Do ICT Skill Shortages Hamper Firms’ Performance? Evidence from UK Benchmarking Surveys 1 John Forth Geoff Mason * National Institute of Economic and Social Research, London September 2006 Abstract Abstract – In light of the increased relative demand for skilled labour associated with Information and Communication Technologies (ICTs), we combine survey data for UK enterprises in 1999 with post-survey financial data for the same enterprises to assess the impact of ICT skill shortages on firms’ financial performance. There is clear evidence that ICT skill shortages have an indirect negative impact on performance through the restrictions that such deficiencies place on ICT adoption and on the intensity of ICT use post-adoption. However, there is only weak evidence of skill shortages impinging directly on performance at given levels of ICT adoption and utilisation.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Do ICT Skill Shortages Hamper Firms’ Performance?
Evidence from UK Benchmarking Surveys1
John Forth Geoff Mason
* National Institute of Economic and Social Research, London
September 2006
Abstract
Abstract – In light of the increased relative demand for skilled labour associated with
Information and Communication Technologies (ICTs), we combine survey data for UK
enterprises in 1999 with post-survey financial data for the same enterprises to assess the
impact of ICT skill shortages on firms’ financial performance. There is clear evidence
that ICT skill shortages have an indirect negative impact on performance through the
restrictions that such deficiencies place on ICT adoption and on the intensity of ICT use
post-adoption. However, there is only weak evidence of skill shortages impinging
directly on performance at given levels of ICT adoption and utilisation.
I. Introduction
There is now a great deal of evidence that the diffusion of Information and
Communication Technologies (ICTs) over recent decades has helped to enhance the
relative demand for skilled labour (Berman, Bound and Machin, 1998; Acemoglu,
2002). Indeed, the skill-biased nature of the ICT ‘technological revolution’ is one
reason why pay-offs to ICT investments at firm and industry level have taken time to
develop, in contrast to some previous new technologies which were complementary
with low-skilled labour and thus capable of being implemented relatively quickly
(Caselli, 1999).
In this context, it was hardly surprising at times of peak demand for ICT-skilled labour
over the last decade, to hear reports of ICT skill shortages which, it was argued, could
have serious negative effects on firms’ commercial performance (IDC, 1999; European
Commission, 2002; ITAA, 2000, 2002). Since 2002 the emphasis on ICT skill
shortages has diminished on both sides of the Atlantic as many ICT-skilled workers
have experienced lay-offs and difficulties in finding ICT-related employment. But even
when forecasts of shortages were made, they attracted criticism on the grounds of
under-estimating potential sources of ICT skills and, especially, taking little account of
the many ways in which employers might respond to periodic skill shortages, for
example, raising salaries, increasing training and introducing labour-saving changes in
work organisation (Lerman, 1998, UEF, 1999; Mason, 2000; Wharton, 2002).
Some years later it remains an open empirical question whether, and to what extent,
ICT skill shortages have negative effects on firm performance. In the present paper we
examine this question by making use of an unusual dataset which combines information
from an ICT benchmarking survey of British enterprises with post-survey financial data
for the same enterprises. The properties of this dataset enable us to test a number of
hypotheses relating to the impact of ICT skill shortages on firm-level performance.
The paper is ordered as follows: Section 2 briefly reviews the separate literatures on
employer responses to skill shortages and the relationship between ICTs and skills
demand in order to develop the theoretical framework from which our central
hypotheses are derived. Section 3 describes the dataset, including the process by which
post-survey financial data were matched to benchmarking survey respondents. Sections
4 and 5 present new summary measures of ICT adoption and utilisation and indicators
of ICT skill shortages at firm level. Section 6 examines the main determinants of ICT
adoption and utilisation. Section 7 presents estimates of the impact of ICT skill
shortages on financial performance at firm level. Section 8 concludes.
II. Discussion and Hypotheses
In assessing the nature and impact of skill shortages, it is important to distinguish
between, on the one hand, employers’ reported difficulties in recruiting skilled workers
on the external labour market and, on the other hand, firms’ ‘internal’ skill deficiencies,
that is, gaps between firms’ current skill levels and their desired or optimum level of
skills (Green and Ashton, 1992). Early analyses of skill shortages tended to focus on
those associated with external recruitment difficulties and were particularly concerned
2
with how and why such phenomena should exist. A common argument was that, in
competitive labour markets, excess demand for skills should put upward pressure on
salaries and thus eventually (all else being equal) help to stimulate an increase in supply
which would eliminate the shortage. In response to this line of thinking Arrow and
Capron (1959) suggested that, even in a competitive labour market, a steady increase in
demand over time for skilled workers (such as engineers and scientists) could produce a
‘dynamic shortage’ due to factors impeding rapid salary increases by employers such as
delays in accepting the needs for such increases, the further time needed to implement
them and a reluctance to incur increased salary costs for existing high-level technical
employees as well as new ones.2
At the same time supply responses to any salary improvements could be slowed down
by the length of time required to educate and train skilled workers. In the case of
engineers and scientists, Freeman (1971, 1976) argued that a slow supply response
could potentially lead to a sequence of imbalances in labour markets whereby, for
example, a delayed increase in supply in response to improved salary prospects could
come on stream just as there was a downward shift in demand; the resulting decline in
relative rates of return to qualification as an engineer or scientist could in turn prompt a
reduction in new entrants to engineering and science courses and thus sow the seeds of
future shortages if demand for engineers and scientists happened to increase at much
the same time as a reduction in new supply entering the market.
Subsequent discussion of the reasons why skill shortages might come to exist and even
to persist over some length of time drew heavily on the concept of ‘internal labour
markets’ (ILMs) which effectively detach firms from external market pressures by
3
confining new recruits to a narrow set of ‘entry jobs’ and relying on internal promotion
to more senior jobs.3 In the standard literature on segmented labour markets, ILMs are
typically associated with relatively high wages and salaries (often linked to seniority) in
‘primary’ sectors of employment as compared to unstable, low-wage employment in
‘secondary’ sectors (Doeringer and Piore, 1971). However, detachment from external
labour market pressures can also limit the extent to which salaries rise in response to
shortages of particular categories of employee. For example, case studies of British
employers of engineers in the 1970’s found that many employers preferred to respond
to recruitment difficulties with non-salary measures such as reorganisation of
workloads and recruitment of less well-qualified personnel rather than disturb the
internal parities on which salaries were based (Mace and Wilkinson, 1977). The
strength of ILM structures in these firms was shown by persistent salary differentials
between the firms for similar categories of engineer (Mace, 1979).
But even in the absence of ILM-type rules and procedures, firms have a range of
potential non-salary responses and ‘coping mechanisms’ available to them when
confronted by actual shortages – such as asking existing employees to work longer
hours, making increased use of subcontractors or retraining existing staff to develop the
skills in shortage. In a study based on data from the 1984 Workplace Industrial
Relations Survey (WIRS), Haskel and Martin (1993a) found no evidence of firms
setting higher wages in response to difficulties in recruiting skilled workers. Indeed,
they cited other UK survey evidence to suggest that salary responses were much less
important than other means of addressing skilled recruitment difficulties.
