as shown in column 4 of Table 8.
As the estimates above cannot allow us to include establishment
fixed effects, we modify our lagged IV to test a third alternative. We
use the 1998 cross section of the WERS. We cannot follow individual
firms from the 2004 and 2011 panel back to 1998 and so cannot link
firm-level delegation between 2004 and 1998. As an alternative, we
simply use the industrial aggregation across the 1998 cross section to
predict individual establishment delegation in the 2004 panel. We match
this by using the industrial aggregation across the 2004 cross section
to predict individual establishment delegation in the 2011 panel. The
results are shown in column 6 of Table 8 and also remain broadly similar
to those in Table 7. While we still cannot exclude the specific
establishment from the aggregation (and so generate within industry
variation in the excluded variable), we can both use a lagged measure
and estimate the fixed effect on the panel.
Thus, all three of these variations in the IV strategy remain
supportive. While one might think of other variations, these give no
indication that the simultaneity in our initial IV spuriously generates
our results.
D. Additional Robustness Tests and Discussion
We now undertake a series of robustness tests that bolster the
empirical results. First, we identify alternative potential measures of
delegation. These come from the Management Questionnaire and so provide
an alternative view to that built up from the actual workers. The first
asks managers: "To what extent would you say that the largest
occupational group here have discretion over how they do their
work?" "A lot," "Some," "Little,"
"None." We identify delegation as if the manager replies
"A lot." This response is limited to the largest occupational
group but we use it as the critical measure in estimates that mimic the
fixed-effect IV estimate in Table 7 (columns 3 and 4). (24) The IV
continues to perform well and suggests that delegation using this
measure is associated with a significant 0.545 of a category increase in
training.
A second alternative asks managers: "To what extent would you
say that the largest occupational group have involvement in decisions
over how their work is organized?" "A lot,"
"Some," "Little," "None." Again, we
identify delegation as when the manager replies "A lot." This
is also limited to the largest occupational group and now emphasizes the
role in decision-making of workers over their own work. While it may be
only an aspect of delegation, it continues to show an association with
training. In the fixed-effect IV estimates, it is associated with a
significant 0.376 of a category increase in training. We have combined
this measure with the previous one using principal components and used
the resulting variable as a delegation measure. (25) It again takes a
meaningfully large coefficient in the fixed-effect IV estimates. All
three of these estimates are presented in Appendix Table B2.
As a related robustness test, we return to the dependent variable
on training and augment it with a follow-up question in the WERS. Those
establishments which provide training to their workforce are asked the
typical number of training days provided in the last year. The responses
are listed in one of six categories with all those that provide no
training placed in the lowest category: "No time" (1),
"Less than 1 day" (2), "1 to less than 2 days" (3),
"2 to less than 5 days" (4), "5 to less than 10
days" (5), and "10 days or more" (6). This provides a
measure of training intensity different from the share of workers being
trained. We repeat the series of estimations using this alternative
measure and present the results with the panel IV estimates with and
without workplace fixed effects in Table 9. (26)
The first stages continue to suggest the absence of weak
instruments and the addition of the fixed-effect estimation suggests
only a modest increase in the coefficient of interest. The final column
in Table 9 suggests that delegation is associated with an increase of
0.529 of a category in the days of training measure. Combining this with
the earlier estimates suggests that workplaces which delegate both train
a larger share of their workers and provide more training time.
We now return to the potential role of incentive schemes. Workplace
incentives vary dramatically in terms of whom they target, what they
reward, and how large a share of compensation they represent. The WERS
provides information on whether the establishment makes use of four
types of performance pay: payment by result, merit pay, profit-related
pay, and employee share ownership schemes. The objective of each may
well differ but it seems likely that payment by result and merit pay tie
individual effort and decisions to compensation in a very immediate way
and so are designed to align the interests of workers with the firm as
suggested in Appendix A. Thus, they potentially inform our earlier
review of the literature, theory, and testing. If they succeed in such
alignment, the investment in training can be anticipated to be greater
once delegation has taken place.
We recognize that incentive schemes are likely to be highly
endogenous as is delegation so we are limited in what we can credibly
test. We present largely as descriptive a simple division of the sample
into those firms that make use of either (or both) of the two individual
incentive schemes and those that do not. We reproduce our initial
fixed-effect IV estimates separately for these two groups of firms in
Table 10.
The "high incentives" group of establishments that use
the individual-level schemes presents the typical results. The IV
strategy remains sensible (no weak instruments) and the results fit the
intuition we have developed. Delegation continues to play an independent
role and is associated with a significant increase of 0.524 of a
training category. While the IV strategy still seems workable in the
"low incentives" establishments (those that do not use the
individual incentives) the coefficient is insignificant. Yet, this does
not seem to be because of a large diminution in coefficient size. The
point estimate admittedly drops to 0.421 but it is imprecision that
seemingly causes this lack of statistical significance. In any event, it
seems appropriate given our earlier description that the result should
be stronger among those establishments using higher powered individual
incentives.
As a final robustness check, we recognize that in some
establishments owner-managers make training and delegation decisions
while in other establishments, hired-managers make these decisions.
While we have assumed that hired-managers act in the owner's
interest, this may not be the case. To examine the empirical pattern, we
divide our sample by a question that asks: "Are the controlling
owners actively involved in day-to-day management of this workplace on a
full-time basis?" The responses identify slightly more than
one-fifth of the sample with an owner-manager and the remainder with a
hired-manager. We repeat the IV exercise on the divided sample and using
our original, preferred measures of training and delegation. The results
for owner-manager establishments reported in columns 5 and 6 in Table 10
reveal a very large and significant role for delegation. Indeed, the
implied increase of virtually a full training category stands as the
largest magnitude of any of our estimates. We see this as the tight fit
with our theoretical model as the delegation decision involves the
owner.
The results for the hired-manager establishments reported in
columns 7 and 8 in Table 10 reveal a smaller but still positive and
significant influence for delegation. The implied increase remains about
one-half of a training category. This attenuation in magnitude may flow
from agency problems between owners and hired managers. Yet, a full
modeling of such a three-tier hierarchy is beyond the scope of this
paper. We, nonetheless, find it reassuring that the relationship remains
intact as it suggests that hired-managers broadly follow the pattern of
owner-managers.
These robustness exercises, together with the original results,
inform the theoretical issue we initially isolated. If the firm
delegates, then it suffers a loss of control but at the same time gives
workers stronger effort incentives. If training reduces the marginal
cost of effort, then the resulting increase in effort can justify the
cost of additional training. Thus, one might anticipate that delegation
generates greater employer-provided training, a result routinely
returned in our empirical investigation.
VI. CONCLUSION
In this paper, we uniquely provide a theoretical illustration and
supporting empirical evidence showing that delegation increases employee
training. We impose delegation or integration exogenously in our model.
This can be sensible for some cases (new outside management) and as a
reflection of structural differences between industries (which we
exploit in our IV estimation). Yet, it will not apply to all cases.
Within this framework, our nonequilibrium approach of looking at
integrating and delegating firms separately provides a sensible and
testable association between delegation and training. We recognize,
however, that it may not help inform those interested in the exact
pathways that result in the endogenous choice of delegation as part of a
simultaneous bundle of management practices.
We assume that employer-provided training reduces the agent's
marginal effort cost of becoming informed about the payoffs of
alternative investment projects. We show that delegation of
decision-making authority increases training if the preferences of the
firm and the worker are sufficiently congruent. When this holds, the
positive effect of delegation on the worker's effort incentives
dominates the loss-of-control effect and therefore the firm becomes
willing to provide additional training.
We test the hypothesis of a positive relationship between
delegation and training on two cross sections and an associated panel of
British establishments. Our preferred measure of delegation is built up
from workers within each establishment and time period. It identifies
delegation when the modal response of the workers is that they have a
lot of influence over their tasks. Indeed, we confirm that those
establishments which delegate also provide training to a larger share of
their workers. This remains true in increasingly more complete
specifications, when accounting for establishment fixed effects, using
alternative functional forms and in a plausible IV exercise that also
controls for establishment fixed effects. The result also proves robust
when altering the dependent variable to capture training intensity in
number of days, and to alternative measures of delegation coming from
the management questionnaire.
This result argues that those workplaces where there may be
particularly good information at the level of the worker will want to
delegate but that they will also want to engage in more training than
those firms which do not delegate. Future "insider"
econometrics might provide important insights that support or refute
this argument. It would be wonderful to identify a specific
establishment that devolves to workers tasks or choices previously
undertaken by the management. Our survey evidence would suggest that
such devolution would be accompanied by increased worker training so
that superior choices would be made. Also developing insights with
survey data from other countries concerning delegation and training
seems a sensible next step. Furthermore, the positive interaction
between incentive pay and delegation revealed by our evidence might
inform a more sophisticated theoretical model in which congruence could
be affected by appropriately designed incentive schemes. The full
exploration of this possibility is an additional task for future
research.
Finally, we recognize limitations of our examination. Delegation is
a subjective employee measure aggregated to the workplace level. While
the alternative measures from the management survey provide some
comfort, we recognize that an objective employee measure may provide
additional insight. Also, we have not modeled or tested a multilevel
hierarchy of owners, managers, and workers nor have we controlled for
endogenous incentive schemes. Despite these open questions, we provide
various alternative measures of delegation and confirm the robustness of
our results to a large number of sensitivity checks, thus establishing
an important contribution on which further work can build.
APPENDIX A
We extend our baseline model to incorporate incentive pay. We
follow Aghion and Tirole (1997) by assuming there are only two relevant
projects yielding nonnegative payoffs to P and A. Specifically, P reaps
payoff B > 0 from one of these two projects and zero from the other.
