Full text of DOI:10.1111/1467-8543.00199

British Journal of Industrial Relations
39:2 June 2001 0007±1080 pp. 257±284
A NewVoice or aWaste ofTime? Wage
Premiums from Using Computers for
Communication in the UK Workplace
Andrew K. G. Hildreth
Abstract
The paper uses the Workplace Employee Relations Survey data on workers to
investigate the wage premium from using e-mail in the workplace against other
more traditional forms of communication. I try to discern whether the exist-
ence of a premium from using e-mail is the product of the workplace encour-
aging worker productivity through voice, or represents unobserved worker
skills in using new technology. Results indicate that any observed premium
from using e-mail is likely to arise from unobserved worker skills, and that any
premium associated with voice in the workplace is likely to result from
management choosing to reward such worker involvement.
1. Introduction: computers and voice within the workplace
A great deal has been made in recent years of the skill-biased technological
change argument for explaining the divergence in wages across individuals.
The basic tenants of the argument are that the increasing demand for
technologically-skilled labour has lead to a wage premium for workers able
to use the new technology; that the new technology makes a worker more
productive than one not using the new technology; and that as a result the
earnings such workers receive are higher. Most of the research in this area
has used data at the individual worker level. This paper proposes to examine
the role of the workplace in providing e-mail, and the impact this has on
certain sections of its work-force.
Estimating the role of the computer in its implications on pay started with
Krueger (1993). Krueger observed that, controlling for the usual range of
demographics and workplace characteristics, workers using a computer at
work earned, on average, 15±20 per cent more than those who did not. The
Andrew K. G. Hildreth is at the CCRDC and the Department of Economics, University of
California-Berkeley. E.mail:hildreth@econ.Berkeley.edu
# Blackwell Publishers Ltd/London School of Economics 2001. Published by Blackwell Publishers Ltd,
108 Cowley Road, Oxford OX4 1JF, UK and 350 Main Street, Malden, MA 02148, USA.

258 British Journal of Industrial Relations
same wage premium existed if the main computer use was e-mail. This was
compared against using the computer to play games, which brought with it a
negative wage premium. The substantial wage premium to productive
computer use was simply the result of being presented with a computer to
use, rather than from any particular selection effect on the part of the
individual workers. Subsequent authors have found that there are returns
arising from increasing the technology available for the work-force. Autor
et al. (1998), Haskel (1999), Machin and Van Reenen (1998) and Van Reenen
(1996) have all found that technology helps increase the wage for workers,
when compared with workers who do not have access to the productivity-
improving machines. Dissenting voices have been given by DiNardo and
Pischke (1997) and Entorf and Kramarz (1997).1
Arguments against the notion of there being a return from using com-
puters (or technology) at the individual worker level have centred around
the selection of individuals and employers to use the new technology. One
way to examine this is to compare the returns to individuals from using
computers with returns from using other workplace tools. DiNardo and
Pischke (1997) use German data on workplace tools, and examine the re-
turns from using computers against those arising from the use of telephones,
pens or pencils. They find that there are similiar returns from using com-
puters as from using pens and pencils. Given the level of literacy in Germany,
it is difficult to believe that the skill of using a pencil is in short supply, and
therefore commands a premium. What is really needed to examine the
problem of a return from technology implementation in the workplace is
data on both workers and employers. Entorf and Kramarz (1997) use such
data, for a sample of French workers and employers, over time. They were
able to examine when an employer adopts a new technology, and which
worker then works with it. They found that it is the more able workers who
work with the new technology.2 These workers earn a premium because they
have higher unobserved ability that is correlated with computer use, and this
correlation gives the observed premium on working with new technology.
Even in the event of no new technology being introduced, workers with a
high unobserved ability would have received a premium.
It is not immediately obvious why the use of e-mail, or any means of
communication for `voice' at the workplace, should provide a wage
premium. `Voice' was described by Hirschman (1970) as `information-rich',
where instructions are able to be given from one side to the other. Where
`exit' has just one outcome Ð the individual leaves the firm Ð `voice', in
terms of a return for the individual, may have two outcomes. The first is that
those who raise their voice may be penalized within the organization if voice
is seen as criticism; in this instance workers may earn less if they use e-mail.
The second is that voice has beneficial effects within the workplace, in that
being able to add their voice to the organization helps to maintain workers'
loyalty and encourages workers to be productive within the firm. Accord-
ingly, the workers receive the benefits of such an exchange in terms of higher
pay.
# Blackwell Publishers Ltd/London School of Economics 2001.

A New Voice or a Waste of Time? 259
More recently, Bresnahan (1999) has called into question the approach of
simply examining who gets a computer and who does not in seeking to
determine if there is a premium to computer use. If computers are used as a
new voice in the workplace (e-mail), then there must be a reason why it is
more efficient to use them. Employers don't simply buy a computer and
place it on someone's desk: the computer, especially if it is to be used for
communication purposes, has to be integrated into the workplace. This
integration will be reflected in workplace performance, as well as the type of
workers the employer now requires and the remuneration for workers in
particular types of job.
Specifically, Bresnahan (1999) suggests that the matter of importance is
not which individual is given a computer for use, but rather, how computers
are used in an organizational context. The organizational computing theory
predicts events at the workplace level. The extent to which we may observe
a return to e-mail across individuals, and across workplaces, depends on
the type of industry, the need for communication within the firm, the type
of employees the company has and the type of employees the company
requires. A high use of e-mail within a workplace would occur only in
two instances. The first is a substitution argument. If the firm is a high-
communication workplace, then the communication needs of the firm have
begun to be met by the introduction of e-mail for some lower-order
cognitive tasks formerly performed by employees. In time, if low-cognitive-
ability type workers are not available, more and more of the tasks that such
workers have performed in the past may be carried out by higher-cognitive-
ability workers using e-mail. For example, managers may be able to co-
ordinate via e-mail rather than using an assistant. Secretarial or clerical
workers would be observed to receive no wage premium from e-mail use.
The second argument is one of complementarity. Once again, for high-
communication workplaces, an observed wage premium would be observed
for workers with people skills using e-mail. For example, managers able to
co-ordinate effectively using e-mail would be observed receiving a wage
premium. In this instance, e-mail cannot replace workers, as the wage
premium indicates a different sort of return from using the computer for
communication. More workers with people skills would have to be hired to
reduce the wage premium to these workers; it is not possible to replace some
of their tasks with the technology.
This paper presents an examination of the returns from using e-mail as a
voice mechanism, and the implications for the workplace, for Great Britain.
Using the WERS (Workplace Employee Relations Survey) for 1998, ques-
tions were presented to a random draw of employees in each workplace on
(among other things) pay, demographics and how information was com-
municated to the work-force. This last subject details e-mail use against
other forms of communication (meetings, noticeboards, newsletters).
Evidence on the information flow through the workplace is examined to
investigate whether information is given by management and used as a
monitoring device, or is shared throughout the workplace. The presence of
# Blackwell Publishers Ltd/London School of Economics 2001.

260 British Journal of Industrial Relations
e-mail, and the potential returns from its use, are examined at the workplace
level. Further analysis is also carried out on the implications for the type of
worker that may receive the wage premium and the implications for the
workplace in which type of worker is being hired. Finally, using other
workplace variables on technology and consultation with the work-force, it
is possible to try and separate out the `voice' mechanism part of computers
from their other uses within the workplace.
The paper is organized in the following manner. Section 2 describes the
data and the model. Section 3 presents the results on the returns from
different methods of communication within the workplace. The results show
that any returns from using e-mail are the product of workers with higher
unobserved skills using computers. These unobserved skills are associated
with different types of job. For example, managers who use e-mail also need
people skills,3 managers earn an observed premium from using e-mail
because it is associated with people skills. Likewise, establishments that have
a high e-mail use also have trouble recruiting managers. From using other
variables on new technology introduction and communication between
workers and management, it appears that e-mail use and new technology
introduction are separable in their effects on pay. What matters in using
e-mail as a voice, if there is any effect at all, is the incentive structure for
supplying information within the workplace.
2. The Workplace Employee Relations Survey (WERS): data and model
Data Description
WERS provides the ideal data set from which to examine the issues
presented in the Introduction because of the relevance of the survey to
establishments, and because of the number of workers surveyed per estab-
lishment.4 The aim of the survey is to provide detailed information on the
organization of management±employee relations within a random sample of
establishments, and on the impact of this on employer performance (see
Cully et al. 1999). Upon completion of the management interview, permission
was sought to survey up to 25 employees per workplace. The information
required from the workers concerned demographic characteristics, educa-
tion, tenure at the workplace, training, occupation, hours, union association
and pay.
The sample used for this paper was restricted to private-sector workplaces.
To test for a wage premium is to suggest that underlying the premium is the
ability to increase worker productivity and to earn rents from doing so. The
proposition that the employer would earn rents is tested in a later section.
The number of workers in private-sector establishments, as shown in Table 1,
is approximately 14,000 once allowance has been made for complete re-
sponses.5 The number of private-sector establishments giving complete
replies was 1509. This allows a large number of workers per establishment.
On average, there were 18 workers per establishment. The identification of
# Blackwell Publishers Ltd/London School of Economics 2001.

