DISCUSSION PAPER SERIES

IZA DP No. 11298

The Effect of Computer Use on Job Quality:

Evidence from Europe

Seetha Menon
Andrea Salvatori
Wouter Zwysen

JANUARY 2018
DISCUSSION PAPER SERIES

IZA DP No. 11298

The Effect of Computer Use on Job Quality:

Evidence from Europe

Seetha Menon
EUI
Andrea Salvatori
OECD, IZA and University of Essex
Wouter Zwysen
ISER, University of Essex

JANUARY 2018

Any opinions expressed in this paper are those of the author(s) and not those of IZA. Research published in this series may
include views on policy, but IZA takes no institutional policy positions. The IZA research network is committed to the IZA
Guiding Principles of Research Integrity.
The IZA Institute of Labor Economics is an independent economic research institute that conducts research in labor economics
and offers evidence-based policy advice on labor market issues. Supported by the Deutsche Post Foundation, IZA runs the
world’s largest network of economists, whose research aims to provide answers to the global labor market challenges of our
time. Our key objective is to build bridges between academic research, policymakers and society.
IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper
should account for its provisional character. A revised version may be available directly from the author.

IZA – Institute of Labor Economics

Schaumburg-Lippe-Straße 5–9 Phone: +49-228-3894-0
53113 Bonn, Germany Email: [email protected] www.iza.org
IZA DP No. 11298 JANUARY 2018

ABSTRACT
The Effect of Computer Use on Job Quality:
Evidence from Europe1
This paper studies changes in computer use and job quality in the EU-15 between 1995 and
2015. We document that while the proportion of workers using computers has increased
from 40% to more than 60% over twenty years, there remain significant differences
between countries even within the same occupations. Several countries have seen a
significant increase in computer use even in low-skilled occupations generally assumed
to be less affected by technology. Overall, the great increase in computer use between
1995 and 2015 has coincided with a period of modest deterioration of job quality in the
EU-15 as whole, as discretion declined for most occupational and educational groups while
intensity increased slightly for most of them. Our OLS results that exploit variation within
country-occupation cells point to a sizeable positive effect of computer use on discretion,
but to small or no effect on intensity at work. Our instrumental variable estimates point to
an even more benign effect of computer use on job quality. Hence, the results suggest that
the (moderate) deterioration in the quality of work observed in the EU-15 between 1995
and 2015 has occurred despite the spread of computers, rather than because of them.

JEL Classification: J21, J23, J24, O33

Keywords: job polarisation, job quality, tasks, discretion, intensity

Corresponding author:
Andrea Salvatori
OECD
2, rue André Pascal
75775 Paris Cedex 16
France
E-mail: [email protected]

1
Andrea Salvatori gratefully acknowledges the support of the UK Economic and Social Research Council (ESRC)
under the grant “Job Polarisation and Job Quality” (grant ES/K001116/1). We thank for comments and discussions
Matias Cortes, Claudio Deiana, Andrea Geraci, and participants in seminars at the University of Essex, EUROFOUND,
OECD, IZA Workshop on the digital economy, EALE 2016, and the University of Lancaster. A previous version of this
paper was circulated in September 2016 with the title “Technology and the labour market: evidence on changes in
tasks and job quality across Europe”.
1 Introduction
There is currently a lively debate both in academic and policy circles on the effect of technology
on the labour market. While the bulk of the evidence point to no overall effect on the quantity of
jobs (D. Autor and Salomons 2017; OECD 2017; Flavio Calvino and Virgillito 2017; OECD 2016) 2, a
number of studies have shown that technology changes the types of jobs in the economy both
through compositional effects and through changes within existing jobs. The compositional
effects arise because of the varying degrees to which workers in different jobs can be
substituted or complemented (Maarten Goos, Manning, and Salomons 2014; Michaels, Natraj,
and Van Reenen 2013; Marcolin, Miroudot, and Squicciarini 2016), while changes within jobs
occur when the adoption of technology leads to changes in the organisation of work, the nature
of the tasks performed and the skills required (F. Green, Felstead, and Gallie 2003; Spitz‐Oener
2006; D. H. Autor 2013; F. Green 2012). A broad body of literature has focused on the
implications of these changes for wages, but less attention has been given to their impact on
non-monetary aspects of job quality. This paper provides novel direct evidence on the impact of
computer use on two important aspects of job quality, namely job discretion (i.e. the extent to
which workers have control over tasks, methods and speed at work) and intensity (i.e. the
extent to which a job involves working at high speeds or tight deadlines).

The focus on job quality is both useful and interesting for at least two reasons. First, surveys
show that workers value non-wage aspects of job quality (Gallie 2013) and this is reflected in
the growing efforts at the national and international level to pursue better quality jobs (OECD
2014). Discretion is strongly correlated with employees’ motivation and job satisfaction (F.
Green 2006) with various measures of psychological wellbeing (Gallie 2013; Wheatley 2017) 3,
while work intensity can lead to negative psychological outcomes, including stress (F. Green and
McIntosh 2001).

Secondly, studying its effect on job quality can provide new insights on how technology changes
work, which can help refine the theoretical frameworks used to study its impact on wages and
employment. As recently discussed in Autor (2015), the current paradigm for thinking about the
effect of technology, which emphasises differences in the extent to which automation can
replace workers in different jobs, faces significant empirical puzzles. Most importantly, the
prediction that lower demand for easy-to-automate routine jobs should lead to lower
employment shares and wage growth has not been borne out by the data in most countries. This
suggests that our understanding of how technology impacts these (and other) jobs might not be

2
Recent papers have found a negative effect of robots on employment in the US (Acemoglu and Restrepo
2017), but not in Germany (Dauth et al. 2017). Graetz and Michaels (2015) do not find an effect of robots on
overall employment exploiting variation over time within country and industries.
3
Several papers have provided both descriptive and causal evidence of the link between wellbeing and
productivity (Böckerman and Ilmakunnas 2012; Oswald, Proto, and Sgroi 2015; Bryson, Forth, and
Stokes 2014).

2
complete. 4 From this perspective, our approach of investigating changes in job quality can be
seen as complementary to the standard approach of focusing on wages and employment.

We focus on computer use for both substantive and practical reasons. The substantive reason is
that computers are the most widely used form of technology in the labour market that are
already recognised in the literature to have played a central role in changing skill demands and
job tasks (D. H. Autor, Levy, and Murnane 2003; Spitz‐Oener 2006; F. Green 2012; Elsayed, de
Grip, and Fouarge 2017). The practical reason is that reliable indicators of actual use of other
forms of technology are not currently available. 5 In the concluding section we argue that, given
how pervasive digital technologies are becoming in all aspects of life and work, there is an
urgent need to verify the effectiveness of standard survey questions on computer use to
measure the penetration of technology in low skill jobs in particular.

Using data from the European Working Conditions Survey covering the EU-15 countries
between 1995 and 2015, 6 we provide direct evidence on the link between computer use and job
quality, exploiting variations within occupations over time. Similar approaches have been used
to study the link between computer adoption and changing skill requirements (F. Green 2012;
Spitz‐Oener 2006), but most of the literature on the effects of technology on other aspects of job
quality has generally taken a more indirect approach. 7 In particular, researchers have
interpreted common patterns in different countries as consistent with technology being a
common driver 8 and have used the direction of aggregate changes in job quality to discriminate
between different theories on the impact of technology on job quality. For example, Gallie
(2013) concludes that the evidence of stable job discretion in several countries in recent
decades is not consistent with theories that predict that new complex technologies raise
discretion. However, the overall trend can be the product of different forces and it is therefore
not necessarily informative of the effect of technology on job quality. Indeed, the fact that
computer adoption has counteracted (rather than contributed to) a modest negative trend in
job quality is one of the main results of our analysis, to which we return below.

4
See, for example, the discussion in Autor and Dorn (2013) who find that the wages of clerical workers
increased robustly in the US between the 1990s and the 2000s in spite of the decline in their employment
shares. They conjecture that this might be due to the fact that technology has a two-fold effect on these
jobs. On one hand, it reduces the demand for these jobs because the tasks involved are relatively easier to
automate. On the other hand, it also changes the nature and organisation of the remaining jobs, possibly
leading to an increase in the productivity of the remaining jobs in these occupations. By contrast, the
recent literature tends to see technology as either complementary or as a substitute to workers in a given
job.
5
Joling and Kraan (2008) propose an indicator of machine use at work based on the EWCS survey. However,
these measures are constructed using questions which are not designed to measure directly the use of
machines at work. For example, one of the questions used is whether the worker is exposed to vibrations from
machines at work. It is perfectly plausible that by this indicator machine use appears to decline as the quality
of the machines improve, resulting in a reduction of vibration. Other contributions have proposed industry-
level indicators of adoption of other forms of technology. These indicators are informative of broad between-
industry differences in technology adoption, but not of actual use by different types of workers within
industries (F. Calvino et al. forthcoming)
6
This appears to be one of the first studies to make use of the most recent wave of data released in 2017.
7
(Joling and Kraan 2008) use data from the 2005 wave of the EWCS to provide a detailed picture of the profile
of workers who use technology and those who do not.
8
A recent example of a paper applying this logic is Green et al. (2013), but the same type of inference if often
invoked well beyond the literature on job quality. See for example Card and Lemieux (2001), Dustman et al.
(2009) and Green and Sand (2015).

