Chapter 3. How technology and globalisation are transforming the labour market1

This chapter documents the impact of two megatrends, technological progress and globalisation, on OECD labour markets over the past two decades, with a focus on the process of job polarisation and de-industrialisation. As both of these phenomena are associated with severe disruption in workers’ lives and rising inequality, they have given rise to growing concerns and uncovering their root causes is of fundamental importance for policy. The chapter begins by presenting key indicators of technology diffusion, participation in global value chains and international trade, and up-to-date evidence on job polarisation. It then analyses the relationship between polarisation and de-industrialisation, and employs econometric techniques to assess the impact of technology and globalisation on these phenomena. Technology displays the strongest association with both polarisation and de-industrialisation. The role of globalisation is less clear-cut, but there is some indication that international trade has contributed to de-industrialisation. Based on this evidence, the chapter outlines the key policy tools to help workers to successfully navigate the ongoing transformation of the labour market and reap the benefits of technological progress.

Key findings

This chapter analyses the impact of technological progress and globalisation on the structure of the labour markets in OECD countries, over the past two decades. In particular, it identifies the effects of these two megatrends on labour market polarisation and de‐industrialisation. Labour market polarisation involves a decline in the share of middle‐skill, middle-pay jobs relative to jobs with higher or lower skill levels and pay. De‐industrialisation is a shift in employment from manufacturing to services. These phenomena have been a major source of anxiety for workers in OECD countries, since they have been associated with job losses, rising inequalities, and the squeezing of the middle class. Understanding their causes has important implications for policy. In particular, the chapter focuses on three key driving factors that have been at the centre of the policy debate in recent years: the diffusion of information and communication technology (ICT); the growing importance of global value chains (GVCs); and the dramatic increase in imports from China following the country’s accession to the World Trade Organization (WTO) in 2001.

The key results are:

Over the past two decades, all regions considered have experienced a process of polarisation away from middle-skill jobs to low- and high-skill employment.
De-industrialisation (the reallocation of employment from manufacturing to services) accounts for about a third of this polarisation. Changes in the occupational structure within sectors explain the remaining two-thirds.
Growing ICT use is associated with an increase in high-skill relative to middle-skill occupations within manufacturing.
The more ICT is used, the slower is overall employment growth in the manufacturing sector. Greater use of ICT does not affect employment in the service sector, and overall has little impact on employment growth in the economy as a whole.
There is no clear evidence that more globalisation (for example, countries’ involvement in global value chains or the penetration of imports from China) causes changes in the occupational mix within industries across the OECD.
There is no evidence that integration in GVCs reduces the relative growth of different industries, but tentative evidence suggests that increasing import penetration from China has contributed to reducing employment in manufacturing, but not in services.
Finally, the chapter finds some evidence that labour market institutions – such as trade unions, minimum wages and the stringency of employment protection legislation (EPL) – may affect the way technology and globalisation impact the structure of the labour market. In particular, the results suggest that stricter EPL increases the effect of both ICT and GVC’s on polarisation, while stronger unions reduce the effect of ICT on bottom polarisation.

Building on these results, the chapter highlights a broad set of policy actions to help workers navigate the ongoing transformations of the labour market. They can be summarised as follows:

Building skills for the future

Policy makers should ensure that initial education, including early education, equips students with solid literacy, numeracy, problem-solving abilities but also basic ICT skills and soft skills, paying particular attention to the most disadvantaged groups who tend to lag behind in skill acquisition, use and adaptation during the working life.
Education and training systems need to better assess and anticipate changing skill needs in order to adapt curricula and guide students towards choices that lead to good labour market outcomes.
It is equally important to recognise that many skills are acquired outside education and training institutions. This emphasises the need for work-based learning opportunities, which has the advantage of linking training provision to a direct expression of employer requirements and workers’ interests, and to provide soft skills that are not easily taught in a classroom environment.
Even when workers have sufficient skills, inefficient use of such competences and skills mismatches may result in lower productivity and competitiveness. Promoting the use of high performance work practices (HPWP) and improved credentialing of skills learned on the job can play a crucial role in this regard.
The large share of workers with few, if any, digital skills, especially among older cohorts, illustrates the more general need to scale up and improve the effectiveness of lifelong learning and training for adults, so that workers are better able to keep up with continuously changing skills needs. This entails offering better incentives for workers and firms to re-skill and up-skill. Training opportunities should be widely available and not necessarily linked to one’s work status or workplace. Particular attention should be dedicated to low-skill workers, who currently tend to be neglected by on-the-job training programmes.
The provision of lifelong learning and adult training can be enhanced by the new opportunities digitalisation opens for innovation in learning infrastructure. MOOCs (massive open online courses) and OERs (open educational resources) are an important new resource, but they remain underutilised and their effectiveness rests on closing gaps in basic digital skills and on adequate investment in digital infrastructure.

Activation and social protection measures to help people face disruptive changes

The provision of welfare benefits should be designed in conjunction with activation measures to maximise the chance of re-employment and minimise disincentives to work, including in the difficult case of mid-career workers who are displaced by structural economic change and need to switch industry or occupation.
An effective activation framework should: i) motivate jobseekers to actively pursue employment; ii) improve their employability; and iii) expand the set of opportunities for them to be placed and retained in appropriate jobs.
As much as possible, activation measures should also be preventive, taking into account ongoing megatrends and the likely risk of job loss in different sectors, and providing workers with adequate information and re-employment support ahead of potential job losses (e.g. during the notice period prior to a mass redundancy).
Adapting social protection systems to the new world of work will require some crucial reforms. In particular, entitlements should be linked to individuals rather than jobs so that they are portable from one job to the next.
An alternative policy option being discussed in some countries is the introduction of a basic income guarantee, i.e. an unconditional income transfer that would replace other forms of public transfers without any means-testing or work requirement. The costs of such a solution, however, could be very large and its effects on work incentives need to be carefully assessed. In some countries, experiments with different forms of basic income guarantees are currently underway or planned that will offer some evidence to help judge the usefulness and feasibility of this kind of scheme.

Introduction

Technological change and globalisation are key forces shaping today’s world. Globalisation consists of a deeper integration of factors of production across countries. New technologies and increased digitalisation profoundly affect many aspects of life and have deeply transformed production processes by complementing workers and allowing the automation of certain tasks. They have also vastly reduced the transaction costs of communicating and co-ordinating globally, enabling a vertical fragmentation of industrial production that takes full advantage of the expertise and comparative advantages of different countries at each stage of production. Overall, by spurring innovation, increasing productivity and decreasing production costs, these two forces have contributed to economic growth and increased overall wellbeing. However, they have also entailed rapid transformations in the labour market, which pose severe challenges for workers, firms and governments.

In recent decades, labour markets across the OECD have experienced profound transformations in their occupational and industrial structures. A process of de‐industrialisation – which has seen significant shifts of employment from manufacturing to services – has taken place alongside one of labour market polarisation, whereby the number of middle-pay, middle-skill jobs has declined relative to the number of low-pay and high-pay jobs. These are fundamental changes, which cause significant disruption in workers’ lives and raise three significant policy challenges. The first is that employment is being reshuffled across occupations and industries, confronting workers with the risk of job loss followed by the need to make a difficult transition to a job in a different occupation or industry. Even workers who are able to stay in the same job are often faced with changing skill demands that require retraining. A second policy challenge arises from the link between the growth of the service sector and the slowdown in productivity growth which can hinder improvements in living standards (OECD, 2015a; Goos et al., 2016). Finally, differential changes in skill demands, driven by changing industrial structures, can affect trends in inequality over time (Acemoglu and Autor, 2011). To formulate adequate policy responses it is necessary to understand what drives changes in the structure of the labour market.

The increasing ability of technology to perform easy-to-codify routine tasks has been singled out in many studies as a key driver of job polarisation (Goos et al., 2014). Similarly, several studies have suggested that an accelerated diffusion of AI-enabled robots could soon lead to many more jobs being destroyed than created and hence to technological unemployment (Brynjolfsson and McAfee, 2011; Mokyr et al., 2015), although Arntz et al. (2016) reached less alarmist conclusions. At the same time, the offshoring of production to countries with lower labour costs has contributed to growing concerns about the negative impacts of globalisation in developed countries. The emergence of new players, increasingly integrated in global value chains (GVC), has heightened these concerns. In particular, China’s transition to a market economy and its entry into the World Trade Organization has benefitted consumers globally through lower prices, but has also been empirically linked to the decline in manufacturing employment in advanced economies (Autor et al., 2016), and to job polarisation in particular (Keller and Utar, 2016). China has recently become the world’s largest exporter, overtaking the United States and Germany (WTO, 2015).

Few studies have considered the relevance of technology and globalisation simultaneously for job polarisation and de-industrialisation, and those that have done so have typically focused on individual countries, neglecting the role of geographical and institutional factors. This chapter exploits industry-level data from 22 OECD countries over the past two decades to explore the relationship between job polarisation and de‐industrialisation, and to assess the importance of technology and globalisation in driving these structural transformations.2 The chapter draws upon a broad literature, which includes recent OECD contributions (e.g. Marcolin et al., 2016; OECD, 2016g). The core of the analysis, however, consists of novel empirical findings that build on the work by Breemersch et al. (2017). Several sources of data are pooled to measure the diffusion of technology and two recent developments in international trading patterns, namely integration in global value chains (GVCs) and the penetration of Chinese imports. Recognising that the effects of technology and trade are not inevitable but can be influenced by policy, the chapter also investigates the potential mediating role of labour market institutions using information on the role of collective bargaining – proxied by union density – as well as the minimum wage and employment protection legislation (EPL).

The remainder of the chapter is structured as follows. Section 1 presents recent evidence on job polarisation, as well as on key indicators of technological change, participation in GVCs and Chinese import penetration across countries. Section 2 then employs econometric techniques to assess the impact of technology and globalisation on job polarisation and de-industrialisation. The final section identifies the key policy tools to help workers to successfully navigate the ongoing transformation of the labour market and reap the benefits of technological progress and deepening international economic integration.

1. The changing structure of the labour market

Over the past decades, the labour markets of OECD countries have experienced a significant change in the occupational structure. One of the most evident transformations is the increased polarisation of employment into high-skill/high-paying jobs on the one hand, and low-skill/low-paying jobs on the other. This has occurred in conjunction with rapid digitalisation and automation, and increased global integration of production processes. This section paints a bird’s eye view of how the occupational structure has evolved in conjunction with technology and globalisation in recent decades, offering a discussion of the complex link between these developments and wage inequality. The following section will further explore the relationship between job polarisation and de‐industrialisation.

The labour market continues to polarise

The polarisation of the labour market into high-skill high-pay jobs and low-skill low-pay jobs has been widely documented in a range of advanced economies. Pioneering research by Autor, Katz and Kearney (2006), Goos and Manning (2007), and Goos, Manning and Salomons (2009) found that the share of employment in occupations in the middle of the skill distribution has declined rapidly in the United States and Europe over the past 30 years. At the same time, the share of employment at the upper and lower ends of the occupational skill distribution has increased. The result has been a hollowing out of the labour market.