4
Hence skill shortages deriving from external recruitment difficulties may persist for
some length of time if responses to them take the form of temporary alleviation of their
effects rather than seeking to increase the supply of skills (for example, through
increased training or raising wages for skilled labour). The same is true of ‘internal’
skill shortages, that is, gaps between current skills and desired skills among the existing
workforce. Employers Skill Survey (ESS) data for England show that the proportion of
establishments reporting internal skill shortages is some 2.7 times greater than the
proportion reporting skill-related external recruitment difficulties.4 At sector level the
persistence of internal skill shortages over time is generally greater than for external
recruitment difficulties (Forth and Mason, 2003). In some cases firms may elect to ‘live
with’ internal skill shortages for periods of time rather than incur the costs of training
and updating workers that would be necessary to bridge the gaps in skills. Such
behaviour could reflect the relatively high costs of bringing lower-skilled workers up to
the required skill standards as compared to employers who are starting from a superior
position in terms of existing skills. It could also reflect imperfect information about the
costs and benefits of training versus other potential responses to internal skill shortages.
Empirical evidence for the UK suggests that many firms choose to tolerate – or fail to
cope with – skill shortages for long enough for negative effects on performance to be
identified. For example, in a study based on CBI survey data, Haskel and Martin
(1993b) found that skilled labour constraints on output had significant negative effects
on labour productivity at industry level.5 They suggested that the main ways in which
skill shortages had these effects were by shifting the employment mix in favour of less
productive low-skilled workers and by reducing the bargaining power of employers in
relation to worker effort. By contrast Nickell and Nicolitsas (2000) focused on the
5
cumulative effects of reduced investments in assets which are complementary to skills,
finding that skilled labour shortages at industry level were significantly and negatively
associated with subsequent reductions in investments in physical capital and R&D at
firm level in the industries concerned. Negative impacts on performance may also occur
through delays in innovation: the Technical Graduates Employers Survey in the UK
found that two thirds of employers which had experienced difficulties in recruiting
high-level skilled personnel reported suffering commercial problems as a result. The
most common problem mentioned was delays in product development and process
improvement projects, impacts which may have no immediate effect on performance
but may contribute to weaker performance in later time periods (Mason, 1999).
The literature on skill shortages thus identifies a number of mechanisms by which
different kinds of shortage may have negative effects on firm performance. The same is
true by implication of the literature on the positive associations between the diffusion of
ICTs and employer demand for skills. Recent studies have focused on the role of skills
in facilitating the effective utilisation of ICTs (for example, Brynjolfsson, Hitt and
Yang, 2002). However, there is also an older literature which highlights the role of
highly-educated or skilled workers in facilitating early adoption of new technologies
(Nelson and Phelps, 1966; Welch, 1970; Schultz, 1975; Bartel and Lichtenberg, 1987).6
In this vein Doms, Dunne and Troske (1997) found that the adoption of
microelectronics-based technologies in US manufacturing plants was positively related
to pre-existing workforce skill levels. Therefore, in the present paper we distinguish
carefully between the adoption of ICTs and the intensity of their use in order to assess
the impact of skill shortages on firm-level performance.
6
Our theoretical framework is centred on profit-maximising firms which, all else being
equal, respond to changes in the relative prices of production inputs by increasing
(reducing) their purchases of those factors which have fallen (risen) in price. In recent
decades ICTs have been characterised by rapid declines in price: performance ratios
which have increased their attractiveness relative to the use of non-ICT capital inputs
and other inputs (such as unskilled labour) which are not complementary to the use of
ICTs.
In principle, ICT investments should help early-adopting firms to achieve higher levels
of performance, for example, by improving the efficiency with which various tasks are
carried out by different sections of the workforce; and /or by facilitating more rapid
monitoring of trends in customer demand and improvements in communications with
suppliers of key components and services. A number of studies have now produced
evidence of ICT investments enhancing firm-level productivity performance (for
example: Lichtenberg and Lehr, 1999; Black and Lynch, 2001; Brynjolfsson and Hitt,
2003). However, in common with some previous new ‘general purpose’ technologies
such as electrification, the short-term impact of ICT investments on firm-level
performance may be small or even negative due to the time and resources needed to
develop complementary production inputs (Helpman and Trajtenberg, 1998). Some
indication of the time needed for these complementary inputs to be developed is given
by Basu et al. (2004) who find that TFP growth at industry level in the US is
significantly related to ICT capital growth with long lags ranging between 5-15 years.
In order to make effective use of ICTs, many firms need to pass through periods of
experimentation and learning, investing in the adaptation or development of software,
7
the implementation of appropriate new modes of work organisation and the
development of new products and services. The development of ICT-related skills is
central to this process of organisational change. Indeed, Bresnahan et al. (2002) suggest
that organisational investments in assets which are complementary to ICTs may
contribute more to raising the relative demand for skilled labour than the diffusion of
ICTs themselves.
In this context we expect that – all else being equal – shortages of ICT-skilled workers
may have negative effects on firm-level performance via several different but related
channels of influence, for example:
restricting their ability to make early use of cost-saving new ICTs;
limiting the improvements in efficiency which might otherwise have been achieved
by utilising ICTs once they have been adopted;
slowing the rate of adaptation to new forms of work organisation that are
complementary with ICTs;
slowing down the rate of development of new products or services which ICTs
make possible.
On the other hand, we may also expect that such effects of skill shortages might be
reduced if the firms affected by shortages respond in ways (such as increasing
workforce training) that help to alleviate the shortages over time.
These expectations regarding the different impacts of ICT skill shortages and ICT-
related training on firm-level performance can be restated as hypotheses to be tested.
We first consider the possibility that ICT skill shortages affect performance indirectly
8
by slowing down ICT adoption and utilisation which would be beneficial for the firm.
This is done by examining the evidence in relation to the following two hypotheses.
Hypothesis 1
All else being equal, reported ICT skill shortages at firm level are negatively
related to measures of ICT adoption and to the intensity of use of ICTs .
Hypothesis 2
All else being equal, these measures of ICT adoption and the intensity of use of
ICTs are positively related to firm-level financial performance.
We then assess the extent to which ICT skill shortages have a direct negative effect on
performance, for example, by limiting firms’ ability to make effective use of ICTs that
have already been installed.
Hypothesis 3
At given levels of ICT adoption and utilisation, ICT skill shortages are
negatively related to firm-level performance.
Finally, we consider the possibility that firms’ efforts to develop skills that are
complementary to the use of ICTs have positive effects on firm performance (thus
helping to offset any negative impact of skill shortages).
Hypothesis 4
At given levels of ICT adoption and utilisation, investments in ICT skills
training contribute positively to firm performance.
9
III. The Matched IBS-Dun & Bradstreet Dataset
Our initial company-level data derive from the UK section of the 1999 International
Benchmarking Survey (IBS), commissioned by the UK Department of Trade and
Industry (DTI), which compared the use of ICTs by British enterprises against that in
other industrialised countries such as France, Germany, the US and Canada (Spectrum,
1999). The target population was UK businesses as a whole, including partnerships and
sole proprietorships as well as private and public limited companies. The sampling
frame was derived from Dun & Bradstreet’s (D&B’s) company financial database
supplemented by companies listed on the Yellow Pages database. A total of 2,410
company respondents were interviewed; our analysis excludes 488 organisations
located in Public Administration, Education and Health in order to focus on the 1,922
private sector organisations.7 Data from the International Benchmarking Surveys of
1997 and 1998 are also called upon in Section IV to provide information on changes in
ICT adoption and utilisation in the late 1990s.