A New Voice or a Waste of Time? 261
unobserved establishment effects is possible only if there are two or more
workers observed at each establishment. Appendix A contains further details.
Most of the sample of workers (about 97 per cent) are of two or more
workers observed at each establishment.
Workplace variables, from the Management Questionnaire, concerned
the ownership and control of the workplace (Section A), the consultation
and communication within the workplace (Section D), performance at the
workplace (Section K) and the degree to which it had implemented any new
technology or management practices in the past five years (Section L). Most
of the Employee Questionnaire was used, especially the question concerning
how a worker keeps up-to-date about the workplace. The question B6 on
the Employee Questionnaire asked: `How helpful do you find the following
in keeping up-to-date about this workplace?' Four communication methods
were listed: noticeboards, e-mail, the workplace newsletter or magazine, and
meetings of managers and employees. To each of these four methods there
were five responses: `very helpful', `helpful', `not very helpful', `not at all
helpful', and `not used here'. The responses were re-coded into a binary
response of `used here' and `not used here'. The degree to which a particular
communication device was useful was not considered.
As the response to the question regarding information acquisition was
made by individuals, the response of `not used here' was taken as `not used
by me'. The results in this paper refer to individual worker wage equations.
Inferring that `not used here' is the same as `not used by me' seems
reasonable in that there is little reason to expect all workers to use the same
communication method. Workers on the shop-floor, or those who are part
of a sales department, might use a different communication method from
administrative or managerial staff.
Table 1 provides a basic set of means for the sample of matched worker±
employer observations that were used. The table illustrates how workers
using e-mail, against other forms of communication, differ in their observed
attributes and the workplaces in which they work. In terms of unconditional
means, workers using e-mail are younger, have more employer training, are
better educated, are more likely to work in white-collar jobs (management,
professionals, associate professionals and technical, and clerical and secre-
tarial jobs), are to be found in certain industries (energy, finance, business/
computing), in larger workplaces and for companies that are financially
more successful.
As the question on the type of communication method used to keep
up-to-date about the workplace referred only to a `top-down' information
flow, further checks were carried out to determine whether the e-mail used
by workers was part of an information flow throughout the workplace.
In particular, three questions at various points of the Management
Questionnaire were combined to form a single index. (1) From Section A
(Background) of the questionnaire, managers were asked about their views
concerning employee relations issues; question 10 concerned decisions
made by management without consulting employees. (2) From Section D
# Blackwell Publishers Ltd/London School of Economics 2001.

262 British Journal of Industrial Relations
TABLE 1
Mean Characteristics of Private-Sector Workers and Employers, by Method of
Communication
Variable
Total e-mail Meetings Noticeboard Newsletter
Predicted weekly paya
Workers less than age 40
Workers with less than 5 years' tenure 0.548
Workers with less than 5 days' training 0.413
Union recognized
Union members
Female
O levels
Degree
271.73 280.34
0.557
234.75
0.567
0.542
0.442
0.455
0.300
0.455
0.270
0.175
0.149
0.109
0.081
0.228
0.089
0.045
0.118
0.118
0.061
0.217
0.069
0.044
0.181
0.058
0.069
0.098
0.121
0.001
0.033
0.075
0.036
0.598
2.27
224.08
0.563
0.539
0.426
0.457
0.301
0.459
0.273
0.160
0.130
0.099
0.075
0.227
0.095
0.045
0.121
0.136
0.071
0.227
0.063
0.043
0.192
0.060
0.074
0.090
0.111
0.001
0.032
0.073
0.035
0.595
2.27
234.40
0.567
0.530
0.454
0.517
0.338
0.448
0.274
0.167
0.138
0.103
0.078
0.241
0.091
0.035
0.133
0.116
0.063
0.199
0.084
0.041
0.199
0.054
0.082
0.107
0.120
0.001
0.026
0.055
0.032
0.667
2.24
0.583
0.537
0.466
0.479
0.293
0.437
0.258
0.230
0.188
0.153
0.107
0.284
0.060
0.019
0.089
0.064
0.035
0.198
0.107
0.032
0.161
0.041
0.068
0.113
0.179
0.001
0.030
0.033
0.037
0.632
2.65
0.426
0.283
0.461
0.271
0.158
0.128
0.096
0.074
0.227
0.098
0.047
0.118
0.130
0.079
0.218
0.059
0.051
0.187
0.059
0.072
0.085
0.125
0.001
0.033
0.071
0.038
0.572
2.28
Managers/senior administrators
Professionals
Associate professional/technical staff
Clerical/secretarial
Craft/skilled services
Personal/protective services
Sales
Operative/assembly work
Other occupations
Manufacturing
Electric/gas/water
Construction
Wholesale retail trade
Hotels/restaurants
Transport/telecommunications
Finance
Business/computing
Public administration
Education
Health/social work
Other
PLC
Financial performance
Labour productivity
Size of workplace
No. of observations
2.42
115
2.42
148
2.41
122
11122
2.41
122
12400
2.38
145
9426
13780
6041
a The pay variable refers to predicted pay, assuming a normal distribution, and conditioned on
age, union association, tenure, training, education, hours/week worked, married, female, health,
race, occupation, industry, size of workplace, franchise, profit-related pay, assessment on indi-
vidual performance and whether a PLC or a single-establishment firm.
(Communication and Consultation) of the questionnaire, there are detailed
questions concerning how managers at a workplace communicate with
employees; the response to one multiple-choice question was that infor-
mation was supplied to employees via a systematic management chain (or
a cascading of information). (3) Finally, from Section K (Performance at
the Workplace), managers could respond to the question on monitoring the
quality of work at the workplace by replying that it was their responsibility.
These three measures were then added to provide an overall measure,
termed a `sinister index': whether managers monitored worker performance,
# Blackwell Publishers Ltd/London School of Economics 2001.

A New Voice or a Waste of Time? 263
consulted with workers, and/or believed that the reported information flow
for management was down to the workers.6
Table 2 provides the results for information flow and the method of
communication used by the workers. The table provides the correlation
coefficients for the method of communication and the information flow
measure. It also reports whether the correlation is significant. The
significant negative correlation in the results in Table 2 shows that e-mail
is used in workplaces where managers do consult with workers, and where
managers do not tell the workers what to do (the `sinister index'). This is
compared with other methods of communication, where there is more
reason to believe that managers do restrict the information flow from the
top down, although other measures of information flow are not significantly
correlated with the information flow measure. While the results are weak, if
anything they indicate, either together or individually, that e-mail (or any of
the other communication techniques) is concerned more with allowing feed-
back than with simply imparting information.
Identifying whether or not using e-mail, or attending meetings, makes a
difference to worker remuneration, both between and within establishments,
is possible, as workers matched to establishments did not give completely
consistent answers. Workers within a workplace did not all attend meetings,
or all use e-mail, look at noticeboards or read newsletters to keep up-to-date
about the workplace. The largest degree of within-establishment variation
was in using e-mail. In total, in 906 establishments, not all the work-force
used e-mail as a means of communication; in 122 workplaces all employees
did use e-mail. The corresponding figures for meetings were 873 and 300.
There was less variation for noticeboards and newsletters. The predicted
TABLE 2
Correlation of Information Flow within the Workplace and Method of Communication
Information flow measure
e-mail
Meetings Noticeboard Newsletter
No consultation with workers by
management
70.076
(0.000)
70.147
(0.000)
70.096
(0.001)
70.125
(0.000)
Information flow from management
down
0.013
(0.119)
0.237
(0.000)
0.223
(0.000)
0.272
(0.000)
Managers monitor worker performance
0.025
(0.377)
70.013
(0.664)
70.014
(0.622)
0.030
(0.307)
`Sinister Index' (additive index of above 70.035
70.029
(0.326)
0.040
(0.164)
0.003
(0.921)
criteria)
(0.000)
Notes
Figures in parentheses are significance values. A value less than 0.05 indicates that the cor-
relation rejects the null of zero correlation between the two measures at the 95% confidence
interval.
No. of observations = 1509.
The method of communication refers to the average number of workers at the workplace using
that method.
# Blackwell Publishers Ltd/London School of Economics 2001.