3
Isolating the causal effect of computer use on job quality remains a difficult task, even in models
that focus on variation over time within occupations. To mitigate concerns that endogeneity
might bias our estimates from the base model in first-differences, we resort to an instrumental
variable approach that exploits the secular declining trend in computing cost for identification
(Acemoglu and Autor 2011; D. H. Autor and Dorn 2013; Nordhaus 2007). We instrument the
change in computer use in one country-occupation cell with the average of the
contemporaneous change in computer use in occupations involving similar tasks in other
countries. While we are not aware of other applications to study the effect of computer use 9, the
approach of using changes in other countries as instruments to exploit common exogenous
trends for identification is increasingly used in related literature. For example, Autor et al.
(2013) use changes in Chinese imports to other high-income countries to instrument changes in
import penetration to US local labour markets, while Acemoglu and Restrepo (2017) instrument
changes in robot penetration in US industries with those in other advanced countries (and
Dauth et al. (2017) apply the same strategy to German data).

A number of recent papers offer comprehensive discussions of the multiple dimensions of job
quality (see, among others, Green (2006), Bustillo et al. (2011), Kalleberg (2011),Green et al.
(2013), and Findlay et al. (2013)). We focus on the effect of computer use on two specific
aspects of job quality, namely job discretion and intensity, that feature prominently in the
debate on the effects of technology on work. To measure these two aspects of job quality we
employ sub-components of the indexes proposed by Green et al. (2013) using earlier waves of
the EWCS. 10 Our measure of work discretion reflects the extent to which individuals have
control over (i) the order in which they perform their tasks, (ii) the methods of work, and (iii)
the speed at which they work. Our measure of work intensity combines the answers to
questions on whether a job involves working (i) at high speeds and (ii) to tight deadlines.

The effect of computer use on discretion is a priori ambiguous. A positive effect might arise if
computers provide workers with a higher degree of flexibility in the organisation of their work
and increase the control they have over it. In particular, researchers have argued that the
upskilling brought about by modern technologies is closely linked to a higher degree of control
over one’s work. However, evidence for the US indicates that the arrival of computers has led to
only a modest increase in skill requirements (Handel 2016; Osterman 2013). In addition, even
in countries like the UK where upskilling has been stronger (F. Green 2012; F. Green et al. 2016),
job discretion has been relatively stable in recent decades suggesting, as pointed out by Gallie
(2013), that this upskilling has not necessarily translated into increased discretion at work.

The negative effect of computers on discretion can arise if technology is used to achieve higher
standardisation of work and monitoring (Weil 2014). The net effect of computer use may differ
across occupations. New technologies might afford workers in cognitive occupations higher
discretion and control over their work, but they can also enable monitoring and closer
management of cognitive tasks, giving rise to a form of “digital Taylorism” in which employees

9
Spitz-Oener (2006) and Green (2012) also exploit variation over time within occupations to study the effect of
computer use on skill requirements and tasks, but neither uses an instrumental variable approach.
10
As far as we are aware, the EWCS is the only dataset to offer harmonised data on these variables over time
for a number of countries. Recently, PIACC has asked similar questions, but these data are currently available
only for a point in time in the early 2010s.

4
enjoy very limited control over their work. 11 Mazmanian et al. (2013) find that interconnected
devices do provide professionals with the possibility of a greater control over the pace and
organisation of their work, but also create an expectation of constant availability by colleagues
and clients that actually reduces their discretion. Workers in more manual occupations might
see a decline in job discretion if technology is mostly used to ensure that they follow precise
procedures or work at a certain pace. For example, computers can be used to provide call centre
operators with precise scripts to reduce the duration of calls, or to automate parts of the
ordering and food-preparation processes in catering, or to shorten health care workers’ visiting
time and to ensure more efficient transfers between visits. While the limited number of
occupations available in our data provide little leverage to obtain precise estimates by
occupation, we do investigate descriptively whether the effect of computer use appears to differ
across different occupations.

The effect of computer use on work intensity is also ambiguous a priori. Technology might be
“effort-biased” in the sense that it complements workers who are able and willing to increase
their effort by making the allocation of tasks more rapid and efficient and by facilitating
monitoring (Green 2006). On the other hand, technology might allow greater flexibility in the
organisation of work, easing the pressure on workers. The net effect of technology on work
intensity may differ depending on the type of tasks performed, as workers performing cognitive
tasks that do not require their physical presence in a given workplace or direct contact with
clients and customers might be better positioned to take advantage of the increased
organisational flexibility allowed by technology. While there is evidence from both the US and
Europe of an increase in the intensity of work in recent decades (Kalleberg 2011; F. Green 2006;
Clark 2005) and that intensity is higher in jobs that use computers more frequently (F. Green
and McIntosh 2001; Gallie 2005), we are not aware of studies that have attempted to isolate the
causal effect of computer use on job intensity.

Our results show that computer use grew substantially between 1995 and 2015 across Europe,
with the share of workers who report using computers at work increasing from 40% to 60%.
However, long after the onset of the PC revolution in the 1980s, countries continued to differ
significantly in the extent to which they used computers in similar occupations. In particular,
Nordic countries have seen large increases in computer use in occupations (such as “service and
sales occupations” and “elementary occupations”) that are typically thought, in the literature, to
be less affected by this technology.

The great increase in computer use between 1995 and 2015 coincided with a period of modest
deterioration in job quality in the EU-15 as whole, as intensity increased for all occupational and
educational groups, while discretion decreased slightly for most of them. Our OLS estimates
point to a sizeable positive effect of computer use on discretion, but to a small or no effect on
intensity at work. Hence, this evidence suggests that the (moderate) deterioration in the quality
of work observed in the EU-15 between 1995 and 2015 has happened despite the spread of
computers, rather than because of them. Our IV estimates point to an even more benign effect of
computer use on job quality, with greater positive effects estimated on discretion and negative
but insignificant effects on intensity. Finally, our descriptive analysis finds little indication of

11
For a discussion of different schools of thoughts on the implications of technology on intrinsic job quality see
Gallie (2013).

5
differences in the effect of computer use on job quality across different occupations. We find no
indication that computer use is associated with a decline in discretion in any occupation.

2 Measuring technology, job discretion and intensity

The European Working Conditions Survey (EWCS) is a 5-yearly survey of workers in the
European Union starting from the year 1990. It is funded, designed and coordinated by the
European Foundation for the Improvement of Living Conditions (EUROFOUND), which is an
agency of the European Union using face-to-face interviews of a randomized representative
sample in each member state. The content of the survey is fairly comprehensive and includes
themes such as employment status, work-life balance and worker participation, which are
relevant to our analysis. We use data from the second wave onwards of the EWCS; year 1995 as
this is the first sample that includes all EU15 countries. We use data up to and including the
latest survey, which was conducted in 2015. Thus our analytical sample consists of data from
the years 1995, 2000 12, 2005 and 2010 and 2015.

2.1 Measuring computer use at work

While the measurement of technology at work is by no means a straightforward task, we rely on

previous literature as a reference to determine our definition of technology use at work, subject
to data availability. We follow previous work by Dhondt et al. ( 2002) and Joling and Kraan
(2008) and exploit information available in the EWCS. Specifically we use the responses to the
question “How often does your main paid job involve each of the following? Working with
computers: PCs, network, mainframe”. This question was asked consistently from 1995 to 2010.
However, in 2015 the question was framed as “Please tell me, does your main paid job involve ...?
working with computers, laptops, smartphones etc.?” to include laptops and smartphones.

In all waves, responses to the computer use question are coded on a scale with 7 categories
ranging from Never to All the time. We create a binary measure of computer use where 0
indicates respondents who never use computers and 1 indicates respondents who reported
some use of computers. As we explain below, to be able to exploit variation over time in our
dataset of repeated cross-sections, our analysis is entirely conducted at the occupation-country-
year level and therefore, effectively, our computer use variable measures the proportion of
workers in each cell that uses computers at least a quarter of the time. Inspection of the change
over time of the variable does not reveal any suspicious differences between the values in 2010
and 2015, in spite of the change in the wording of the underlying question. In any case, we verify
the robustness of our results to the exclusion of the 2015 data throughout the analysis.
Furthermore, in our regression analysis we check whether there is any indication that the
relationship between computer use and job quality changes over time, either (i) as a result of
the change in what the variable captures due to different wording or (ii) as a result of changes in
the capabilities of computers over time.

A possible limitation of this measure of computer use which is rarely recognised in the
literature is that it might be more effective in capturing the adoption and use of digital

12
In the survey for the year 2000, no questions regarding education levels were recorded. For our main
analysis we construct a noisy measure of education by extrapolating data. We also check if our results are
robust to the exclusion of the year 2000 and, reassuringly, find qualitatively similar results.

6
technology in some occupations than in others. For example, Dhondt et al. (2002) raise concerns
that standard computer use questions might not effectively capture the use of digital
technologies in lower-skilled manual occupations. We return to this issue and its possible
implications for our results in Section 5.

Figure 1 reports average computer use by occupation in all countries at the beginning and the
end of our observation period. A clear contrast emerges between the occupations typically
characterised by cognitive tasks (from managers to clerks) and others. Within this former group
of occupations, differences in computer use across countries have shrunk substantially. Even
Greece, which in 1995 stood out as a clear outlier with PC use below 50% in all cognitive
occupations, had reached values above 70% in all of them (except managers) in 2015. In
countries where the figure was high to start with – such as Denmark, Sweden, and Finland –
computer use was approaching saturation by 2015, with figures above 90% in several
occupations.

However, the convergence in computer use across countries is not seen in all the remaining
occupations. To the contrary, for crafts, machine operatives and elementary occupations the
range of values across countries increased. Hence, long after the onset of the PC revolution in
the 1980s, countries continued to differ significantly in the extent to which they used computers
in similar occupations.