Figure 3.1 shows the most recent available evidence on job polarisation across the OECD, between 1995 and 2015. Occupations are ranked by wage level following Autor and Dorn (2013) and Goos et al. (2014) and the results are presented by broad geographical area.3 The figure shows that all areas considered have experienced a decline in the share of middle-skill jobs relative to both high-skill and low-skill jobs. The country-specific results reported in Figure 3.A1.1 in Annex 3.A1 confirm that the decline in the share of middle-skill jobs is a pervasive phenomenon affecting all countries with only two exceptions in Central Europe (Hungary and the Czech Republic).4 Among the macro regions in Figure 3.1, only in Japan have low-skill occupations outgrown high-skill jobs, albeit only slightly, while in Northern Europe, Southern Europe, Western Europe and North America the employment shares lost in the middle have mostly been acquired by top occupations.

While a global analysis is beyond the scope of this chapter, it is worth noting that job polarisation tends to be lower or absent in emerging economies. For example, in China, there has been strong growth in both middle- and high-skill employment between 2000 and 2010, but an even larger increase in low-skill employment has resulted in the overall share of both medium and high-skill occupations falling (see Figure 3.A1.2 in Annex 3.A1). In India, the shares of low and medium skill occupations have decreased relative to high-skill occupations over the same period. The share of occupations that could experience automation in coming decades will be larger in emerging economies.5 Even in these countries, therefore, the risk of polarisation is significant and will depend to a large extent on the speed at which new technologies will be adopted (World Bank, 2016; Maloney and Molina, 2016). While lower wage costs have played a key role in attracting offshored jobs and containing the spread of automation, sustained real wage growth in emerging economies might contribute to some re-shoring of jobs in the coming decades, as well as providing incentives for the adoption of labour-replacing technology.

Technology and globalisation are advancing fast

One of the most commonly-identified drivers of labour market polarisation is the fact that the effect of technology varies across the skill distribution depending on the main tasks characterising different jobs. In particular, ICT is seen as complementing high-skill workers who perform the types of complex cognitive tasks typically found in managerial and professional occupations. On the other hand, middle-skill clerical and production jobs are typically characterised by “routine” tasks, i.e. the ones that can be executed following a precise set of instructions and are therefore easier to automate given current technological capabilities. Finally, low-skill jobs (such as those in catering and cleaning occupations, and other personal services) tend to involve non-routine manual tasks that, for example, require more manual dexterity and hand-eye co-ordination (which have so far proven more difficult to automate on a large scale). This so-called routine-biased technological change (RBTC), therefore, results in lower demand for middle-skill jobs relative to both high-skill and low‐skill ones, giving rise to the polarisation of occupational structures documented in advanced countries.

The decline in the share of middle-skill jobs has also been linked to increasing globalisation in at least two ways. First, the reductions in transaction and monitoring costs brought about by new technologies have contributed to the spread of global value chains which often entail the offshoring of the production of intermediate inputs and back office services that are typically provided by middle-skill workers (e.g. Oldenski, 2014). Second, the growth of international trade in final products has been concentrated in manufacturing sectors that traditionally account for a significant share of middle-skill/middle-pay jobs in advanced countries. For instance, the growth in Chinese import competition has been shown to have reduced manufacturing employment in the United States and Denmark (Autor et al., 2013; Keller and Utar, 2016).

A related literature has highlighted that international trade can alter the composition of labour demand by generating incentives for firms to innovate and adopt new technologies. Bloom, Draca and Van Reenen (2016) have shown evidence that the increase in trade with China has induced European firms to innovate significantly while also driving low-tech firms out of the market.6 This has increased the demand for high-skill workers in European firms and might therefore have contributed to the significant reallocation of employment from middle to top occupations that characterises the polarisation process in most countries.7

These points illustrate that trade and technology are mutually reinforcing and interact in complex ways in shaping the structure of labour markets. ICT tends to reduce transaction and monitoring costs that hamper international trade and GVCs, while in turn the competitive pressure arising from the increasing globalisation can induce firms to innovate and adopt technology which itself changes the demand for different skills.8

Empirical studies that have compared the explanatory power of alternative theories of polarisation in individual countries have generally concluded that technology and globalisation are the two main forces at play (Acemoglu and Autor, 2011; Goos et al., 2014). The jury is still out, however, on their relative importance. Before addressing this question, it is useful to set the scene by providing some descriptive evidence of how technology and globalisation have advanced in OECD countries.

Technology increasingly permeates the world of work

The growth in ICT use in the workplace provides a clear indication of how fast technology has permeated the world of work over the past three decades. From 1995 to 2007, the level of ICT capital services per hour worked at least doubled in every country analysed (Figure 3.2). There is, however, substantial cross-country heterogeneity, indicating that different countries experience very different paces of technology adoption. While in Hungary, Japan, and Slovenia, ICT levels increased by just over 150% over the period, the increase was as much as 300% in the Netherlands, the Czech Republic, Ireland, and Germany and above 350% in the United States, Belgium and the United Kingdom. For the period after 2007, the data are only available for selected countries and show that the growth rate of ICT slowed down in most countries (with the exception of Spain) following the recession.

Cross-country differences in the speed of technology diffusion have important implications for any predictions about the rate at which automation will contribute to job destruction going forward. Recent estimates of the share of jobs at risk of automation, discussed in detail in Box 3.3, are based on an assessment by experts of the likelihood that engineering obstacles to the automation of different tasks will be overcome in the near future (Frey and Osborne, 2013; Arntz et al., 2016). However, if there is substantial variation across (and within) countries, industries and occupations in the speed at which existing technologies are adopted, some countries may feel the effects of automation much later than others.

Large differences in the speed of technology adoption also exist between different sectors. While all industries have been impacted by fast penetration of new technologies, some economic activities have been affected more heavily than others (Figure 3.3). Across the countries analysed, for instance, “Total manufacturing” has seen the largest increase in ICT intensity, experiencing a growth of around 230% between 1995 and 2007. “Agriculture, hunting, forestry and fishing”, “Hotels and restaurants”, and “Wholesale and retail trade” have also recorded increases of around 200%. Even the sectors with the lowest growth rates – “Transport and storage and communication” and “Construction” – have nevertheless doubled their ICT intensity between 1995 and 2007.

Production processes are increasingly global

In parallel to the diffusion of ICT (and partly thanks to it), industrial production has become increasingly fragmented and internationalised. In particular, the world economy is increasingly organised in global value chains (GVCs) whereby the different stages of the production process are spread across countries and regions.

Figure 3.4 demonstrates the growing importance of GVCs by presenting the share of a country’s exports that is accounted for by foreign value added, as captured in the trade in value added (TiVA) dataset. It indicates the extent to which countries rely on intermediate products from abroad in their production processes (for a description of the dataset, see Box 3.1).9 Almost all countries have experienced increasing integration between 1995 and 2011, some of them at a very fast pace (e.g. the Slovak Republic, the Czech Republic, Hungary, Korea and Luxembourg). The global financial crisis caused a major slow-down in the integration process (not shown for conciseness), but all regions of the world have since returned to an upward trend.

Box 3.1. Mapping global value chains: The TiVA dataset

International trade increasingly involves global value chains (GVCs) whereby services, raw materials, parts and components are exchanged across countries before being incorporated in final products that are shipped to consumers all over the world. Exports from one country to another now reflect increasingly complex interactions among a variety of domestic and foreign suppliers and create income for firms and workers in widely separated locations. Trade is increasingly determined by the international strategies of firms that engage in foreign outsourcing and foreign direct investment so as to carry out their production activities or source their inputs wherever the necessary skills and materials are available at competitive cost and quality. The OECD has undertaken comprehensive data work that sheds new light on the scale, nature and consequences of international production sharing (OECD, 2013b).

In order to better account for the internationalisation and fragmentation of production, new trade statistics have been developed that identify the value added by each country in GVCs (http://oe.cd/tiva). These value added calculations are decomposed into foreign and domestic components, allowing for an in-depth examination of trade flows. The TiVA database encompasses a wide variety of trade measures, including: trade balances, domestic and foreign demand, re-imports, re-exports, service value added, and value added by source country and industry. These statistics build upon the OECD’s Inter-Country Input-Output (ICIO) tables and are expressed in millions of current USD, or as percentages. Reported variables are available by industry.

The most recent version of the TiVA database includes 61 economies covering OECD, EU28, G20, most East and South-east Asian economies and a selection of South American countries. The industry list has been expanded to cover 34 unique industrial sectors, including 16 manufacturing and 14 service sectors. The years covered are 1995, 2000, 2005 and 2008 to 2011.

Source: OECD (2015), “Trade Policy Implications of Global Value Chains”, available at: www.oecd.org/tad/trade-policy-implications-gvc.pdf.

China is an increasingly important global player

One of the most striking features of the past decades of rapid globalisation has been the rapid penetration of Chinese goods in the global economy. Several other countries have experienced rapid export growth, but given the scale of the Chinese economy, they probably have not had as large an impact on the labour markets of importing countries and definitely have not attracted the same interest in the international debate. Since China’s accession to the WTO in 2001, the share of Chinese imports in total domestic absorption of the average OECD country has grown from 1.4% to 4.8%, with peaks of 6.1% in North America and Central Europe (Figure 3.5). This has attracted the attention of policy makers concerned about the impact of Chinese competition on domestic labour markets, and it has motivated a growing body of academic research (e.g. Autor et al., 2013; Keller and Utar, 2016). The analysis in the next section continues in the same vein and includes Chinese import penetration among the variables whose impact on the labour market will be tested.

Overall, the effect of GVCs on the labour market is complex (Marcolin et al., 2016). On the one hand, as the importance of GVCs grows, part of domestic production is offshored and certain skills may no longer be needed, leading to potential displacement of workers and substantial labour reallocation across occupations and sectors. This may exacerbate the process of de-industrialisation and of job polarisation, since middle-skill jobs with a higher routine content have a greater potential to be offshored (Goos et al., 2014).10 On the other hand, as firms change their production structures to take part in GVCs, they adopt new processes that may have positive effects on productivity and competitiveness, and thus beneficial implications for wages and job quality. Moreover, international trade may have direct positive effects on overall employment. It has been estimated that between 30% and 40% of jobs in the business sector in most European countries in 2011 were sustained by consumers in foreign markets (OECD, 2016a).

The effects of GVCs are likely to be highly heterogeneous across economies, depending on their level of development. In less developed countries, low labour costs may attract offshored jobs and discourage offshoring of domestic jobs, but also slow down the adoption of technology that permits automation, leading to a slower process of polarisation. Labour market institutions may also play an important role, by cushioning (or amplifying) some of the effects of these megatrends on the labour market.11 These considerations pose a challenge for the empirical analysis in the next section, which will estimate average effects on a global scale and should be kept in mind when interpreting the results. They also motivate the analysis of heterogeneous effects across different regions and institutional settings, described below.