264 British Journal of Industrial Relations
conditional wage variable (see Hildreth 1999) shows that, for private-sector
establishments using e-mail and meetings, the greatest degree of wage vari-
ation comes from within the establishment rather than between them. Higher
wages are associated with communicating by e-mail within an establishment,
compared with workers within the same workplace who were not using the
e-mail system. For establishments using noticeboards and newsletters, there
was little difference in the wage variation within the workplace, compared
with between different places of work.
The Econometric Model
As WERS provides a great deal of detail on both the employer and the
worker, the basic model for the work presented in this paper is an extended
human capital wage equation:
Ln(wif) = Xif 'b+fcif+eif ,
(1)
where i = 1,...,N and denotes the number of workers per establishment
(1 ꢀ N ꢀ 25); f = 1,...,F, Ln(wif) is the log of wages for individual i in work-
place f, and Xif is a matrix of observable worker and employer character-
istics that are thought to affect a worker's wage. Worker characteristics
included in the model are age, gender, race, education, married, occupation,
hours worked in the week, health, tenure at the workplace, training received
at the workplace, whether the job was permanent and union membership.
Such variables can usually be found in most types of wage analysis; see
Polachek and Siebert (1993) for a review.
Wage equations using individual worker data are usually limited to in-
cluding only size of workplace and industry as employer variables (see
Brown and Medoff 1989; and Krueger and Summers 1988). Employer
variables have been shown in other studies to influence the remuneration
to workers. Agency theory predicts that there should be a difference in
remuneration to workers between the case where the owners are present
at the workplace, and that where ownership is separated from the control
of the firm. Krueger (1991) presents a test using fast-food restaurants. The
Xif matrix also included variables indicating the ownership and control
structure of the establishment, as well as variables indicating the size of the
workplace (number of employees), the industry, and union recognition.
Such ownership and control variables would also affect the use of and need
for communication methods at the workplace. Larger, more diversified
firms require that employees communicate more effectively. Omitting these
variables may lead to a larger or smaller coefficient because the variable of
interest is correlated with a workplace or worker variable.
Using the large sample of workers reporting wages as a censored
(grouped) variable allows more general inference on the distribution of
wages between and within establishments. The pay variable was asked in
terms of workers reporting their weekly or yearly pay within a specified
range. For the work presented in this paper, the wage was taken as the log
# Blackwell Publishers Ltd/London School of Economics 2001.

A New Voice or a Waste of Time? 265
weekly wage. The number of hours worked per week was conditioned as a
separate regressor in the wage equation. The pay variable is grouped; an
individual belongs to a wage group. Asking an individual to report pay in a
grouped manner is useful for two reasons. The first is that it increases the
number of responses in recording pay for each worker. The second is that it
may help reduce measurement error. Provided that appropriate assumptions
can be made about the underlying distribution of the variable, for example
that pay across individuals is normally distributed, the grouped variable
gives the same inference as asking for actual pay.
Finally, we can suppose that the error terms can be broken down into
three unobserved elements: eif = mi + lf + nif . Respectively, the three
elements are individual worker effects (mi), establishment effects (lf) and
a random error term (nif). The worker and establishment effects may be
correlated with any of the variables in X, or, for the purposes of this paper,
with e-mail use (cif). Appendix A presents different methods of estimation
that condition on the unobservable elements to varying degrees of success.
The variable of interest for this paper Ð different methods of communi-
cation within the workplace, and in particular e-mail use Ð is represented
by the cif variable in equation (1). Each method of communication (cif) is
indicated separately by a discrete variable, which takes the value of 1 if that
method was used, and 0 otherwise. According to equation (1), the estimated
parameter (f) provides an indication of the returns from using a particular
method of communication. It provides a comparison of the remuneration of
workers using, say, meetings against workers who do not, conditional all the
variables included in xif.
Provided that all the variables that influence an individuals' wage are
contained in the wage equation, estimating equation (1) will produce con-
sistent coefficients on the elements of Xif and cif. Complete details of the
estimation procedure are given in Appendix A. Estimation of equation (1) is
complicated by the fact that the dependent variable is grouped. This requires
a maximum likelihood estimator and an assumption about the underlying
distribution of pay within the sample. While allowing for observable char-
acteristics, for both worker and workplace, involves standard estimation
techniques, there may be unobservable characteristics of the worker or
workplace that are correlated with the variable of interest, biasing the result.
The basic hypothesis is that there are unobservable factors about workers
(for example, other skills not associated with education) or the workplace
(for example, a building that easily allows worker interaction) that may bias
the returns to different types of communication. Appendix A details two
different methods to attempt to condition on these unobserved factors, (a)
between establishments and (b) within establishments.
If the unobserved factors are important between workplaces, this indi-
cates that any observed returns from a communication method are more
relevant at the workplace level. The establishment may rely on efficient and
informative means of communicating with its work-force; similarly, the
establishment's work-force is more inclined to communicate more effectively.
# Blackwell Publishers Ltd/London School of Economics 2001.

266 British Journal of Industrial Relations
If the within-establishment unobserved effects are more important, then
worker attributes are more important; workers are perhaps more motivated,
or better able to use information and to communicate with their fellow
workers, to increase their earnings.
Not all differences between workplaces are captured in unobservable
factors for the set of establishments as a whole. The estimates are average
effects, and in averaging across the set of establishments important differ-
ences in the observed returns to communication may be hidden. As a further
examination of what may lie behind a return arising from communication
and a potential correlation with unobserved worker skills, use was made of
predictions from Bresnahan's (1999) organizational theory of computing, by
looking at the type of workplace, and the observed wage premium/labour
demand difficulties associated with different occupational classifications of
workers. As a further test, e-mail use was interacted with establishments that
communicate or introduce new technology to establish if it is the new
technology that counts, or the attitude to communication expressed by
management.
3. Results
The Basic Results
Table 3 presents the results from estimating equation (1) for the private-
sector establishments. The table divides up the estimates by estimation
method. What is apparent from this table is that there are large and well
defined (significant at the 95 per cent level of confidence) returns to using
e-mail, except when unobserved heterogeneity within establishments is con-
ditioned out of the estimating equation (column (5)). The point estimates
indicate that there is approximately a 10 per cent return from using e-mail
and a 5 per cent return from attending meetings. The returns from using
e-mail are similar in magnitude to those cited in Krueger (1993). There
appears to be no well-defined effect from reading the company newsletter or
noticeboard.
Most of the coefficients on the other worker and workplace variables are
of the expected sign and magnitude, given other work with household data
(Hildreth 1999). The return from age shows the usual quadratic curve; there
are increasing returns from remaining at the same workplace; there are also
increasing returns from training and education. Female workers earn less, as
do non-white workers and those with health problems. Married workers,
union members and workers in a permanent job all earn well-defined wage
premiums. The workplace variables are all well determined, and most have
the expected sign on the coefficient. The variables with the expected sign are
size of workplace (positive), industry effects, being a franchise (negative),
being part of a public limited company (PLC), and the workplace being
a single-establishment firm (negative). However, the sign of the coefficient
on the workplace recognition of a union is negative. It appears that the
# Blackwell Publishers Ltd/London School of Economics 2001.