Similarly, A's benefit is b > 0 from one of the two relevant
projects and zero from the other. The degree of congruence is
represented by the probability a = [beta] [member of] (0,1 ] that both P
and A prefer the same project. Incentive pay is introduced through the
assumption that A receives wage w [greater than or equal to] 0 if
P's payoff is B and zero otherwise. Suppose also that w < B and
w < b (if the latter assumption is violated, wage incentives are so
strong that A always recommends P's preferred project--i.e.,
incentives are aligned). In this context, the payoffs under integration
and delegation become
(A1) [u.sup.n.sub.p] = E x (B- w) + (1 -E)e x a(B- w)-c(I)
(A2) [u.sup.n.sub.A] = E x (w + ab) + (1 - E)e x (b + aw) - g(e,I)
(A3) [u.sup.d.sub.p] = e x a(B-w) + (1 - e)E x (B - w) - c (I)
(A4) [u.sup.d.sub.A] = e x (b + aw) + (1 - e)E x (w + ab) -g(e,I).
In the case of integration, A chooses e to maximize [u.sup.n.sub.A]
in Equation (A2) implying the first-order condition (1 - E) x (b + aw) -
[partial derivative]g/[partial derivative]e = 0, which yields the
interior solution:
(A5) e* (I, w) = ((1 - E)(b + aw) I) /[rho].
Since [partial derivative]e*/[partial derivative]I > 0 and
[partial derivative]e*/[partial derivative]w > 0 both training and
monetary incentives can now be used by the employer as alternative
instruments to increase A's effort contribution. Anticipating
Equation (A5), P sets w and I so as to maximize [u.sup.n.sub.p] in
Equation (A1) subject to A's participation constraint. Letting
[[mu].sub.3] denote the multiplier of this constraint in the associated
Lagrangian ([L.sub.3]), we can write the first-order conditions:
(A6) [mathematical expression not reproducible]
(A7) [mathematical expression not reproducible]
(A8) [mathematical expression not reproducible]
where [mathematical expression not reproducible] The
principal's choices of I and w balance the marginal benefit
associated with increased worker effort against the marginal cost
resulting from increased training and wage expenditures, respectively,
while also taking into account that higher values of w and I raise
A's utility (thus making it easier to satisfy the latter's
participation constraint). Solving Equations (A6) to (A8) yields the
intensity of training ([I.sup.n]) and wage incentives ([w.sup.n]) which
can then be substituted back into Equation (A5) to derive A's
effort ([e.sup.n]) for the case of integration.
Similarly, under delegation A maximizes [u.sup.d.sub.A] in Equation
(A4) with respect to e and the first-order condition b(1 - aE) + (a -
E)w - [partial derivative]g/[partial derivative]e = 0 implies the
interior solution:
(A9) [??](I,w) = [b)1-aE) + (a-E)w]xI/[rho].
A direct comparison of Equations (A5) and (A9) reveals that [??](I,
w) > e* (I, w) when incentives are not aligned (i.e., for w < b).
Given the levels of training and wage, A is relatively more willing to
provide effort under delegation (as in the case without incentive pay).
Since [partial derivative][??]/ [partial derivative]I > [partial
derivative]e*/[partial derivative]I and [partial derivative][??]/
[partial derivative]w < [partial derivative]e*/[partial derivative]w,
the positive impact of training on effort is relatively stronger under
delegation but the positive impact of monetary incentives on effort is
relatively stronger under integration. Moreover, for a < E (i.e.,
when the probability that P and A prefer the same project is lower than
the probability that P becomes informed) an increase in w reduces
A's marginal benefit from effort provision and thus weakens his
incentives to become informed himself ([partial derivative][??]/[partial
derivative]w < 0).
Anticipating Equation (A9), P now selects the levels of w and I
that maximize [u.sup.d.sub.p] in Equation (A3) subject to A's
participation constraint. If we denote by [[mu].sub.4] the multiplier of
this constraint in the associated Lagrangian ([L.sub.4]), we get the
following first-order conditions:
(A10) [mathematical expression not reproducible]
(A11) [mathematical expression not reproducible]
(A12) [mathematical expression not reproducible]
Again, P optimally selects I and w by taking into account their
impact on A's effort, the marginal cost resulting from increased
training and wage expenditures as well as the need to keep A's
participation constraint satisfied. Solving Equations (A10) to (A 12)
yields the levels of training ([I.sup.d]) and wage ([w.sup.d]) under
delegation, which can then be used to compute worker effort ([e.sup.d])
from Equation (A9). In principle, the outcome ([I.sup.d], [w.sup.d],
[e.sup.d]) can be compared to ([I.sup.n], [w.sup.n], [e.sup.n]) as
already done in the main body of the article without the presence of
incentive pay. While we leave such a thorough investigation of the
relationship between delegation, training, and incentive pay as an open
question for future research, in Table 10 of the main paper (columns
1-4), we examine the presence of highpowered incentives (individual
payment by results or merit pay) or not. The results show that
delegation has a large and significant impact on training only for
establishments with the higher powered incentive pay.
APPENDIX B
TABLE B1
Descriptive Statistics of All Variables
WERS 2004 WERS 2011
Variable M SD M SD
Log number of employees 2.640 0.918 2.667 0.885
Workplace operates more than 5 years 0.844 0.363 0.867 0.340
Part of a larger organization 0.589 0.492 0.508 0.500
Single independent workplace 0.390 0.488 0.473 0.500
UK owned/controlled (predominantly 0.861 0.346 0.874 0.332
UK owned 51% or more)
% of employees using computers 0.523 0.403 0.607 0.398
% of female employees 0.524 0.324 0.525 0.319
% of part-time employees 0.330 0.302 0.304 0.301
% union membership 0.083 0.214 0.037 0.134
% of employees on fixed-term 0.040 0.157 0.066 0.206
contract
% of employees on temporary contract 0.015 0.069 0.014 0.077
% of employees who quitted last year 0.179 0.215 0.112 0.145
% of employees dismissed/redundant 0.027 0.069 0.029 0.069
last year
% of employees with "O" levels, 0.108 0.155 0.122 0.194
grades D-E
% of employees with "0" levels, 0.197 0.201 0.325 0.290
grades A-C
% of employees with "A" levels 0.107 0.150 0.113 0.156
% of employees with first degree 0.093 0.143 0.137 0.197
(BA, BSc, BEd, etc.)
% of employees with higher degree 0.025 0.088 0.042 0.113
(MSc, MA, MBA, PhD)
% of employees with other academic 0.281 0.226 0.190 0.228
qualification
% of managers/senior officials 0.142 0.108 0.168 0.117
% of professional staff 0.046 0.132 0.082 0.182
% of technical staff 0.057 0.148 0.094 0.197
% of sales staff 0.244 0.348 0.180 0.300
% of operative and assembly staff 0.091 0.212 0.072 0.191
% of clerical and secretarial staff 0.131 0.200 0.112 0.178
% of craft and skilled staff 0.098 0.226 0.077 0.182
% of personal service staff 0.077 0.232 0.116 0.280
Largest occupational group: 0.048 0.215 0.082 0.274
professional
Largest occupational group: 0.063 0.244 0.116 0.321
technical
Largest occupational group: 0.106 0.308 0.098 0.298
administrative
Largest occupational group: skilled 0.125 0.330 0.106 0.308
Largest occupational group: caring, 0.098 0.297 0.144 0.351
leisure
Largest occupational group: sales 0.296 0.457 0.234 0.424
Largest occupational group: 0.129 0.336 0.102 0.303
operatives
Payment by result 0.280 0.449 0.194 0.395
Merit pay 0.077 0.267 0.135 0.342
Profit-related pay 0.343 0.475 0.314 0.464
Employee share schemes (SIP, SAYE, 0.120 0.325 0.091 0.288
EMI, CSOP, other)
Few competitors 0.334 0.472 0.396 0.489
Many competitors 0.602 0.490 0.578 0.494
Current state of the market: growing 0.467 0.499 0.309 0.462
Current state of the market: mature 0.238 0.426 0.200 0.400
Current state of the market: 0.136 0.343 0.152 0.359
declining
Current state of the market: 0.159 0.365 0.339 0.473
turbulent
Manufacturing 0.130 0.336 0.111 0.314
Utilities (electricity, gas, 0.000 0.021 0.000 0.019
and water)
Construction 0.047 0.213 0.053 0.224
Wholesale and retail 0.296 0.457 0.263 0.441
Hotels and restaurants 0.081 0.273 0.094 0.292
Transport and communication 0.049 0.216 0.037 0.190
Financial services 0.057 0.232 0.004 0.063
Other businesses 0.161 0.368 0.196 0.397
Education 0.009 0.096 0.048 0.213
Health 0.108 0.311 0.133 0.340
North East 0.045 0.208 0.048 0.214
North West 0.111 0.315 0.099 0.298
East Midlands 0.075 0.263 0.071 0.258
West Midlands 0.122 0.328 0.110 0.313
East Anglia 0.048 0.215 0.053 0.223
South East 0.302 0.460 0.310 0.463
South West 0.078 0.268 0.119 0.324
Wales 0.033 0.178 0.031 0.172
Scotland 0.092 0.289 0.089 0.285
Owner-manager firm 0.204 0.403 0.245 0.430
Management gives employees 0.482 0.499 0.497 0.500
information about investment plans
Training intensity: no time 0.261 0.439 0.194 0.395
Training intensity: less than 1 day 0.050 0.218 0.047 0.211
Training intensity: 1 to less than 0.193 0.395 0.199 0.400
2 days
Training intensity: 2 to less than 0.273 0.445 0.334 0.472
5 days
Training intensity: 5 to less than 0.114 0.318 0.128 0.334
10 days
Training intensity: 10 days or more 0.108 0.310 0.095 0.294
Dummy for missing firm age 0.042 0.200 0.026 0.160
Dummy for missing % union membership 0.039 0.193 0.049 0.215
Dummy for missing % of employees on 0.009 0.095 0.000 0.000
fixed-term contract
Dummy for missing % of employees on 0.006 0.077 0.000 0.021
temporary contract
Dummy for missing % of employees 0.040 0.197 0.023 0.149
quitted last year
Dummy for missing % of employees 0.041 0.199 0.019 0.138
dismissed/redundant last year
Observations 994 1,012
Panel 2004-2011
Variable M SD
Log number of employees 3.403 0.942
Workplace operates more than 5 years 0.955 0.207
Part of a larger organization 0.552 0.498
Single independent workplace 0.438 0.497
UK owned/controlled (predominantly 0.894 0.308
UK owned 51% or more)
% of employees using computers 51.762 39.023
% of female employees 0.573 0.314
% of part-time employees 0.406 0.312
% union membership 0.057 0.157
% of employees on fixed-term 0.047 0.163
contract
% of employees on temporary contract 0.011 0.045
% of employees who quitted last year 0.153 0.191
% of employees dismissed/redundant 0.038 0.094
last year
% of employees with "O" levels, 0.131 0.175
grades D-E
% of employees with "0" levels, 0.251 0.214
grades A-C
% of employees with "A" levels 0.117 0.169
% of employees with first degree 0.105 0.132
(BA, BSc, BEd, etc.)