Interestingly, Figure 1 also shows that in 2015 some countries were making extensive use of
computers in occupations that, in the economics literature, are generally thought to be less
affected by this technology. For example, in Denmark, Luxembourg, The Netherlands, Sweden,
Finland, Belgium and Austria over 50% of sales and service workers were already using a
computer in 2015 – a share similar or higher to that found in several countries in 1995 among
professionals and technicians, i.e. occupations typically thought to benefit from strong
complementarities with computers. Even for elementary occupations, the proportion using
computers was at least 20% in 5 countries in 2015, with the highest value of just under 40%
recorded in Denmark.

2.2 Measuring job discretion and intensity

Job quality is a multi-dimensional concept which has been the subject of many studies both from
a theoretical and empirical point of view. For recent in-depth discussions of this subject see,
among others, Green (2006), Bustillo et al. (2011), Kalleberg (2011) and Green et al. (2013).
Our aim is not to document broad trends in job quality in general, but to address the question as
to how computer use affects two specific aspects of job quality, namely job discretion and
intensity, that feature prominently in the debate on the effects of technology on work. The focus
on particular aspects of job quality rather than an overall aggregate index (as, for example in
Bustillo et al. (2011)) allows a more intuitive interpretation of the regression results.

As an indicator for job discretion, we use a subcomponent of the Work Quality indicator of
Green et al. (2013), which uses the answers to the following questions: “Are you able to
choose or change;

1. Your order of tasks

2. Your methods of work
3. Your speed or rate of work”

7
Our indicator of work intensity is also a subcomponent 13 of that in Green et al. (F. Green et al.
2013) and uses the answers to “Does your job involve;

1. Working at very high speeds

2. Working to tight deadlines”

For both of these indicators, the items are conceived as heterogeneous manifestations of the
relevant aspect of job quality rather than as variables reflecting an underlying single construct.
We choose these subcomponents as they are available for all years in our analysis and construct
the job quality indicators using a principle component analysis with a polychoric correlation
matrix. The indices are calculated by pooling the EU15 countries across years together. The
proportion of variance explained by the first component is 0.84 and 0.82 for the discretion and
intensity indicators respectively.

Figure 2 plots average discretion by country for each occupation in 1995 and 2015. The graph
shows no clear sign of decreasing dispersion across countries over time, even in cognitive
occupations (with the exception of managers) that have seen some convergence in computer
use. Noticeably, Greece remains a clear outlier in terms of discretion in most cognitive
occupations, despite the significant catch-up in computer use seen in Figure 1. Among clerks, the
range of values reported has increased, even when Greece is excluded. The picture also shows
that high-skilled cognitive occupations tend to have both higher average discretion and lower
variation across countries. Ranges across countries are generally around half of a standard
deviation for managers, professionals and technicians, but closer to a full standard deviation for
occupations such as crafts, machine operatives, and elementary occupations. 14

The higher cross-country dispersion of discretion in lower-skilled occupations is mostly due to

the fact that these occupations have particularly low levels of discretion in countries generally
found at the lower end of the discretion ranking. In other words, there is more inequality in
discretion within countries with generally lower levels of discretion across occupations. In
particular, Denmark, Sweden, Finland and the Netherlands have high average discretion and
lower dispersion across occupations 15, while Austria, Portugal, Greece and Germany have a
relatively low average discretion with greater differences between occupations 16.

Figure 3 shows that there are also sizeable differences in reported job intensity for a given
occupation across countries. However, unlike discretion, the dispersion in intensity across
countries does not appear to be systematically different for cognitive occupations. All
occupations, except elementary ones, saw a decline in the dispersion of intensity across

13
We exclude some subcomponents of the index of Green et al. (2013) that are often used as task indicators in
related literature. For example, whether the pace of work depends on direct demands from people such as
customers, passengers, pupils, patients etc.
14
The differences in the range of average discretion by country between the cognitive occupations and the
others remain, even if one ignores Greece, which tends to have particularly low levels of discretion in most
occupations.
15
Their average discretion across occupations (unweighted) is between 101 and 103 and the standard
deviation (across occupations) is always below 2.8.
16
For the first group of countries the (unweighted) average discretion is between 101 and 103 and the
standard deviation (across occupations) is always below 2.8, while for the second group the average is always
below 100 (around 98 for Germany, and below that for Greece) and the standard deviation always in excess of
4.

8
countries over the two decades. This is mostly because across all occupations intensity has
increased among the countries that reported the lowest levels in 1995. In some occupations –
such as professionals and clerks – the highest values have also become smaller.

Hence, overall, while the past twenty years have seen convergence among countries in the use
of computers in some occupations (notably the ones involving more cognitive tasks), there is
little indication that this has coincided with a period of increasing homogeneity in the quality of
work. The increased homogeneity in terms of intensity that we find does not seem to be
concentrated in occupations that have seen convergence in computer use and it is mostly driven
by increasing intensity in countries with initial low levels of intensity across the board. The
weak relation between dispersion in job quality and dispersion in computer use across
countries is confirmed more formally when we run a regression (not reported here) of the
variance across countries of the (occupation-level) job quality indicators on the variance of
computer use.

Overall these descriptive results are suggestive that technology is not a clearly dominating
determinant of labour market conditions across countries, a result that might appear surprising,
given the central role that technology has played in the recent debates on the ongoing changes
in the labour market. To gain further insights on the link between computer use and job quality
in Europe, we now turn to the central question of this paper and exploit variation over time
within country-occupation cells to estimate the effect of computer use on job quality at the
mean.

3 Trends in job quality and computer use across Europe

Figure 4 plots changes in job quality and computer use between 1995 and 2015 in the EU-15.
Computer use (on the right-hand side scale) increased considerably, rising from just above 40%
to just above 60%. Job intensity increased by 0.15 standard deviations, with most of this
increase occurring in the first decade. At the end of the two decades under consideration,
discretion was at a level just below that of 1995, having almost fully recovered the loss of 10%
of a standard deviation that occurred in the first decade.

The break down by three education levels in the top panel of Table 1 shows that the increase in
intensity took place within all education groups, but was larger for those with high and low
education – exceeding 20% of a standard deviation. Discretion decreased slightly for the two
groups with lower education only.

Computer use increased within all education groups. In fact, workers with the lowest level of
education saw the greater proportional increase (+77%) as the share using computers
increased from 0.18 to 0.32. Nevertheless, computer use remains much higher among workers
with higher levels of education, having reached 0.54 among those with secondary education and
0.86 among those with tertiary education.

The break down by computer use shows that in 1995 workers who did not use computers had
both lower intensity and lower discretion at work. Over the two successive decades, intensity
increased by 25% of a standard deviation for non-pc users, but remained substantially stable for
PC-users. Hence, the aggregate increase in intensity is driven by non-PC-users. Similarly,
discretion declined by 13% of a standard deviation for non-PC-users but by less than 5% of a

9
standard deviation for PC-users, (both changes are statistically significant (p<0.01)). As a result,
the gap in intensity between users and non-users of PCs has all but closed, while that in
discretion has slightly increased.

The lower part of the table reveals that at a given point in time there are greater differences
between occupations in terms of discretion than intensity. For example, in 2015 there is more
than a standard deviation difference in discretion between managers and machine operatives,
but for intensity the range is less than 40% of a standard deviation. Discretion declines almost
monotonically as one moves down the occupational classification, but intensity exhibits a more
complex pattern. In particular, throughout the period, crafts and machine operatives exhibit the
highest levels of intensity, but professionals and technicians report levels similar to those of
workers in service and elementary occupations.

Intensity increased in all occupations between 1995 and 2015, but not statistically significantly
(p<0.1) so among managers and plant and machine workers. Both of these latter groups had
already some of the highest levels of intensity in 1995. The greatest increase (30% of a standard
deviation) is seen in elementary occupations, but some middle-skilled (such as crafts) and high-
skilled (professionals and technicians) occupations also saw increases in excess of 20% of a
standard deviation.

Discretion has decreased slightly in most occupations, with the greatest decline of 17% of a
standard deviation recorded among service and sales occupations. The decrease is only
statistically significant (p<0.05)for technicians, those in service and sales occupations and plant
and machine workers. Only crafts saw a statistically significant increase in discretion which
exceeded 10% of a standard deviation.

Computer use has increased significantly (p<0.01) across the board, albeit at different rates. In
1995 the fraction using computers was above 75% only among clerks, with the second highest
figure (found among professionals) a distant 20 percentage points lower. By 2015, the cognitive
occupations (i.e. the first four occupations in the table) had computer use rates above 80% and
spanning a range of only 5 percentage points (pp). Computer use did grow significantly in all
other occupations as well – including elementary occupations, which saw a proportional
increase of over 60%. Nevertheless, in 2015 the fraction using computers was generally 45pp
lower in non-cognitive occupations than in cognitive ones.

Overall, therefore, the great increase in computer use between 1995 and 2015 coincided with a
period of modest deterioration in job quality in the EU-15 as whole, as intensity increased for all
occupational and educational groups while discretion decreased slightly for most of them.

In Figure 5 we plot changes in job quality against changes in the proportion using computers for
each country-occupation pair in our dataset, using all five waves available between 1995 and
2015. The regression lines fitted through the scatter plots (which are weighted by cell size)
indicate a positive relationship between computer use and discretion which is statistically
significant at the 1% level, but a positive and statistically insignificant one between changes in
computer use and changes in intensity. 17 Taken at face value, these results suggest that
computer use might have contributed to the increase in intensity but counteracted the decline

17
We check if the results are consistent to the exclusion of 2015 data to ensure our estimates are not driven
by the new wording of the question on computer use and find similar results.

10
in discretion over our observation period in the EU-15. However, these simple bivariate
correlations are likely to be affected by endogeneity. In the next section, we discuss the strategy
we adopt to tackle this issue.