TABLE 3
Estimates of the Returns to Using Different Methods of Communication
Dependent variable = grouped
pay
(1)
(2)
(3)
(4)
(5)
e-mail
0.097
(0.007)
0.053
0.102
(0.007)
0.098
(0.034)
0.047
(0.035)
Meetings
(0.009)
70.024
(0.011)
70.002
(0.008)
0.068
(0.008)
70.012
(0.008)
70.163
(0.008)
0.110
Noticeboard
Newsletter
Union member
Union recognition
Female
0.063
(0.008)
70.014
(0.008)
70.167
(0.008)
0.110
0.068
(0.008)
70.013
(0.008)
70.164
(0.008)
0.110
70.077
(0.039)
0.066
(0.080)
70.096
(0.041)
0.118
70.113
(0.039)
0.097
(0.078)
70.096
(0.041)
0.037
O levels
(0.010)
0.259
(0.010)
0.252
(0.010)
0.252
(0.037)
0.214
(0.040)
0.199
Degree
(0.013)
0.809
(0.017)
0.742
(0.013)
0.772
(0.017)
0.706
(0.013)
0.779
(0.017)
0.712
(0.054)
0.336
(0.066)
0.549
(0.057)
0.221
(0.066)
0.117
Managers/senior administrators
Professionals
(0.018)
0.618
(0.018)
0.427
(0.015)
0.371
(0.017)
0.085
(0.018)
0.583
(0.018)
0.397
(0.015)
0.370
(0.017)
0.081
(0.018)
0.586
(0.018)
0.398
(0.015)
0.371
(0.017)
0.084
(0.101)
0.239
(0.070)
0.097
(0.061)
0.206
(0.057)
70.154
(0.096)
0.284
(0.130)
0.216
(0.070)
70.013
(0.064)
0.101
(0.057)
70.023
(0.097)
0.270
Associate professional/technical
staff
Clerical/secretarial
Craft/skilled services
Personal/protective services
Sales
(0.020)
0.193
(0.020)
0.180
(0.020)
0.185
(0.017)
0.175
(0.016)
70.015
(0.007)
0.031
(0.003)
756.19
156.46
95.46
450.85
13,069
(0.017)
0.180
(0.016)
70.016
(0.007)
0.025
(0.003)
760.08
151.51
64.69
417.69
13,069
(0.016)
0.181
(0.016)
70.016
(0.007)
0.025
(0.003)
757.38
151.92
76.29
422.09
13,069
(0.075)
0.093
(0.049)
(0.084)
0.059
(0.047)
Operative/assembly work
PLC
Log (size of workplace)
(employees)
w2 (6) test for age dummies
w2 (4) test for tenure dummies
w2 (4) test for training dummies
w2 (11) test for industry dummies
No. of observations
42.01
14.68
11.63
9.68
29.90
3.05
12,756
8924
Notes
The number in parentheses is the standard error.
Columns (1), (2) and (3) were estimated using ordinary least squares (OLS). Column (4) was
estimated as a between-establishment fixed-effects model. Column (5) was estimated as a within-
establishment fixed-effects model. See Appendix A for details of the estimation procedure.
Other variables included in the OLS regressions were: age dummies, tenure dummies, training
dummies, union recognition, hours per week, whether the job is permanent, race, gender, CSE,
A levels, vocational qualifications, health restrictions, married, whether the workplace is part of
a PLC, whether it is a single-plant firm, whether it is part of a franchise, whether it has profit-
related pay, whether workers are assessed on individual performance, whether there has been
ownership change, industry dummies.
Other variables included in the fixed-effects models were: age dummies, tenure dummies,
training dummies, hours per week, whether the job is permanent, race, gender, CSE, A levels,
vocational qualifications, health restrictions, married.
# Blackwell Publishers Ltd/London School of Economics 2001.

268 British Journal of Industrial Relations
workplace recognizing a union lowers the average wage by about 1.5 per
cent.7
Although we might expect a difference between individuals in their ability
to use a computer, it is difficult to imagine, given that most people can listen
in meetings, that there should be a wage premium associated with these
types of communication. It is difficult to believe that there are returns from
attending meetings, so there might be unobserved characteristics, of either
the workplace or the workers, or both, that explain why there might be these
observed wage premiums.8 The results in the next two sub-sections discuss
the effect of unobserved heterogeneity on the returns to using e-mail.
Between-Establishment Effects
Column (4) of Table 3 gives the result from estimating the wage premium
for e-mail, conditional on the between-establishment effects (see Appendix A).
Between-establishment fixed effects can be conditioned out of the estimating
equations by transforming the variables. By averaging across workers in
each individual workplace, one can then subtract the mean from the reported
values to remove the unobserved establishment effect. By estimating equa-
tion (1) on the transformed variables, the estimates report a between-
establishment variation in e-mail use and the effect on pay, independent of
unobserved establishment and average worker effects (see Abowd and
Kramarz 1999).
What is noticeable about the results in Table 3, column (4), is that the
coefficient on the use of e-mail shows a positive, well defined return. There
appears to be a 10 per cent return from using e-mail.9 This indicates that it
is not the workplace, or the average quality of the workers they hire, that
determines the returns from communicating by e-mail. Potentially, there are
still unobservable worker effects that are correlated with e-mail use that
could explain the wage premium.
Within-Establishment Effects
Column (5) of Table 3 shows the within-establishment returns to e-mail
(see Appendix A for details). The within-establishment fixed effects were
generated by averaging across workers within pay groups, and then aver-
aging across workers within establishments. This technique removes more
variation due to workers that is not captured by the observed variables such
as age, education and the variables included in the estimation of column (3).
Column (5) shows that the within-establishment estimates of using e-mail
are not now well defined, the coefficient is not well determined, and statis-
tically, at a 5 per cent level of significance, the t-test on the e-mail use
coefficient indicates that there is no difference from zero.10
The difference in the returns from communication between columns (4)
and (5) can be interpreted in the following manner. The between-establishment
effects in column (4) show that, despite conditioning on observed and un-
# Blackwell Publishers Ltd/London School of Economics 2001.

A New Voice or a Waste of Time? 269
observed workplace characteristics, as well as an average of the unobserved
worker characteristics, there were still significant returns to using e-mail.
However, when some allowance was made for individual unobserved
heterogeneity within establishments, while also including an average of the
unobserved establishment effect, there was no discernable return to using
e-mail. This implies that it is not the workplace per se that matters, but the
unobserved quality of the work-force that uses e-mail. This result will be
explored in a later section to the paper when examining Bresnahan's (1999)
organizational computing theory.
Other results that are important to note are that the within-establishment
(pay group) effects diminish the significance of age effects, occupation and
training in the wage equation. The returns from education and from the
within-firm tenure remain well determined. Remuneration, both within pay
groups and within the firm, is mainly the result of individuals' qualifications
and the time they have been employed at the particular establishment. The
other result that is of some importance concerns the union wage effects of
membership and of recognition. Once unobserved individual and establish-
ment heterogeneity have been accounted for, the sign and significance of the
coefficients change. The coefficients indicate that any wage premium comes
from recognition rather than membership, and that membership tends to
be a negative effect. These results are different from those in the literature
that use household data (Hildreth 1999), but are similar to those reported in
Hildreth and Pudney (1999a). Hildreth and Pudney construct a complex
model to explain the union choice. A negative wage differential for union
membership was the result of individual choice for non-pecuniary union
benefits; a positive wage differential for union-recognized workplaces was
the result of retaining better-quality workers.
Conditioning on unobserved worker and establishment effects in this
manner indicates whether the average return from some factor has un-
observed elements included within it. It does not indicate what factors may
indicate the possible difference in workplaces or workers that gave rise to
the observed premium in the first place. Providing that the variables are
observed and recorded, it is possible to indicate what separates out the
successful from the less successful firms, and what constitutes `unobserved
factors'.
Spurious Correlation or Wage Premium?
As a check on whether the wage premium observed between establishments
in Table 3 was spurious, or was in fact a return to using e-mail as a more
productive method of communication, establishment-level regressions were
carried out where the dependent variable was an ordinal variable that
described how well the establishment was performing, in terms of financial
performance or labour productivity, against other establishments. The
results of estimating how e-mail affects the performance of the workplace
are given in Table 4. All variables are aggregated at the establishment level,
# Blackwell Publishers Ltd/London School of Economics 2001.