% of employees with higher degree 0.022 0.063
(MSc, MA, MBA, PhD)
% of employees with other academic 0.243 0.203
qualification
% of managers/senior officials 0.113 0.075
% of professional staff 0.059 0.142
% of technical staff 0.062 0.153
% of sales staff 0.228 0.344
% of operative and assembly staff 0.076 0.187
% of clerical and secretarial staff 0.112 0.187
% of craft and skilled staff 0.077 0.170
% of personal service staff 0.160 0.315
Largest occupational group: 0.059 0.236
professional
Largest occupational group: 0.079 0.270
technical
Largest occupational group: 0.088 0.284
administrative
Largest occupational group: skilled 0.112 0.315
Largest occupational group: caring, 0.193 0.395
leisure
Largest occupational group: sales 0.256 0.437
Largest occupational group: 0.097 0.296
operatives
Payment by result 0.217 0.413
Merit pay 0.097 0.296
Profit-related pay 0.318 0.466
Employee share schemes (SIP, SAYE, 0.122 0.328
EMI, CSOP, other)
Few competitors 0.355 0.479
Many competitors 0.598 0.491
Current state of the market: growing 0.353 0.478
Current state of the market: mature 0.220 0.414
Current state of the market: 0.153 0.360
declining
Current state of the market: 0.272 0.445
turbulent
Manufacturing 0.079 0.270
Utilities (electricity, gas, 0.000 0.000
and water)
Construction 0.038 0.193
Wholesale and retail 0.340 0.474
Hotels and restaurants 0.070 0.255
Transport and communication 0.028 0.165
Financial services 0.000 0.000
Other businesses 0.136 0.343
Education 0.021 0.145
Health 0.214 0.411
North East 0.075 0.263
North West 0.215 0.411
East Midlands 0.046 0.209
West Midlands 0.123 0.328
East Anglia 0.030 0.170
South East 0.284 0.451
South West 0.099 0.299
Wales 0.017 0.128
Scotland 0.052 0.223
Owner-manager firm 0.209 0.407
Management gives employees 0.517 0.500
information about investment plans
Training intensity: no time 0.211 0.408
Training intensity: less than 1 day 0.065 0.247
Training intensity: 1 to less than 0.239 0.427
2 days
Training intensity: 2 to less than 0.255 0.436
5 days
Training intensity: 5 to less than 0.132 0.338
10 days
Training intensity: 10 days or more 0.096 0.295
Dummy for missing firm age 0.013 0.113
Dummy for missing % union membership 0.073 0.260
Dummy for missing % of employees on 0.000 0.000
fixed-term contract
Dummy for missing % of employees on 0.000 0.013
temporary contract
Dummy for missing % of employees 0.034 0.181
quitted last year
Dummy for missing % of employees 0.022 0.147
dismissed/redundant last year
Observations 474
Notes: Means and standard deviations for each variable are reported
for the two cross sections and the panel samples. Means are
weighted using workplace weights. Means for variables with missing
observations are estimated on nonmissing observations. SIP, Share
Incentive Plan; SAYE, Save As You Earn; EMI, Enterprise Management
Incentives; CSOP, Company Share Option Plan.
TABLE B2
Instrumental Variable (IV) Results: Panel Data 2004-2011,
Alternative Measures of Delegation
(1) (2)
Discretion
With Workplace
Fixed Effects
First Stage Second Stage
Delegation Training
Delegation 0.545 **
(0.247)
Instrument for delegation 15.883 ***
(2.348)
F-statistic ([H.sub.0]: 128.44
instrument is weak) p val. = .000
Observations 474 474
Industry dummies Yes Yes
Region dummies No No
Workplace characteristics Yes Yes
Workforce characteristics Yes Yes
Occupational composition Yes Yes
Largest occupational Yes Yes
group dummies
PRP dummies Yes Yes
Competition dummies Yes Yes
Market state dummies Yes Yes
Missing dummies Yes Yes
(3) (4)
Involvement
With Workplace
Fixed Effects
First Stage Second Stage
Delegation Training
Delegation 0.376 **
(0.189)
Instrument for delegation 12.246 ***
(1.574)
F-statistic ([H.sub.0]: 142.21
instrument is weak) p val. = .000
Observations 474 474
Industry dummies Yes Yes
Region dummies No No
Workplace characteristics Yes Yes
Workforce characteristics Yes Yes
Occupational composition Yes Yes
Largest occupational Yes Yes
group dummies
PRP dummies Yes Yes
Competition dummies Yes Yes
Market state dummies Yes Yes
Missing dummies Yes Yes
(5) (6)
Principal Component
Analysis
With Workplace
Fixed Effects
First Stage Second Stage
Delegation Training
Delegation 0.265 **
(0.117)
Instrument for delegation 9.324 ***
(1.452)
F-statistic ([H.sub.0]: 95.42
instrument is weak) p val. = .000
Observations 474 474
Industry dummies Yes Yes
Region dummies No No
Workplace characteristics Yes Yes
Workforce characteristics Yes Yes
Occupational composition Yes Yes
Largest occupational Yes Yes
group dummies
PRP dummies Yes Yes
Competition dummies Yes Yes
Market state dummies Yes Yes
Missing dummies Yes Yes
Notes: The dependent variable is the proportion of experienced
employees in the largest occupational group who have been given
time off from their normal daily work to undertake training over
the last 12 months. In columns 1 and 2, delegation is measured from
the Management Questionnaire from the following question: "Using
the scale on this card, to what extent would you say that the
largest occupational group here have discretion over how they do
their work?" "A lot," "Some," Little," "None." We code as
delegation if managers replied "A lot." In columns 3 and 4,
delegation is measured from the management questionnaire from the
following question: "Using the scale on this card, to what extent
would you say that the largest occupational group have involvement
in decisions over how their work is organized?" "A lot," "Some,"
"Little," "None." We code as delegation if managers replied "A
lot." In columns 5 and 6, delegation is constructed using the first
principal component of discretion and involvement. For reasons of
brevity, we only present estimates of the variable of interest.
Other controls are those shown in column 4 of Table 3 in the main
paper, as well as a year dummy. The estimates for the rest of the
covariates are available upon request. Bootstrap standard errors
using 1,000 replications with replacement and clustered at industry
cells are reported in parentheses. Estimates are weighted using
workplace weights. PRP, performance-related pay.
*** p <.01, ** p <.05.
ABBREVIATIONS
2SLS: Two-Stage Least Squares
AT: Aghion and Tirole (1997)
COPE: Colloquium of Personnel Economics
IV: Instrumental Variable
OLS: Ordinary Least Squares
WERS: Workplace Employment Relations Surveys
doi: 10.1111/ecin.12515
Online Early publication October 27, 2017
REFERENCES
Acemoglu, D., and J. S. Pischke. "Why Do Firms Train? Theory
and Evidence." Quarterly Journal of Economics, 113, 1998, 79-119.
--. "Beyond Becker: Training in Imperfect Labour
Markets." Economic Journal, 109, 1999, F112-42.
Aghion, P., N. Bloom, and J. Van Reenan. "Incomplete Contracts
and the Internal Organization of Firms." Journal of Law, Economics.
& Organization, 30, 2014, 37-63.
Aghion. P., and J. Tirole. "Formal and Real Authority in
Organizations." Journal of Political Economy, 105, 1997, 1-29.
Allgulin, M., andT. Ellingsen. "Monitoring and Pay
"Journal of Labor Economics, 20, 2002, 201-16.
Arulampalam, W., and A. Booth. "Training and Labour Market
Flexibility: Is There a Trade-Off?" British Journal of Industrial
Relations, 36, 1998,521-36.
Baker, G.. R. Gibbons, and K. Murphy. "Informal Authority in
Organizations." Journal of Law, Economics, and Organization, 15,
1999, 56-73.
Barron, J. M., M. C. Berger, and D. A. Black. "How Well Do We
Measure Training?" Journal of Labor Economics, 15, 1997, 507-28.
Bester, H., and D. Krahmer. "Delegation and Incentives."
The RAND Journal of Economics, 39, 2008, 664-82.
Bilanakos, C., C. P. Green, J. S. Heywood, and N. Theodoropoulos.
"Do Dominant Firms Provide More Training?" Journal of
Economics and Management Strategy, 26. 2017,67-95.
Black, D. A., B. J. Noel, and Z. Wang. "On-the-Job Training
Establishment Size, and Firm Size: Evidence for Economies of Scale in
the Production of Human Capital." Southern Economic Journal, 66,
1999, 82-100.
Bloom, N., C. Genakos, R. Sadun. and J. Van Reenan.
"Management Practices across Firms and Countries." Academy of
Management Perspectives, 26, 2012a, 12-33.'