In order to explore the determinants of company performance, we sought to match
survey responses from the 1999 IBS against post-survey financial data up to 2001
provided by Dun and Bradstreet (D&B). Usable data on employment, sales and capital
assets were obtained for some 459 companies, representing 24 per cent of all private
sector organisations which participated in the 1999 IBS. The attrition which occurred
was primarily due to gaps in D&B records along with the exclusion of some companies
which either failed our matching checks or for which insufficient information was
available to enable such checks to be carried out. In addition, holding companies were
10
excluded from the analysis because of uncertainty over the correspondence between the
interviewed unit and that which provided financial data to D&B.
As a result of this process, smaller companies (with fewer than 50 employees) are
under-represented in the matched dataset as compared to the original IBS sample,
largely due to limited coverage of those size-groups by D&B (Table 1). And in terms of
sectors, manufacturing companies turn out to be somewhat over-represented at the
expense of service companies. However, these selection biases are smaller than might
have been feared and we can expect any potential impact on our regression parameters
to be largely mitigated by the inclusion of size-group and industry dummies. In
addition, the matched dataset still benefits from a fairly wide spread of companies
across size-groups and sectors.
* Table 1 about here *
IV. ICT Adoption and Utilisation
The findings from successive IBS surveys between 1997 and 1999 highlight the rapid
growth in take-up of ICTs over a relatively short period. By 1999 some 88 per cent of
IBS respondents made use of at least one computer - this proportion had not changed
much in the previous two years (Table 2). However, this was a period of rapid growth
in the use of email (65 per cent of businesses in 1999, up from 37 per cent in 1997), the
Internet (53 per cent in 1999, up from 26 per cent) and the development of companies’
own websites (reported by 42 per cent of respondents in 1999 compared to only 17 per
cent in 1997). In addition, during this period there were marked increases in the
11
proportions of businesses using Intranets or Extranets and those facilitating remote
access to company computer systems by employees.
* Table 2 about here *
These data permit a summary measure of the extent of ICT adoption to be derived from
a simple count of the different types of ICT equipment and facilities in use in each
company at the time of the annual surveys. As shown in Table 3, Part A, by 1999 only
10 per cent of IBS respondents did not use at least one of the eight listed types of ICT
while 35 per cent used six or more. The rapid increase in ICT take-up can be seen by
the position of the median enterprise in each year which changed progressively from
using two different types of ICT equipment or facilities in 1997 through to using four
different types in 1999.
* Table 3 about here *
In addition we define two different summary measures of the intensity of ICT
utilisation:
(i) the proportion of employees making use of computers, available for 1998-99
(Table 3, Part B)
(ii) an index ranging from 0 to 6 which captures both the extent of usage of three
key ‘connectivity technologies’ (networks, Internet, company websites) and the
extent to which these technologies are used to communicate with and engage in
on-line transactions with customers and suppliers (Table 3, Part C)
12
As indicated in the notes to Table 3, Part C, a score of three out of six on the measure
of intensity of use of connectivity technologies corresponds to, for example, using
networks for local communications only, and using the Internet for email, information
and marketing purposes but not for on-line purchasing or sales, and using a company
website for advertising, marketing, etc., but not for on-line sales. In 1997 some 91 per
cent of IBS respondents were rated at point 3 or lower on this scale but by 1999 the
equivalent proportion had dropped to 67 per cent with the remaining 33 per cent all
rated at points 4-6, that is, possessing wide area networks and/or using the Internet or
their own websites for on-line commercial transactions.8
V. ICT Skill Shortage Measures
The main question on skill shortages in the 1999 IBS captures information about
internal skill gaps among existing employees. It was phrased as follows: ‘Do your
employees have sufficient understanding of the ICTs available in your company to
enable them to maximise the competitive advantage that these technologies bring?’
Only 20 per cent of enterprises replied that all employees possessed sufficient
understanding. A further 40 per cent replied that ‘Yes, some’ employees fitted this
description while the remaining 41 per cent simply answered ‘No’. 9
The 1999 survey also asked, firstly, about the extent to which ‘Information and
communication technology skills in the workplace’ were ‘influential in dictating your
business’s uptake of [ICTs]’ and, secondly, how much of a ‘barrier’ these skills had
been to ICT adoption. Since the responses to both these questions were allocated to 1-5
point scales, we defined a new ‘Net skills influence’ variable defined as ‘Influence
13
score’ less ‘Barrier score’ which ran from a minimum of –4 (=1-5) through to +4 (=5-
1). High positive scores (+3 or +4) on this scale suggest that the companies concerned
were well-endowed with ICT skills while low negative scores (–3 or –4) point to a
severe ICT skills constraint. About 21 per cent of respondents had negative scores
implying that ICT skills had been more of a barrier than a positive influence on the
uptake of ICTs. Just under half (48 per cent) had positive scores, implying the reverse.
And 31 per cent had zero scores, implying that the effect of ICT skills had been largely
neutral. This ‘net skills influence’ variable turned out to be significantly negatively
correlated with the ICT skills shortage measure (r = -0.23, p<0.01), which is a form of
validation of managers’ assessments of the presence or absence of ICT skill
deficiencies.
VI. The Determinants of ICT Adoption and Utilisation
In order to examine the relative importance of different factors influencing the extent of
ICT adoption and utilisation, we estimated ordered logistic regression equations, taking
the three measures presented in Table 3 as dependent variables. The analysis was
confined to private sector organisations responding to the 1999 survey, which gathered
more comprehensive information on participating firms than did the 1997-98 surveys.
Independent variables in these equations comprised a vector of company-level
characteristics including information on employee size-group, sector and regional
location, as well as a range of other factors which might conceivably operate as either
positive or negative influences on the level of ICT utilisation:
14
Skill shortages: Ordered three-point measure (on a 0-2 scale) of whether or not
enterprises reported deficiencies in employees’ ICT skills (Section V refers).
Training: based on responses to a question ‘Does your company provide
training for its employees in the use of ICTs, such as those described in this
survey? If YES, how often?’, with a 4-point scale ranging from ‘rarely’ to
‘frequently’. Given the well-established positive association between training
provision and formal qualifications (Dickerson and Wilson, 2003), this training
variable serves in part as a proxy indicator of human capital endowments in the
model.
Single-site operations: whether enterprises were based on one site only or on
more than one site
Multinational operations: whether enterprises had ‘more than one site
internationally’
Competitiveness: a measure of the perceived importance of ICTs to each
company’s present and future competitiveness (comprising the mean value of
responses on a 1-5 scale to two separate questions about the links between ICTs
and competitiveness).
Use of external information sources: a measure of companies’ ‘openness’ to
external sources of information and advice on ICTs (eg, from technology
suppliers, external consultants or government business support organisations).
A number of measures indicating the sensitivity of the enterprise to a variety of
environmental factors, calculated in the same way as the ‘net skills’ indicator discussed
in Section V:
15
Firm culture: A measure of the extent to which the ‘cultural willingness of …
senior management to accept and use ICTs’ and the ‘cultural willingness to
accept and use ICTs within [each] business as a whole’ had served as a positive
influence on ICT uptake rather than as a barrier.