270 British Journal of Industrial Relations
TABLE 4
Estimates of Workplace Performance in Using Different Methods of Communication
Dependent variable = financial
performance
Dependent variable = labour productivity
(1)
(2)
(3)
(4)
e-mail
0.223
(0.105)
70.032
(0.221)
0.126
0.199
(0.105)
0.213
(0.158)
0.020
(0.232)
70.168
(0.234)
0.049
0.195
(0.151)
Meetings
Noticeboard
(0.151)
0.256
Newsletter
(0.227)
70.232
(0.229)
0.092
(0.203)
0.160
(0.161)
0.515
(0.237)
70.033
(0.211)
0.129
Union member
Female
70.242
(0.229)
0.079
(0.203)
0.203
0.503
(0.237)
70.052
(0.211)
0.203
O levels
(0.333)
0.221
(0.331)
0.186
(0.344)
0.241
(0.343)
0.169
Degree
(0.404)
70.069
(0.375)
70.505
(0.447)
0.286
(0.420)
0.154
(0.310)
0.113
(0.403)
70.034
(0.374)
70.467
(0.445)
0.294
(0.418)
0.147
(0.308)
0.091
(0.431)
70.099
(0.396)
70.663
(0.472)
70.001
(0.427)
0.377
(0.321)
70.222
(0.363)
0.352
(0.429)
70.070
(0.393)
70.709
(0.470)
70.099
(0.424)
0.337
(0.319)
70.263
(0.362)
0.296
Managers/senior administrators
Professionals
Associate professional/technical staff
Clerical/secretarial
Craft/skilled services
Personal/protective services
Sales
(0.355)
0.156
(0.309)
0.369
(0.354)
0.127
(0.326)
0.436
(0.337)
0.449
(0.335)
0.478
(0.323)
0.069
(0.320)
0.084
(0.320)
0.120
(0.316)
0.104
(0.336)
0.091
(0.329)
0.076
(0.334)
0.073
(0.326)
0.106
Operative/assembly work
PLC
(0.085)
0.071
(0.036)
9.43
2.44
2.96
14.81
1509
(0.084)
0.087
(0.035)
8.76
2.23
2.85
14.71
1509
(0.089)
0.014
(0.036)
11.99
6.64
8.90
12.95
1509
(0.088)
70.023
(0.036)
11.20
6.24
10.18
Log size of workplace
w2 (6) test for age dummies
w2 (4) test for tenure dummies
w2 (4) test for training dummies
w2 (11) test for industry dummies
No. of observations
12.45
1509
Notes
The number in parentheses is the standard error.
Other variables included in the regressions were: establishment averages of: age dummies,
tenure dummies, training dummies, union recognition, hours per week, whether the job is
permanent, race, gender, CSE, A levels, vocational qualifications, health restrictions, married,
whether the workplace is part of a PLC, whether it is a single-plant firm, whether it is part of
a franchise, whether it has profit-related pay, whether workers are assessed on individual
performance, whether there has been ownership change, industry dummies.
# Blackwell Publishers Ltd/London School of Economics 2001.

A New Voice or a Waste of Time? 271
so that the worker variables (e.g. e-mail, female or occupational category)
refer to the proportion of workers in that category within the workplace.
Table 4 shows that, even when allowing for a number of workplace
variables, as well as the average characteristics of the workers who work
there, there is an observed positive effect from using e-mail, both in terms of
financial performance and in terms of labour productivity. Although not
well determined at normal levels of significance, the coefficients on e-mail
were stronger than with any other method of communication. Other
methods of communication had no such productivity-enhancing effects that
could be observed at the workplace level. What is also (perhaps) surprising
is the lack of any role for average observable worker characteristics in
determining workplace performance. While the workplace variables such as
size (in terms of the number of employees) and industry were well
determined, average worker characteristics such as education, age or tenure
appeared to play no role.
As a further check on the possible link of unobserved effects (the work-
place plus the average for the worker) to workplace performance, the estimated
between-establishment effects and performance variables were correlated
with each other. The correlation coefficient for the unobserved effects and
the financial performance variable was positive (0.035) and significant at a
5 per cent level of significance (a standard error of 0.016). This implies a link
between some part of the unobserved element and the financial performance
of the workplace that is independent of the observable characteristics of the
workers or the workplace. In keeping with conclusions from other papers on
rent-sharing (see Hildreth and Oswald 1997), employers determine a large
extent of the wage distribution.11
Organizational Computing and Worker Characteristics
The results in Table 3 show that unobserved worker characteristics, correlated
with e-mail use, were likely to be generating the observed wage premium.
While conditioning on unobserved worker variation in the manner described
in Appendix A produces consistent estimates of the effects of e-mail use on
pay, it does not describe what the unobserved effects are. The use of com-
puters within an organization depends on the nature of the workplace, what
it produces, how many employees there are, and what type of employee it
needs. Bresnahan (1999), in setting out an organizational theory of computer
use (as described in the Introduction), notes that a firm does not simply buy
computers for the workforce and just put them it on the workers' desks: it
needs to integrate the machines, and the users, into the company network.
Different types of worker will benefit (or not) from using e-mail within the
firm. Bresnahan's theory suggests two different types of unobserved worker
skill that would be correlated with e-mail use. The first is people skills, i.e.
being able to deal with other people as a complement to using e-mail. The
second is cognitive skills (for example, co-ordinating communication with
different people), which are a substitute for e-mail use. Different occupations
# Blackwell Publishers Ltd/London School of Economics 2001.

272 British Journal of Industrial Relations
across the establishment will require different mixes of the substitute or
complement skills. If the substitution argument dominates, then, if the firm
should have trouble hiring this type of worker, the worker could be
substituted out of the workplace, and this would have negative correlation
between wages and e-mail use. If the complement argument dominates, and
the firm has trouble hiring workers in that particular occupation, there will
be a premium to such workers and to the use of information technology.
Table 5 examines Bresnahan's (1999) theory of organizational computing
as a means of examining how the unobserved ability may vary across types
of worker and in different jobs and workplaces. The table is divided into
four sections, reflecting employers with potentially different needs for
workers with high e-mail use. The divisions are by industrial sector,
production or service, and size of workplace Ð small (those firms below the
average workplace size) and large (the average workplace size and over).
According to Bresnahan's theory, the large service-sector employers would
be the ones that would most need e-mail use in conjunction with workers
with people skills (managers, administrators, professionals and technical
workers), and the most likely to remove workers that are replaceable by
e-mail (clerical and secretarial).
Table 5 was constructed in the following manner. Using equation (1), the
e-mail variable was interacted with the worker's occupation title. The rest of
the variables included in the regression equation were the same as those for
the results shown in Table 3, column (3). The wage premium associated with
e-mail now depends upon which occupation the individual is working in.
There is an average effect from using e-mail, given by the coefficient on the
e-mail variable, added to the effect from working in a particular occupation.
The vacancy difficulty numbers were simply the mean (proportion of
workplaces that reported `yes') for that particular occupation/industry
sector/size of workplace cell, correlated with the same cell e-mail use. The
vacancy difficulty numbers consist of a correlation coefficient, with the
number in parentheses reporting the significance of the correlation (with
the null hypothesis being that there is zero correlation).
The results in Table 5 are not altogether strong in supporting Bresnahan's
(1999) theory, but there is some evidence. The first result to note is that the
overall wage premium for e-mail use does vary by size of workplace and by
industrial sector. In both the production and service sectors, for small
employers there is no significant (at usual levels) premium for e-mail use.
This reflects the fact that its main effect for productive use lies with large
organizations. Indeed, Table 5 shows that for large workplaces there is a
well-defined premium associated with e-mail use. In terms of specifics, in
large organizations in both production and service sectors, for professional
workers there is a positive wage premium associated with e-mail use; also,
the workplace was having difficulty hiring workers for that type of job.
Workers who were able to use e-mail effectively to communicate and to
inform their fellow workers were the ones receiving a wage premium. This
can be contrasted with the `clerical and secretarial' occupation group, where
# Blackwell Publishers Ltd/London School of Economics 2001.

TABLE 5
Estimates of Occupational Returns to Using e-Mail and Difficulties in Filling Occupational
Vacancies, by Type of Workplace
Dependent variable = grouped pay
Production sector:
small employers
Production sector:
large employers
Returns
coefficient
Vacancy
difficulty
Returns
coefficient
Vacancy
difficulty
e-mail
70.008
(0.088)
0.109
(0.096)
0.119
0.184
(0.073)
70.060
(0.082)
0.119
Managers/senior administrators*
Professionals*
0.018
(0.389)
0.038
0.013
(0.428)
0.105
(0.100)
0.071
(0.105)
0.066
(0.095)
0.073
(0.094)
(0.073)
0.110
(0.000)
70.033
(0.113)
0.155
(0.087)
70.094
(0.085)
70.102
(0.079)
70.117
(0.077)
0.047
(0.275)
70.054
(0.116)
70.132
(0.080)
2599
(0.000)
70.016
(0.335)
0.067
(0.001)
70.045
(0.008)
70.011
(0.000)
70.027
(0.106)
70.136
(0.000)
Associate professional/technical staff*
Clerical/secretarial*
Craft/skilled services*
Personal/protective services*
Sales*
(0.000)
70.065
(0.140)
0.022
(0.095)
1749
0.043
(0.044)
0.040
Operative/assembly work*
No. of observations
(0.061)
Service sector: small
employers
Service sector: large
employers
e-mail
0.079
(0.059)
0.097
(0.046)
Managers/senior administrators*
Professionals*
0.069
(0.067)
0.063
70.020
(0.144)
0.113
0.042
(0.068)
0.111
0.070
(0.000)
0.113
(0.077)
0.038
(0.000)
0.076
(0.000)
0.056
(0.001)
70.086
(0.000)
70.077
(0.000)
70.015
(0.280)
70.040
(0.004)
(0.069)
0.006
(0.076)
70.134
(0.052)
0.033
(0.083)
70.001
(0.088)
0.025
(0.056)
70.084
(0.065)
3162
(0.000)
70.033
(0.034)
0.120
(0.000)
70.101
(0.000)
70.102
(0.000)
70.033
(0.030)
70.109
(0.000)
Associate professional/technical staff*
Clerical/secretarial*
Craft/skilled services*
Personal/protective services*
Sales*
(0.087)
70.009
(0.062)
70.134
(0.088)
0.131
(0.096)
0.016
(0.064)
70.041
(0.088)
3842
Operative/assembly work*
No. of observations
Notes
* Refers to the occupation title interacted with e-mail use.
For the Returns coefficients, the number in parentheses is the standard error. Other variables
included in the regression were: age dummies, tenure dummies, training dummies, union member-
ship, union recognition, hours per week, whether the job is permanent, race, gender, education
qualifications, health restrictions, married, occupation dummies, whether the workplace is part
of a PLC, whether it is a single-plant firm, whether it is part of a franchise, whether it has profit-
related pay, whether workers are assessed on individual performance, whether there has been
ownership change, industry dummies, size of workplace (number of employees).
`Vacancy difficulty' refers to the cell mean correlation between the occupation title and use of
e-mail in that occupation in the workplace. The figures in parentheses are significance values. A
value less than 0.05 indicates that the correlation does not reject the null of zero correlation
between the two measures at the 95% confidence level.