Bloom, N., R. Sadun, and J. Van Reenan. "Americans Do IT
Better: US Multinationals and the Productivity Miracle." American
Economic Review, 102, 2012b, 167-201.
Boedker. C., R. Vidgen, K. Meahgher, J. Cogin, J. Mouritsen, and J.
M. Runnnalls. Leadership, Culture and Management Practices of High
Performing Workplaces in Australia: High Performing Workplace Index.
Sydney: Society for Knowledge Economics, 2011.
Boheim, R., and A. Booth. "Trade Union Presence and
Employer-Provided Training in Greater Britain." Industrial
Relations, 43, 2004, 520-44.
Bolton, P., and M. Dewatripont. "Authority in Organizations: A
Survey," in The Handbook of Organizational Economics, edited by R.
Gibbons and J. Roberts. Princeton, NJ: Princeton University Press, 2013,
342-72.
Booth, A. "Job Related Formal Training: Who Receives It and
What Is It Worth?" Oxford Bulletin of Economics and Statistics, 53,
1991, 281-94.
Booth, A., and M. Bryan. "Testing Some Predictions of Human
Capital Theory: New Training Evidence from Britain." Review of
Economics and Statistics, 87, 2005, 391-4.
Brunello, G., and F. Gambarotto. "Do Spatial Agglomeration and
Local Labor Market Competition Affect Employer-Provided Training?
Evidence from the UK." Regional Science and Urban Economics, 37,
2007, 1-21.
Caroli, E., N. Greenan, and D. Guellec. "Organizational Change
and Skill Accumulation." Industrial and Corporate Change, 10, 2001,
481-506.
Chaplin, J., J. Mangla, S. Purdon, and C. Airey. "The
Workplace Employment Relations Survey 2004." Technical Report
(Cross Section and Panel Surveys), National Centre for Social Research,
London, 2005.
Colombo. M., and M. Delmastro. "Delegation of Authority in
Business Organizations: An Empirical Test." Journal of Industrial
Economics, 52, 2004, 53-80.
De Paola, M., and V. Scoppa. "Organizational Design, Project
Selection, and Incentives." Journal of Institutional and
Theoretical Economics, 162, 2006, 424-49.
De Varo, J., and F. A. Kurtulus. "An Empirical Analysis of
Risk, Incentives and the Delegation of Worker Authority."
Industrial and Labor Relations Review, 64, 2010, 641-61.
De Varo, J., and S. Prasad. "The Relationship between
Delegation and Incentives across Occupations: Evidence and Theory."
Journal of Industrial Economics, 63, 2015, 279-312.
Dessein. W. "Authority and Communication in
Organizations." Review of Economic Studies, 69, 2002, 811-38.
Dustmann, C., and U. Schonberg. "Training and Union
Wages." Review of Economics and Statistics, 91, 2007, 363-76.
Fisman, R., and J. Svensson. "Are Corruption and Taxation
Really Harmful to Growth? Firm Level Evidence." Journal of
Development Economics, 83, 2007, 63-75.
Foss, N. J., and K. Laursen. "Performance Pay and Delegation
under Uncertainty and Innovativeness: An Empirical Investigation."
Journal of Economic Behavior & Organization, 58, 2005, 246-76.
Garicano, L., and L. Rayo. "Why Organizations Fail: Models and
Cases." Journal of Economic Literature. 54, 2016, 137-92.
Green, F., and L. Zanchi. "Trends in the Training of Male and
Female Workers in the United Kingdom." British Journal of
Industrial Relations, 35, 1997, 635-44.
Green, F., S. Machin, and D. Wilkinson. "Trade Unions and
Training Practices in British Workplaces." Industrial and Labor
Relations Review, 52, 1999, 179-95.
Greene, W. H. "Estimating Econometric Models with Fixed
Effects." Discussion Paper, Department of Economics, Stern School
of Business, New York University, 2001.
Grossman, S. J., and O. D. Hart. "The Costs and Benefits of
Ownership: A Theory of Vertical and Lateral Integration." Journal
of Political Economy, 94,1986,691 -719.
Hart, O. D., and B. Holmstrom. "A Theory of Firm Scope."
Quarterly Journal of Economics, 125. 2010, 483-513.
Hart, O. D., and J. Moore. "Property Rights and the Nature of
the Firm." Journal of Political Economy, 98, 1990, 1119-58.
Hilbe, J. M., and W. H. Green. "Count Regression Models,"
in Handbook of Statistics (Epidemiology and Medical Statistics), Vol.
27, Chapter 7, edited by C. R. Rao, J. P. Miller, and D. C. Rao.
Amsterdam: North-Holland Publishing, 2008, 210-52.
Holmstrom, B. "On the Theory of Delegation," in Bayesian
Models in Economics Theory, edited by M. Boyer and R. Kihlstrom. New
York: North Holland, 1984, 115-41.
Holtmann, A. G., and T. L. Idson. "Employer Size and on the
Job Training Decisions." Southern Economic Journal, 58, 1991,
339-55.
Ichniowski, C., K. Shaw, and G. Prennushi. "The Effects of
Human Resource Management Practices on Productivity." American
Economic Review, 87, 1997, 291-313.
Itoh, H., T. Kikutani, and O. Hayashida. "Complementarities
among Authority, Accountability and Monitoring: Evidence from Japanese
Business Groups." Journal of the Japanese and International
Economies, 22, 2008, 207-28.
Jensen, M., and W. Meckling. "Specific and General Knowledge
and Organizational Structure," in Contract Economics, edited by L.
Werin and H. Wijkander. Oxford: Blackwell Publishers, 1992, 251-74.
Katz, E., and A. Ziderman. "Investment in General Training:
The Role of Information and Labour Mobility." The Economic Journal
100. 1990, 1147-58.
Kersley, B., A. Carmen, J. Forth, A. Bryson, H. Bewley, G. Dix, and
S. Oxenbridge. Inside the Workplace: Findings from the 2004 Workplace
Employment Relations Survey. Oxford: Routledge, Taylor & Francis
Group, 2006.
Krueger, A. B. "How Computers Have Changed the Wage Structure:
Evidence from Micro Data, 1984-1989." Quarterly Journal of
Economics, 108, 1993, 33-60.
Lai, T.-K., and T. Ng. "The Impact of Product Market
Competition on Training Provision: Evidence from Canada." The
Canadian Journal of Economics, 47, 2014, 856-88.
Lo, D. (H.-F.), W. Dessein, M. Ghosh, and F. Lafontaine.
"Price Delegation and Performance Pay: Evidence from Industrial
Sales Forces." Journal of Law, Economics, and Organization, 32,
2016, 508-44.
Manning, A. Monopsony in Motion: Imperfect Competition in Labor
Markets. Princeton, NJ; Oxford: Princeton University Press, 2003.
Meagher, K. J., and A. Wait. "Delegation of Decisions about
Change in Organizations: The Roles of Competition, Trade, Uncertainty,
and Scale." Journal of Law, Economics, & Organization, 30,
2014, 709-33.
Murphy, K. M., and R. P. Topel. "Estimation and Inference in
Two-Step Econometric Models." Journal of Business & Economic
Statistics, 3, 1985, 370-79.
Nagar, V. "Delegation and Incentive Compensation." The
Accounting Review, 77, 2002, 379-95.
Osterman, P. "How Common Is Workplace Transformation and Who
Adopts It?" Industrial and Labor Relations Review, 47, 1994,
173-88.
Pagan, A. "Econometric Issues in the Analysis of Regressions
with Generated Regressors." International Economic Review, 25,
1984, 221-47.
Rajan, G. R., and J. Wulf. "The Flattening Firm: Evidence from
Panel Data on the Changing Nature of Corporate Hierarchies." Review
of Economics and Statistics, 88, 2006, 759-73.
Stein, J. "Information Production and Capital Allocation:
Decentralized versus Hierarchical Firms." Journal of Finance, 57,
2002, 1891-921.
Van Wanrooy, B., H. Bewley, A. Bryson, J. Forth, S. Freeth, L.
Stokes, and S. Wood. "The 2011 Workplace Employment Relations
Survey: First Findings." 2013. Accessed September 15, 2017.
https://www.gov.uk/government/uploads/sy
stem/uploads/attachment_data/file/210103/13-1010-WERSfirst-
findings-report-third-editionmay-2013.pdf
Zabojnik, J. "Centralized and Decentralized Decision Making in
Organizations." Journal of Labor Economics, 20, 2002, 1-22.
SUPPORTING INFORMATION
Additional Supporting Information may be found in the online
version of this article:
Appendix S1.
Table S1. Panel Data 2004-2011: Alternative Functional Forms
Table S2. Instrumental Variable (IV) Results, Cross Sections
Table S3. A Parsimonious IV Specification
Table S4. Panel Data 2004-2011: Alternative Measures of Delegation
Table S5. Alternative Dependent Variable (Days of Training)
CHRISTOS BILANAKOS, JOHN S. HEYWOOD, JOHN G. SESSIONS and NIKOLAOS
THEODOROPOULOS*
* We acknowledge the Department of Trade and Industry, the Economic
and Social Research Council, the Advisory, Conciliation and Arbitration
Service, and the Policy Studies Institute as the originators of the
Workplace Employee Relations Survey data, and the Data Archive at the
University of Essex as the distributor of the data. None of these
organizations bears any responsibility for our analysis or
interpretation. We thank participants of the 20th Colloquium of
Personnel Economics (COPE, Zurich), the Leeds Festival of Economics,
Democracy and the Workplace, and seminar attendees at the University of
Crete. We also thank the associate editor and two anonymous referees.