The complex link between inequality and the labour market

One of the main concerns with rising job polarisation is its potential implication for wage inequality. Indeed, the change in occupational structure documented above has coincided with a period of increasing wage inequality in a number of OECD countries (Figure 3.6). The link between polarisation and overall inequality, however, is complex. In the simple scenario where the polarisation of employment is entirely demand driven (for example, as a result of technology replacing middle-skill workers), one would expect to observe polarisation in wage growth as well, since the wages in low-skill and high-skill occupations would tend to grow at a faster pace than wages in middle-skill occupations. This is in fact what was observed in the United States in the 1990s, when lower tail inequality decreased and upper tail inequality increased (Acemoglu and Autor, 2011). However, wage polarisation has not been found in later decades in the United States (Mishel et al., 2013; Autor, 2015), nor at any point in time in any other country where job polarisation has occurred.12 Rather, most countries have seen an increase in the gap between top and median wages, and either a stable or increasing gap between median and bottom wages (Figure 3.6).

In discussing this apparent puzzle, Autor (2015) highlights that wage growth in bottom occupations can be hindered by the fact that these occupations generally do not benefit from significant complementarities with new technologies while also facing a very elastic labour supply, given their low skill requirements. This latter supply issue can be exacerbated if the decline in middle-skill job opportunities also means that some middle-skill workers have to settle for lower-skilled jobs. At the other end of the occupational, skill distribution, occupations requiring advanced cognitive skills typically see their productivity boosted by new ICTs and are characterised by a less elastic labour supply given the time necessary to acquire the education typically required for these jobs. As for middle-skill occupations, Autor (2015) emphasises that complementarity and substitution can coexist. So, while computers might be replacing some workers in performing routine tasks, they can complement those who remain in these occupations, therefore raising their productivity and, potentially, their wage growth.13 These mechanisms imply that job polarisation need not lead to wage polarisation and can instead contribute to growing inequality across the board (OECD, 2015a).

2. Estimating the effects of technology and globalisation on the labour market

The analysis in this section uses econometric techniques to estimate the effect of technology and globalisation on polarisation, and on another related aspect of labour market transformation: the process of de-industrialisation. It first discusses the relationship between the two phenomena, showing that polarisation is in fact partly the result of the shift of employment from manufacturing to services. The analysis covers 19 European countries, as well as the United States, Canada and Japan, between 1995 and 2007. For a subset of countries, where data are available through to 2015, more recent estimates are provided in Annex 3.A2.

Clarifying the relationship between polarisation and de-industrialisation

As a first step, it is important to clarify the relationship between job polarisation and the decline in manufacturing (de-industrialisation). To do that, it is useful to begin by distinguishing transformations that have occurred inside individual industries (i.e. within-industry polarisation) from changes due to the reallocation of employment from less polarised manufacturing sectors to more polarised service sectors (i.e. between-industry polarisation).

Middle-skill jobs have declined within all sectors

Figure 3.7 documents within-industry polarisation. The share of middle-skill occupations in total employment has declined in almost all sectors of the economy between 1995 and 2015. In most industries, these declines have been entirely offset by the growth in top occupations. This is particularly the case for those sectors where the decline in middle-skill occupations has been the largest across the OECD. This includes manufacturing industries (such as “Pulp, paper, paper products, printing and publishing”, “Chemicals and chemical products”, and “Transport equipment manufacturing”), as well as services (such as “Finance and insurance”, and “Real estate and business services”). Two service industries have seen a clear shift of employment towards the bottom of the skill distribution (“Hotels and restaurants” and “Wholesale and retail trade; repairs”). Figure 3.A3.1 in Annex 3.A3 (available online at OECD, 2017b) documents variations in the pattern of within industry polarisation across different regions. Northern Europe, Southern Europe, Western Europe and North America all exhibit a clear pattern of polarisation in all industries with a shift of employment from middle-skill jobs predominantly directed towards top occupations. Japan, which is excluded from Figure 3.7 due to a structural break in the data, also shows a similar pattern up to 2010. Central Europe stands out for the pronounced shift of employment away from low-skill occupations within most sectors, in line with the aggregate pattern in Figure 3.1.

The manufacturing sector has shrunk significantly…

The overall shrinking of the manufacturing sector has further contributed to the loss of middle-skill jobs. Figure 3.8 reports the percentage change in employment by industry, and the process of de-industrialisation is very clear. Only 2 of the 13 manufacturing sectors have seen their employment grow slightly, while 5 of them have experienced reductions of 30% or more. Most service sectors have increased their share of employment, with the largest growth recorded in “Real estate and business services” (+70%). The two sectors for which polarisation has meant a shift from middle- and high-skill jobs to low-skill jobs – as seen in Figure 3.7 – have increased their employment levels (“Wholesale and retail trade”, “Hotels and restaurants”). In particular, “Hotels and restaurants” was the second fastest growing sector with an increase of total employment in excess of 45%. Annex 3.A3 (available online at OECD, 2017b) documents variations across regions in changes in industry-level employment. The decline in manufacturing is clear in all areas except for Central Europe, where industries such as “Transport equipment manufacturing” have grown substantially. The fastest growing service sector is “Real estate and business services” in all regions except Central Europe and Japan, where it is the second fastest growing sector.14

… but polarisation mostly occurs within sectors, and not as a consequence of shrinking manufacturing

To understand the relative importance of between- and within-industry effects, one can also apply a formal decomposition of the change in overall polarisation over the period analysed into between- and a within-industry components (Goos et al., 2014).15 The results are reported in Table 3.1. Across all countries considered, the share of top and bottom occupations in total employment increased on average by about 5 percentage points between 1995 and 2007, from 58% to 63%. The last row shows that 62% of this increase is explained by changes in polarisation within industries, while the remaining 38% is accounted for by changes in the relative size of different industries. The positive between-industry component is the result of the fact that overall employment has shifted towards industries with higher polarisation. On top of that, within most sectors, polarisation has increased. As a result of these two forces, the business services sector emerges as the industry making the largest contribution to aggregate polarisation (50% of the overall increase).

Table 3.1. Industrya contributions to within- and between-industry polarisation, b 1997 to 2007c
Industry	Within	Industry	Between
Manufacturing, all	0.951	Manufacturing, all	-1.435
Agriculture	0.048	Agriculture	-0.253
Electricity, gas, water	0.089	Electricity, gas, water	-0.134
Mining	0.015	Mining	-0.043
Transport & communication	0.253	Transport & communication	-0.158
Wholesale and retail trade	0.203	Wholesale and retail trade	-0.045
Education	0.113	Education	0.071
Finance and insurance	0.341	Finance and insurance	-0.056
Public administration	0.449	Public administration	-0.067
Construction	0.113	Construction	0.269
Other services	0.121	Other services	0.279
Hotels and restaurants	-0.026	Hotels and restaurants	0.429
Health and social work	0.159	Health and social work	0.851
Business Services	0.393	Business Services	2.226
Total	3.221	Total	1.934
← a. Industries are classified according to ISIC Rev.3 2 digit classifications. The groupings are as follows: Agriculture (1 to 5), Business services (70 to 74), Construction (45), Education (80), Electricity, gas, water (40 to 41), Finance and insurance (65 to 67), Health and social work (85), Hotels and restaurants (55), Manufacturing, all (15 to 37), Mining (10 to14), Other services (90 to 93), Public administration (75), Transport and communication (60 to 64), Wholesale and retail trade (50 to 52).
← b. In this table, overall polarisation is calculated as the sum of high- and low-skill workers over total employment. Within-sector polarisation is the increase in the share of high- and low-skill jobs within an industry, while between-sector polarisation is the reallocation of employment towards more highly polarised industries. Within-industry polarisation is calculated as the change in polarisation by industry over the time period, multiplied by the average share of employment of that industry. Between-industry polarisation is calculated as the change in the employment share of an industry over the time period, multiplied by the average polarisation of that industry.
← c. Some countries were missing observations in 1995 and 1996, and so 1997 was taken as the beginning of the period with the exception of the Slovak Republic, which uses data from 1998. There was a revision in the ISIC industry classification in 2008, limiting the analysis to 2007.
d. Averages are calculated at a country level.
Source: OECD calculations based on the European Labour Force Survey; labour force surveys for Canada (LFS), Japan (LFS) and the United States (CPS MORG).
https://doi.org/10.1787/888933478175

Figure 3.9 shows that the prevalence of the within-industry component is a pattern observed in most countries, with some notable exceptions including the Czech Republic, Japan, the Slovak Republic, the Netherlands, Hungary, Germany and Portugal, where the decline of specific sectors has played a major role in the loss of middle-skill jobs relative to high- and low-skill occupations.

The effects of technological change and globalisation on within and between polarisation are theoretically ambiguous

Technology can affect overall polarisation both through within- and between-industry adjustments (Goos et al., 2014), but its overall effects are theoretically ambiguous and deserve empirical scrutiny. To see why, consider a sector which initially makes high use of routine jobs and is therefore relatively unpolarised. The adoption of new labour-saving technology will cause polarisation within the sector. It will also slow employment growth in this sector and thus raise aggregate polarisation by increasing the relative share of more polarised sectors. However, as technological advances trigger greater productivity and lower prices for consumers, greater demand for the sector’s output may partly offset the first-order effect of automation. Similarly, offshoring of middle-skill jobs might induce both higher polarisation and slower employment growth in less polarised sectors. If it leads to higher productivity, however, some of these negative effects may be partly offset. Import penetration is akin to offshoring as it contributes to the displacement of middle-skill workers and to the overall decline of manufacturing sectors that make intensive use of such workers (Keller and Utar, 2016). However, trade models that incorporate firm heterogeneity also predict that trade can induce adjustments within industries as production switches towards more productive firms (Melitz, 2003). To the extent to which such firms also have a more polarised occupational structure (perhaps as a result of technology adoption), this means that trade could affect aggregate polarisation through within-industry adjustments as well. And, again, these efficiency gains within industries might lead to stronger employment growth for the more polarised sectors, therefore further contributing to the overall polarisation of the labour market.

There are of course, a number of other factors that can contribute to polarisation within and across industries – either independently or by interacting with the megatrends that are the focus of this chapter. For example, the fortunes of different industries can be driven by changes in consumer preferences, and firms can adjust their production technology and occupational composition to changes in the composition of the workforce in terms of skills, gender and immigration status.16 While existing studies have suggested that these factors might have played a role in explaining some aspects of the polarisation process in at least some countries, they are beyond the scope of this chapter.

What drives polarisation within industries?

Given the relative importance of within-industry polarisation, the first goal of the econometric analysis is to investigate how changes in technology and integration in GVCs affect job polarisation within individual sectors. For this purpose, within-industry polarisation can be split into two complementary indicators: i) the share of high-skill relative to medium-skill occupations, by industry and country, can be used to capture polarisation at the top; while ii) the share of low-skill relative to medium-skill occupations can be used to capture polarisation at the bottom. These two indicators are used as the dependent variables in the empirical model below (full details are provided in Box 3.2 and in Breemersch et al., 2017).17 In addition, the model relies on a set of proxies for technology and globalisation.