274 British Journal of Industrial Relations
workers experienced a negative wage effect associated with e-mail use, and
where the employer reported no trouble in hiring workers of this type. In
this instance the cognitive ability is negatively correlated with e-mail use,
and some of the tasks these workers perform are instead being carried out
by the new technology.
Separating Out High-Technology or High-Voice Establishments
The results from examining how unobserved ability may vary across occu-
pations within establishments were far from conclusive. As an alternative
to asking why there might be a premium associated with an ability to use
e-mail within a workplace, Table 6 uses the questions on whether managers
have a procedure for rewarding workers for effective communication, and
whether the establishment has introduced new technology within the last
five years. The variable called `rewards' equals 1 if the management finan-
cially rewards employees who offer helpful suggestions, 0 otherwise. The
variable `new technology' equals 1 if the establishment has introduced new
technology into the workplace in the last five years, or 0 otherwise.12
TABLE 6
Estimates of the Returns to e-mail for High-Technology and Voice-Rewarding Workplaces
(1)
(2)
e-mail
0.133
(0.009)
0.106
(0.055)
70.028
(0.012)
70.143
(0.086)
0.033
0.082
(0.052)
70.004
(0.027)
70.031
(0.048)
0.033
New technology
Financial rewards
e-mail*New technology
e-mail*Financial rewards
(0.049)
0.100
(0.016)
1011.96
216.74
117.78
3371.89
717.54
(0.018)
4.52
9.60
7.35
13.48
w2 (6) test for age dummies
w2 (4) test for tenure dummies
w2 (4) test for training dummies
w2 (8) test for occupation dummies
w2 (11) test for industry dummies
Notes
The number in parentheses is the standard error.
Column (1) was estimated using ordinary least squares (OLS).
Column (2) was estimated as a within-establishment fixed-effects model; see Appendix A for
details of the estimation procedure.
Other variables included in the OLS regression: age dummies, tenure dummies, training
dummies, union recognition, hours per week, whether the job is permanent, race, gender, CSE,
A levels, vocational qualifications, health restrictions, married, whether the workplace is part of
a PLC, whether it is a single-plant firm, whether it is part of a franchise, whether it has profit-
related pay, whether workers are assessed on individual performance, whether there has been
ownership change, industry dummies.
Other variables included in the fixed-effects model: age dummies, tenure dummies, training
dummies, hours per week, whether the job is permanent, race, gender, CSE, A levels, vocational
qualifications, health restrictions, married.
# Blackwell Publishers Ltd/London School of Economics 2001.

A New Voice or a Waste of Time? 275
To try to `net' the effects of communication against the effects of tech-
nology from e-mail use, interaction terms were used that were the same as
the method described for the results in Table 5. The results in Table 6 show
an interesting pattern. First, introducing new technology has a separable
effect from the use of e-mail. This implies that it is the communication itself,
rather than the technological disposition of the establishment, that matters.
The coefficients on both e-mail and new technology show well-defined
effects in the basic cross-section equations. The second implication of the
results in Table 6 is that it is not e-mail that matters at the workplace, but
how managers choose to reward the introduction of useful advice by
workers. The fact that e-mail has a separate (positive) effect tends to indicate
that this variable is a proxy for some other unobserved characteristics of
workers, as indicated in Table 3. This positive effect is correlated (to some
degree) with the type of firm and occupation in which the individual works.
4. Conclusions
Using WERS to estimate the returns arising from the employment of
different methods of communication has shown that any such observed
returns in the data are probably a product of unobserved characteristics of
the work-force at the establishment. However, some of the results on the use
of e-mail suggests that there may be other factors associated with this mode
of communication that are more akin to worker skills in using e-mail than to
e-mail as a mode of communication per se. There was some evidence that a
premium is associated with certain occupations being able to combine using
the new technology with other unobserved skills that are a complement to
e-mail use. What appears to matter within the workplace is how man-
agement chooses to reward information provision, rather than how com-
munication is made. Further, using e-mail and introducing new technology
are separable within the workplace. Overall, the results suggest that any
observed returns from using e-mail are probably indicating workers' un-
observed skills, for which they receive a premium against their fellow
workers. E-mail may be a new voice in the workplace, but it does not realize
any measurable pecuniary gain to its users unless the management chooses
to structure information use in that manner.
Appendix A: Estimation with Grouped Data in Matched
Employer±Employee Cross-Sections
In estimating the returns to using e-mail against other forms of communi-
cation, there are a number of problems to be overcome. One is that the
dependent variable is grouped. It is not a continuous variable, as pay is
often recorded as part of household surveys. Workers respond by reporting
that their pay falls within a `range' (£141±£180 per week, for example). Least
# Blackwell Publishers Ltd/London School of Economics 2001.

276 British Journal of Industrial Relations
squares estimation under such conditions is biased (Stewart 1983), entailing
that a least squares dummy variable approach to allow for establishment
specific effects will also be biased. Stewart (1983) proposes using a maxi-
mum likelihood method with starting values being given by a `corrected' set
of least squares estimates.
The starting point is the sample. Observations on the workers are the result
of a two-stage process. A stratified random sample was drawn for establish-
ments of over 10 employees, from which a random sample of workers was
drawn. A maximum of 25 employees were to be selected by the interviewer
from a list of all employees provided by the management. As the sample of
employees was a random draw from a stratified random draw, the sample
should be approximately representative of both employers and workers in
the cross-section. Hildreth and Pudney (1999a,b) discuss various issues of
sampling in matching workers and employers. The panel element in WERS
arises from multiple observations on workers per establishment. This allows
the removal of unobservable establishment effects as given in equation (1).
The basic model (where the method of communication variable Ð cif Ð
is subsumed into the X vector) can be defined as:
wif = Xif b+eif ,
(A1)
where i = 1, ..., N; f = 1, ..., F, and eif can be decomposed into three
elements: an individual effect (mi), an establishment-specific effect (lf) and
a random error term (nif * N (O, sn2)):
eif = mi+lf+nif.
From a single matched cross-section of workers and employers, the terms
for mi and lf cannot be separately identified (see Abowd and Kramarz 1999
for further discussion). An aggregate firm effect can be identified because of
the existence of more than one worker observation per establishment (see
Figure A1). Defining the transformation matrix (Zm) : Zm = iN : IF , where
iN is a vector of ones of dimension N, IF is an identity matrix of dimension
F6F, and : is the Kronecker product, the `within' transformation matrix
-
-
can be written as IF = Z(Z'Z)71 Z', and P = JN7IF, where JN is a matrix
of ones of dimension N. The within-establishment transformation is then:
U
Wp = INF7P. However, as neither establishments nor workers are
followed across time, it is possible only to identify an average workplace
effect that includes an average of the worker unobserved effect. Multiplying
(A1) through with UWp eliminates the lf, along with a workplace average of
the mi. The fixed-effect that is actually removed is:
Ð
xf = lf+mf ,
which removes the incidental parameters, but does not allow an examin-
ation of whether establishment or individual effects are potentially corre-
lated with any of the observable worker or workplace characteristics, or in
particular with the variables of interest.
# Blackwell Publishers Ltd/London School of Economics 2001.