4 Empirical strategy
The first-difference transformation underlying these plots in Figure 5 accounts for any time-
invariant omitted variables at the country-occupation level which might make computer use
endogenous. 18 However, endogeneity could still arise if changes in computer use are correlated
with occupation-country shocks. To address this concern, we first move beyond the simple
bivariate correlation of Figure 5 to include controls at the occupation-country level which can
capture some of the confounding changes. In particular, we estimate the following model in
stacked-differences:

∆𝑦𝑦𝑜𝑜𝑜𝑜𝑜𝑜 = 𝛼𝛼 + 𝛽𝛽1 ∆𝑃𝑃𝐶𝐶𝑜𝑜𝑜𝑜𝑜𝑜 + 𝛽𝛽2 ∆𝑋𝑋𝑜𝑜𝑜𝑜𝑜𝑜 + ∑3𝑖𝑖=2 𝑇𝑇𝑖𝑖 + ∆𝜖𝜖𝑜𝑜𝑜𝑜𝑜𝑜 (1)

Where ∆ is the difference operator between t and t-1, and the subscripts o and c refer to (1-
digit) occupations and countries respectively. PC is our binary computer use indicator and X is a
vector of controls which includes the within-occupation share of education, gender and age
groups, the share of employment of a given occupation-country pair in three broadly defined
industries (non-services, personal services, and other services), the share on temporary
contracts and the share of self-employed. 19 We include these latter two controls because these
groups might differ both in terms of job quality and computer use, but they will also help
capture business cycles effects to some extent. Russel and McGinnity (2014) argue that
organisational changes implemented dring the recession in Ireland led to a higher work
pressure. Since some of these changes might be correlated with computer adoption, we want to
control for the the business cycle to try and purge our estimates of these confounding effects.
To this end, we also include the country-level unemployment rate and include time dummies (
∑3𝑖𝑖=2 𝑇𝑇𝑖𝑖 ) that capture temporary deviations from the linear trends in levels implied by the
inclusion of the constant in this model in first-differences.

The OLS estimates of equation 1 will still be biased if time-variant determinants of computer
use and job quality are omitted. For example, a strand of literature emphasises that significant
changes in the organization of work have taken place in recent decades which are often
correlated with technology adoption but have effects on workers’ outcomes over and above
those of technology (Caroli and Reenen 2001; F. Green 2012, 2004). More generally, exogenous
changes in the conditions (e.g. in wages) of labour markets can alter the incentives facing firms
to adopt technology.

To mitigate these remaining concerns, we instrument ∆𝑃𝑃𝐶𝐶𝑜𝑜𝑜𝑜𝑜𝑜 with the average of the
contemporaneous change in computer use in occupations involving similar tasks in other

18
Similarly, when studying the impact of computer use on skill requirements and tasks, Spitz-Oener (2006)
uses first-differences at the occupational level with German data and Green (2012) uses a fixed-effect model
at the occupational level with British data. We focus on the occupational level analysis as there is no gain in
terms of identification of the effect of interest in using individual level data.
19
The sample used throughout the analysis reported here includes the self-employed, but we obtain very
similar results if we exclude them from the sample.

11
countries. 20 Here we define as similar those occupations that fall within the same group of the
classification proposed by Acemoglu and Autor (Acemoglu and Autor 2011) (AA henceforth)
and widely used in subsequent literature. 21

The rationale for our instrument is that (i) the major driver of the pervasive increase in
computer use in recent decades is the secular decline in the price of computing and that (ii)
occupations involving similar tasks will have similar rates of computer adoptions across
countries for purely technological reasons. Both parts of this argument are commonly made in
related literature. Nordhaus (2007) finds that, during the 1980s and 1990s, the rate of decline of
computing costs was on average 64% per year. A large body of literature spurred by Autor et al.
(2003) argues that computer adoption occurs differentially across occupations depending on
the extent to which they involve tasks that are easier to automate with the new technology
(Acemoglu and Autor 2011; D. H. Autor and Dorn 2013).

Our IV strategy aims at isolating the exogenous variation in computer adoption driven by the
secular decline in computing costs and uncorrelated with the occupation-country specific
shocks. This approach is conceptually similar to that of Autor et al. (2013) who attempts to
isolate the increase in import penetration into US industries driven by the arguably exogenous
expansion of the Chinese economy by using changes in Chinese import penetration into the
same industries in other high-income countries. More recently, Acemoglu and Restrepo (2017)
have instrumented changes in robot penetration in US industries with changes in robot
penetration in the same industries in other advanced countries. As these authors point out,
while not a panacea against all sources of endogeneity, this strategy enables the researcher to
focus on the variation that results solely from industries (or occupations in our case) in which
the change in the potential endogenous variable has been concurrent in most advanced
economies, attenuating endogeneity concerns arising from potential unobserved country-
industry (occupation) shocks.

20
In the construction of the instrument, we weight each occupation-country cell by its size. We also use the
average change in computer use in the same occupation in other countries and obtained very similar results
which are not reported here. We prefer the version excluding the own-occupation from the instrument as we
think it makes the assumption of no correlation across countries more plausible. As we discuss in the text, a
possible threat to this assumption arises from common shocks to industries in which an occupation is
concentrated in different countries. Using different occupations with involve similar tasks reduces the chances
of correlation due to common industry shocks and increases the chances that any correlation is driven by the
common level of exposure to the effect of technology. We discuss possible threats to this assumptions in the
main text. Furthermore, we considered a different instrument: we used measures of changes in ICT intensity at
the country-industry level from the EUKLEMS dataset and apportion that to our occupation-country level
observations using the proportion of occupational employment found in a given industry at the beginning of
our sample period (1995) within each country. This method is similar to that in Ebenstein et al. (2014), and
measures the exposure of an occupation to changes in ICT intensity at the industry level. As we are interested
in isolating exogenous variations driven by the secular decline in computing prices, we use variation in ICT
intensity in countries not included in our sample, namely US, Australia and Japan. This approach is similar to
that followed by Bloom et al. (2015) to instrument import penetration. However, we find that this instrument
is always too weak in our first-stage regressions for changes in computer use to provide reliable IV estimates.
21
We map the ISCO88 1-digit codes available in the data as follow: legislators (1), professionals (2) and
technicians and associate professionals (3) are non-routine cognitive; clerks (4) are routine cognitive;
service workers and shop and market sales workers (5) and elementary occupations (9) are non-routine
manual; craft and related trade workers (7) and plant and machine operators and assemblers (8) are routine
manual.

12
A threat to the exogeneity of the instrument arises from the possible cross-country correlation
in shocks to job quality between occupations in the same occupational group. The time trend
and dummies will capture changes that affect job quality in all occupations and countries.
Nevertheless, correlation in shocks to similar occupations could arise from global shocks to
industries in which such occupations are concentrated across countries. Plausible sources of
international industry-level changes over our sample period are the growth in international
trade (D. H. Autor, Dorn, and Hanson 2013) and changes in output demand due to demographic
changes or wealth effects (Mazzolari and Ragusa 2013; Moreno-Galbis and Sopraseuth 2014).

All our specifications control for the share of employment of a given occupation-country in
broadly-defined industries. To the extent that industry shocks lead to changes in the
distribution of occupational employment across such industries (for example a shift away from
manufacturing and towards personal services), this will help address the issue. Similarly, the
controls for demographics should alleviate the concerns relating to the growth in the number of
graduates and older workers. We further investigate the robustness of our results in three
different ways.

First, we verify the robustness of our results to the inclusion of EU-wide occupation-specific
trends and country-specific trends. This is a demanding specification which effectively assumes
that only deviations from these linear trends can be attributed to the secular decline in
computing costs which our instrument exploits.

Second, we consider a different version of the instrument which for any occupation-country pair
excludes the data from bordering countries. This is a useful approach if the correlation between
different spatial units is weaker the further apart they are, as commonly assumed in spatial
econometrics (Gibbons and Overman 2012).

Finally, we use data from a different dataset to control for changes in wages at the occupation-
country level in our IV models. In this approach, changes in wages are treated as a proxy for
shocks affecting different occupations across countries since the significant increase in
international trade over our sample period has been documented to affect wages differentially
across industries (D. H. Autor, Dorn, and Hanson 2013). Unfortunately, the data we need to
perform this check are only available to us for the period up to 2010 and not for every country
in every year. 22 For this reason we do not report these results here, but they broadly align with
the results of our other robustness checks.

4.1 Results
In Table 2 we report our OLS and IV estimates from models in first-difference in which each
country-occupation observation is weighted by their average size between t and t-1. 23 Panel A

22
The wage data come from the ECHP and EU-SILC and are not available for all observations in our sample. In
particular, we do not have Finland 1995; Sweden and Netherlands 1995-2000; Luxembourg 2000; Greece
2010; and half of the occupations are missing in France 2000. Due to these issues, and the fact that it is unclear
whether wages are a “good” control (Angrist and Pischke 2009) in our main regressions since they might be
one of the channels through which computer adoption affects job quality, we do not control for wages in our
preferred specifications but use them only in our robustness checks.
23
We note that as a default STATA uses weights from time t in first difference models. This is also the
approach taken in other related papers using models in first difference with aggregate data. When we do that,
the statistical significance of all our estimates for job discretion improves, while the estimates remain
statistically insignificant in the regressions for job intensity.

13
reports the results for job discretion. The first column includes only time dummies and implies
that, between 1995 and 2015, job discretion decreased slightly across the EU-15 by 0.14 points
– or just over 1% of a standard deviation. 24 The change conditional on observable
characteristics implied by the estimates in column 2 is greater (-1.8 or 18% of a standard
deviation). Hence, compositional changes have tended to counteract the decline in discretion
over the sample period. Computer use appears to have played a significant role in this sense: the
variable attracts a positive and statistically significant coefficient, which implies an increase in
the discretion index for the average occupation of over 12% of a standard deviation. 25 This is a
large effect when compared to the overall conditional decline in discretion of 18% of a standard
deviation: in the average occupation, the spread of computers is associated with a reduction in
the decline in discretion of over 60%. 26 To test whether the effect of computer use has changed
over time (perhaps as consequence of the growth in computing power or the increase in
interconnectivity), we also run separate regressions for each of the four time periods covered
by our data. These results – not reported here – show stable coefficients over time.