Bilanakos: Postdoctoral Researcher, Department of Management
Science and Technology, Athens University of Economics and Business,
Athens, 10434, Greece. Phone +30 6939694996, Fax +30 2106987459, E-mail
[email protected]
Heywood: Distinguished Professor, Department of Economics,
University of Wisconsin-Milwaukee, Milwaukee, WI 53211. Phone (414)
229-4310, Fax (414) 229-5915, E-mail
[email protected]
Sessions: Professor of Economics, Department of Economics and IZA,
University of Bath, Bath, BA2 7AY, UK. Phone +44 1225 384517, Fax +44
1225 383423, E-mail
[email protected]
Theodoropoulos: Assistant Professor of Economics, Department of
Economics, University of Cyprus, Nicosia, CY-1678, Cyprus. Phone +357
22893715, Fax +357 22895028, E-mail
[email protected]
(1.) We refer to the terms "firm," "workplace,"
and "establishment" interchangeably throughout the paper.
(2.) Empirical studies discussing the growth of delegation and its
consequences include Osterman (1994), Caroli, Greenan, and Guellec
(2001), and Rajan and Wulf (2006).
(3.) Another strand of the literature investigates how a
comprehensive collection of human resource practices influences
productivity. This literature is uniformly empirical--see, for example.
Ichniowski, Shaw, and Prennushi (1997).
(4.) Bester and Krahmer (2008) also vary AT by modeling a situation
in which the agent's job is to complete rather than to identify a
project, showing that delegation becomes less attractive in this case.
In another extension, Stein (2002) argues that delegation is most likely
to dominate when information is "soft"--that is, not
verifiable. Relatedly, Dessein (2002) builds a model with asymmetric
information to show that delegation might be the most efficient way for
the principal to extract the agent's private knowledge on
projects' payoffs.
(5.) For empirical evidence on the determinants of firm-sponsored
general training see, for example, Katz and Ziderman (1990), Krueger
(1993), Acemoglu and Pischke (1998), and Booth and Bryan (2005).
(6.) De Paola and Scoppa (2006) argue that delegation, thanks to
on-the-job learning and the possible expropriation of resources, might
well increase an agent's outside option. This increase, in turn,
might increase the quit propensity of the agent and thus the turnover
costs of the firm. Thus, firms should be less likely to delegate the
higher are turnover costs and the lower the degree of firm-specific
human capital.
(7.) We also assume that for each party, there exists at least one
project generating a loss of such magnitude that both P and A prefer
inaction to implementing a random project in the absence of information
about payoffs.
(8.) The assumption that the cross derivative of g(.) has a
negative sign--that is, that A's marginal cost of becoming informed
decreases with training--is a critical driving force of our results,
since it implies that training is complementary to effort and therefore
generates the rationale for positive human capital investment on the
part of the employer. In essence, our formulation assumes that training
increases the worker's productive efficiency in acquiring
information about projects' payoffs. While acknowledging the
possibility of alternative specifications, we consider such a
conceptualization of training as a productivity-enhancing (or,
equivalently, cost-reducing) investment to be reasonable.
(9.) The outcome in Equation (9) holds for
[theta]>[[theta].sup.n] [equivalent to] [b.sup.2][(1 - E).sup.3] (2aB
+ b)/2[[rho].sup.2] (so that [e.sup.n] <1) and sufficiently high
values of u to ensure that [w.sup.n] > 0.
(10.) The results in Equation (14) hold for [theta] =
[[theta].sup.d] [equivalent to] [b.sup.2] [(1 - [beta]E).sup.2][2(a -
E)B + b(1 - [beta]E)/2[[rho].sup.2] (so that [e.sup.d]<1), B>b (1
-[beta]E)/2E (implying [[alpha].sub.0] > 0) and high enough values of
[bar.u] guaranteeing [w.sup.d] > 0.
(11.) We are grateful to an anonymous referee for illuminating this
identity.
(12.) Appendix A extends our model by introducing incentive pay as
an additional instrument available to the firm, thus enabling the latter
to use both training and the appropriate design of wage incentives to
elicit more productive effort from the worker.
(13.) We have experimented with employment weights and the results
remain robust.
(14.) The response rates of the employee questionnaire for 2004 and
2011 were 60% and 54% yielding 22,451 and 21,981 employees,
respectively.
(15.) There were 989 workplaces in WERS 2004 that also participated
in WERS 2011. The response rate of the panel questionnaire was 52%.
(16.) The experiment noted that firms which provide information to
workers about potential investments engage in a full category greater
training on average. We then restricted our delegation measure to apply
only to those firms that provide investment information (we assumed
there was no meaningful delegation without such information). The
estimations with this narrower measure remain very similar and are
available upon request.
(17.) We experimented with the mean and the median of this measure
and results remain robust. More fundamentally, we also imagined
retaining the mode but changing the cutoff so that either reports of
"A lot" or "Some" were identified as delegation. We
also imagined simply entering three dummies for whether the mode was
"A little," "Some," or "A lot." Neither of
these reasonable alternatives to structuring the critical independent
variable change the fundamental results we report and they are each
available upon request.
(18.) Appendix Table B1 reports the descriptive statistics of all
the variables used in the analysis.
(19.) While Bilanakos et al. (2017) present UK evidence that
dominant firms do more training, Meagher and Wait (2014) present
Australian evidence that delegation itself is associated with more
competitive product markets. Thus, while initially controlling for these
critical variables, we ultimately tackle the implied concern with the
endogeneity of delegation.
(20.) All fixed-effects specifications in our analysis include
industry controls. Even though reported industry switches from 2004 to
2011 are rare, they do happen, so the models are estimated with the
industry controls. If the industry controls are dropped from our
analysis, the results remain essentially unchanged.
(21.) We also searched for a proxy to the theoretical product
market uncertainty, E. The WERS question closest asks if the product
market is "growing," "mature,"
"declining," or "turbulent." If we identify
uncertain environments as "turbulent," then this variable and
its interaction with delegation take insignificant coefficients. If we
identify uncertain environments as both turbulent and declining markets,
then the interaction takes a negative and significant coefficient that
essentially eliminates the influence of delegation. These results are
available upon request. Thus, if one thought high E was associated with
turbulent or declining product markets, the suggestion that delegation
has no influence might be seen as broadly fitting with the theory. Yet,
we emphasize that the available question is not a particularly good
theoretical fit. Lo et al. (2016) proxy environmental uncertainty using
two measures, rapid technological change, and industry demand
uncertainty. They find that both measures of environmental uncertainty
have a negative but not always statistically significant effect on price
delegation.
(22.) These IV estimates on the cross section are available as
Table S2 in Appendix S1.
(23.) While not formal tests, two results support our strategy.
First, if delegation is predicted using training in a workplace
fixed-effect estimate, it is simply irrelevant hinting that reverse
causation may not be an issue. Second, when the assumed exogenous
industrial aggregation is added to a single stage workplace fixed-effect
estimate of training, it emerges with a very small and insignificant
coefficient, less than 0.01 (t-stat of 0.60), even as the coefficient on
delegation remains large and highly significant. This suggests we have
introduced meaningful exogenous variation.
(24.) Pooled ordered probit and OLS panel estimates without
workplace fixed effects that mimic columns 1 and 2 of Table 6 are
reported in Table S4 in Appendix S1.
(25.) The eigenvalue of 1.4 between discretion and involve ment
exceeds the rule of thumb of 1.0. Moreover, the first principal
component explains over 70% of the common variance of the two measures.
In addition to principal component analysis, we also created an
aggregate standardized measure of delegation by creating and adding
together the associated Z-scores. The results remain robust and are
available upon request.
(26.) Cross section and panel estimates of this exercise without
instrumenting are available in Table S5 in Appendix S1.
Caption: FIGURE 2 The Top Panel Shows Training Intensity, the
Middle Panel Shows Worker's Effort and the Bottom Panel Shows the
Wage Level with and without Delegation
TABLE 1
Distribution of Training
WERS 2004 WERS 2011
M SD Obs. M SD Obs.
None (0%) 0.237 0.426 131 0.190 0.393 112
Just a few 0.152 0.359 163 0.149 0.356 127
(1%-19%)
Some (20%-39%) 0.010 0.300 127 0.099 0.299 109
Around half 0.099 0.298 99 0.069 0.254 90
(40%-59%)
Most (60%-79%) 0.058 0.234 80 0.051 0.219 79
Almost all 0.063 0.243 118 0.093 0.290 130
(80%-99%)
All (100%) 0.291 0.454 276 0.349 0.477 365
Total 994 1,012
observations
Panel 2004-2011
M SD Obs.
None (0%) 0.174 0.380 53
Just a few 0.191 0.394 80
(1%-19%)
Some (20%-39%) 0.087 0.282 55
Around half 0.099 0.298 54
(40%-59%)
Most (60%-79%) 0.075 0.264 39
Almost all 0.081 0.274 50
(80%-99%)
All (100%) 0.293 0.455 143
Total 474
observations
Notes: The training question reads as follows: "What proportion of
experienced employees in the largest non-managerial occupational
group have been given time off from their normal daily work duties
to undertake training over the past 12 months?" The two
cross-section samples consist of private trading sector workplaces
and exclude workplaces where the largest occupational group is
managerial/senior official staff as the training question does not
apply to this group. For the panel dataset, we apply the same
restriction as in the two cross sections and keep workplaces we
observe twice. Thus, the panel is balanced, and we observe 237
workplaces generating 474 observations. Means are weighted using
workplace weights and sum to 100%.
TABLE 2
Distribution of Delegation
WERS 2004 WERS 2011
M SD Obs M SD Obs
None 0.114 0.318 114 0.051 0.221 54
A little 0.169 0.375 118 0.067 0.250 42
Some 0.417 0.493 461 0.396 0.489 417
A lot 0.300 0.459 301 0.485 0.500 499
Total observations 994 1,012
Panel 2004-2011
M SD Obs
None 0.054 0.226 41
A little 0.117 0.322 36
Some 0.437 0.496 229
A lot 0.392 0.488 168
Total observations 474
Notes: The delegation question is obtained from the employee
questionnaire and reads as follows: "In general, how much influence
do you have about the range of tasks you do in your job?" Responses
are recorded on a four-point scale: 1 "None," 2 "A little," 3
"Some," and 4 "A lot." We aggregate the worker responses to the
workplace level by taking the modal worker response, ala De Varo
and Kurtulus (2010). We code employee delegation to take the value
of 1 if the modal response is "A lot" and 0 if the modal response
is "Some," "A little," and "None." Means are weighted using
workplace weights and sum to 100%.