A New Voice or a Waste of Time? 277
FIGURE A1
Number of Workers per Establishment for the Whole Sample: WERS 1998
0.1
0.08
0.06
0.04
0.02
1
2
5
10
15
20
25
num_work
To try to obtain an indication of the importance of the individual
heterogeneity within an establishment, the division of the dependent
variable (pay) into groups within establishments was used to obtain an
average worker effect per establishment. The `true' dependent variable is not
directly observed; instead, a grouped indicator of what pay an individual
receives is observed. What is observed about pay (wi Ð ignoring the con-
ditioning on f ) is that it is given in certain intervals of a real line. The real
line is divided into K intervals and the real line is exhausted by these
intervals such that the kth interval is defined as (Ak71, Ak), and the left and
right parts of the distribution are defined as A0 = 7? and AK = +?. An
indicator variable ki (14 ki 4 K) is observed for each i.
Alternatively, to try and condition on the individual fixed effect, a
transformation is suggested based on the K division of the dependent
variable. Identification will require that more than one person per wage
group is present at the workplace. Figure A2 illustrates the difference in
the sample from this restriction. The transformation is basically a within-
pay-group one, across establishment. Define a matrix Zl = iN : IK : IF ,
where IK is a matrix of dimension K6K. The within-pay-group transfor-
-
-
mation is then: UWk = INF7Q, with Q = JN7IKF and IKF = Z(Z'Z)71Z'.
If equation (A1) is now multiplied through with UWk , this eliminates the mi
±
along with the establishment effect averaged within K: lK. The fixed-effect
removed has the form
±
aK = mi+lK
# Blackwell Publishers Ltd/London School of Economics 2001.

278 British Journal of Industrial Relations
FIGURE A2
Number of Workers per Establishment for Establishments with more than Two Workers
in Any Pay Group: WERS 1998
0.12
0.1
0.08
0.06
0.04
0.02
2
5
10
15
20
25
num_work
which removes an average individual effect across K that incorporates part
of the averaged establishment effect. Once again, it is not possible to
separate out the individual fixed effect for the worker from the establish-
ment effect. However, the fixed effect, as defined as a within-pay-group,
across-establishment effect, does allow a more refined definition of the
establishment fixed effect (lf) if a two-way structure is assumed as part of
the model.
One-way between establishment effects
The one-way model is simply the one that concentrates on the establishment
effect as given by xf . As part of a fixed-effects model that is analogous to the
least-squares dummy variable model, the effects on wages resulting from
communication structure can be estimated allowing for the establishment
effect, plus the average of the workers' unobserved heterogeneity at that
workplace. The model to be estimated is
UWp {wif = Xif b+eif }.
(A2)
Given the grouped nature of the dependent variable, the one-way fixed-
effects model can be estimated by a basic log-likelihood function that allows
for the left and right censoring of the grouped data:
# Blackwell Publishers Ltd/London School of Economics 2001.

A New Voice or a Waste of Time? 279
ꢁ
ꢂ
ꢀ
ꢀ
ꢁꢃ
X
Ç
WpAif;k
ÇWpXif ꢀ
Ç
WpAif;k
ÇWpXif ꢀ
1
l ˆ
log
È
Ç
È
ꢁꢂ
ꢁꢂ
k2G
ꢀ
ꢁ
ꢀ
ꢁ
X
X
WpAif;k
Ç
WpXif ꢀ
Ç
WpAif;k
ÇWpXif ꢀ
‡
log È
‡
log È
;
ꢁꢂ
ꢁꢂ
k2L
k2R
provided that UWpAif, k4UWpAif, k71, where F is the standard normal
cumulative distribution function, k2G if the observation lies on the interval
(Ak71, Ak); k2L if the observations are left-censored; and k2R if the
observations are right-censored.13
Assuming fixed effects implies that there are factors specific to the unit of
observation that are unobserved, which the model should condition on.
Assuming fixed effects implies that there are factors correlated with the
observable characteristics of the worker and workplace which, if not
conditioned on separately, would bias the estimated coefficients. An
alternative view on the nature of unobserved heterogeneity is that the
unobserved effects are random across observational units. Assuming a
random effects structure implies that all factors are unconditional, or
marginal, to the population of possible outcomes. In the case of technology,
if the use of computers is correlated with either worker or employer
characteristics, then, allowing for fixed effects, there would be an expected
change in the coefficient on technology for either the worker or the
workplace. If the unobserved heterogeneity is assumed random in nature,
then there is the underlying assumption that the distribution of both
technology and unobserved worker and employer characteristics are
random across the sample.
A one-way random-effects model implies making a further assumption:
that xf * N (0, sx2). Random effects requires the formulation of a general-
ized component-weighting matrix. Following Maddala (1971), the weighting
matrix for a panel data set can be written as V = sn2IN + sx2JN. The inverse
of the weighting matrix is then:
"
#
ꢁꢃ2
1
ꢁꢂ2
1
V
ˆ
IN
JN
:
ꢁ2ꢂ ‡ Nꢁ2ꢃ
This is the weighting matrix used in the likelihood in the following manner:
Ã
X
plus terms for left and right-censored observations.
Â
1
l ˆ
log fÈk Èk 1gV fÈk Èk 1g ;
k2G
One-way within establishment effects
The estimation for the within-establishment effect is essentially the same
as for the between-establishment model. For the fixed effects, the UWk
# Blackwell Publishers Ltd/London School of Economics 2001.

280 British Journal of Industrial Relations
transformation was used, so that (A1) is now written as
UWk {wif = Xif b+eif }.
(A3)
For the random effects model, the weighting matrix is then
ꢂ
ꢃ
1
ꢁꢂ2
ꢁꢄ2
ꢁ2ꢂ ‡ Nꢁ2ꢄ
1
V
ˆ
IN
JN
:
Appendix B: List of Variables by WERS Data Set
Seq98.dta (Employee Questionnaire)
serno, serial, a1, a2, a3, b2, b6a, b6b, b6c, b6d, c1, c5, d1, d2, d3, d4, d5, d6,
d7, d8, d9, d10code, d11
Mq98.dta (Management Questionnaire)
serno, zallemps, asic, asingle, astatus, afranch, acontrol, aconhead, ahow-
long, aownchan, dcircles, dperfor1, fvarpay1, fvarpay2, fvarpay3, fvarpay4,
fmeasur1, fmeasur2, fmeasur3, fmeasur4, jnonem08, lmancha1, lmancha2,
lmancha3, lmancha4, lmancha5, lmancha6, lmancha7, limpcha, ltricha1,
ltricha2, ltricha3, ltricha4, ltricha5, ltricha6, lunacha, lynoch01, lynoch02,
lynoch03, lynoch04, lynoch05, aphras10, dconsul1, dconsul2, dconsul3,
dconsul4, dconsul5, dconsul6, khowmon1, klabscal, kestper1, kestper2,
cvacdif1-cvacdif9
Final version accepted 29 November 2000.
Acknowledgements
The data are available on request to the Data Archive at the University of
Essex. All programmes are available on request to the author. My thanks
to John Van Reenen for providing helpful comments over a beer at the
Bison Brewery on an initial draft of the paper, which helped greatly in
this revision. I would also like to thank participants, in particular Willy
Brown, at the British Journal of Industrial Relations WERS98 Conference
at Cumberland Lodge, 20±21 September 1999, for a lively and frank dis-
cussion that led to the improvements in the paper. Thanks also to two
anonymous referees whose careful reading of the paper led to a number of
improvements.
# Blackwell Publishers Ltd/London School of Economics 2001.