Column 3 presents our IV estimates using our baseline specification from column 2. The
instrument is strongly correlated with computer use, as shown by the test statistic reported at
the bottom of the table. 27 The coefficient on PC decreases slightly in absolute value and retains
statistical significance at the 10% level and the implied increase in discretion for the average
occupation (11%) is very similar to that found in column 3.

Column 4 adds occupation and country linear trends to our baseline IV specification. Neither
sets of linear trends are jointly statistically significant as indicated by the tests reported at the
bottom of the column, but the instrument remains strong. The computer use variable now
attracts a larger coefficient which is itself statistically significant at the 5% level. Its size implies
an increase in discretion in the average of occupation of 19% of a standard deviation which
offsets just over 60% of the entire conditional decline implied by the coefficients on the time
dummies in column 4. Finally, column 5 reports the estimates obtained excluding bordering
countries from the computation of the instrument. The instrument performs well in the first
stage and returns a coefficient for computer use which is again statistically significant at the 5%
level and implies an increase in discretion in the average occupation of over 15% of a standard
deviation.

Column 1 of Panel B shows that over the sample period work intensity increased across Europe
by just over 17% of a standard deviation. The estimated increase is greater (at about 25% of a
standard deviation) in column 2 where we condition on observable characteristics. Hence, as we
have already seen for discretion, compositional changes in general appear to have counteracted

24
Since this is a model in first difference, including a constant and a dummy for all but one changes, the total
estimated change is computed as the sum of 3 times the constant and the two coefficients on the time
dummies.
25
Computer use increased by 20pp between 1995 and 2015. Multiplying this change by the coefficient on the
PC variable yields: 0.20*6.065=1.14.
26
Results obtained excluding controls for the share of temporary contracts and self-employed indicate even
larger effects.
27
We use the command xtivreg2 (Schaffer 2010) in STATA to compute our IV estimates, which provides the
Kleibergen-Paap rk Wald F statistic for weak identification when using robust standard errors. Critical values
for such statistics are not available but the software reports those for the Cragg-Donald F statistic with i.i.d.
errors for different levels of tolerated relative bias above 10%. In all cases in which we refer to our IV as strong,
the reported (robust) test statistic is above each of those critical values.

14
the underlying trend in work intensity. Computer use, however, attracts a positive and
statistically insignificant coefficient, which implies a small effect of less than 1% of a standard
deviation for the average occupation. In regressions not reported here we find similar results
throughout the period covered by our data. The IV estimates in the remaining columns are
negative and greater in absolute value but are also statistically insignificant.

To summarise, we find that the overall modest decline in job quality has occurred in spite of
compositional changes that have tended to counteract this trend. As for computer use, the OLS
estimates point to a sizeable positive effect on discretion, but to small or no effect on intensity at
work. Hence, this evidence suggests that the (moderate) deterioration in the quality of work
observed in the EU-15 between 1995 and 2015 has happened despite the spread of computers,
rather than because of them.

Our IV estimates point to an even more benign effect of PC on job quality, with mostly greater
positive effects estimated on discretion and negative but insignificant effects on intensity. These
estimates are conditional on changes in demographics and in the distribution of occupational
employment across industries which, as we discussed above, should capture some of the
potential correlation across countries that might confound the instrument. Moreover, our
attempts to increase the plausibility of the exogeneity of the instrument, while again resulting in
statistically insignificant estimates in the intensity regression, paint a consistent picture overall:
they suggest more benign effects of computer use in the form of larger positive coefficients for
job discretion and larger negative ones for intensity.

4.1.1 Regressions by occupation groups

Recent contributions emphasise that the effect of computers on jobs depends on the type of
tasks they involve (D. H. Autor, Levy, and Murnane 2003; D. H. Autor 2015). In particular, this
literature argues that workers performing cognitive tasks benefit from strong
complementarities with computers, while those performing more routine tasks are more likely
to be substituted by current technology. Furthermore, low-skilled occupations involving non-
routine manual tasks are generally thought to offer little scope for either complementarity or
substitution with technology. This argument suggests that the effect of computer use on job
quality might also differ across occupations, perhaps being more pronounced in occupations
involving cognitive tasks.

To investigate differences in the effect of computer use across occupations, in Table 3 we

present OLS estimates obtained separately for different types of occupations. This is a simple
exploratory analysis as the small samples used in each regression make it difficult to obtain
statistically precise estimates. Moreover, our instrument does not offer enough variability to be
applied in this context. As in our main analysis, we stack the five-year differences between 1995
and 2015 together and include a constant and a full set of time dummies in all specifications.
Each observation is again weighted by the average cell size for each difference.

Following several previous studies, we group occupations as in Acemoglu and Autor (2011),
based on their task content. The tasks are classified along two dimensions: routine vs. non-
routine, and cognitive vs manual. Non-routine cognitive occupations are high-skilled
managerial, professional and technical occupations requiring problem-solving, intuition and
creativity. Routine cognitive occupations include clerical jobs and involve tasks including
organising, storing, retrieving and manipulating information. Routine manual occupations are

15
those involving repetitive production work. Finally, non-routine manual occupations include
personal service jobs and typically require situational adaptability, visual and language
recognition and in-person interactions. 28

Panel A shows the results for two specifications for job discretion, one including only time
dummies and the other including the same controls used in the main analysis. The estimated
constants and time dummies indicate that the change in discretion has been small in all groups
and negative in all except routine manual (+6% of a standard deviation). However, the
conditional change was negative for all groups and larger than the unconditional one for all
except routine cognitive occupations. There is no indication in these results that computer use
reduces job discretion in any occupational group: the coefficient on computer use is positive for
all occupations and is statistically significant at least the 10% level for all except the routine
cognitive group.

The large positive coefficients found for manual occupations are somewhat surprising. In
existing studies, non-routine manual occupations, in particular, are generally assumed not to be
affected in substantial ways by existing technologies. By contrast, the coefficients in Table 3
imply that a 10 percentage point increase in computer use is associated with an increase in job
discretion of 9% of a standard deviation in non-routine manual occupations, but only of 4% of a
standard deviation in routine cognitive jobs. Such a great difference might in part be due to the
fact that technology might have less of an impact in occupations (such as the cognitive ones)
that enjoy higher initial levels of discretion. Nevertheless, the finding of a strong association
between computer use and discretion in lower-skilled occupations is an interesting one, which
warrants a more careful consideration of the relationship between technology and employment
at the lower end of the skill spectrum.

The standard argument in the existing literature is that computers do not easily substitute or
complement workers in performing the main tasks that characterised non-routine manual
occupations. But even if computers do not lead to substantial changes in the type of tasks
performed by workers, they could lead to changes in the organisation of work that affect the
management and organisation of tasks – rather than the nature of the tasks themselves – in a
way that attributes a higher degree of discretion to the employee.

The results from the intensity regressions by occupational groups are reported in Panel B of
Table 3. The estimated time dummies and constant imply that intensity has increased across the
board and that compositional changes have partially counteracted the increasing trend in
routine cognitive and non-routine manual occupations. Computer use has a positive and
significant coefficient only in the regression for routine cognitive occupations. The estimated
coefficient means that a 10pp increase in computer use is associated with an increase in
intensity of just 7% of a standard deviation. In non-routine manual occupations, on the other
hand, computer use attracts a negative and statistically insignificant coefficient.

28
For the exact grouping of ISCO codes see footnote 21.

16
5 Discussion and conclusions
This paper uses harmonised data from across the EU-15, spanning the period 1995-2015, to
study the relationship between computer use and two aspects of job quality, namely discretion
and intensity. The main empirical contributions of the paper are two-fold. First, the analysis
provides an up-to-date picture of differences in computer use and job quality in the same
occupations across different countries documenting large differences between countries in
computer adoption within lower-skill occupations. Second, the paper directly investigates the
impact of computer use on job quality, using an identification strategy that exploits variation
over time within occupation-country cells and an instrumental variable approach that exploits
the variation in computer adoption generated by the arguably exogenous secular decline in the
cost of computing power. This set of results indicate that computer use has a large positive
effect on job discretion and that the recent modest decline in discretion has occurred in spite of
the diffusion of computers rather than because of it. In the remainder of this section we offer
some further discussion of these results.

Computer use has grown substantially between 1995 and 2015 across Europe, with the share of
workers who report using computers at work increasing from 40% to 60%. However, long after
the onset of the PC revolution in the 1980s, countries continued to differ significantly in the
extent to which they used computers in some occupations. In particular, Nordic countries have
seen great increases in computer use in occupations (such as “service and sales occupations”
and “elementary occupations”) that are typically thought to offer little scope for either
complementarity or substitution with technology (D. H. Autor 2015).