TABLE 3
Dependent Variable: Categorical Measure of Share Trained (WERS 2004)
(1) (2) (3)
Ordered Ordered Ordered
Probit Probit Probit
Delegation 0.184 ** 0.213 *** 0.242 ***
(0.073) (0.079) (0.081)
Log number of employees 0.079 *** 0.071 ** 0.098 ***
(0.021) (0.029) (0.030)
Workplace operates more than -0.119 -0.182 -0.188
5 years (0.130) (0.144) (0.147)
Part of a larger organization 0.328 * 0.311 0.221
(0.180) (0.190) (0.208)
Single independent workplace 0.041 0.148 0.100
(0.191) (0.202) (0.221)
UK owned -0.109 -0.180 * -0.232 **
(0.088) (0.103) (0.108)
% of employees using computers 0.389 *** 0.420 ***
(0.140) (0.141)
% of female employees 0.747 *** 0.824 ***
(0.206) (0.239)
% of part-time employees -0.754 *** -0.818 ***
(0.210) (0.223)
% union membership 0.432 *** 0.389 **
(0.158) (0.171)
% of employees with a 0.341 * 0.401 *
fixed-term contract (0.205) (0.221)
% of employees with a -0.155 -0.212
temporary contract (0.461) (0.491)
% of employees who quit -0.153 -0.207
last year (0.252) (0.255)
% of employees dismissed/ -0.700 -0.570
redundant last year (0.430) (0.457)
Payment by result
Merit pay
Profit-related pay
Employee share schemes
Few competitors
Many competitors
Market growing
Market mature
Market declining
Cutoff 1 -0.697 *** -0.065 -0.331
(0.228) (0.371) (0.432)
Cutoff 2 -0.101 0.596 0.346
(0.228) (0.372) (0.432)
Cutoff 3 0.255 0 993 *** 0.752 *
(0.228) (0.373) (0.431)
Cutoff 4 0.526 ** 1.298 *** 1.064 **
(0.229) (0.374) (0.433)
Cutoff 5 0 723 *** 1.518 *** 1.287 ***
(0.229) (0.374) (0.432)
Cutoff 6 1.044 *** 1.867 *** 1.642 ***
(0.230) (0.375) (0.434)
Constant
Observations 994 994 994
[R.sup.2]
Educational composition No Yes Yes
Occupational composition No Yes Yes
Largest occupational groups No Yes Yes
Industry dummies No No Yes
Region dummies No No Yes
Missing dummies Yes Yes Yes
(4)
Ordered (5)
Probit OLS
Delegation 0.236 *** 0.422 ***
(0.082) (0.142)
Log number of employees 0.083 *** 0 153 ***
(0.031) (0.052)
Workplace operates more than -0.160 -0.208
5 years (0.150) (0.259)
Part of a larger organization 0.192 0.398
(0.219) (0.393)
Single independent workplace 0.069 0.252
(0.233) (0.410)
UK owned -0.236 ** -0.433 **
(0.108) (0.192)
% of employees using computers 0.360 ** 0.651 ***
(0.146) (0.242)
% of female employees 0.808 *** 1.439 ***
(0.243) (0.406)
% of part-time employees -0.847 *** -1.281 ***
(0.230) (0.368)
% union membership 0.444 ** 0.729 **
(0.179) (0.306)
% of employees with a 0.379 * 0.673 *
fixed-term contract (0.225) (0.395)
% of employees with a -0.342 -0.469
temporary contract (0.503) (0.792)
% of employees who quit -0.264 -0.345
last year (0.262) (0.438)
% of employees dismissed/ -0.818 * -1.792 **
redundant last year (0.458) (0.806)
Payment by result 0.012 0.081
(0.095) (0.166)
Merit pay 0.010 0.065
(0.113) (0.205)
Profit-related pay 0.218 ** 0.400 ***
(0.088) (0.153)
Employee share schemes 0.119 0.201
(0.106) (0.188)
Few competitors -0.285 * -0.292 *
(0.147) (0.165)
Many competitors -0.342 ** -0.357 **
(0.144) (0.163)
Market growing 0.175 * 0.339 *
(0.106) (0.185)
Market mature 0.031 0.058
(0.118) (0.208)
Market declining -0.238 -0.350
(0.160) (0.274)
Cutoff 1 -0.287
(0.488)
Cutoff 2 0.399
(0.488)
Cutoff 3 0.812 *
(0.488)
Cutoff 4 1.128 **
(0.489)
Cutoff 5 1.354 ***
(0.488)
Cutoff 6 1.712 ***
(0.490)
Constant 1.079
(0.841)
Observations 994 994
[R.sup.2] 0.248
Educational composition Yes Yes
Occupational composition Yes Yes
Largest occupational groups Yes Yes
Industry dummies Yes Yes
Region dummies Yes Yes
Missing dummies Yes Yes
Notes: Bootstrap standard errors using 1,000 replications with
replacement are clustered at workplace cells and reported in
parentheses. Estimates use workplace weights.
*** p < .01, ** p <.05, p <. 1.
TABLE 4
Dependent Variable: Categorical Measure of Share Trained (WERS 2011)
(1) (2) (3)
Ordered Ordered Ordered
Probit Probit Probit
Delegation 0.236 *** 0.229 *** 0.245 ***
(0.070) (0.073) (0.075)
Log number of employees 0.070 *** 0.063 ** 0.076 **
(0.021) (0.032) (0.034)
Workplace operates more than -0.014 -0.045 -0.045
5 years (0.151) (0.157) (0.163)
Part of a larger organization 0.258 0.205 0.221
(0.204) (0.204) (0.212)
Single independent workplace -0.085 -0.077 -0.063
(0.217) (0.220) (0.228)
UK owned -0.097 -0.189 * -0.195 *
(0.0%) (0.112) (0.114)
% of employees using computers 0.445 *** 0.464 ***
(0.145) (0.149)
% of female employees 0.568 *** 0.443 **
(0.201) (0.220)
% of part-time employees -0.432 ** -0.524 ***
(0.182) (0.194)
% union membership 0.783 *** 0.848 ***
(0.183) (0.206)
% of employees with a 0.741 ***
fixed-term contract (0.208) (0.211)
% of employees with a 0.227 0.161
temporary contract (0.413) (0.452)
% of employees who quit -0.632 * -0.463
last year (0.357) (0.369)
% of employees dismissed/ -0.526 -0.520
redundant last year (0.340) (0.374)
Payment by result
Merit pay
Profit-related pay
Employee share schemes
Few competitors
Many competitors
Market growing
Market mature
Market declining
Cutoff 1 -0.894 *** 0.223 0.450
(0.264) (0.474) (0.532)
Cutoff 2 -0.351 0.825 * 1.063 **
(0.265) (0.475) (0.531)
Cutoff 3 0.043 1.274 *** 1.518 ***
(0.264) (0.475) (0.530)
Cutoff 4 0.269 1.529 *** 1.778 ***
(0.263) (0.476) (0.530)
Cutoff 5 0.482 * 1.763 *** 2.017 ***
(0.262) (0.475) (0.530)
Cutoff 6 0.832 *** 2.142 *** 2.402 ***
(0.261) (0.476) (0.531)
Constant
Observations 1,012 1,012 1.012
[R.sup.2]
Educational characteristics Yes Yes Yes
Occupational composition No Yes Yes
Largest occupational group No Yes Yes
dummies
Industry dummies No No Yes
Region dummies No No Yes
Missing dummies Yes Yes Yes
(4) (5)
Ordered OLS
Probit
Delegation 0.245 *** 0.488 ***
(0.077) (0.135)
Log number of employees 0.069 ** 0.141 **
(0.035) (0.059)
Workplace operates more than -0.045 -0.179
5 years (0.166) (0.278)
Part of a larger organization 0.207 0.471
(0.214) (0.395)
Single independent workplace -0.056 0.104
(0.232) (0.425)
UK owned -0.210 * -0.460 **
(0.117) (0.209)
% of employees using computers 0.430 *** 0.857 ***
(0.152) (0.255)
% of female employees 0.453 ** 0.824 **
(0.224) (0.398)
% of part-time employees -0.532 *** -1.206 ***
(0.194) (0.329)
% union membership 0.860 *** 1.590 ***
(0.214) (0.359)
% of employees with a 0.747 *** 1.226 ***
fixed-term contract (0.212) (0.305)
% of employees with a 0.122 0.032
temporary contract (0.466) (0.773)
% of employees who quit -0.473 -0.941
last year (0.375) (0.617)
% of employees dismissed/ -0.680 * -0.695 *
redundant last year (0.390) (0.412)
Payment by result 0.021 0.124
(0.101) (0.181)
Merit pay 0.099 0.169
(0.106) (0.184)
Profit-related pay 0.175 ** 0.336 **
(0.087) (0.151)
Employee share schemes 0.072 0.095
(0.129) (0.233)
Few competitors -0.428 ** -0.498 **
(0.213) (0.250)
Many competitors -0.439 ** -0.502 **
(0.214) (0.247)
Market growing 0.195 ** 0.338 **
(0.098) (0.166)
Market mature 0.196 * 0.337 *
(0.113) (0.195)
Market declining -0.148 -0.281
(0.129) (0.226)
Cutoff 1 0.692
(0.567)
Cutoff 2 1.312 **
(0.566)
Cutoff 3 1.772 ***
(0.566)
Cutoff 4 2.032 ***
(0.565)
Cutoff 5 2.271 ***
(0.565)
Cutoff 6 2.657 ***
(0.567)
Constant -0.195
(0.994)
Observations 1,012 1,012
[R.sup.2] 0.240
Educational characteristics Yes Yes
Occupational composition Yes Yes
Largest occupational group Yes Yes
dummies
Industry dummies Yes Yes
Region dummies Yes Yes
Missing dummies Yes Yes
Notes: Bootstrap standard errors using 1,000 replications with
replacement are clustered at workplace cells and reported in
parentheses. Estimates use workplace weights.