Costs of ICT adoption: A measure of the extent to which ‘pricing of ICTs’ had
served as a positive influence on ICT uptake rather than as a barrier.
Infrastructure: A measure of the extent to which ‘access to ICT infrastructure’
had served as a positive influence on ICT uptake rather than as a barrier.
Regulation: A measure of the extent to which the ‘regulatory framework’ had
served as a positive influence on ICT uptake rather than as a barrier.
The main results of our ordered logistic regression analyses are shown in Table 4. The
analyses of ICT adoption are based on all private sector survey respondents, whereas
those for the intensity of ICT usage relate to ICT users only.
As expected, both the extent of ICT adoption and the intensity of use of connectivity
technologies are positively and significantly related to enterprise size-group (Table 4,
columns 1-2 and 5-6). Equally unsurprisingly, the intensity of computer use – measured
as the proportion of the workforce using computers – tends to be higher in small and
medium-sized enterprises than in larger organisations (column 3-4). Other control
variables to perform much as expected are those for single-site enterprises –
significantly less likely to make intensive use of connectivity technologies – and the
multinational indicator which is significantly positively related to both ICT adoption
and intensity of use.
16
With these controls in place, the measure of internal ICT skill shortages is found to
have a strong negative and significant association with all three ICT measures (p<0.01
for intensity of computer use and of connectivity; p=0.05 for the extent of ICT
adoption; see columns 1, 3 and 5 of Table 4). When variables representing a number of
other prospective influences on ICT usage are introduced, the skills deficiency measure
is no longer a significant determinant of the extent of ICT adoption (Table 4, column 2)
but remains negative and statistically significant in the connectivity and computer use
equations (columns 4 and 6).
ICT training is significantly positively related to all three ICT measures and the
coefficients remain positive and very well-defined when the additional control variables
are added to the model. Of course, in this case there is very likely to be two-way
causation, with some firms increasing their ICT training provision in response to needs
which become apparent after ICT equipment has been purchased.10 The regression
results also highlight other influences on the level and intensity of ICT usage. For
example, the measures of the perceived importance of ICTs to firm competitiveness are
strongly positive and significant throughout. Other positive influences include ‘access
to ICT infrastructure’ (in all three models) and openness to external sources of
information on ICTs in the models for ICT adoption and connectivity.
Nearly all these findings provide support for Hypothesis 1, namely that, all else being
equal, reported ICT skill shortages at company level are negatively related to measures
of ICT adoption and the intensity of use of ICTs (once installed). In one specification
relating to ICT adoption (Table 4, column 2) the coefficient on the skills shortage
indicator is no longer statistically significant although it remains negatively signed.
17
However, in similar specifications the two measures of ICT utilisation are significantly
negatively related to skill shortages. We now go on to test the support for Hypotheses
2-4, making use of the matched financial dataset described in Section III.
* Table 4 about here *
VII. The Impact of ICT Skill Shortages on Firm-Level Performance
A. Estimation Methodology
We base our analysis on an augmented Cobb-Douglas production function containing
annual sales (Q), assets (K), employment (L) and a variety of additional variables (Z),
including sector and region identifiers, indicators of ICT-related skills or training and
indicators of ICT utilisation. With a single type of labour this production function takes
the form:
(1) Q=AK L Z .
Taking logs we obtain:
(2) Ln(Q) = ln(A) + ln(K) + ln(L) + ln(Z).
Equation (2) represents the basis for our analysis of the level of sales. The coefficients
on those parts of Z which are indicators of ICT-related skills and training will enable us
to examine whether these have an impact on sales for given capital and labour inputs.
18
However, we first need to discuss various practical issues that arise in the estimation of
equation (2), in particular, the likely endogeneity of ICT investments with respect to
company performance. On the one hand, it is conceivable that ICT investments may be
prompted by poor performance, with such technologies being seen as a means of
making production or service delivery more efficient and thereby as a means of helping
the company to become more competitive. Alternatively, investments in ICT may be
facilitated by the profits generated by previous above-average performance. Either
scenario raises the possibility that the decision to invest in ICT may be at least partially
determined by company performance. If this is indeed the case in practice, estimation
via Ordinary Least Squares (OLS) will lead us to biased estimates of the impact of ICT
investments on company performance.
A standard means of dealing with endogeneity issues is through the use of instrumental
variables (IV) regression. Brynjolfsson and Hitt (1996) used once-lagged values of ICT
investments as instruments for current investments in an analysis of the returns to
information systems spending among US organizations. They found that the estimated
returns to ICT investments were twice as large under the IV specification, but the
Hausman specification test (Hausman, 1978) did not support the rejection of their more
efficient OLS estimates. In a more recent analysis of the impact of ICT investments on
productivity among German establishments, Zwick (2003) used two external
instruments – a variable identifying establishments that expected an increase in their
demand for qualifications and training, and one identifying establishments that
expected an increase in the importance of formal external training. The coefficient on
19
Zwick’s binary indicator of ICT investments increased ten-fold under his IV
specification.
In this paper we address the potential endogeneity of ICT adoption and utilisation
through the use of external instruments. The precise sets of instruments vary according
to the particular indicator of ICT usage that is being used, but they broadly comprise
indicators of: the anticipated impact of ICT on future competitiveness; company
attitudes to the use of ICT; and the importance of particular environmental factors in
facilitating or inhibiting the company’s uptake of ICT. Each of these variables is found
to be highly correlated with the relevant potentially endogenous regressors (measures of
ICT adoption or utilisation) while at the same time being orthogonal to error processes
in the firm-level performance regression equations.
Other practical issues concern the specific nature of our data. Values for total sales in
1999, 2000 and 2001 were adjusted to constant prices using output price deflators for
the United Kingdom that were assembled at a broad sectoral level from the National
Accounts (OECD, 2003). Total sales were also adjusted to remove distortions caused
by changes in a company’s accounting period. 11 Small numbers of cases in the
matched IBS-financial dataset were excluded because of outlying values on the
financial variables.
In what follows, we first present cross-sectional estimates for IBS 1999 and then report
on panel estimates over the period 1999-2001. In principle, estimating equation (2) via
panel regression methods should remove estimation biases caused by unobserved time-
invariant characteristics such as management quality. However, the full application of
20
such an approach relies on the availability of repeat observations for all of the variables
of interest. The IBS 1999 survey offers only a single year of data on ICT investments
and many other characteristics of interest. Nonetheless, we have been able to match
longitudinal data on sales (Q), assets (K) and employment (L) to survey respondents.
We are therefore able to employ a two-stage methodology, previously adopted by Black
and Lynch (2001) among others. Under this methodology, a fixed-effects estimator is
first used to provide unbiased estimates of α and β. The average residual (the firm fixed
effect) is then regressed on Z in order to examine whether the components of Z are
related to having above or below-average levels of performance over a specified
period.12
B. Results
Table 5 shows the results of cross-sectional regressions of the determinants of log sales
in IBS 1999, highlighting the key regressors relevant to our assessment of the impact of
ICT skill shortages on company performance. In each equation, the coefficients on
capital and labour are positive, well-defined and broadly within the range of expected
values.
Table 5, Part A shows the results when our measure of ICT adoption is taken as a
determining variable, with three different IV specifications presented alongside their
OLS counterparts. Parts B and C of this table repeat the exercise for, respectively, our
measures of intensity of computer use and intensity of use of connectivity technologies.