When considered in relation to the recent literature in economics on the effects of technology in
the labour market, these results lend themselves to two considerations. First, the finding of
differences in computer use in similar occupations across countries – which are particularly
large for some occupations – calls into question the common assumption that occupations are
homogenous across countries in terms of their task content and organisation. This assumption
is often explicitly or implicitly made in the literature on the effects of technology on the
occupational structure and underlies the use of task measures built from one country in the
analysis for a different country. 29

Second, these findings show that computer use is increasingly reaching into segments of the
labour markets that have so far widely been considered exempt from a direct impact in the
economics literature (M. Goos and Manning 2007; Acemoglu and Autor 2011; D. H. Autor 2015).
For example, in Autor and Dorn (2013)’s paper on the rise of low-skilled occupations in the US,
technology affects service occupations only indirectly through complementarities in
consumption with goods produced by occupations that are directly affected by automation.
Other contributions have highlighted that technology has the potential to significantly impact
the organisation and the content of low-skilled occupations, but while useful and insightful
these discussions have mostly been limited to anecdotal evidence or case studies (Weil 2014).
Cortes and Salvatori (2016) have also provided evidence that the use of computers in firms that
employ low-skilled workers grew significantly in the UK over the 2000s, while Kulkarni et al.
(2017) report a strong increase in the use of digital technologies in low skill occupations in the

29
For example, task measures constructed for the US are often used in the analysis for European
countries (Maarten Goos, Manning, and Salomons 2014). See Salvatori (2015) for further discussion of these
issues in the context of the UK.

17
US between 2002 and 2016. This evidence points to the need to develop a better understanding
of what technology does at the lower end of the skill spectrum in future research.

The large increase in computer use between 1995 and 2015 coincided with a period of modest
deterioration in job quality in the EU-15 as whole, as intensity increased for all occupational and
educational groups while discretion decreased slightly for most of them. Other studies that have
used earlier data have also reported limited time variation in job quality in Europe (F. Green et
al. 2013; Eurofound 2007). However, the results of our occupational-level analysis caution
against interpreting these results as indicative of a small negative effect of computer use on job
quality. To the contrary, our main empirical results suggest that the modest deterioration in job
quality has occurred in spite of the spread of computers rather than because of it. Our OLS
estimates suggest a sizeable positive effect of computer use on discretion, but small or no effect
on intensity at work. Our IV estimates point to an even more benign effect of computer use on
job quality, with greater positive effects estimated on discretion and negative but insignificant
effects on intensity. In addition, we find little indication of differences in the effect of computer
use on job quality across different occupations. In particular, we find no indication that
computer use is associated with a decline in discretion in any occupation.

Hence, our results lend support to the theories that emphasise the potential positive effects of
computer use on job quality through increased flexibility and control over one’s work. In
addition, they also illustrate that computer adoption has not been the dominant driver of the
evolution of working conditions within occupations. This is a somewhat surprising result, given
that technology has taken the centre-stage in the policy debate on the ongoing changes in the
labour market. While our data do not allow us to investigate this hypothesis directly, a number
of earlier studies have argued that organisational change plays an important role affecting job
quality over and above technology (F. Green 2012; Caroli and Reenen 2001). In their review of
the UK literature, Green et al. (2016) refer to changes in “management culture” as their
preferred explanation for the decline in job discretion over the past 10-15 years. Bryson et al.
(2016) show that job quality is affected by managerial practices (and other workplace
characteristics) over and above individual and job characteristics. Understanding the relative
importance of these factors vis-à-vis technology in driving aggregate trends in job quality
remains an important challenge for future research, with significant policy implications.

A possible limitation of our study is that our computer use variable might be more effective in
capturing the adoption and use of digital technology in some occupations than in others. 30
Dhondt et al. (2002) raise concerns that standard computer use questions might not effectively
capture the use of digital technologies in lower-skilled manual occupations. For example, it is
not obvious that workers at a fast-food restaurant who execute orders displayed on a monitor
would report using a computer at work. Yet, the pace and content of the job for (some of) these
workers is largely determined (if not entirely driven) by digital machines (Fitzgerald 2007;
Orleck 2017). More generally, workers might be less likely to report the use of computers if this
is confined to peripheral tasks relating to organisational and monitoring aspects of their jobs (as
in the case of a cook receiving orders through digital devices).

30
A similar concern may be raised within occupations, as workers who perform more cognitive tasks that are
typically associated with the use of standard computers might be more likely to report using a computer at
work.

18
If these reporting biases do exist, the growth in computer use in low-skilled occupations that we
have reported might be an underestimate, while our estimates of the effect of computer use on
job quality might disproportionally reflect the effect for workers who perform more cognitive
tasks (within occupations). We are not aware of any discussion of these issues in the literature,
but they would have implications for any of the many studies that use standard questions to
look at the impact of computers on a broad variety of outcomes. We would argue that, given
how pervasive digital technologies are becoming in all aspects of life and work, there is an
urgent need to verify the effectiveness of the survey instruments currently available to measure
the use of technology in the workplace and to develop new and improved ones if necessary.

19
6 References
Acemoglu, Daron, and David Autor. 2011. “Chapter 12 - Skills, Tasks and Technologies: Implications
for Employment and Earnings.” In Handbook of Labor Economics, edited by David Card and
Orley Ashenfelter, Volume 4, Part B:1043–1171. Elsevier.
http://www.sciencedirect.com/science/article/pii/S0169721811024105.
Acemoglu, Daron, and Pascual Restrepo. 2017. “Robots and Jobs: Evidence from US Labor Markets.”
w23285. Cambridge, MA: National Bureau of Economic Research.
https://doi.org/10.3386/w23285.
Angrist, Joshua David, and Jörn-Steffen Pischke. 2009. Mostly Harmless Econometrics: An Empirict’s
Companion. Princeton University Press.
Autor, David H. 2013. “The ‘Task Approach’ to Labor Markets: An Overview.” Working Paper 18711.
National Bureau of Economic Research. http://www.nber.org/papers/w18711.
Autor, David H. 2015. “Why Are There Still So Many Jobs? The History and Future of Workplace
Automation.” Journal of Economic Perspectives 29 (3):3–30.
https://doi.org/10.1257/jep.29.3.3.
Autor, David H, and David Dorn. 2013. “The Growth of Low-Skill Service Jobs and the Polarization of
the US Labor Market.” American Economic Review 103 (5):1553–97.
https://doi.org/10.1257/aer.103.5.1553.
Autor, David H., David Dorn, and Gordon H. Hanson. 2013. “The China Syndrome: Local Labor Market
Effects of Import Competition in the United States.” American Economic Review 103
(6):2121–68. https://doi.org/10.1257/aer.103.6.2121.
Autor, David H., Frank Levy, and Richard J. Murnane. 2003. “The Skill Content of Recent
Technological Change: An Empirical Exploration.” The Quarterly Journal of Economics 118
(4):1279–1333. https://doi.org/10.1162/003355303322552801.
Autor, David, and Anna Salomons. 2017. “Does Productivity Growth Threaten Employment?” In ECB
Forum on Central Banking, Sintra, Portugal, 26–28.
Bloom, Nicholas, Mirko Draca, and John Van Reenen. 2015. “Trade Induced Technical Change? The
Impact of Chinese Imports on Innovation, IT and Productivity.” The Review of Economic
Studies, September, rdv039. https://doi.org/10.1093/restud/rdv039.
Böckerman, Petri, and Pekka Ilmakunnas. 2012. “The Job Satisfaction-Productivity Nexus: A Study
Using Matched Survey and Register Data.” Industrial & Labor Relations Review 65 (2):244–
62. https://doi.org/10.1177/001979391206500203.
Bryson, Alex, Christine Erhel, and Zinaida Salibekyan. 2016. “Job Quality.” In Comparative Workplace
Employment Relations, edited by Thomas Amossé, Alex Bryson, John Forth, and Héloïse
Petit, 179–209. London: Palgrave Macmillan UK. https://doi.org/10.1057/978-1-137-57419-
0_6.
Bryson, Alex, John Forth, and Lucy Stokes. 2014. Does Worker Wellbeing Affect Workplace
Performance? Department for Business, Innovation & Skills, UK Government.
https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/366637/bi
s-14-1120-does-worker-wellbeing-affect-workplace-performance-final.pdf.
Bustillo, Rafael Munoz de, Enrique Fernanndez-Macias, Jose-Ignacio Anton, and Fernando Esteve.
2011. Measuring More than Money: The Social Economics of Job Quality. Edward Elgar Pub.
Calvino, F., C. Criscuolo, L. Marcolin, and M. Squicciarini. forthcoming. “A Taxonomy of Digital
Sectors.” STI Working Paper. Paris: OECD Publishing.