To capture technological change, the model relies on two different variables. First, expenditure on ICT capital services per hour worked is used as an indicator of ICT penetration in the labour market. Goos et al. (2016) report a positive correlation between the intensity of ICT capital use and job polarisation. Second, R&D intensity is used as a proxy for technological change, as commonly done in the literature studying the effects of process and product innovation at firm level on employment changes (e.g. Klette and Forre, 1998). Bogliacino et al. (2012) find that R&D is a good proxy for innovation not only in manufacturing industries but also in service industries, corroborating the strategy adopted here.18

To measure integration in GVCs, the analysis uses data from the trade in value added (TiVA) dataset published by the OECD and the WTO (2015). The data is derived from the 2015 version of OECD Inter-Country Input-Output (ICIO) Database (a description of the dataset is provided in Box 3.1). The main indicator used in the estimation is the share of the foreign component of value added in gross exports by industry and country. A higher share implies that an industry relies more on international specialisation and the international fragmentation of the production process.19 This is a measure of backward participation in GVCs, since the domestic industry is assumed to be in the middle of the global value chain.20

In order to further investigate the effects of international trade on the labour market, a measure of Chinese import penetration in the domestic economy is added to the analysis.21 This is captured by the share of Chinese imports in total industry domestic absorption, calculated on the basis of the WIOD database (Timmer et al., 2015).22 A higher value of this variable indicates greater importance of Chinese goods in overall domestic consumption in a given industry. If Chinese imports compete with domestic output, they may directly lead to job losses in industries that are most exposed causing changes in the relative size of different industries (Keller and Utar, 2016). In addition, the competitive pressures arising from increasing international competition can lead to a shift of production towards more productive firms (Melitz, 2003). If these firms use production processes that make greater use of high-skill workers, this could lead to higher polarisation within industries.

Finally, the analysis includes several country-level indicators to capture the effect of a range of labour market institutions that may mitigate (or amplify) the impact of technology and integration in GVCs on the labour market.23 In particular, the emphasis is placed on employment protection legislation (EPL), union density and the level of the minimum wage.24

Box 3.2. Estimating the effects of technology and globalisation on the labour market

To estimate the effects of technology and globalisation on within-sector polarisation, the empirical strategy rests on two reduced form equations modelling, respectively, the shares of workers in high- and low-paid occupations relative to middle-paid occupations:

where the subscripts i, c and t refer to industry, country and year, respectively. qic captures fixed effects that are specific to each industry in a given country, while jct captures effects that are specific to a certain country in a given time period. eict and hict are idiosyncratic error terms. These specifications permit analysing how the employment structure within industries has on average been affected by the megatrends of interest. All the variables are measured in logarithms to facilitate the interpretation of the results.

The analysis of between-industry polarisation is carried out by estimating the following specification in differences:

where ΔlnEict is the percentage change in the employment of industry i between two periods, while ΔlnICTict, ΔlnR&D intensityict, ΔlnTiVAict, and capture changes in the independent variables. dct is a country x period fixed effect. wict is an idiosyncratic error term.

Technological change has a stronger effect than globalisation on the labour market

The first set of results, reported in Table 3.2, concentrates on manufacturing industries, the sector that has been most heavily affected by labour-saving technologies. In light of previous studies pointing to technology as the main driver of the polarisation process, the first specification in the table includes only the ICT variable while the successive columns progressively add the other variables of interest. The coefficient on the ICT variable is rather stable across specifications, displaying a stronger correlation than globalisation with the extent to which labour markets polarise. The coefficients imply that a 10% change in ICT intensity is correlated with an increase in high-skill employment polarisation of 1.5%. There is no clear evidence, on the other hand, of a correlation between integration in GVCs and polarisation.

Table 3.2. Unpacking polarisation in manufacturing
Explaining polarisation using manufacturing sector data (ISIC two-digit) in the period 1995 to 2007
	(1) top	(2) bottom	(3) top	(4) bottom	(5) top	(6) bottom
	(1) top	(2) bottom	(3) top	(4) bottom	(5) top	(6) bottom	ICT	0.16**	-0.03	0.15**	-0.03	0.15**	-0.03
	(0.06)	(0.06)	(0.06)	(0.06)	(0.06)	(0.06)
R&D intensity			0.04	-0.03	0.04	-0.03
			(0.02)	(0.03)	(0.03)	(0.03)
TiVA					-0.10	-0.02
					(0.12)	(0.26)
Imp.pen^CHN					0.01	0.06
					(0.02)	(0.04)
N	2 496	2 488	2 496	2 488	2 496	2 488
← Standard errors in parentheses. *, , * statistically significant at 1%, 5% and 10% levels respectively.
“ICT” is the ratio of ICT capital services per hour worked. “R&D intensity” is the ratio of research and development expenditure over value added. “TiVA” is the ratio of foreign value added of exports over total exports. “Imp.pen^CHN” is the ratio of Chinese imports over total domestic absorption. Standard errors are clustered at the country level. All sets of analysis include dummies for country by year fixed effects, and also country by industry fixed effects. All the variables are converted to a logarithmic scale. Observations are weighted by the industry share of total employment within each country. Data after 2007 is not included in the analysis due to a lack of ICT intensity observations for a majority of countries.
Source: OECD calculations based on the European Labour Force Survey; labour force surveys for Canada (LFS), Japan (LFS) and the United States (CPS MORG); the World Input-Output Database (WIOD); the Trade in Value Added (TiVA) database; the EU KLEMS growth and productivity accounts; and the OECD Research and Development Statistics database.
https://doi.org/10.1787/888933478183

When the analysis is extended through to 2015 for a subset of countries, these conclusions are largely confirmed, despite the decrease in statistical precision that derives from considerably lower sample sizes (Table 3.A2.1).

And the effects are predominantly on polarisation at the top

Further, while the results show that technology increases top polarisation, they show no significant correlation between technological change and polarisation at the bottom. Industries with higher penetration of ICT tend to have more high-skill workers, but not more low-skill ones, relative to middle-skill workers. This indicates that the effect of technology on polarisation within industries is not only through substitution of middle-skill workers – which would lead to both top and bottom polarisation. Instead, the result suggests that part of the effect of technology might be through complementarities with high-skill jobs, or through substitution of lower-skill workers as well.25 Globalisation also shows little sign of correlating with bottom polarisation. The penetration of Chinese imports is positively correlated with the increase of low-skill jobs relative to middle-skill jobs, but the estimate is statistically insignificant.

The low statistical precision of the trade estimates is not due to the inclusion of the country-specific time trends which are quite demanding of data in this setting. In fact, further analysis (not reported here) fails to pick up significant effects for import penetration even when the country-specific trends are excluded or the effect of import penetration is allowed to change before and after China’s accession to the WTO in 2001. Previous studies have generally emphasised that trade is more likely to affect aggregate polarisation through its impact on the relative size of industries that differ in the intensity of use of middle-skill workers, rather than through within-industry effects (Keller and Utar, 2016). The analysis returns to this point below.

Within non-manufacturing sectors the effects are more difficult to estimate

Turning to the non-manufacturing sector, the analysis reveals that the effects of interest are harder to estimate with statistical precision. The lower number of non-manufacturing industries available in the data results in a sharp drop in sample size, which reduces the statistical precision of the results.26 The analysis uncovers a positive correlation between ICT intensity and polarisation which is larger at the top than at the bottom (similar to the results for manufacturing), but which is not statistically significant in either case (Table 3.3). However, when the analysis is extended to 2015 for a subset of countries where data are available, the effect of technology on top polarisation becomes significant and its magnitude is consistent with the effect uncovered in manufacturing (Table 3.A2.2).

Table 3.3. Unpacking polarisation in non-manufacturing
Explaining polarisation using non-manufacturing sector data (ISIC one-digit) in the period 1995 to 2007
	(1) top	(2) bottom	(3) top	(4) bottom	(5) top	(6) bottom
	(1) top	(2) bottom	(3) top	(4) bottom	(5) top	(6) bottom	ICT	0.11	0.04	0.13	0.12	0.08	0.04
	(0.12)	(0.13)	(0.12)	(0.12)	(0.06)	(0.08)
R&D intensity			-0.07	-0.17*	0.02	-0.05
			(0.06)	(0.10)	(0.03)	(0.04)
TiVA					0.13	0.27*
					(0.14)	(0.14)
Imp.pen^CHN					0.01	-0.04**
					(0.01)	(0.02)
N	1 399	1 399	1 104	1 104	950	950
Standard errors in parentheses. *, , * statistically significant at 1%, 5% and 10% levels respectively.
“ICT” is the ratio of ICT capital services per hour worked. “R&D intensity” is the ratio of research and development expenditure over value added. “TiVA” is the ratio of foreign value added of exports over total exports. “Imp.pen^CHN” is the ratio of Chinese imports over total domestic absorption. Standard errors are clustered at the country level. All sets of analysis include dummies for country by year fixed effects, and also country by industry fixed effects. All the variables are converted to a logarithmic scale. Observations are weighted by the industry share of total employment within each country. Data after 2007 is not included in the analysis due to a lack of ICT intensity observations for a majority of countries.
Source: OECD calculations based on the European Labour Force Survey; labour force surveys for Canada (LFS), Japan (LFS) and the United States (CPS MORG); the World Input-Output Database (WIOD); the Trade in Value Added (TiVA) database; the EU KLEMS growth and productivity accounts; and the OECD Research and Development Statistics database.
https://doi.org/10.1787/888933478198

Both TiVA and import penetration are positively correlated with top polarisation, but neither estimates are statistically significant. On the other hand, the two coefficients differ in sign when looking at bottom polarisation, with TiVA showing a positive effect and import penetration a negative one. Introducing TiVA and Chinese import penetration in the model reduces the size of the coefficients on the proxies for technology, which suggests that some of the effects of trade may occur by inducing technological change.

The effects of technology have been particularly strong in some regions

Focusing on the manufacturing sector, which has displayed the clearest impacts of technology, Table 3.4 investigates whether the correlation of the megatrends with top and bottom polarisation varies across different regions. The first column reports the correlation of each megatrend with polarisation in Western Europe, which is chosen as the reference. The successive columns show the difference in the estimates between each of the other regions and Western Europe, as indicated by the column headings. Hence, the total effect for each of the regions is given by the sum of i) the coefficient in the first column and ii) the coefficient in the region-specific column.