In Columns (3) and (4) indicators of ICT skill shortages and ICT training provision are
added to the base specification in each case. In Columns (5) and (6) an interacted model
21
is estimated in order to explore what effects, if any, the presence of ICT skill shortages
and ICT training provision have on the relationship between ICT adoption and
utilisation and company sales performance.
Three variables are used as external instruments in the IV regressions: a variable
indicating the likely importance of ICT investments to company competitiveness over
the next two years; a variable indicating whether access to ICT infrastructure was a
positive or negative influence on the uptake of ICT within the company; and a general
indicator of company attitudes to ICT.13 As it turns out, in each IV regression shown in
Table 5, the Sargan test statistics support the validity of the chosen instruments but
Hausman tests for the endogeneity of ICT adoption and utilisation suggest that the more
efficient OLS estimates should be preferred to the more consistent IV estimates.
In the case of the ‘extent of ICT adoption’ and the ‘intensity of computer use’, the
relevant coefficients are positive and significant throughout. The same applies to two of
the three (preferred) OLS equations for the ‘intensity of use of connectivity
technologies’. Taking the first of these three equations in each case (column 2 of Table
5), we can estimate that the marginal effect of increasing the extent of ICT adoption
score by one standard deviation from its mean value (i.e. from 4.13 to 6.58) would be to
raise the value of sales by 27 per cent (95 per cent confidence interval: 15 per cent - 39
per cent), all else being equal. The marginal effect of increasing the intensity of
computer use by one standard deviation from its mean value (i.e. from 2.12 to 3.52)
would be to increase sales by 21 per cent (95 per cent confidence interval: 12 per cent -
31 per cent). And the effect of increasing the intensity of use of connectivity
technologies by one standard deviation from its mean value (i.e. from 2.45 to 4.28)
22
would be to increase sales by 20 per cent (95 per cent confidence interval: 10 per cent -
31 per cent). By contrast, ICT skill shortages and ICT training provision are not found
to have any statistically significant effects on sales performance after controlling for
levels of ICT adoption or utilisation (Columns 3-4). Similarly, in the interacted models
the relevant coefficients lack significance in all three preferred OLS specifications
(Column 6).
Before considering the implications of these findings for our main hypotheses, we first
examine the principal coefficients of interest from our panel analyses for IBS 1999. As
for our cross-sectional analyses, three separate sets of equations are presented, with
each of our measures of ICT adoption and utilisation taken in turn as a determining
variable. Column (1) of Table 6 presents the coefficients on K and L from the fixed-
effects panel regression that forms the first stage of our two-step procedure for all three
sets of equations. Columns (2), (4) and (6) present the coefficients on our indicators of
ICT adoption and utilisation, skills and training from the IV regression that forms the
second stage in each case. The contrasting OLS estimates for similar second-stage
specifications are shown in Columns (3), (5) and (7).
Column (1) shows that we obtain reasonable coefficients on assets (K) and employment
(L) in our first-stage regression. The coefficient on K is lower than one might anticipate
– and lower than that found in the cross-sectional regressions reported in Table 5 – but
this is a common outcome of the application of panel methods (Griliches and Mairesse,
1997). A favoured solution in the literature is to apply general method of moments
(GMM) estimators (see, for example: Arellano and Bond, 1991; Blundell and Bond,
1998). However, such approaches rely on the availability of lagged values of both the
23
levels and changes over time of sales, assets and employment for use as instruments. In
our dataset, we have financial data for only three time periods. This proves insufficient
to yield sensible results and so we do not discuss this option further.
Columns (2) and (3) show the impact of each measure of ICT adoption and utilisation
when they are entered in turn into the second-stage IV and OLS regressions alongside
our control variables. In Columns (4)-(7) ICT skill shortage and training variables are
entered in the same way as described for the cross-section analysis above. Since the
White test statistics indicate the presence of heteroscedasticity throughout the panel
regressions, we follow Baum, Shaffer and Stillman (2003) and rely on (1) Hansen J
statistics to test the validity of our chosen instruments and (2) C statistics which test the
null hypothesis that the potentially endogenous regressors (here measures of ICT
adoption and utilisation) are in fact exogenous.14 While instrument validity is supported
for all equations shown, the C statistics indicate that the OLS specifications should be
preferred to their IV counterparts in all cases except for the first model of intensity of
computer use (Column 2) and each of the interacted models (Column 6).
In all preferred models each of the three measures of ICT utilisation have positive and
statistically significant coefficients. In the models for extent of ICT adoption (Table 6,
Part A) and intensity of use of connectivity technologies (Part C), the coefficients and
associated estimates of marginal effects on sales performance are similar to those
identified in the cross-sectional estimates above. However, in the preferred IV
specification relating to intensity of computer use (Part B), the relevant coefficient is
over twice as high as that estimated in the cross-sectional analysis.15 Taking the panel
and cross-sectional estimates together, therefore, we find clear support for Hypothesis 2
24
that company-level performance is positively related to measures of ICT adoption and
utilisation. Since we have already established support in Section VI for Hypothesis 1 –
that skill shortages restrict both the extent of ICT adoption and the intensity of use of
ICTs – we conclude that ICT skill shortages have a clear indirect negative impact upon
company performance.
However, in order to find out whether our evidence supports Hypothesis 3 – that ICT
skill shortages have a direct negative effect on firm-level performance – we need to
consider whether such shortages have some additional impact on company performance
over and above any indirect impact that occurs through the restriction of ICT
utilisation. In other words, we need to examine whether ICT skill constraints are
significantly negatively related to company performance even after controlling for the
extent of ICT adoption and the intensity of ICT use. Columns (4)-(5) in Table 6 show
the effect of adding an indicator of whether the company has an ICT skills shortage to
the second-stage regression. In both the IV and OLS regressions, the relevant
coefficients are small and poorly defined in respect of the three measures of ICT
adoption and utilisation and thus provide no support for Hypothesis 3.
To what extent are the potential effects of ICT skill shortages on ICT adoption and
utilisation alleviated by employers’ provision of training for their employees?
Regression analysis of the determinants of training provision, with similar controls to
those in Table 4, finds that the ICT skills shortage measure is significantly negatively
related to training provision (p<0.01). The relative absence of training by firms
reporting skill shortages highlights the potential for those shortages to persist over some
length of time in the firms concerned.
25
The equations reported in Table 6, Columns (4)-(5) also contain indicators of ICT
training provision that enable us to test Hypothesis 4 that, at given levels of ICT
adoption and intensity of use, investments in ICT skills training contribute positively to
firm performance. ICT skills training is expected to be an important complementary
investment to spending on ICTs themselves, as it should help employees to make more
productive and efficient use of investments in ICT hardware and software. However,
the coefficients on this indicator in Columns (4)-(5) turn out to be not significantly
different from zero. And replacing this indicator of the frequency of training with a
dichotomous variable which indicates whether any training is provided yields the same
outcome.