A New Voice or a Waste of Time? 281
Notes
1. It might be noted that Chennels and Van Reenen (1998) use the 1984 and 1990
WIRS (Workplace Industrial Relations Survey) and find that cross-section
correlations show that the introduction of technology increases wages (especially
for the higher skilled workers). However, when identifying the existence of
technology using gender composition at the plant, there is no significant effect of
technology on wages.
2. The method used by Entorf and Kramarz is to difference out the unobserved
worker and firm effects. This is done by identifying through the use of `switchers'
(workers who change) between firms. If a sufficient number of workers switching
between the set of firms are observed, it becomes possible to `difference out' the
unobserved elements Ð the same worker at two different firms, two different
workers at the same firm, for example. The effects of technology are estimated
both with and without the unobserved employer effects, and the difference be-
tween the two sets of estimates reveals whether unobserved worker or employer
effects are correlated with the worker or firm observed variables.
3. Bresnahan (1999) describes `people skills' as being associated with supervising
and being supervised.
4. Throughout the paper, the terms `workplace' and `establishment' are used to
mean the same entity as described in the WERS98 survey description. The
principal unit of analysis for WERS98 is the establishment or workplace. A
workplace is defined as comprising the activities of a single employer at a single
set of premises. Examples include a single branch of a bank, a car factory or a
school. This is the same definition as used by the Office for National Statistics
(ONS) when surveying business units for the Annual Business Inquiry (ABI),
formerly known as the Annual Census of Production. See Hildreth and Pudney
(1999b).
5. A complete response was defined as a response to all the variables included in the
estimation.
6. Note that the sinister index is something of an ad hoc construct. Equal weighting
is given to each element that comprises the index. The questions used were
apparently the only ones within the questionnaire that dealt with information
provision and use within the workplace. As a means of examining which elements
of the index were important, each component part was correlated against each
communication method. Table 2 presents the results.
7. See Andrews et al. (1998). If union membership and union recognition dummy
variables are both included in the same estimating equation, then the coefficient
on the union recognition represents the coefficient on union coverage only; the
union membership effect is the joint product of the coefficients on membership
and recognition. In the results in Table 3 (columns (1)±(3)), the union recognition
effect is slightly negative. This accords with other studies using British data
(Andrews et al. 1998). However, both union effects are potentially biased because
of the selection of different types of worker into union status and union
employers.
8. The premium to meetings was not a consideration, as the paper is concerned
with the use of technology for information within the workplace. The between-
establishment coefficient on a wage premium for meetings is very small and not
significantly different from zero (even at high levels of significance). However, as
one referee pointed out, it may very well be the case that within workplaces some
# Blackwell Publishers Ltd/London School of Economics 2001.

282 British Journal of Industrial Relations
workers will take part in meetings, or managers may hold meetings only with
workers who come up with the best ideas. This may be true, but, as meetings
have been a forum for information at the workplace for some time and do not
involve a change in technology at the workplace, holding a meeting does not
necessarily require the introduction of new technology, whereas communicating
by e-mail does.
9. It might be noted that such a return is not too different from other estimates of
using computers for British workers (Bell 1996).
10. However, it should be noted that using data in this form reduces the number
of observations by a large amount (see Table 3, column (5)). The OLS results
from this subset of data are not substantively different from using the whole
sample.
11. Hildreth (1998) noted that the distribution of workers across profitable estab-
lishments was such that a high recorded rent-sharing parameter could explain
the whole wage distribution as higher qualified workers were sorted into the
lowest profit workplaces. The problem here is that the financial performance
question is a relative measure (as is labour productivity). However, if we under-
take the same exercise for WERS98, there is little pattern of more qualified
workers being employed in poorer performing workplaces. The relative financial
performance of workplaces is not explained by the observable traits of workers.
12. My thanks to an anonymous referee who pointed out that these two items are
not strictly comparable. Financial rewards were asked only of managers oper-
ating `quality circles or something similar'. It may be that workplaces that do not
operate quality circles do reward employees who offer helpful suggestions. The
results as presented here may be biased. There is no way to correct for this within
the confines of the WERS questionnaire.
13. An alternative method that has been used before would be to condition on
establishment fixed effects using the establishment specific dummy variables
(Cahuc and Kramarz 1997). Although theoretically such an approach should
work in this case, the fact that the dependent variable is grouped, and that
so many observed characteristics are discrete, introducing establishment fixed
effects into the likelihood gives it a high dimensionality. Consequently, the like-
lihood had trouble converging.
References
Abowd, J. M. and Kramarz, F. (1999). `The analysis of labor markets using matched
employer±employee data'. In O. Ashenfelter and D. Card (eds.), Handbook of
Labor Economics, Vol. III. Amsterdam: North-Holland.
Andrews, M. J., Stewart, M. B., Swaffield, J. K. and Upward, R. (1998). `The
estimation of union wage differentials and the impact of methodological choices'.
Labour Economics, 5: 449±74.
Autor, D. H., Katz, L. F. and Krueger, A. B. (1998). `Computing inequality: have
computers changed the labor market?' Quarterly Journal of Economics, 113: 1169±
1214.
Bell, B. D. (1996). `Skill-biased technical change and wages: evidence from a
longitudinal data set'. Unpublished paper, Nuffield College, Oxford.
Bresnahan, T. F. (1999). `Computerisation and wage dispersion: an analytical
reinterpretation'. Economic Journal, 109: F390±F415.
# Blackwell Publishers Ltd/London School of Economics 2001.

A New Voice or a Waste of Time? 283
Brown, C. and Medoff, J. (1989). `The employer-size wage effect'. Journal of Political
Economy, 97: 1027±59.
Cahuc, P. and Kramarz, F. (1997). `Voice and loyalty as a delegation of authority:
a model and a test on matched worker±firm panels'. Journal of Labor Economics,
15: 658±88.
Chennels, L. and Van Reenen, J. (1998). `Establishment level earnings, technology
and the growth of inequality: evidence from Britain'. Economics of Innovation and
New Technology, 5: 139±64.
Cully, M., Woodland, S., O'Reilly, A. and Dix, G. (1999). Britain at Work: As
depicted by the 1998 Workplace Employee Relations Survey. London: Routledge.
DiNardo, J. E. and Pischke, J-S. (1997). `The returns to computer use revisited:
have pencils changed the wage structure too?' Quarterly Journal of Economics, 112:
291±304.
Disney, R., Gosling, A. and Machin, S. (1995). `British unions in decline: deter-
minants of the 1980's fall in union recognition'. Industrial and Labor Relations
Review, 48: 403±19.
Doms, M., Dunne, T. and Troske, K. R. (1997). `Wages, workers, and technology'.
Quarterly Journal of Economics, 112: 253±90.
Entorf, H. and Kramarz, F. (1997). `Does unmeasured ability explain the higher
wage of new technology workers?' European Economic Review, 41: 1489±510.
Haskel, J. (1999). `Small firms, contracting out, computers and wage inequality:
evidence from UK manufacturing'. Economica, 66: 1±21.
Hildreth, A. K. G. (1998). `Rent-sharing and wages: product demand or technology
driven premia?' Economics of Innovation and New Technology, 5: 199±226.
ÐÐ (1999). `What has happened to the union wage differential in Britain in the
1990's?' Oxford Bulletin of Economics and Statistics, 61: 5±37.
ÐÐ and Oswald, A. J. (1997). `Rent-sharing and wages: evidence from company
and establishment panels'. Journal of Labor Economics, 15: 318±37.
ÐÐ and Pudney, S. E. (1999a). `Employers, workers, and unions: analysis of a
matched firm±worker panel with endogenous sampling and survey response'.
Unpublished paper, University of California-Berkeley.
ÐÐ and ÐÐ (1999b). `Econometric issues in the analysis of linked worker±
employer cross-section surveys'. In J. C. Haltiwanger, J. Lane, J. Theeuwes and
K. Troske (eds.), The Creation and Analysis of Employer and Employee Matched
Data. Amsterdam, North-Holland.
Hirschman, A. (1970). Exit, Voice, and Loyalty. Cambridge, Mass.: Harvard Unver-
sity Press.
Krueger, A. B. (1991). `Ownership, agency, and wages: an examination of
franchising in the fast food industry', Quarterly Journal of Economics, 106: 75±102.
ÐÐ (1993). `How computers have changed the wage structure: evidence from
microdata, 1984±89'. Quarterly Journal of Economics, 108: 33±60.
ÐÐ and Summers, L. S. (1988). `Efficiency wages and the inter-industry wage
structure'. Econometrica, 56: 259±93.
Machin, S. and Van Reenen, J. (1998). `Technology and changes in skill structure:
evidence from seven OECD countries'. Quarterly Journal of Economics, 113: 1215±
44.
Maddala, G. S. (1971). `The likelihood approach to pooling cross-section and time-
series data'. Econometrica, 39: 939±54.
Polachek, S. W. and Siebert, W. S. (1993). The Economics of Earnings, Cambridge:
Cambridge University Press.
# Blackwell Publishers Ltd/London School of Economics 2001.

284 British Journal of Industrial Relations
Stewart, M. B. (1983). `On least squares estimation when the dependent variable is
grouped'. Review of Economic Studies, 50: 737±54.
Van Reenen, J. (1996). `The creation and capture of rents: wages and innovation in a
panel of UK companies'. Quarterly Journal of Economics, 111: 195±226.
# Blackwell Publishers Ltd/London School of Economics 2001.