Table 3.4. The impacts of technology and globalisation on polarisation in different regions
Manufacturing sector polarisation in the period 1995 to 2007
	top					bottom
	WE	NA-WE	NE-WE	SE-WE	CE-WE	WE	NA-WE	NE-WE	SE-WE	CE-WE
ICT	0.11**	-0.04	0.20***	0.40***	-0.08	-0.00	-0.11	-0.08	0.32*	-0.08
	(0.05)	(0.05)	(0.05)	(0.13)	(0.09)	(0.08)	(0.08)	(0.23)	(0.16)	(0.12)
R&D	-0.03	0.07*	0.12**	0.17***	0.06	-0.03	0.04	-0.04	-0.01	0.00
	(0.04)	(0.04)	(0.05)	(0.04)	(0.04)	(0.04)	(0.04)	(0.12)	(0.06)	(0.05)
TiVA	-0.27	0.15	0.19	-0.60*	0.64	0.22	-0.15	-0.84	-0.58	-0.24
	(0.18)	(0.20)	(0.40)	(0.33)	(0.37)	(0.39)	(0.39)	(1.13)	(1.30)	(0.62)
Imp.pen^CHN	0.05***	-0.06***	-0.07	-0.07**	-0.06**	0.19**	-0.19**	-0.20**	-0.18	-0.21**
	(0.02)	(0.02)	(0.05)	(0.03)	(0.03)	(0.08)	(0.08)	(0.09)	(0.12)	(0.09)
N	2 353					2 349
Standard errors in parentheses. *, , * statistically significant at 1%, 5% and 10% levels respectively.
“ICT” is the ratio of ICT capital services per hour worked. “R&D” is the ratio of research and development expenditure over value added. “TiVA” is the ratio of foreign value added of exports over total exports. “Imp.pen^CHN” is the ratio of Chinese imports over total domestic absorption. Standard errors are clustered at the country level. Observations are weighted by the industry share of total employment within each country. Both sets of analysis include dummies for country by year fixed effects, and also country by industry fixed effects. “SE” represents Southern Europe and contains Spain, Greece, Italy and Portugal. “WE” represents Western Europe and contains Austria, Belgium, Germany, France, Ireland, the Netherlands and the United Kingdom. “CE” represents Central Europe and contains the Czech Republic, Hungary, the Slovak Republic and Slovenia. “NE” represents Northern Europe and contains Denmark, Finland, Norway and Sweden. “NA” represents North America and consists of Canada and the United States. Results for Japan not reported, as limited data availability reduces the reliability of the estimates. All the variables are converted to a logarithmic scale. Data after 2007 is not included in the analysis due to a lack of ICT intensity observations for a majority of countries.
Source: OECD calculations based on the European Labour Force Survey; labour force surveys for Canada (LFS), Japan (LFS) and the United States (CPS MORG); the World Input-Output Database (WIOD); the Trade in Value Added (TiVA) database; the EU KLEMS growth and productivity accounts; and the OECD Research and Development Statistics database.
https://doi.org/10.1787/888933478202

The results show that ICT tends to increase the share of high-skill relative to middle-skill occupations in all countries, but the effect is significantly larger in Northern and Southern Europe – where a 10% increase in ICT intensity is associated with a 3% and 5% increase in top polarisation respectively.27 Similarly, the correlation of R&D with top polarisation – which in the aggregate results of Table 3.2 appears statistically insignificant – is found positive and significant in these two regions in Table 3.4.

While for TiVA the estimates are generally statistically insignificant across regions, the bottom row reveals that the lack of an overall clear correlation between Chinese import penetration and polarisation documented above masks considerable heterogeneity across regions. In particular, import penetration is correlated with both top and bottom polarisation in Western Europe. In addition, it is worth stressing that these estimates are obtained after controlling for technology adoption and that they indicate a stronger impact on bottom polarisation than on top polarisation (a 10% increase in import penetration is associated with a 2% and 0.5% increase in bottom and top polarisation respectively). Hence, they suggest that import penetration in Western Europe has affected middle-skill occupations directly, rather than by providing incentives for firms to adopt new technologies that may have led to an upskilling of the workforce (Bloom et al., 2016). However, the remaining columns show that similar effects are not detected in any of the other macro regions considered here, as indicated by the sum of the coefficient reported in the first column and those in the remaining columns.

Labour market institutions may influence the effect of technology

Next, the chapter turns to analysing the role of institutions in affecting the impact of the megatrends of interest on the labour market. Breemersch et al. (2017) show that controlling for labour market institutions, such as the strength of trade unions, the strictness of employment protection legislation (EPL) and the minimum wage (measured by the Kaitz index), in a very similar model to the one estimated above does not uncover strong relationships between those variables and polarisation across industries. However, even if these institutions do not have a direct effect on polarisation, they might alter the effect of technology and globalisation on the labour market, although the direction of the effect is theoretically ambiguous.

On the one hand, stricter employment protection and stronger unions might be expected to slow employment adjustments caused by the megatrends considered in this chapter (e.g. Causa et al., 2016 suggest that stronger EPL is effective in protecting low- and middle-skill workers). Under this hypothesis, countries with high EPL and union density can be expected to have lower polarisation, at least temporarily. Similarly, a higher minimum wage can slow the reallocation of employment towards the lower end of the earnings distribution, attenuating the effect of the megatrends on bottom polarisation. This, however, might be achieved at the cost of higher unemployment.

On the other hand, firms might be more likely to use technology to replace workers when facing the rigidities imposed by stricter regulations or stronger unions. Previous literature has shown that the higher costs generated by overly strict labour market regulations can induce firms to increase their capital intensity (e.g. Alesina and Zeira, 2006; Cingano et al., 2015). In addition, it is plausible that even for a given level of capital intensity, firms facing rigidities generated by regulation or unions might be more likely to use technology to replace rather than complement workers. Under this hypothesis, therefore, stricter EPL and stronger unions might be associated with a stronger effect of technology on polarisation.

Table 3.5 reports the estimates of an augmented model of polarisation that includes interactions between the variables of interest and indicators capturing strong institutions (i.e. above median levels of union density, the Kaitz index and the EPL stringency index). Perhaps unsurprisingly, given the limited variability available for estimation once country-specific time trends are included, most coefficients are estimated with little statistical precision.28 However, the interaction of ICT with high EPL attracts positive and statistically significant coefficients in the regressions for both top and bottom polarisation, providing support for the hypothesis that stricter regulations induce firms to use technology to replace workers. Stricter EPL is also associated with a stronger impact of integration in GVCs on top polarisation, but the estimate is statistically insignificant for bottom polarisation. On the contrary, high union density appears to dampen the effect of ICT on bottom polarisation but not on top polarisation.

Overall, therefore, a mixed picture emerges from Table 3.5. While most interactions of interest are estimated with low statistical precision, there is an indication that stricter EPL amplifies the effect of both ICT and GVCs on polarisation, while stronger unions reduce the effect of ICT on bottom polarisation.

Table 3.5. The role of labour market institutions
Manufacturing sector polarisation in the period 1995 to 2007
	(1) Top	(2) Top	(3) Top	(4) Bottom	(5) Bottom	(6) Bottom
Institutions →	Union Den.	Min. Wage	EPL	Union Den.	Min. Wage	EPL
ICT	0.15**	0.16**	0.11*	0.06	-0.03	-0.11
	(0.07)	(0.06)	(0.06)	(0.08)	(0.07)	(0.07)
ICT x Strong institution	0.01	-0.01	0.09*	-0.16***	0.01	0.17***
	(0.04)	(0.03)	(0.05)	(0.05)	(0.05)	(0.05)
R&D	0.04*	0.04	0.04**	-0.02	-0.03	-0.08*
	(0.02)	(0.03)	(0.02)	(0.03)	(0.03)	(0.04)
R&D x Strong institution	0.00	-0.01	-0.01	-0.01	-0.01	0.09*
	(0.02)	(0.02)	(0.03)	(0.03)	(0.03)	(0.05)
TiVA	-0.07	-0.10	-0.27**	-0.01	-0.04	-0.59
	(0.12)	(0.13)	(0.11)	(0.21)	(0.26)	(0.50)
TiVA x Strong institution	-0.04	0.04	0.28**	-0.02	0.12	0.94
	(0.08)	(0.09)	(0.11)	(0.16)	(0.14)	(0.63)
N	2 496	2 496	2 496	2 488	2 488	2 488
Standard errors in parentheses. *, , * statistically significant at 1%, 5% and 10% levels respectively.
“ICT” is the ratio of ICT capital services per hour worked. “R&D” is the ratio of research and development expenditure over value added. “TiVA” is the ratio of foreign value added of exports over total exports. Each column reports the results of a different estimation, where the variables of interest are interacted with a dummy equal to 1 if a particular institution is stronger than the median. Estimating the same model with all the institution dummies and interaction terms in a single regression does not change the conclusions. All sets of analysis include dummies for country by year fixed effects, and also country by industry fixed effects. Standard errors are clustered at the country level. Observations are weighted by the industry share of total employment within each country. Variables with the suffix “x Strong institution” represent data for which the Institution of interest is above the median value. The variable EPL is an index indicator of the employment protection legislation for permanent workers. All the variables are converted to a logarithmic scale. Data after 2007 is not included in the analysis due to a lack of ICT intensity observations for a majority of countries.
Source: OECD calculations based on the European Labour Force Survey; labour force surveys for Canada (LFS), Japan (LFS) and the United States (CPS MORG); the World Input-Output Database (WIOD); the Trade in Value Added (TiVA) database; the EU KLEMS growth and productivity accounts; the OECD Labour Force Statistics Database; and the OECD Research and Development Statistics database.
https://doi.org/10.1787/888933478216

What drives de-industrialisation?

This section investigates the role that technology and globalisation play in fostering the growth and decline of different sectors, as documented in Figure 3.9. In particular, it is crucial to understand to what extent these megatrends have contributed to the process of de-industrialisation that has affected advanced economies, with employment shrinking in the manufacturing sector while growing in industries such as business services, health and social services. As discussed at the beginning of Section 2, these changes have contributed to about a third of the increase in overall polarisation across the countries considered here.

To achieve this objective, the analysis turns to the statistical link between changes in employment by industry and changes in the same variables used to capture technology and globalisation in the previous section. The full empirical specification is detailed in Box 3.2.

Greater technology use is associated with lower employment in manufacturing

Table 3.6 suggests a small negative effect of increased technology use on employment in manufacturing. The coefficients imply that an increase in ICT use of 10% is associated with a fall in employment in manufacturing of 0.5% which is consistent with the hypothesis that new technologies in this sector are to some extent labour replacing. Conversely, no negative effect of technology on employment in service sectors is detected (and the overall impact of ICT penetration on the economy as a whole, estimated when pooling both manufacturing and non-manufacturing sectors together, is negligible). This is in line with existing studies which have generally found no clear negative association between technology adoption and aggregate employment using firm, occupation, industry and individual level data (Bessen, 2015; Graetz and Michaels, 2015; Gaggl and Wright, 2015; Cortes and Salvatori, 2016; Gregory et al., 2016), with the recent exception of Acemoglu and Restrepo (2017) who find large and robust negative effects of robots on employment across commuting zones in the United States. Box 3.3 considers the available evidence about whether automation will become a major driver of job losses in the coming decades.