In order to subject Hypotheses 3 and 4 to further scrutiny we also consider the results of
specifications where indicators of ICT skill shortages and ICT training provision are
interacted with each of our measures of ICT adoption and utilisation (Table 6, Columns
6-7). In the case of the adoption measure (Table 6, Part A), this enhanced model still
provides no support for either Hypothesis 3 or Hypothesis 4. However, in the case of
the intensity of computer use measure (Part B), the coefficient on the computer use /
skills shortage interacted variable is negatively-signed and statistically significant in the
IV regression, suggesting that the performance of companies making more intensive
use of computers is negatively affected by ICT skill shortages. (The relevant C statistic
in this case suggests that the IV estimates should be preferred to the OLS variants).
And in the case of the intensity of use of connectivity technologies measure (Table 6,
Part C), the results shed some light on the relationship between ICT skills training and
26
firm-level performance. Firstly, the coefficient on the training provision variable is
positive and significant in the IV regression; secondly, the coefficient on the interacted
connectivity / training variable is negatively-signed and significant. Taken together,
these findings do provide support for Hypothesis 4 but the coefficient on the product
term suggests that the advantage derived from training is greatest at low levels of
connectivity and tails off as connectivity rises.
To recapitulate, we have found strong evidence that internal ICT skill shortages – skill
gaps among existing employees – have negative indirect effects on firm-level
performance because of the ways in which such skill deficiencies restrict companies
both in terms of ICT adoption and the intensity of use of ICTs once they have been
installed. However, the evidence for direct effects of skill constraints – and of ICT-
related training – on financial performance at given levels of ICT adoption and
utilisation is more limited. This may reflect the relatively short run of post-survey
financial data available to us. Basu et al. (2003) suggest that UK firms invested heavily
in inputs complementary to ICTs during the second half of the 1990s and that this
diversion of resources helps to explain the slowdown in total factor productivity growth
in the UK during this period. It is possible that these complementary investments will
have positive lagged effects on future performance and that, in this context, future
analysis may reveal more evidence of a systematic relationship between skills and
financial performance at given levels of ICT adoption and utilisation.
27
VIII. Summary and Assessment
It is now widely recognised that the technological revolution associated with ICTs is
skill-intensive, not just because of the skill requirements associated with ICT adoption
but also due to the lengthy periods of experimentation and learning that many firms
need to go through in order to make effective use of ICTs. Indeed, the process of
investing in assets which are complementary to ICTs (such as software, business
reorganisation and new product development) may contribute more to raising the
relative demand for skilled labour than the diffusion of ICTs themselves. In this context
it is widely assumed that shortages of ICT skills will have negative effects on firms’
ability to make early and effective use of ICTs.
In this paper we make use of benchmarking survey data for UK enterprises in 1999,
combined with post-survey financial data for the same enterprises, in order to assess the
impact of ICT skill shortages on financial performance at firm level. Overall, the results
provide very clear evidence for the UK about the positive impacts on firm-level
performance of the rapid adoption and deployment of ICTs during the late 1990s, and
the negative effects on performance experienced by those companies in which ICT skill
shortages inhibited the adoption or intensive utilisation of ICTs.
We define three different ICT measures at firm level, summarising the extent of ICT
adoption, the intensity of computer use and the intensity of usage of three key
connectivity technologies (networks, Internet and company web-sites). After
controlling for employee size-group, sector and regional location and other pertinent
28
enterprise characteristics, a measure of internal ICT skill shortages is found to have a
negative and significant impact on both measures of ICT utilisation and (in most
specifications) on the extent of ICT adoption as well. At the same time, cross-sectional
and panel estimates of the determinants of company sales performance suggest that all
three measures of ICT utilisation are positively and significantly related to sales
performance after controlling for capital assets, labour inputs, sector, region and a
number of other company characteristics. These findings are robust to the use of
instrumental variable techniques to allow for the likely endogeneity of ICT investments
with respect to company performance.
Taken together these two sets of results clearly show that ICT skill shortages have an
indirect negative impact upon company performance. However, there is only limited
evidence in this dataset to support a hypothesis that ICT skill shortages have a direct
negative effect on firm-level performance, that is, some additional impact on company
performance over and above the indirect impact that occurs through the restriction of
ICT adoption and utilisation.
These findings in respect of the direct links between ICT skills and performance at firm
level may reflect the relatively short run of post-survey financial data available to us. If
UK firms have been investing heavily in inputs complementary to ICTs during the
second half of the 1990s, then these complementary investments may have positive
lagged effects on performance. In this context, future analysis of longer periods of post-
sample financial data may reveal more evidence of a systematic relationship between
skills and financial performance at given levels of ICT adoption and utilisation.
29
REFERENCES
Acemoglu, Daron, “Technical change, inequality and the labor market,” Journal of
Economic Literature, 40 (2002), 7-72.
Arellano, Manuel, and Stephen Bond, “Some tests of specification for panel data:
Monte Carlo evidence and an application to employment equations”, Review
of Economic Studies, 58:2 (1991), 277-297.
Arrow, Kenneth and William Capron, “Dynamic shortages and price rises: the
engineer-scientist case”, Quarterly Journal of Economics, 73 (1959), 292-308.
Bartel, Ann and Frank Lichtenberg, “The comparative advantage of educated workers
in implementing new technology”, Review of Economics and Statistics, 69:1
(1987), 1-11.
Basu, Susanto, John Fernald, Nicholas Oulton and Sylaja Srinivasan, “The case of the
missing productivity growth: or, does information technology explain why
productivity accelerated in the United States but not in the United Kingdom?”,
in NBER Macroeconomics Annual 2003, Mark Gertler and Kenneth Rogoff,
eds. (Cambridge, MA: The MIT Press, 2004).
Baum, Christopher, Mark Shaffer and Steven Stillman, “Instrumental variables and
GMM: Estimation and testing”, Boston College Department of Economics
Working Paper No. 545, 2003.
Berman, Eli, John Bound and Stephen Machin, “Implications of skill-biased
technological change: international evidence”, Quarterly Journal of
Economics, 113:4 (1998), 1245-1279.
Black, Sandra and Lisa Lynch, “How to compete: the impact of workplace practices
and information technology on productivity”, Review of Economics and
Statistics, 83:3 (2001), 434-445.
Blundell Richard and Stephen Bond “GMM estimation with persistent panel data: an
application to production functions”, Institute for Fiscal Studies Working
Paper No. W99/4, 1998.
Bresnahan, Timothy, Erik Brynjolfsson and Lorin Hitt, “Information technology,
workplace organisation and the demand for skilled labour: firm-level
evidence”, Quarterly Journal of Economics, 117 (2002), 339-376.
Brynjolfsson, Erik and Lorin Hitt “Paradox lost? Firm-level evidence on the returns to
information systems spending”, Management Science, 42:4 (1996), 541-558.
30
Brynjolfsson Erik, Lorin Hitt and Shinkyu Yang, “Intangible assets: computers and
organisational capital”, Brookings Papers on Economic Activity, I (2002),
137-199.
Caselli Francesco “Technological revolutions”, American Economic Review, 89:1
(1999), 80-102.
Chun, Hyunbae “Information technology and the demand for educated workers:
disentangling the impacts of adoption versus use”, Review of Economics and
Statistics, 85:1 (2003), 1-8.
Dickerson, Andrew and Robert Wilson “The determinants of the incidence and
intensity of off the job training”, in Employers Skill Survey: New Analyses and
Lessons Learned, Geoff Mason and Robert Wilson, eds. (Nottingham:
Department for Education and Skills, 2003).