20
Calvino, Flavio, and Maria Enrica Virgillito. 2017. “The Innovation-Employment Nexus: A Critical
Survey of Theory and Empirics.” Journal of Economic Surveys, January, n/a-n/a.
https://doi.org/10.1111/joes.12190.
Card, David, and Thomas Lemieux. 2001. “Can Falling Supply Explain the Rising Return to College for
Younger Men? A Cohort-Based Analysis.” The Quarterly Journal of Economics 116 (2):705–
46. https://doi.org/10.1162/00335530151144140.
Caroli, Eve, and John Van Reenen. 2001. “Skill-Biased Organizational Change? Evidence from A Panel
of British and French Establishments.” The Quarterly Journal of Economics 116 (4):1449–92.
https://doi.org/10.1162/003355301753265624.
Clark, Andrew E. 2005. “Your Money or Your Life: Changing Job Quality in OECD Countries.” British
Journal of Industrial Relations 43 (3):377–400. https://doi.org/10.1111/j.1467-
8543.2005.00361.x.
Cortes, Guido Matias, and Andrea Salvatori. 2016. “Delving into the Demand Side: Changes in
Workplace Specialization and Job Polarization.” 10120. IZA Discussion Paper. Bonn: IZA.
http://www.iza.org/en/webcontent/publications/papers/viewAbstract?dp_id=10120.
Dauth, Wolfgang, Sebastian Findeisen, Jens Südekum, and Nicole Woessner. 2017. “German Robots -
The Impact of Industrial Robots on Workers.” 12306. CEPR Discussion Papers. C.E.P.R.
Discussion Papers. https://ideas.repec.org/p/cpr/ceprdp/12306.html.
Dhondt, Steven, Karolus Kraan, and Guurtje van Sloten. 2002. Work Organisation, Technology and
Working Conditions. Luxembourg: Publications Office.
Dustmann, Christian, Johannes Ludsteck, and Uta Schonberg. 2009. “Revisiting the German Wage
Structure.” Quarterly Journal of Economics 124 (2):843–81.
Ebenstein, Avraham, Ann Harrison, Margaret McMillan, and Shannon Phillips. 2014. “Estimating the
Impact of Trade and Offshoring on American Workers Using the Current Population
Surveys.” Review of Economics and Statistics 96 (4):581–595.
Elsayed, Ahmed, Andries de Grip, and Didier Fouarge. 2017. “Computer Use, Job Tasks and the Part-
Time Pay Penalty: Computer Use, Job Tasks and the PTPP.” British Journal of Industrial
Relations 55 (1):58–82. https://doi.org/10.1111/bjir.12175.
Eurofound. 2007. Fifteen Years of Working Conditions in the EU: Charting the Trends. Dublin:
Eurofound.
Findlay, Patricia, Arne L Kalleberg, and Chris Warhurst. 2013. “The Challenge of Job Quality.” Human
Relations 66 (4):441–51. https://doi.org/10.1177/0018726713481070.
Fitzgerald, Michael. 2007. “Making Fast Food Even Faster.” The New York Times, October 28, 2007.
http://www.nytimes.com/2007/10/28/technology/28proto.html.
Gallie, Duncan. 2005. “Work Pressure in Europe 1996-2001: Trends and Determinants.” British
Journal of Industrial Relations 43 (3):351–75. https://doi.org/10.1111/j.1467-
8543.2005.00360.x.
———. 2013. “Skills, Job Control and the Quality of Work: The Evidence from Britain (Geary Lecture
2012).” The Economic and Social Review 43 (3, Autumn):325–341.
Gibbons, Stephen, and Henry G. Overman. 2012. “Mostly Pointless Spatial Econometrics?” Journal of
Regional Science 52 (2):172–91. https://doi.org/10.1111/j.1467-9787.2012.00760.x.
Goos, M., and A. Manning. 2007. “Lousy and Lovely Jobs: The Rising Polarization of Work in Britain.”
The Review of Economics and Statistics 89 (1):118–133.
Goos, Maarten, Alan Manning, and Anna Salomons. 2014. “Explaining Job Polarization: Routine-
Biased Technological Change and Offshoring †.” American Economic Review 104 (8):2509–26.
https://doi.org/10.1257/aer.104.8.2509.
Graetz, Georg, and Guy Michaels. 2015. “Robots at Work.”
Green, David A., and Benjamin M. Sand. 2015. “Has the Canadian Labour Market Polarized?”
Canadian Journal of Economics/Revue Canadienne D’économique, November, n/a-n/a.
https://doi.org/10.1111/caje.12145.

21
Green, Francis. 2004. “Why Has Work Effort Become More Intense?” Industrial Relations 43 (4):709–
41. https://doi.org/10.1111/j.0019-8676.2004.00359.x.
———. 2006. Demanding Work: The Paradox of Job Quality in the Affluent Economy. Princeton
University Press.
———. 2012. “Employee Involvement, Technology and Evolution in Job Skills: A Task-Based
Analysis.” Indus. & Lab. Rel. Rev. 65:36.
Green, Francis, Alan Felstead, and Duncan Gallie. 2003. “Computers and the Changing Skill-Intensity
of Jobs.” Applied Economics 35 (14):1561–1576.
Green, Francis, Alan Felstead, Duncan Gallie, and Golo Henseke. 2016. “Skills and Work Organisation
in Britain: A Quarter Century of Change.” Journal for Labour Market Research, February, 1–
12. https://doi.org/10.1007/s12651-016-0197-x.
Green, Francis, and Steven McIntosh. 2001. “The Intensification of Work in Europe.” Labour
Economics, First World Conference of Labour Economists, EALE-SOLE, 8 (2):291–308.
https://doi.org/10.1016/S0927-5371(01)00027-6.
Green, Francis, Tarek Mostafa, Agnès Parent-Thirion, Greet Vermeylen, Gijs Van Houten, Isabella
Biletta, and Maija Lyly-Yrjanainen. 2013. “Is Job Quality Becoming More Unequal?” Industrial
& Labor Relations Review 66 (4):753–84.
Handel, Michael J. 2016. “What Do People Do at Work?” Journal for Labour Market Research 49
(2):177–97. https://doi.org/10.1007/s12651-016-0213-1.
Joling, Catelijne, and Karolus Kraan. 2008. Use of Technology and Working Conditions in the
European Union. Luxembourg: Office for Official Publ. of the European Communities.
Kalleberg, Arne L. 2011. Good Jobs, Bad Jobs: The Rise of Polarized and Precarious Employment
Systems in the United States, 1970s to 2000s. Russell Sage Foundation.
Kulkarni, Mark, Sifanu Liu, Jacob Whiton, and Siddharth Kulkarni. 2017. “Digitalization and the
American Workforce.” Brookings (blog). November 15, 2017.
https://www.brookings.edu/research/digitalization-and-the-american-workforce/.
Marcolin, Luca, Sébastien Miroudot, and Mariagrazia Squicciarini. 2016. “Routine Jobs, Employment
and Technological Innovation in Global Value Chains.”
Mazmanian, Melissa, Wanda J. Orlikowski, and JoAnne Yates. 2013. “The Autonomy Paradox: The
Implications of Mobile Email Devices for Knowledge Professionals.” Organization Science 24
(5):1337–57. https://doi.org/10.1287/orsc.1120.0806.
Mazzolari, Francesca, and Giuseppe Ragusa. 2013. “Spillovers from High-Skill Consumption to Low-
Skill Labor Markets.” Review of Economics & Statistics 95 (1):74–86.
Michaels, Guy, Ashwini Natraj, and John Van Reenen. 2013. “Has ICT Polarized Skill Demand?
Evidence from Eleven Countries over Twenty-Five Years.” Review of Economics and Statistics
96 (1):60–77. https://doi.org/10.1162/REST_a_00366.
Moreno-Galbis, Eva, and Thepthida Sopraseuth. 2014. “Job Polarization in Aging Economies.” Labour
Economics 27 (April):44–55. https://doi.org/10.1016/j.labeco.2013.12.001.
Nordhaus, William D. 2007. “Two Centuries of Productivity Growth in Computing.” The Journal of
Economic History 67 (01):128–159. https://doi.org/10.1017/S0022050707000058.
OECD. 2014. OECD Employment Outlook 2014. Paris: OECD Publishing.
http://public.eblib.com/choice/PublicFullRecord.aspx?p=1825934.
———. 2016. “ICTs and Jobs: Complements or Substitutes?” OECD Digital Economy Papers 259.
https://doi.org/10.1787/5jlwnklzplhg-en.
———. 2017. OECD Employment Outlook 2017. OECD Employment Outlook. OECD Publishing.
https://doi.org/10.1787/empl_outlook-2017-en.
Orleck, Annelise. 2017. “McDonald’s and the Global Revolution of Fast Food Workers.” The
Conversation. 2017. http://theconversation.com/mcdonalds-and-the-global-revolution-of-
fast-food-workers-64574.

22
Osterman, Paul. 2013. “Introduction to the Special Issue on Job Quality: What Does It Mean and How
Might We Think about It?” ILR Review 66 (4):739–52.
https://doi.org/10.1177/001979391306600401.
Oswald, Andrew J., Eugenio Proto, and Daniel Sgroi. 2015. “Happiness and Productivity.” Journal of
Labor Economics 33 (4):789–822. https://doi.org/10.1086/681096.
Russell, Helen, and Frances McGinnity. 2014. “Under Pressure: The Impact of Recession on
Employees in Ireland: Work Pressure in Recession.” British Journal of Industrial Relations 52
(2):286–307. https://doi.org/10.1111/bjir.12018.
Salvatori, Andrea. 2015. “The Anatomy of Job Polarisation in the UK.” IZA Discussion Papers.
https://opus.zbw-kiel.de/dspace/handle/10419/114081.
Schaffer, M.E. 2010. “xtivreg2: Stata Module to Perform Extended IV/2SLS, GMM and AC/HAC, LIML
and K-Class Regression for Panel Data Models.”
http://ideas.repec.org/c/boc/bocode/s456501.html.
Spitz‐Oener, Alexandra. 2006. “Technical Change, Job Tasks, and Rising Educational Demands:
Looking Outside the Wage Structure.” Journal of Labor Economics 24 (2):235–70.
https://doi.org/10.1086/499972.
Weil, David. 2014. The Fissured Workplace: Why Work Became so Bad for so Many and What Can Be
Done to Improve It. http://alltitles.ebrary.com/Doc?id=10844271.
Wheatley, Daniel. 2017. “Autonomy in Paid Work and Employee Subjective Well-Being.” Work and
Occupations 44 (3):296–328. https://doi.org/10.1177/0730888417697232.