Table 3.6. What has been driving the fall in manufacturing, and the rise of service sector employment?
Explaining employment growth using manufacturing and non-manufacturing sector data in the period 1995 to 207
	(1)	(2)	(3)	(4)
	Manufacturing		Non-manufacturing
	Δ ln emp	Δ ln emp	Δ ln emp	Δ ln emp
ICT	-0.06*	-0.05*	-0.01	0.01
	(0.03)	(0.03)	(0.02)	(0.03)
		(0.07)		(0.11)
Imp.pen^CHN		-0.02**		0.01
		(0.01)		(0.00)
N	2 619	2 477	1 399	908
Standard errors in parentheses. *, , * statistically significant at 1%, 5% and 10% levels respectively.
“ICT” is the ratio of ICT capital services per hour worked. “Imp.pen^CHN” is the ratio of Chinese imports over total domestic absorption. Not shown are results including controls for the share of foreign value added in total exports and research and development intensity. Standard errors are clustered at the industry level and observations are weighted by the employment share of each industry at the first year of the analysis. The analysis includes dummies for country by year fixed effects. The estimation is based on a regression of annual differences between 1995 and 2007. Δ ln emp captures the change in the log of employment. All the other variables are also converted to a logarithmic scale. Data after 2007 is not included in the analysis due to a lack of ICT intensity observations for a majority of countries.
Source: OECD calculations based on the European Labour Force Survey; labour force surveys for Canada (LFS), Japan (LFS) and the United States (CPS MORG); the World Input-Output Database (WIOD); the Trade in Value Added (TiVA) database; the EU KLEMS growth and productivity accounts; and the OECD Research and Development Statistics database.
https://doi.org/10.1787/888933478220

Box 3.3. The risk of automation in the next 10-20 years

The analysis presented in this chapter relies on historical data and, as such, it is only directly informative about past trends. A complementary body of research focuses on the effects of technological change going forward, building on evidence gathered through foresight exercises. Recent OECD work in this area has concentrated on estimating the share of jobs at medium and high risk of automation. The analysis, detailed in Arntz et al. (2016), builds on previous work by Frey and Osborne (2013), who estimate that almost half of all jobs in the United States are at risk of being substituted by computers or algorithms within the next 10 to 20 years. These estimates are constructed using experts’ assessment of the probability that the main task in a given occupation will be automated. Critics of these alarming estimates argue that occupations as a whole are unlikely to be automated, as each occupation consists of a set of tasks that often differ significantly in their degree of automatibility (Autor and Handel, 2013). Similarly, two workers in the same occupation may not perform the same tasks. For example, if their work is organised differently, one of them may require more face-to-face interaction or autonomy than the other.

An alternative approach to estimate the number of jobs at risk of automation is to directly analyse the task content of individual jobs instead of the average task content within each occupation. This can be done using the OECD Adult Skills Survey (Programme for the International Assessment of Adult Competencies, PIAAC), which has produced a dataset that allows for a detailed breakdown of workers’ tasks. This results in lower figures for the share of jobs at high risk of automation (i.e. those with a probability of being automated of at least 70%) which Arntz et al. (2016) estimate to be 9% across the OECD. The figures for individual countries range from 12% in Austria, Germany and Spain to around 6% in Finland and Estonia (the results are presented in Figure 3.10, which also includes new data from countries in the second PIAAC round).1 A far larger share of jobs (25%), however, is estimated to have a lower risk of automation (50-70%) but a significant risk of seeing the majority of the tasks they entail changed by technology.

The analysis also shows that the tasks most at risk of being substituted by technology are those involving basic exchange of information, buying and selling and simple manual dexterity. On the other hand, occupations that entail creative tasks, those that involve inter-personal relationships and greater socio-emotional skills are at lower risk.

Finally, the risk of automation is particularly severe for workers from the most disadvantaged socio-demographic groups, who are most likely to be in low-skill occupations. The analysis shows that while 40% of workers with a lower secondary degree are in jobs with a high risk of automation, less than 5% of workers with a tertiary degree are. Policy makers should pay particular attention to these differences, as automation could reinforce existing disadvantages faced by some workers.

← 1. Cross-country differences reflect, to some extent, the degree to which technology has already permeated the labour market (Figure 3.2 showed significant heterogeneity in this respect).

The impact of globalisation is less clear cut. On the one hand, the variable measuring GVC integration is never statistically significant (and not reported in the table). Import penetration from China, on the other hand, shows a small negative correlation with employment growth in manufacturing. The coefficient in Table 3.6 implies that a 10% increase in import penetration leads to a slow-down in employment growth of about 0.2%. Further checks not reported here indicate that the statistical significance of this estimate is quite sensitive to modelling choices and in particular to the length of the differences used to compute changes in employment. However, the indication of a negative effect of import penetration from China on employment in manufacturing is consistent with the findings of a number of studies which have applied alternative empirical strategies to data from individual countries, including the United States (see Autor et al., 2016 for a review), Norway (Balsvic, 2015), Spain (Donoso et al. 2014), Germany (Dauth et al., 2014), France (Malgouyres, 2016), and Australia (Tuhin, 2015).

Overall, these results suggest that both technology and trade might have contributed to the between-industry component of job polarisation by slowing down employment growth in manufacturing but not in services. The result on the potential role of trade integration is consistent with that of Keller and Utar (2016) who look explicitly at the effect of import penetration from China on polarisation and conclude that the trade-induced shift of employment from manufacturing to services has contributed significantly to the polarisation of the labour market in Denmark. Similarly, Autor and Dorn (2015) find that rising Chinese import penetration has contributed to the polarisation of the US labour market by reducing employment in manufacturing for non-college workers.

3. Policy options to help workers withstand labour market transformations

The empirical findings in this chapter have important implications for policy. The fact that ICT is found to be an important force behind labour market transformations suggests that renewed efforts are needed to help workers to withstand the disruption caused by the digital revolution, while allowing them to reap the benefits of technological change. Furthermore, the evidence that, at least in some regions, the growth of trade has contributed to the shift of employment from middle- to low-skill jobs demonstrates the need for a policy framework to ensure that the workers affected have access to adequate learning and training opportunities, and receive adequate support to improve their chances of career progression. Effective activation measures, designed in conjunction with social protection, are especially important to ensure that displaced workers can make successful transitions between occupations and sectors. A comprehensive policy strategy to achieve the full potential gains from new technologies and globalisation while assuring that no one is left behind will need to embrace a wide range of economic policy areas, such as innovation, trade and tax policy. It would also need to be tailored to the specific needs of each country. However, the scope of this section is limited to outlining the general policy principles to be considered in the domains of skills, activation and social protection.

Building skills for the future

The existing evidence suggests that some countries may be ill-prepared to embrace the rapid technological transformation brought about by digitalisation. According to the OECD Survey of Adult skills (PIAAC), more than 50% of the adult population on average in 28 OECD countries, can only carry out the simplest set of computer tasks, such as writing an email and browsing the web, or have no ICT skills at all (OECD, 2016b). Only around 30% of workers have the more advanced cognitive skills that enable them to evaluate problems and find solutions using digital technologies (Figure 3.11 and OECD, 2013a). As a result, many workers use ICTs regularly without adequate ICT skills: on average, over 40% of those using software at work every day do not have the skills required to use digital technologies effectively (OECD, 2016c).

Furthermore, Figure 3.11 shows that ICT skill levels differ significantly across countries and age groups. Most importantly, it highlights that while ICT skills among older workers are relatively low in all countries, the competencies of younger workers vary significantly across the OECD. The top four countries (Finland, Sweden, Japan and Denmark) have more than twice as many young people with higher ICT competencies than the bottom four countries (Lithuania, Chile, Greece and Turkey). This raises the prospect of further divergence in these countries’ ability to reap the benefits of technological progress in the future. A comprehensive policy strategy to bridge these gaps should build on four pillars (OECD, 2016b).

First, policy makers should ensure that initial education, including early education, equips all students with basic ICT skills, as well as solid literacy, numeracy, problem-solving abilities, and soft skills (e.g. the ability to communicate, work in teams, lead, self-organise, etc.).29 School curricula should be adapted accordingly, but it is equally important to recognise that many of these skills are acquired outside education and training institutions. This emphasises the need for work-based learning opportunities, which has the advantage of linking training provision to a direct expression of both employers’ requirements and workers’ interests, and to provide soft skills that are not easily taught in a classroom environment. Building a solid system of workplace training poses a number of challenges. First, it rests on reliable mechanisms of quality assurance and on adequate incentives for employers’ engagement. The provision of financial incentives, including direct subsidies, tax breaks and special arrangements to share the burden of training among enterprises, are some of the measures countries adopt to overcome this hurdle. Second, work-based learning options should be attractive enough to potential apprentices, who should be able to afford their direct costs (e.g. tuition fees) and indirect costs (e.g. foregone earnings). Government grants or subsidies can be helpful in this respect, as well as special provisions to give workers the possibility to take leave for training and educational purposes. Finally, effective recognition systems for competencies gained at work and, more generally, outside formal channels are crucial.

Second, education and training systems need to better assess and anticipate changing skill needs in order to adapt curricula and guide students towards choices that lead to good labour market outcomes. Big data can be harnessed to complement existing labour market information systems and monitor changing skill needs (OECD, 2016c). All the relevant stakeholders should be included in skill assessment exercises, to ensure that the information collected is useful and that policies respond to actual needs (OECD, 2016d). The information obtained should be made available to students, workers and employers, to help them make informed decisions about their education, investment and career choices.

Third, even when workers have sufficient skills, inefficient use of such competences, and skills mismatches may result in lower productivity and competitiveness. The use of skills, such as reading and writing, numeracy, problem solving and ICT, varies substantially across countries (OECD, 2016d). A key factor driving this variation is the use of high performance work practices (HPWP) relating both to the way work is organised and to the management practices adopted by firms. More specifically, HPWP involve an emphasis on team work, autonomy, task discretion, mentoring, job rotation and applying new learning. These practices can increase firms’ internal flexibility to adapt job tasks to the skills of new hires, while also promoting a better allocation of the workforce to required tasks. They can also provide incentives for workers to deploy their skills at work more fully through, for instance, bonus pay, training provision and flexibility in working hours. Many countries have taken policy initiatives to promote better skills utilisation through workplace innovation and to foster the skills needed to support these practices. The background to most interventions is the recognition that many firms, if offered expert advice and encouragement to adopt more effective managerial practices, can better utilise existing skills and reap the ensuing productivity gains. Good labour market institutions, such as effective systems of collective bargaining, can also improve skills use at work (OECD, 2016f).

Fourth, the large share of workers with little if no digital skills and, more generally, the increasing need of workers to be able to re-train in the face of structural transformations, stresses the need to scale up and improve the effectiveness of lifelong learning and training for adults, so that workers are better able to keep up with continuously changing skills needs. This entails offering better incentives for workers and firms to re-skill and up‐skill. Training opportunities should be widely available and not necessarily linked to one’s work status or workplace. France recently introduced the Compte personnel d’activité which allows workers to preserve accumulated training rights throughout their careers, even when they switch employer. Indeed, the rise of non-standard work and the diffusion of “on-demand” jobs on digital platforms places increased responsibility on individuals for managing their own skills development (OECD, 2016a). Yet, in the absence of adequate and widely accessible training opportunities, workers may be unable to invest sufficiently in their human capital accumulation, and the problem may be particularly acute among the most disadvantaged groups. Currently, throughout the OECD, low- and medium-skill workers are the least likely to receive training, even though they may be facing the greatest risk of job loss (OECD, 2013a). This is partly the reflection of limited opportunities offered to these groups, and partly the result of lower returns to training which weaken the incentives for workers’ participation. An index of readiness to learn calculated by the OECD in Education at a Glance (OECD, 2016h) shows how the low-skilled are the least well prepared for further participation in learning.30 Low-skill workers also face specificbarriers to participation, including financial constraints. Improving basic skills and removing such barriers is important to avoid exacerbating existing inequalities.