Doeringer, Peter and Michael Piore, Internal Labor Markets and Manpower Analysis,
(Lexington, MA.: Heath and Company, 1971).
Doms, Mark, Timothy Dunne and Kenneth Troske, “Workers, Wages and
Technology”, Quarterly Journal of Economics, 112:1 (1997), 253-290.
European Commission, E-Business and ICT Skills in Europe: Benchmarking Member
State Policy Initiatives, (Brussels: European Commission ICT Skills
Monitoring Group, 2002).
Forth John and Geoff Mason, “Persistence of skill deficiencies across sectors, 1999-
2001”, in Employers Skill Survey: New Analyses and Lessons Learned, Geoff
Mason and Robert Wilson, eds. (Nottingham: Department for Education and
Skills, 2003).
Green, Francis and David Ashton, “Skill shortage and skill deficiency – a critique”,
Work, Employment & Society, 6:2 (1992), 287-301.
Griliches Zvi and Jacques Mairesse, “Production functions: the search for
identification”, in Econometrics and Economic Theory in the Twentieth
Century, Steinar Strom, ed. (Cambridge: Cambridge University Press, 1997).
Haskel Jonathan and Christopher Martin, “The causes of skill shortages in Britain”,
Oxford Economic Papers, 45:4 (1993a), 573-88.
Haskel Jonathan and Christopher Martin, “Do skill shortages reduce productivity?
Theory and evidence from the United Kingdom”, Economic Journal, 103
(1993b), 386-394.
31
Hausman Jerry, “Specification tests in econometrics”, Econometrica, 46:3 (1978),
1251-1271.
Helpman, Elhanan and Manuel Trajtenberg, “A time to sow and a time to reap: growth
based on general purpose technologies”, in General Purpose Technologies
and Economic Growth, Elhanan Helpman, ed. (Cambridge, MA: MIT Press,
1998).
IDC, Europe’s Growing IT Skills Crisis, (International Data Corporation: London,
1999).
ITAA, Bridging the Gap: Information Technology Skills for a New Millennium,
(Arlington, VA: Information Technology Association of America, 2000).
ITAA, Bouncing Back: Jobs, Skills and the Continuing Demand for IT Workers,
(Arlington, VA: Information Technology Association of America, 2002).
Lerman, Robert, “Emerging trends in the Information Technology job market: How
should the public and private sectors respond?”, Testimony before the
Subcommittee on Education and the Workforce, US House of
Representatives, 23rd April 1998.
Mace, John, “Internal labour markets for engineers in British industry”, British Journal
of Industrial Relations, 17:1 (1979), 50-63.
Mace, John and Geoffrey Wilkinson, “Are labour markets competitive? A case study of
engineers”, British Journal of Industrial Relations, 15:1 (1977), 1-17.
Mann, Ranjiv and Sudhir Junankar, “40 years on: how do companies respond to the
CBI’s Industrial Trends Survey?”, Confederation of British Industry
Economic Situation Report, 1998.
Mason, Geoff, “The Labour Market for Engineering, Science and IT Graduates: Are
There Mismatches Between Supply And Demand?”, Department for
Education and Employment Research Report No. 112, 1999.
Mason, Geoff, Key Issues in IT Skills Research in the UK, Report to the Department for
Education and Employment. (London: National Institute of Economic and
Social Research, 2000).
Nelson, Richard and Edmund Phelps, “Investment in humans, technological diffusion,
and economic growth”, American Economic Review, 56:2 (1966), 69-75.
Nickell Stephen and Daphne Nicolitsas, “Human capital investment and innovation:
what are the connections?”, in Productivity, Innovation and Economic
32
Performance, Ray Barrell, Geoff Mason and Mary O’Mahony, eds.
(Cambridge: Cambridge University Press, 2000).
OECD, National Accounts of OECD Countries: Main Aggregates Volume 1 1990-2001.
(Paris: OECD, 2003).
Schultz, Theodore, “The value of the ability to deal with disequilibria”, Journal of
Economic Literature, 31 (1975), 199-225.
Siebert, W. Stanley and John Addison, “Internal labour markets: causes and
consequences”, Oxford Review of Economic Policy, 7:1 (1991), 76-92.
Spectrum, Moving into the Information Society 1999: International Benchmarking
Study, (London: Department of Trade and Industry, Information Society
Initiative, 1999).
Taubman, Paul and Michael Wachter, “Segmented labor markets”, in Handbook of
Labor Economics, Orley Ashenfelter and Richard Layard, eds. (Amsterdam:
North-Holland, 1986).
Wachter, Michael and Randall Wright, “The economics of internal labor markets”,
Industrial Relations, 29:2 (1990), 240-262.
Welch, Finis “Education in production”, Journal of Political Economy, LXXVIII
(1970), 35-59.
UEF, IT Workforce Data Project, Reports I-IV, (New York, NY: United Engineering
Foundation, 1999).
Zwick, Thomas “The impact of ICT investment on establishment productivity”,
National Institute Economic Review, 184 (2003), 99-110.
33
NOTES
1. This paper is based on research which was kindly supported by the UK Department of Trade and
Industry, Department for Education and Skills and HM Treasury; none of these departments are responsible for any views expressed in the paper. We would like to thank Ian Brades (Dun & Bradstreet), Graham Craigmile (HI Europe, formerly Romtec) and Chris Kay (NOP) for their help in matching Dun & Bradstreet data to the IBS datasets. We are also grateful to Kate Robinson, Michela Vecchi and other colleagues at NIESR for comments and advice, and to Ana Rincon-Aznar and Lucy Stokes for excellent research assistance. Responsibility for any errors is ours alone.
2. Arrow and Capron define the term ‘shortage’ as ‘a situation in which there are unfilled vacancies in positions where salaries are the same as those currently being paid in others of the same type and quality’ (1959, p.301).
3. The main reasons advanced why it is efficient for some firms, particularly larger ones, to operate ILM’s centre on the benefits to employee motivation, the cost savings from lower labour turnover and firms’ efforts to maximise returns from job-specific and company-specific training (Taubman and Wachter, 1986; Wachter and Wright, 1990; Siebert and Addison, 1991).
4. In 2001 some 16 per cent of English establishments with 5 or more employees reported internal skill gaps, defined as employees lacking ‘full proficiency’ in their current jobs. The equivalent proportion reporting skill-related hard-to-fill vacancies was 6 per cent (Forth and Mason, 2003).
5. The CBI data on skill shortages are based on employers’ responses to a question about whether constraints on ‘skilled labour’ in general are likely to limit their output in the following four-month period. A follow-up study of participants in this survey suggested that about 60 per cent of respondents interpreted the question as referring primarily to external recruitment difficulties while 45 per cent thought it referred instead (or as well) to deficiencies in the skills possessed by their existing workforce (Mann and Junankar, 1998).
6. In a recent study of the relationship between information technology and the demand for educated workers at industry level in the US, Chun (2003) distinguishes carefully between the adoption and use effects of information technology and finds that both have contributed substantially to the increased relative demand for college graduates.
7. For details of sampling procedures and response rates, see Spectrum (1999). 8. We experimented with different versions of this ICT connectivity measure which made use of other
survey responses on the ways in which email, EDI and Extranets were used. However, these other data were not available for all three years and the alternative measures were in any event all highly correlated with the connectivity measure described in Table 3, Part C. Hence we elected to proceed with this as our preferred measure of the intensity of use of connectivity technologies.