*** p<.01, ** p<.05, * p<.1.
TABLE 5
Marginal Effects
WERS 2004 WERS 2011
(1) (2)
Delegation Delegation
Training ME SE ME SE
Cutoff 1: none -0.039 *** 0.012 -0.030 *** 0.010
Cutoff 2: just a few -0.037 *** 0.013 -0.034 *** 0.011
Cutoff 3: some -0.016 *** 0.006 -0.024 *** 0.008
Cutoff 4: around half -0.003 0.002 -0.008 *** 0.003
Cutoff 5: most 0.004 ** 0.002 0.002 * 0.001
Cutoff 6: almost all 0.014 *** 0.005 0.008 *** 0.002
Cutoff 7: all 0.077 *** 0.027 0.090 *** 0.028
Notes: Entries are marginal effects obtained from a weighted
ordered probit model based on the estimates reported in column 4
of Table 3 (WERS 2004) and in column 4 of Table 4 (WERS 2011),
respectively. We only report the marginal effects of the variable of
interest. Marginal effects for all the other covariates are available
upon request. Robust standard errors are obtained using a bootstrap
exercise with 1,000 replications with replacement and are clustered at
workplace cells.
*** p < .01, ** p < .05, * p < 1.
TABLE 6
Panel Data 2004-2011
(1) (2) (3) (4)
Ordered Probit OLS OLS Poisson
without FE without FE with FE with FE
Delegation 0.214 ** 0.379 ** 0.389 ** 0.246 **
(0.092) (0.164) (0.192) (0.121)
Log-likelihood -658.424 -170.542
[R.sup.2] 0.341 0.895
Observations 474 474 474 474
Industry dummies Yes Yes Yes Yes
Region dummies Yes Yes No No
Workplace Yes Yes Yes Yes
characteristics
Workforce Yes Yes Yes Yes
characteristics
Occupational Yes Yes Yes Yes
composition
Largest Yes Yes Yes Yes
occupational
group dummies
PRP dummies Yes Yes Yes Yes
Competition Yes Yes Yes Yes
dummies
Market state Yes Yes Yes Yes
dummies
Missing dummies Yes Yes Yes Yes
Notes: For information on the sample and on the main variables of
interest, see Notes in Tables 1 and 2. The dependent variable is
the proportion of experienced employees in the largest occupational
group who have been given time off from their normal daily work to
undertake training over the last 12 months. For reasons of brevity,
we only present estimates of the main independent variable of
interest. Other controls are those shown in column 4 of Table 3, as
well as a year dummy. The estimates for the rest of the covariates
are available upon request. Bootstrap standard errors using 1,000
replications with replacement and clustered at workplace cells are
reported in parentheses. Estimates are weighted using workplace
weights. PRP, performance-related pay.
** p < .0.5.
TABLE 7
Instrumental Variable (IV) Results, Panel 2004-2011
Without Workplace Fixed Effects
First Stage Second Stage
(1) (2)
Delegation Training
Delegation 0.504 ** (0.243)
Instrument for delegation 0.068 *** (0.004)
Log number of employees 0.014 (0.015) 0.105 * (0.056)
Workplace operates more 0.067 (0.100) 0.009 (0.198)
than 5 years
Part of a larger 0.095 ** (0.045) 0.122 (0.286)
organization
Single independent 0.082 * (0.047) 0.078 (0.256)
establishment
UK owned 0.037 (0.043) -0.099 (0.114)
% of employees using 0.057 ** (0.035) 0.223 * (0.125)
computers
% of female employees -0.027 (0.062) 0.380 * (0.214)
% of part-time employees -0.067 (0.092) -0.396 ** (0.172)
% union membership 0.040 (0.101) 0.142 * (0.078)
% of employees with a 0.158 (0.105) 0.028 (0.159)
fixed-term contract
% of employees with a -0.143 (0.270) -0.659 (0.473)
temporary contract
% of employees who quit -0.110 (0.123) -0.084 (0.241)
last year
% of employees dismissed/ -0.205 (0.115) -0.464 ** (0.231)
redundant last year
Payment by result 0.042 * (0.024) 0.015 (0.095)
Merit pay 0.041 (0.052) 0.196 ** (0.100)
Profit-related pay 0.027 (0.040) 0.172 ** (0.080)
Employee share schemes 0.037 (0.058) -0.075 (0.117)
Few competitors 0.110 (0.088) -0.129 ** (0.065)
Many competitors 0.167 * (0.095) -0.187 ***
(0.071)
Market growing 0.045 ** (0.022) 0.207 ** (0.096)
Market mature 0.012 (0.055) 0.050 (0.094)
Market declining -0.080 (0.056) -0.197 (0.214)
F-statistic ([H.sub.0]: 250.90 p val. = .000
instrument is weak)
Observations 474
Educational composition Yes Yes
Occupational composition Yes Yes
Largest group occupational Yes Yes
dummies
Industry dummies Yes Yes
Region dummies Yes Yes
Missing dummies Yes Yes
With Workplace Fixed Effects
First Stage Second Stage
(3) (4)
Delegation Training
Delegation 0.538 *** (0.169)
Instrument for delegation 0.092 *** (0.006)
Log number of employees 0.052 (0.092) 0.032 (0.095)
Workplace operates more 0.028 (0.198) -0.102 (0.270)
than 5 years
Part of a larger 0.170 * (0.095) 0.319 (0.378)
organization
Single independent 0.078 (0.062) 0.454 (0.372)
establishment
UK owned 0.092 (0.158) -0.458 ** (0.237)
% of employees using 0.020 (0.012) 0.152 (0.186)
computers
% of female employees -0.011 (0.067) 0.855 ** (0.401)
% of part-time employees -0.156 (0.225) -0.775 * (0.435)
% union membership 0.160 (0.324) 0.542 * (0.314)
% of employees with a 0.224 (0.229) 0.168 (0.294)
fixed-term contract
% of employees with a -0.036 (0.572) -0.865 * (0.479)
temporary contract
% of employees who quit -0.270 (0.320) -0.095 (0.370)
last year
% of employees dismissed/ -0.038 (0.321) -0.262 * (0.142)
redundant last year
Payment by result 0.132 * (0.069) 0.020 (0.110)
Merit pay 0.020 (0.128) 0.275 ** (0.137)
Profit-related pay 0.085 (0.089) 0.070 (0.094)
Employee share schemes 0.019 (0.085) 0.205 (0.179)
Few competitors 0.030 (0.110) -0.330 * (0.172)
Many competitors 0.140 (0.189) -0.430 * (0.240)
Market growing 0.075 * (0.042) 0.072 * (0.037)
Market mature 0.052 (0.126) 0.026 (0.130)
Market declining -0.080 (0.149) -0.152 (0.189)
F-statistic ([H.sub.0]: 175.69 p val. = .000
instrument is weak)
Observations 474
Educational composition Yes Yes
Occupational composition Yes Yes
Largest group occupational Yes Yes
dummies
Industry dummies Yes Yes
Region dummies No No
Missing dummies Yes Yes
Notes: The instrument for delegation is the proportion of
workplaces in industry cells reporting a "lot of' delegation after
removing the given workplace from the industry cell. The estimation
method is a 2SLS. Bootstrap standard errors using 1,000
replications with replacement and clustered at workplace cells
(columns 1 and 2) and at industry cells (columns 3 and 4) are
reported in parentheses. Estimates are weighted using workplace
weights. Estimates for the other control variables are available
upon request.
*** p < .01, ** p <. 05, * p < 1.
TABLE 8
Robustness Tests Using Lagged Instrumental Variables
WERS--Panel 2011
First Stage Second Stage
(1) (2)
Delegation Training
Delegation 0.728 **
(0.344)
Instrument for delegation 1 029 ***
(0.046)
Workplace fixed effects No No
F-statistic ([H.sub.0]: 501.09
instrument is weak) p val. = .000
Observations 202
Firm characteristics Yes Yes
Workforce characteristics Yes Yes
Occupational composition Yes Yes
Largest occupational Yes Yes
group dummies
Incentive pay controls Yes Yes
Competition controls Yes Yes
Current market state Yes Yes
controls
Missing dummies Yes Yes
Region dummies Yes Yes
Industry dummies Yes Yes
WERS--Cross Section 2011
First Stage Second Stage
(3) (4)
Delegation Training
Delegation 0.452 *
(0.251)
Instrument for delegation 0.010 **
(0.004)
Workplace fixed effects No No
F-statistic ([H.sub.0]: 12.27
instrument is weak) p val. = .012
Observations 853
Firm characteristics Yes Yes
Workforce characteristics Yes Yes
Occupational composition Yes Yes
Largest occupational Yes Yes
group dummies
Incentive pay controls Yes Yes
Competition controls Yes Yes
Current market state Yes Yes
controls
Missing dummies Yes Yes
Region dummies Yes Yes
Industry dummies No No
WERS--Panel 2004-2011
First Stage Second Stage
(5) (6)
Delegation Training
Delegation 0.559 ***
(0.178)
Instrument for delegation 0.036 ***
(0.009)
Workplace fixed effects Yes Yes
F-statistic ([H.sub.0]: 14.29
instrument is weak) p val. = .000
Observations 462
Firm characteristics Yes Yes
Workforce characteristics Yes Yes
Occupational composition Yes Yes
Largest occupational Yes Yes
group dummies
Incentive pay controls Yes Yes
Competition controls Yes Yes
Current market state Yes Yes
controls
Missing dummies Yes Yes
Region dummies No No
Industry dummies Yes Yes
Notes: The instrument in column 1 is lagged industry-level
delegation constructed using 2004 observations only from the panel
sample. We estimate this specification only for the 2011 panel
data. In column 3, the instrument is lagged industry-level
delegation constructed using the 2004 cross section for those firms
that have been in operation for more than 7 years. In column 5, the
instrument is composed of lagged 1998 industry-level delegation for
2004 observations, and lagged 2004 industry-level delegation for
2011 observations. The estimation method is a 2SLS. Bootstrapped
standard errors using 1,000 replications with replacement are
clustered at the workplace cells (columns 1-4) and at industry
cells (columns 5 and 6) and are reported in parentheses. Estimates
are weighted using workplace weights. Estimates for the other
control variables are available upon request. In columns 5 and 6,
we also add a year dummy. The number of observations for all three
IV robustness checks is smaller compared to those where the
instrument is contemporaneous as we lose observations when merging
in lagged industry-level delegation.