In the process of overhauling lifelong learning, countries should take advantage of the new opportunities digitalisation opens for innovation in learning infrastructure and approaches. MOOCs (massive open online courses) and OERs (open educational resources) are an important new resource, but they remain underutilised. Take-up is low due to the low perceived quality of these forms of learning, lack of incentives and limited recognition of the competencies acquired through these and other non-formal means. To this end, alternative certification methods (e.g. OpenBadge) have begun to appear (ITU, 2014). In addition, a number of technology companies such as Microsoft, CISCO, HP, Samsung, Apple, and Google, offer certificates that MOOC participants can earn directly online (OECD, 2016b). Since learning through MOOCs necessitates basic digital skills, the diffusion and effectiveness of such tools rests crucially on closing existing skill gaps, especially among the most disadvantaged social groups. It also necessitates adequate investment in digital infrastructure to ensure that all workers, including those from poorer backgrounds or living in remote areas, have adequate access to online resources.

Activation and social protection to withstand disruptive change

As the megatrends analysed in this chapter will inevitably generate further disruption in the labour market, it is essential to provide workers who are displaced with a safety net to ensure that they and their families do not fall into poverty, and to provide them with the means necessary to find a new job. The provision of welfare benefits should be designed in conjunction with activation measures to maximise the chance of re-employment and minimise disincentives to work, including in the difficult case of mid-career workers who are displaced by structural economic change and need to switch industry or occupation. As highlighted in recent OECD work, an effective activation framework should: i) motivate jobseekers to actively pursue employment; ii) improve their employability; and iii) expand the set of opportunities for them to be placed and retained in appropriate jobs (OECD, 2015c). As much as possible, activation measures should also be preventive, taking into account ongoing megatrends and the likely risk of job loss in different sectors, and providing workers with adequate information, counselling and re-employment support ahead of their potential displacement (e.g. during the notice period prior to a mass redundancy). Using statistical profiling techniques to provide tailored support on the basis of workers’ characteristics and interests can increase the effectiveness of these measures. Social partners can play an important role in providing adjustment assistance to workers who will be displaced, tailoring the support offered to the specific needs of the affected workers and already beginning to deliver that assistance during the notification period prior to the workers becoming unemployed. That is the case, for instance, in the Job Security Councils in Sweden, which represent one of the most successful examples of re‐employment assistance for displaced workers (OECD, 2015d).

The changes in the occupational structure discussed in this chapter and the process of de-industrialisation have also been accompanied, in a number of countries, by a growing incidence of non-standard forms of work (fixed-term employment, self-employment, part-time). These new ways of working are setting significant challenges for existing social security systems, which are still largely predicated on the assumption of a full-time, regular, open-ended contract with a single employer. As a result of these challenges, large numbers of workers risk falling through the cracks. In most OECD countries, for instance, self-employed workers are not eligible for unemployment benefits (OECD, 2015a). In the European Union, a recent study estimated that 54.5% of the self-employed were at risk of not being entitled to unemployment benefits in 2014, and 37.5% of the self-employed were at risk of not being entitled to sickness benefits (Matsaganis et al., 2016).

Adapting social protection systems to the new world of work will require some crucial reforms. In particular, entitlements should be linked to individuals rather than jobs, and they should be portable from one job to the next. Such an approach should allow workers to transition more smoothly across jobs and sector. In doing so, it should encourage labour mobility, as current arrangements may effectively lock individuals in their existing job out of fear that moving would result in a loss of their entitlements. It could also make independent work more attractive.

A crucial challenge countries will face in trying to set up a sustainable system of social protection is that new forms of work and the rise of self-employment hinder the ability of employment offices to enforce the principle of mutual obligations on unemployment benefit recipients, as it becomes more difficult to monitor work activity. At the same time, the rise of work through digital platforms provides a unique opportunity, albeit still in its infancy, to obtain information on workers’ activity that was not previously available, and overcome the monitoring challenge. Activation might also become more difficult if more frequent interruptions in workers’ careers result in a larger share of the unemployed not being eligible for unemployment benefits and, hence, not being in contact with public employment services. Revising the rules of benefit eligibility to ensure adequate coverage for workers with fragmented work histories and broadening the scope of activation measures beyond the standard link with unemployment benefits will be a step in the right direction.

Another policy option being discussed in some countries is the introduction of a basic income guarantee – i.e. an unconditional income transfer that would replace other forms of public transfers without any means-testing or work requirement. This approach would provide all workers with the basic means to withstand the potential disruptions – e.g. job displacement, unemployment – caused by automation and digitalisation. It would also offer a simpler alternative to the complex mixture of in- and out-of-work benefits, which suffer from the monitoring problems outlined above. However, the costs of such a solution could be very large and its effects on work incentives need to be carefully assessed. On the one hand, if countries aimed to introduce a basic income without reducing existing transfers that are based on specific needs (e.g. disability, child benefits, etc.), its implementation would typically require a large increase in social spending. On the other hand, a basic income that is budget neutral (and thus replaces many of the cash transfers that are currently in place) would typically correspond to an income level below the poverty line, while exposing some of the most vulnerable groups to a higher risk of poverty (OECD, 2017a). In some countries, experiments with different forms of basic income guarantees are currently underway or planned (e.g. Finland; the Canadian Province of Ontario; Oakland [United States]; and several municipalities in the Netherlands). While those schemes differ significantly in their structure, their evaluation might offer some evidence to help judging the usefulness and feasibility of this kind of scheme.

Conclusions

This chapter analyses the impact of technological progress and globalisation on the structure of employment in OECD countries over the past two decades. In particular, it attempts to identify the effects of these two megatrends on job polarisation and the process of de-industrialisation that has characterised most advanced economies. As both of these phenomena may lead to job displacement and rising inequality, a better understanding of their causes has important implications for designing adequate labour market and social policies.

All of the regions considered have experienced a decline in the share of middle-skill, middle-pay jobs relative to that of high-skill and low-skill jobs. The analysis shows that this process of occupational polarisation is linked to but also broader than de‐industrialisation per se. In particular, the reallocation of employment from manufacturing to services accounts for about a third of aggregate polarisation, while changes in the occupational structure within sectors explain the remaining two-thirds.

Of the different megatrends analysed in this chapter, fast technological change displays the strongest association with both polarisation within industries and the shift of employment from manufacturing to services. In particular, growing ICT use is associated with an increase in high-skill relative to middle-skill occupations within manufacturing and with weaker employment growth in the manufacturing sector but not in services.

By contrast, the evidence of an effect of globalisation on polarisation is weaker. Neither the involvement in global value chains nor the penetration of Chinese imports (except for Western Europe) are clearly correlated with increasing polarisation within industries – which accounts for most of the overall polarisation in the economy. On the other hand, tentative evidence supports the hypothesis that increasing import penetration from China has contributed to overall polarisation through a small negative effect on employment growth in manufacturing. This is consistent with the empirical evidence from recent studies that also finds a negative effect of imports from China on employment in manufacturing in a number of advanced countries. Recent work by the OECD offers a more comprehensive analysis of the effects of globalisation, showing that increasing international trade has boosted firms’ productivity and consumers’ welfare, while also imposing a cost on some workers in particular geographical areas and contributing to higher earnings inequalities (OECD, 2017c).

Finally, the chapter finds some tentative evidence that labour market institutions – such as trade unions, minimum wages and the stringency of employment protection legislation – may affect the way trade and globalisation impact the structure of the labour market. In particular, the results suggest that stricter EPL amplifies the effect of both ICT and GVCs on polarisation, while stronger unions reduce the effect of ICT on bottom polarisation.

These results have important implications for public policy. Most importantly, they imply that policy efforts should be concentrated on helping workers to reap the benefits of technological progress and withstand the disruptive changes that globalisation and digitalisation are causing in the labour market. A comprehensive policy strategy should aim to strengthen initial education by fine-tuning education and training curricula in light of changing labour market needs. It should provide incentives for adult learning and remove the obstacles that prevent participation in education and training of the most disadvantaged workers. Recognition of non-formal qualifications obtained outside the education system will also need to play an increasingly important role. Finally, skills policies should be coupled with strengthened activation measures and modern social protection systems that account for the increased fragmentation of working life, so as to foster flexibility and facilitate transitions between jobs.

Further analysis can deepen our understanding of how the labour market is being reshaped and thus help to refine these policy recommendations. First, it will be important to shed light on the impact of other major megatrends on the labour market. For instance, population ageing is associated with changes in both the skills endowment of the workforce and the consumption patterns driving the growth of different sectors, and may therefore play an important role in driving some of the patterns analysed here. Second, while this chapter has focused on the quantity of jobs and their distribution across sectors, the megatrends of interest may also affect job quality and the types of jobs available, especially since these forces contribute to reshaping the content of occupations and the nature of employment relationships. Understanding what drives the emergence of new forms of work will be crucial to designing effective policies to capitalise on the opportunities generated by the new world of work, while ensuring that no worker is left behind.

References

Acemoglu, D. and D. Autor (2011), “Skills, Tasks and Technologies: Implications for Employment and Earnings”, Handbook of Labor Economics, Vol. 4, pp. 1043-1171.

Acemoglu, D. and P. Restrepo (2017), “Secular Stagnation? The Effect of Aging on Economic Growth in the Age of Automation”, NBER Working Paper, No. w23077, National Bureau of Economic Research, Cambridge, United States.

Acharya, R.C. and W. Keller (2009), “Technology Transfer through Imports”, Canadian Journal of Economics/Revue canadienne d’économique, Vol. 42, pp. 1411-1448, https://doi.org/10.1111/j.1540-5982.2009.01550.x.

Alesina, A. and J. Zeira (2006), “Technology and Labor Regulations”, Harvard Institute of Economic Research Discussion Paper, No. 2123.

Amiti, M. and C. Freund (2010), “The Anatomy of China’s Export Growth”, in R.C. Feenstra and S.J. Wei (eds.), China’s Growing Role in World Trade, NBER, Cambridge, United States.

Antonczyk, D., T. DeLeire and B. Fitzenberger (2010), “Polarization and Rising Wage Inequality: Comparing the US and Germany”, IZA Discussion Paper, No. 4842, Bonn.

Arntz, M., T. Gregory and U. Zierahn (2016), “The Risk of Automation for Jobs in OECD Countries: A Comparative Analysis”, OECD Social, Employment and Migration Working Papers, No. 189, OECD Publishing, Paris, https://doi.org/10.1787/5jlz9h56dvq7-en.

Autor, D. (2015), “Why Are There Still So Many Jobs? The History and Future of Workplace Automation”, Journal of Economic Perspectives, Vol. 29, No. 3, pp. 3-30.

Autor, D. (2010), “The Polarization of Job Opportunities in the US Labor Market: Implications for Employment and Earnings”, Center for American Progress and the Hamilton Project.

Autor, D., Dorn and G.H. Hanson (2015), “Untangling Trade and Technology: Evidence from Local Labour Markets”, Economic Journal, Vol. 125, No. 584, pp. 621-646.