9. No further information was gathered about the types or levels of skills and knowledge in shortage or the occupations most affected.
10. Due to a lack of suitable instruments we do not investigate this issue further. Rerunning the equations in Table 4 without the potentially endogenous training variable does not materially alter the other findings presented here, for example, the findings relating to skill shortages.
11 Such changes in accounting periods mean that a small number of companies recorded total sales over a period of less than, or more than, 12 months; sales figures were adjusted pro rata to a 12-month basis as appropriate.
12. This method is also employed by Zwick (2003). The Hausman specification test (Hausman, 1978) can be used to determine whether the fixed effects estimator is more appropriate than a random effects model. Application of the Hausman test justifies use of a fixed effects estimator for the IBS 1999 data. Nonetheless, the use of a random effects estimator in place of the two-stage methodology does not substantively change the results of the analysis.
13. The latter of these three variables was only used as an instrument for ‘the intensity of connectivity technologies’, as it was not found to be significantly associated with ‘the extent of ICT adoption’ or ‘the intensity of computer use’.
14. Note that we account for heteroscedasticity in the estimation itself by using the robust (Huber/ White/ sandwich) variance estimator.
15. The panel results suggest that the marginal effect of increasing the intensity of computer use by one standard deviation from its mean value (i.e. from 2.12 to 3.52) would be to increase sales by as much as 47 per cent. However, the standard error attached to this estimate is relatively large (95 per cent confidence interval: 15 per cent - 78 per cent).
34
TABLE 1
Comparison of Original IBS Sample, 1999, and Dataset Matched to Post-Survey Dun & Bradstreet Data, Analysed by Employee Size-Group and Sector
(a) Scored as 1= use and 0=non-use for each of the following: computers/PCs; networks; remote access; email; EDI; Internet; company website; Intranet and/or Extranet. B: Intensity of Use of Computers, 1998-99
1998 1999 Proportions of employees making use of computers/PCs:
Column percentages
No computer use 15 12 Used by up to a quarter of employees 32 39 Used by 26-50 per cent of employees 12 14 Used by 51-75 per cent of employees 13 11 Used by more than 75 per cent of employees 29 25
TOTAL 100 100 Continued on next page…
37
TABLE 3 (continued)
C: Intensity of Use of Connectivity Technologies, 1997-99
1997 1998 1999 1997 1998 1999 Intensity of use of connectivity technologies on 0-6 scale(a):
TOTAL 100 100 100 Notes: (a) This index is defined as the sum of network, Internet and website use scores which are allocated as follows: Network: 0=No network; 1=Local network only; 2=Networked between national sites or wider area [1997: 1= Networked within site or between local sites (within 20 miles); 2= Networked between national sites or international sites;1998-99: 1= Networked within a site; 2= Networked between sites or with suppliers or with business customers] Internet/WWW use: 0=No use of Internet/WWW; 1= Internet/WWW used mainly for email, advertising, marketing and/or information purposes; 2=Internet/WWW used for purchasing and/or making sales on-line Website: 0=No website; 1= Website used mainly for advertising, marketing and/or information distribution; 2=Website used for making sales on-line Base: Private sector organisations. Source: International Benchmarking Surveys, 1997-1999
38
TABLE 4
Determinants of ICT Adoption and Utilisation (Ordered Logistic Regressions) – Private Sector Organisations
Extent of ICT
adoption Intensity of
computer use Intensity of use of
connectivity technologies
(1) (2) (3) (4) (5) (6) Number of employees: (Ref. 1-9)
Notes: 1. Robust standard errors in parentheses; * significant at 10 per cent; ** significant at 5 per cent; ***
significant at 1 per cent 2. Equations 1, 3 and 5: Estimated via instrumental variables, where the endogenous variables are the
respective measures of ICT adoption and utilisation. Summary statistics from the initial equations are specified, along with the additional instruments used, below:
A Extent of ICT adoption: Instruments: Anticipated impact of ICT on competitiveness; Net influence of
infrastructure Summary statistics: Adjusted R2 – 0.17; Partial R2 of excluded instruments – 0.11; F
test of excluded instruments – 17.09***. B Intensity of computer use:
Instruments: Anticipated impact of ICT on competitiveness; Net influence of infrastructure
Summary statistics: Adjusted R2 – 0.14; Partial R2 of excluded instruments – 0.09; F test of excluded instruments – 14.09***.
C Intensity of connectivity technologies: Instruments: Anticipated impact of ICT on competitiveness; Net influence of
infrastructure; Company attitude to ICT Summary statistics: Adjusted R2 – 0.11; Partial R2 of excluded instruments – 0.08; F
test of excluded instruments – 7.96***. 4. Equations 2, 4 and 6: OLS estimates. Regressions also include six industry dummies, ten region dummies and dummies for: single-site enterprise-site companies, multi-national companies; and use of external information sources on new technologies. Full results are available on request from the authors.
42
TABLE 6
Determinants of (Log) Firm Sales, 1999-2001 – Panel Regressions
(0.203) (0.048) Computer use * ICT training provision -0.030 -0.013
(0.077) (0.024) Adjusted R2 0.10 0.10 0.10 Hansen J test of over-identification 0.16 0.03 2.02
P-value 0.69 0.87 0.57 White test of heteroscedasticity 340.55 460.28 694.86
P-value 0.00 0.00 0.00 Test of endogeneity of computer use (C-statistic)
3.09 1.84 8.78
P-value 0.08 0.18 0.03 Observations 902 902 896 896 896 896 No. of companies 312 312 310 310 310 310 C: The impact of use of connectivity technologies, ICT skill shortages and ICT training provision See
Table 6 (continued) Notes: 1. Robust standard errors in parentheses; * significant at 10 per cent; ** significant at 5 per cent; ***
significant at 1 per cent 2. Equation 1 (first stage): Estimated using panel (fixed effects) regression. Regression also includes
two year dummies plus six industry dummies interacted with these year dummies. 3. Equations 2, 4 and 6: Second stage regressions estimated via instrumental variables, where the
endogenous variables are the respective measures of ICT adoption and utilisation. Summary statistics from the initial equations are specified, along with the additional instruments used, below:
A Extent of ICT adoption: Instruments: Anticipated impact of ICT on competitiveness; Net influence of
infrastructure Summary statistics: Adjusted R2 – 0.21; Partial R2 of excluded instruments – 0.12; F
test of excluded instruments – 56.30**. B Intensity of computer use:
Instruments: Anticipated impact of ICT on competitiveness; Net influence of infrastructure
Summary statistics: Adjusted R2 – 0.18; Partial R2 of excluded instruments – 0.08; F test of excluded instruments – 40.26***.
C Intensity of connectivity technologies: Instruments: Anticipated impact of ICT on competitiveness; Net influence of
infrastructure; Company attitude to ICT Summary statistics: Adjusted R2 – 0.11; Partial R2 of excluded instruments – 0.08; F
test of excluded instruments – 27.66***. 4. Equations 3, 5 and 7: Second stage OLS estimates. 5. Regressions also include six industry dummies, ten region dummies and dummies for: single-site
enterprise-site companies, multi-national companies; and use of external information sources on new technologies. Full results and tables of descriptive statistics are available on request from the authors.