*** p < .01, ** p <. 05, * p <.1.
TABLE 9
Alternative Dependent Variable (Days of Training),
Instrumental Variable Results, Panel 2004-2011
Without Workplace Fixed Effects
First Stage Second Stage
(1) (2)
Delegation Training
Delegation 0.520 ** (0.261)
Instrument for delegation 0.030 *** (0.003)
Log number of employees 0.025 (0.021) 0.281 *** (0.062)
Workplace operates more 0.247 ** (0.120) 0.122 (0.360)
than 5 years
Part of a larger 0.203 (0.240) 0.385 (0.368)
organization
Single independent 0.113 (0.242) 0.566 (0.369)
establishment
UK owned 0.008 (0.080) 0.099 (0.210)
% of employees using 0.045 (0.100) 0.459 * (0.268)
computers
% of female employees -0.369 ** (0.167) 0.521 * (0.292)
% of part-time employees -0.085 (0.144) -0.890 ** (0.359)
% union membership 0.057 (0.139) 0.253 (0.362)
% of employees with a 0.202 (0.161) 0.403 (0.499)
fixed-term contract
% of employees with a -0.196 (0.451) -0.041 (0.341)
temporary contract
% of employee; who quit -0.030 (0.186) -0.403 (0.499)
last year
% of employes? dismissed/ -0.306 (0.247) -0.239 (0.749)
redundant last year
Payment by result 0.044 * (0.026) 0.087 (0.187)
Merit pay 0.011 (0.067) 0.351 (0.182)
Profit-related pay 0.009 (0.062) 0.089 (0.178)
Employee share schemes 0.158 (0.189) 0.012 (0.243)
Few competitors 0.105 (0.134) -0.105 ** (0.045)
Many competitors 0.108 (0.135) -0.175 ** (0.078)
Market growing 0.011 (0.067) 0.244 * (0.142)
Market mature 0.027 (0.074) 0.002 (0.220)
Market declining 0.084 (0.082) 0.028 (0.238)
F-statistic ([H.sub.0]: 145.46p val. = .000
instrument is weak)
Observations 474
Educational composition Yes Yes
Occupational composition Yes Yes
Largest group occupational Yes Yes
dummies
Industry dummies Yes Yes
Region dummies Yes Yes
Missing dummies Yes Yes
With Workplace Fixed Effects
First Stage Second Stage
(3) (4)
Delegation Training
Delegation 0.529 ** (0.267)
Instrument for delegation 0.045 *** (0.004)
Log number of employees 0.012 (0.080) 0.568 * (0.337)
Workplace operates more 0.090 (0.270) -0.795 (0.600)
than 5 years
Part of a larger 0.071 (0.190) 0.555 (0.564)
organization
Single independent 0.036 (0.172) 0.589 (0.670)
establishment
UK owned 0.360 (0.320) -0.346 (0.305)
% of employees using 0.062 (0.110) 0.450 (0.340)
computers
% of female employees -0.450 * (0.249) 0.859 * (0.508)
% of part-time employees -0.170 (0.192) -0.750 (0.439)
% union membership -0.380 (0.340) 0.690 * (0.350)
% of employees with a 0.120 (0.168) 0.070 (0.629)
fixed-term contract
% of employees with a -0.430 (0.670) -0.936 (0.780)
temporary contract
% of employee; who quit -0.055 (0.279) -0.010 (0.205)
last year
% of employes? dismissed/ -0.304 (0.545) -0.522 ** (0.258)
redundant last year
Payment by result 0.080 (0.109) 0.150 (0.149)
Merit pay 0.095 (0.089) 0.210 (0.260)
Profit-related pay 0.039 (0.072) 0.470 (0.160)
Employee share schemes 0.101 (0.134) 0.190 (0.135)
Few competitors 0.240 (0.230) -0.420 * (0.248)
Many competitors 0.254 (0.232) -0.460 * (0.259)
Market growing 0.049 (0.201) 0.310 (0.410)
Market mature 0.025 (0.242) 0.264 (0.289)
Market declining 0.015 (0.130) -0.039 (0.602)
F-statistic ([H.sub.0]: 127.42 p val. = .000
instrument is weak)
Observations 474
Educational composition Yes Yes
Occupational composition Yes Yes
Largest group occupational Yes Yes
dummies
Industry dummies Yes Yes
Region dummies No No
Missing dummies Yes Yes
Notes: The instrument for delegation is the proportion of
workplaces in industry cells reporting a "lot of' delegation after
removing the given workplace from the industry cell. The estimation
method is a 2SLS. Bootstrap standard errors using 1,000
replications with replacement and clustered at workplace cells
(columns 1 and 2) and at industry cells (columns 3 and 4) are
reported in parentheses. Estimates for the other control variables
are available upon request.
*** p <.0l, ** p <.05, * p <.1.
TABLE 10
High Incentives versus Low Incentives and Owner-Managers versus
Hired-Managers, WERS Panel 2004-2011
High Incentives
First Stage Second Stage
(1) (2)
Delegation Training
Delegation 0.524 **
(0.235)
Instrument for delegation 13.149 ***
(0.716)
Workplace fixed effects Yes Yes
F-statistic ([H.sub.0]: 237.03 p
val. = .000
instrument is weak)
Observations 206
Firm characteristics Yes Yes
Workforce characteristics Yes Yes
Occupational composition Yes Yes
Largest occupational group Yes Yes
dummies
Incentive pay controls No No
Competition controls Yes Yes
Current market state controls Yes Yes
Missing dummies Yes Yes
Region dummies No No
Industry dummies Yes Yes
Low Incentives
First Stage Second Stage
(3) (4)
Delegation Training
Delegation 0.421
(0.320)
Instrument for delegation 10.244 ***
(0.771)
Workplace fixed effects Yes Yes
F-statistic ([H.sub.0]: 176.46 p
val. = .000
instrument is weak)
Observations 268
Firm characteristics Yes Yes
Workforce characteristics Yes Yes
Occupational composition Yes Yes
Largest occupational group Yes Yes
dummies
Incentive pay controls No No
Competition controls Yes Yes
Current market state controls Yes Yes
Missing dummies Yes Yes
Region dummies No No
Industry dummies Yes Yes
Owner-Manager
First Stage Second Stage
(5) (6)
Delegation Training
Delegation 0.947**
(0.481)
Instrument for delegation 0.076 ***
(0.025)
Workplace fixed effects No No
F-statistic ([H.sub.0]: 20.80 p
val. = .000
instrument is weak)
Observations 100
Firm characteristics Yes Yes
Workforce characteristics Yes Yes
Occupational composition Yes Yes
Largest occupational group Yes Yes
dummies
Incentive pay controls Yes Yes
Competition controls Yes Yes
Current market state controls Yes Yes
Missing dummies Yes Yes
Region dummies Yes Yes
Industry dummies Yes Yes
Hired-Manager
First Stage Second Stage
(7) (8)
Delegation Training
Delegation 0.432 **
(0.219)
Instrument for delegation 0.069 ***
(0.007)
Workplace fixed effects No No
F-statistic ([H.sub.0]: 75.09 p
val. = .000
instrument is weak)
Observations 374
Firm characteristics Yes Yes
Workforce characteristics Yes Yes
Occupational composition Yes Yes
Largest occupational group Yes Yes
dummies
Incentive pay controls Yes Yes
Competition controls Yes Yes
Current market state controls Yes Yes
Missing dummies Yes Yes
Region dummies Yes Yes
Industry dummies Yes Yes
Notes: The instrument for delegation is the proportion of
workplaces in industry cells reporting a "lot of' delegation after
removing the given workplace from the industry cell. The estimation
method is a 2SLS. Bootstrapped standard errors using 1,000
replications with replacement are clustered at industry cells in
columns 1-4 and at workplace cells in columns 5-8 and are reported
in parentheses. Estimates are weighted using workplace weights.
Estimates for the other control variables as well as for a year
dummy are available upon request.
*** p < .01, ** p < .05.
FIGURE 1
Time Sequence of Actions
Stage 1 Stage 2 Stage 3
P offers a P chooses the A chooses the
contract wage, w, and level of effort, e,
allocating the level of devoted to
formal training, 1, learning
authority to provided to A. candidate
either herself projects' payoffs
(Integration) or and becomes
to A informed with
(Delegation) corresponding
over the future probability e,
choice of whereas P
projects. becomes
informed with
exogenous
probability E.
Stage 4 Stage 5
The party who The party who
does not have has formal
formal authority selects
authority a project (or
recommends a none) based on
project (or his or her own
none) on the information as
grounds of well as on the
information recommendation
acquired in the made by the
previous stage. other party.
COPYRIGHT 2018 Western Economic Association International
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2018 Gale, Cengage Learning. All rights reserved.