23
Tables
Table 1 - Job quality and computer use by education, technology use and occupations over time.
Intensity Discretion Computer Use
1995 2015 % change 1995 2015 % change 1995 2015 change
Lower Secondary Qualifications 98.19 100.4 2.25*** 98.38 97.83 -0.56** 0.18 0.32 0.14***
Upper Secondary Qualifications 99.61 100.53 0.92*** 99.78 99.41 -0.37** 0.41 0.54 0.13***
Tertiary Qualifications 98.55 100.6 2.08*** 102.5 102.63 0.13 0.56 0.86 0.30***

No PC 97.59 100.12 2.59*** 98.85 97.51 -1.36***

PC 101.02 100.79 -0.23 102.45 101.97 -0.47***

Managers 100.61 100.63 0.02 104.88 104.85 -0.03 0.56 0.81 0.25***
Professionals 97.72 100.09 2.43*** 103.19 102.96 -0.22 0.57 0.89 0.32***
Technicians 97.68 99.96 2.33*** 102.28 101.75 -0.52** 0.52 0.86 0.34***
Clerks 99.18 100.24 1.07*** 100.37 100.2 -0.17 0.76 0.86 0.10***
Service/Sales 97.71 99.56 1.89*** 100.13 98.39 -1.74*** 0.33 0.41 0.08***
Craft/Trade 100.69 103.26 2.55*** 98.77 99.78 1.02*** 0.15 0.30 0.15***
Plant/Machine 101.57 102.12 0.54 94.26 93.55 -0.75* 0.19 0.33 0.14***
Elementary 97.31 100.28 3.05*** 97.45 98.02 0.58 0.10 0.16 0.06***
*** p<0.01, ** p<0.05, * p<0.1, weighted estimates and changes over waves and significance of the difference is tested in bivariate regression of the difference from 1995 to 2015 within each
category.

24
Table 2 - First-difference job quality regression using occupation-country observations from the EU-15,
1995-2015.

(1) (2) (3) (4) (5)

OLS OLS IV 1 IV 1 IV 2
Panel A – Dependent Variable - Job Discretion

6.065*** 5.705* 9.389** 7.313**

D.Computer Use
(1.172) (3.305) (3.907) (3.445)

0.308 -0.105 -0.0901 -0.262 -0.158

D.2005
(0.368) (0.315) (0.324) (0.341) (0.327)

1.076*** 0.828** 0.825** 0.757** 0.838**

D.2010
(0.384) (0.401) (0.399) (0.375) (0.398)

1.575*** 1.037*** 1.036*** 1.047*** 1.039***

D.2015
(0.338) (0.333) (0.329) (0.319) (0.325)

-0.775*** -0.911*** -0.896*** -1.153** -0.962***

Constant
(0.247) (0.254) (0.297) (0.563) (0.298)

Composition controls (a) No Yes Yes Yes Yes

Occupational and country trend No No No Yes No
F-Test of joint significant of country trends (df=13) 0.531
F-Test of joint significant of occupational trends (df=6) 0.758
Observations 480 480 480 480 480
0.087 0.338 0.338 0.351 0.334
R-squared
26.68 22.23 25.54
Kleibergen-Paap rk Wald F from first-stage

Panel B: Dependent Variable – Job Intensity

0.335 -3.809 -5.724 -3.418
D.Computer Use
(1.691) (3.794) (4.393) (4.073)

0.700 0.763 0.937* 1.042** 0.920*

D.2005
(0.468) (0.469) (0.495) (0.462) (0.493)

-0.931** -1.021** -1.053** -0.988** -1.050**

D.2010
(0.426) (0.483) (0.482) (0.441) (0.483)

-0.116 0.118 0.111 0.184 0.112

D.2015
(0.460) (0.460) (0.463) (0.434) (0.462)

0.526 0.670* 0.840** 1.241* 0.824**

Constant
(0.341) (0.350) (0.367) (0.683) (0.372)

Composition controls (a) No Yes Yes Yes Yes

Occupational and country trends No No No Yes No
F-Test of joint significant of country trends (df=13) 0.000
F-Test of joint significant of occupational trends (df=6) 0.289
Observations 480 480 480 480 480

25
 0.055 0.164 0.132 0.182 0.138
R-squared
26.68 22.23 25.54
Kleibergen-Paap rk Wald F from first-stage
First difference models with country-occupations weighted by average cell size.

(a): share of education, gender and age groups within each occupation-country cell; share of employment of a given
occupation-country pair in non-services, personal services, and other services, share of self-employed, share of those in
temporary contracts, and overall unemployment rate by country-year.

All regressions use data from five waves of the EWCS (1995, 2000, 2005, 2010 and 2015) and include data on employees and
self-employed workers. IV 1 uses change in PC use in similar occupations in all other countries as instrument. IV 2 excludes
bordering countries from the computation of the instrument.

Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

26
Table 3 - OLS estimates of first-difference models by occupational group for all employed.

Non Routine Routine Routine Non Routine

Cognitive Cognitive Manual Manual
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A: Dependent Variable: Job Discretion
D.Computer 4.622* 3.817 4.302* 9.407***
Use
(2.645) (3.104) (2.400) (1.796)

-0.710* -0.326 -0.817 -0.310 -0.186 -0.168 2.975*** 0.483

D.2005
(0.387) (0.381) (0.639) (0.715) (0.633) (0.761) (1.000) (0.902)

0.392 0.563 0.285 0.518 1.517** 1.265* 2.244** 2.091**

D.2010
(0.534) (0.607) (0.492) (0.653) (0.764) (0.695) (0.960) (0.859)

0.232 1.081** 1.965*** 1.250** 2.353*** 1.488** 3.055*** 1.129

D.2015
(0.411) (0.481) (0.520) (0.556) (0.659) (0.683) (0.806) (0.822)

0.0105 -0.747* -0.400* -0.763 -0.769* -1.027* -2.293*** -1.549**

Constant
(0.225) (0.417) (0.237) (0.597) (0.402) (0.594) (0.697) (0.747)

Controls (a) No Yes No Yes No Yes No Yes

Observations 180 180 60 60 120 120 120 120
0.050 0.307 0.306 0.702 0.199 0.485 0.209 0.566
R-squared

Panel B: Dependent Variable: Job Intensity

4.255 7.170** 3.150 -2.551
Computer Use
(3.663) (3.217) (3.055) (2.667)

-0.149 -0.335 2.397*** 2.057** 1.075 1.887** 0.870 1.022

D.2005
(0.771) (0.802) (0.850) (0.770) (0.886) (0.721) (1.014) (1.130)

-
-1.138 -1.119 0.362 -1.311 -1.937*** -0.424 -0.939
D.2010 1.710**

(0.726) (0.825) (0.867) (0.871) (0.714) (0.799) (0.942) (0.925)

-0.362 0.0966 0.772 1.100 -0.898 -0.0107 0.395 0.430

D.2015
(0.778) (0.775) (1.069) (0.773) (0.892) (0.767) (0.924) (0.860)

0.787 0.656 -0.578 0.379 0.873 0.122 0.389 0.715

Constant
(0.588) (0.658) (0.643) (0.593) (0.679) (0.615) (0.696) (0.672)

Controls (a) No Yes No Yes No Yes No Yes

Observations 180 180 60 60 120 120 120 120
0.037 0.249 0.158 0.543 0.209 0.442 0.032 0.211
R-squared

First difference models with country-occupations weighted by average cellsize.

27
(a): share of education, gender and age groups within each occupation-country cell; share of employment of a
given occupation-country pair in non-services, personal services, and other services, share of self-employed,
share of workers on a temporary contract, and the overall unemployment rate by country-year.

All regressions use data from five waves of the EWCS (1995, 2000, 2005, 2010, and 2015) and include data on
employees and self-employed workers..

Robust standard errors in parenthesis. *** p<0.01, ** p<0.05, * p<0.1

28
7 Figures

Figure 1 - Average computer use by country, occupation and wave. 1995-2015

Average computer use by country, occupation and wave

1995
Managers 2015

1995
Professionals 2015

1995
Technicians 2015

1995
Clerks 2015

1995
Service 2015

1995
Crafts 2015

1995
Plant_Machine 2015

1995
Elementary 2015

0 .2 .4 .6 .8 1
BE DK DE GR ES FR IE IT
LU NL AT PT FI SE UK

EWCS weighted data.

29
Figure 2 - Average discretion by country, occupation and wave

Average autonomy by country, occupation and wave

1995
Managers 2015

1995
Professionals 2015

1995
Technicians 2015

1995
Clerks 2015

1995
Service 2015

1995
Crafts 2015

1995
Plant_Machine 2015

1995
Elementary 2015

90 95 100 105 110

BE DK DE GR ES FR IE IT
LU NL AT PT FI SE UK

EWCS weighted data.

Figure 3 - Average intensity by country, occupation and wave

Average intensity by country, occupation and wave

1995
Managers 2015

1995
Professionals 2015

1995
Technicians 2015

1995
Clerks 2015

1995
Service 2015

1995
Crafts 2015

1995
Plant_Machine 2015

1995
Elementary 2015

90 95 100 105 110

BE DK DE GR ES FR IE IT
LU NL AT PT FI SE UK

EWCS weighted data.

30
Figure 4 - Job quality and computer use in the EU-15 between 1995 and 2015

101 Job quality and PC-use over time

.6
Job quality (standardised)
100.5

.55
PC-use (%)
100

.5
99.5

.45
99

.4
1995 2000 2005 2010 2015

Intensity Discretion PC-use

Source: EWCS 1995-2015, weighted average of job quality and PC-use in EU-15
wording of question on computer use was expanded in 2015 to include laptops and smartphones

Figure 5 - Correlation between changes in job quality and computer use at the country-occupation level.

Changes in job quality and technology use, '95-'15

Discretion and Computer Intensity and Computer
10
10

5
5
sd change Discretion

sd change Intensity
0
0

-5
-5

-10
-10

-40 -20 0 20 40 60 -40 -20 0 20 40 60

p.p. change in Computer p.p. change in Computer
slope: .056 (.016), p-value=0.000 slope: .024 (.017), p-value=0.156

Source: EWCS for EU-15, changes 1995-2000; 2000-2005; 2005-2010; 2010-2015 by occupation-country
relation between standardized job quality and p.p. of change in technology use
slope and standard error from regression clustered by country and weighted by cellsize
wording of question on computer use was expanded in 2015 to include laptops and smartphones

31