Autor, D. and D. Dorn (2013), “The Growth of Low Skill Service Jobs and the Polarization of the U.S. Labor Market”, American Economic Review, Vol. 103, pp. 1553-1597.

Autor, D. and M. Handel (2013), “Putting Tasks to the Test: Human Capital, Job Tasks, and Wages”, Journal of Labor Economics, Vol. 31, No. 2, S59-S96.

Autor, D. and L. Katz (1999), “Changes in the Wage Structure and Earnings Inequality”, in O. Ashenfelter and D. Card (eds.), Handbook of Labor Economics, Vol. 3A, pp. 1463-1555.

Autor, D., D. Dorn and G.H. Hanson (2016), “The China Shock: Learning from Labor-Market Adjustment to Large Changes in Trade”, Annual Review of Economics, Vol. 8, pp. 205-240.

Autor, D., D. Dorn and G.H. Hanson (2013), “The China Syndrome: Local Labor Market Effects of Import Competition in the United States”, American Economic Review, Vol. 103, No. 6, pp. 2121-2168.

Autor, D., L. Katz and M. Kearney (2006), “The Polarization of the US Labor Market”, American Economic Review, Vol. 96, No. 2, pp. 189-194.

Balsvik, R., S. Jensen and K.G. Salvanes (2015), “Made in China, Sold in Norway: Local Labor Market Effects of an Import Shock”, Journal of Public Economics, Vol. 127, pp. 137-144.

Bessen, J. (2015), “How Computer Automation Affects Occupations: Technology, Jobs and Skills”, Law and Economics Research Paper, Boston University School of Law.

Bloom, N., M. Draca and J. Van Reenen (2016), “Trade Induced Technical Change? The Impact of Chinese Imports on Innovation, IT and Productivity”, Review of Economic Studies, Vol. 83, No. 1, pp. 87-117.

Bogliacino, F., M. Piva and M. Vivarelli (2012), “R&D and Employment: An Application of the LSDVC Estimator Using European Microdata”, Economics Letters, Vol. 116, No. 1, pp. 56-59.

Breemersch, K., J.P. Damijan and J. Konings (2017), “Labor Market Polarization in Advanced Countries: Impact of Global Value Chains, Technology, Import Competition from China and Labor Market Institutions”, forthcoming.

Brynjolfsson, E. and A. McAfee (2011), Race Against the Machine: How the Digital Revolution is Accelerating Innovation, Driving Productivity, and Irreversibly Transforming Employment and the Economy, Digital Frontier Press.

Causa, O., M. Hermansen and N. Ruiz (2016), “The Distributional Impact of Structural Reforms”, OECD Economics Department Working Papers, No. 1342, OECD Publishing, Paris, https://doi.org/10.1787/5jln041nkpwc-en.

Cingano, F. et al. (2016), “Employment Protection Legislation, Capital Investment and Access to Credit: Evidence from Italy”, Economic Journal, Vol. 126, pp. 1798-1822, https://doi.org/10.1111/ecoj.12212.

Cortes, G.M. and A. Salvatori (2016), “Delving into the Demand Side: Changes in Workplace Specialization and Job Polarization”, IZA Discussion Paper, No. 10120, Bonn.

Dauth, W., S. Findeisen and J. Suedekum (2014), “The Rise of the East and the Far East: German Labor Markets and Trade Integration”, Journal of the European Economic Association, Vol. 12, No. 6, pp. 1643-1675.

Donoso, V., V. Martín and A. Minondo (2015), “Do Differences in the Exposure to Chinese Imports Lead to Differences in Local Labour Market Outcomes? An Analysis for Spanish Provinces”, Regional Studies, Vol. 49, No. 10, pp. 1746-1764.

Dustmann, C., J. Ludsteck and U. Schönberg (2009), “Revisiting the German Wage Structure”, Quarterly Journal of Economics, Vol. 124, No. 2, pp. 843-881.

Frey, C.B. and M. Osborne (2013), “The Future of Employment: How Susceptible Are Jobs to Computerisation?”, Oxford Martin Programme on Technology and Innovation Working Paper.

Gaggl, P. and G. Wright (2015), “A Short-run View of What Computers Do: Evidence from a UK Tax Incentive”, SKOPE Research Paper, No. 123.

Goos, M. and A. Manning (2007), “Lousy and Lovely Jobs: The Rising Polarization of Work in Britain”, Review of Economics and Statistics, Vol. 89, No. 1, pp. 118-133.

Goos, M., J. Konings and E. Rademakers (2016), “Future of Work in the Digital Age: Evidence from OECD Countries”, Flexibility@work2016.

Goos, M., A. Manning and A. Salomons (2014), “Explaining Job Polarization: Routine-Biased Technological Change and Offshoring”, American Economic Review, Vol. 104, No. 8, pp. 2509-2526.

Goos, M., A. Manning and A. Salomons (2009), “Job Polarization in Europe”, American Economic Review, pp. 58-63.

Graetz, G. and G. Michaels (2015), “Robots at Work”, CEPR Discussion Paper, No. DP10477.

Green, D.A., and B.M. Sand (2015), “Has the Canadian Labour Market Polarized?”, Canadian Journal of Economics/Revue canadienne d’économique, Vol. 48, No. 2, pp. 612-646.

Gregory, T., A. Salomons and U. Zierahn (2016), “Racing with or Against the Machine? Evidence from Europe”, Discussion Paper Series, No. 16-05, Utrecht School of Economics.

ITU – International Telecommunication Union (2014), “Digital Opportunities: Innovative ICT Solutions for Youth Employment”, ITU, Geneva.

Keller, W. and H. Utar (2016), “International Trade and Job Polarization: Evidence at the Worker Level”, Technical Report, No. 22315, National Bureau of Economic Research.

Klette, J. and S.E. Forre (1998), “Innovation and Job Creation Small Open Economy-evidence from Norwegian Manufacturing Plants 1982-92”, Economics of Innovation and New Technology, Vol. 5, No. 2-4, pp. 247-272.

Lopez Gonzalez, J., P. Kowalski and P. Achard (2015), “Trade, Global Value Chains and Wage-income Inequality”, OECD Trade Policy Papers, No. 182, OECD Publishing, Paris, https://doi.org/10.1787/5js009mzrqd4-en.

Malgouyres, C. (2016), “The Impact of Chinese Import Competition on the Local Structure of Employment and Wages: Evidence from France”, Journal of Regional Science, https://doi.org/10.1111/jors.12303.

Maloney, W.F. and C. Molina (2016), “Are Automation and Trade Polarizing Developing Country Labor Markets Too?”, World Bank Policy Research Papers, No. 7922, World Bank Group, Washington, DC.

Marcolin, L., S. Miroudot and M. Squicciarini (2016), “GVCs, Jobs and Routine Content of Occupations”, OECD Trade Policy Papers, No. 187, OECD Publishing, Paris, https://doi.org/10.1787/5jm0mq7kr6s8-en.

Matsaganis, M. et al. (2016), “Non-standard Employment and Access to Social Security Benefits”, European Commission Research Notes, No. 8/2015.

Mazzolari, F. and G. Ragusa (2013), “Spillovers from High-skill Consumption to Low-skill Labor Markets”, Review of Economics and Statistics, Vol. 95, No. 1, pp. 74-86.

McKinsey Global Institute (2017), A Future that Works: Automation, Employment, and Productivity, McKinsey&Company.

Melitz, M.J. (2003), “The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity”, Econometrica, Vol. 71, pp. 695-1725, https://doi.org/10.1111/1468-0262.00467.

Mishel, L., H. Shierholz and J. Schmitt (2013), “Don’t Blame the Robots: Assessing the Job Polarization Explanation of Growing Wage Inequality”, EPI-CEPR Working Paper, November 19.

Mokyr, J., C. Vickers and N. Ziebarth (2015), “The History of Technological Anxiety and the Future of Economic Growth: Is this Time Different?”, Journal of Economic Perspectives, Vol. 29, pp. 31-50.

OECD (2017a), “Basic Income As a Policy Option: Can It Add Up?”, Policy Brief on The Future of Work, OECD Publishing, Paris, May.

OECD (2017b), “How Technology and Globalisation Are Transforming the Labour Market”, Annexes of Chapter 3 of the OECD Employment Outlook 2017, available online at www.oecd.org/employment/oecd-employment-outlook-19991266.htm.

OECD (2017c), OECD Economic Outlook Vol. 2017, Issue 1, No. 101, OECD Publishing, Paris, www.oecd.org/eco/economicoutlook.htm.

OECD (2016a), “New Markets and New Jobs”, OECD Digital Economy Papers, No. 255, OECD Publishing, Paris, https://doi.org/10.1787/5jlwt496h37l-en.

OECD (2016b), “Skills for a Digital World”, Policy Brief on the Future of Work, OECD Publishing, Paris, www.oecd.org/els/emp/Skills-for-a-Digital-World.pdf.

OECD (2016c), “Skills for a Digital World: 2016 Ministerial Meeting on the Digital Economy Background Report”, OECD Digital Economy Papers, No. 250, OECD Publishing, Paris, https://doi.org/10.1787/5jlwz83z3wnw-en.

OECD (2016d), Getting Skills Right: Anticipating and Responding to Changing Skill Needs, OECD Publishing, Paris, https://doi.org/10.1787/9789264252073-en.

OECD (2016f), OECD Employment Outlook 2016, OECD Publishing, Paris, https://doi.org/10.1787/empl_outlook-2016-en.

OECD (2016g), “Global Value Chains and Trade in Value-Added: An Initial Assessment of the Impact on Jobs and Productivity”, OECD Trade Policy Papers, No. 190, OECD Publishing, Paris, https://doi.org/10.1787/5jlvc7sb5s8w-en.

OECD (2016h), Education at a Glance 2016: OECD Indicators, OECD Publishing, Paris, https://doi.org/10.1787/eag-2016-en.

OECD (2015a), In It Together: Why Less Inequality Benefits All, OECD Publishing, Paris, https://doi.org/10.1787/9789264235120-en.

OECD (2015b), The Future of Productivity, OECD Publishing, Paris, https://doi.org/10.1787/9789264248533-en.

OECD (2015c), “Activation Policies for More Inclusive Labour Markets”, OECD Employment Outlook 2015, OECD Publishing, Paris, https://doi.org/10.1787/empl_outlook-2015-7-en.

OECD (2015d), Back to Work: Sweden: Improving Re-employment Prospects of Displaced Workers, OECD Publishing, Paris, https://doi.org/10.1787/9789264246812-en.

OECD (2013a), OECD Skills Outlook 2013: First Results from the Survey of Adult Skills, OECD Publishing, Paris, https://doi.org/10.1787/9789264204256-en.

OECD (2013b), “Measuring Trade in Value Added”, Interconnected Economies: Benefiting from Global Value Chains, OECD Publishing, Paris, https://doi.org/10.1787/9789264189560-4-en.

OCDE (2011), Divided We Stand: Why Inequality Keeps Rising, OECD Publishing, Paris, https://doi.org/10.1787/9789264119536-e.

Oesch, D. (2013), Occupational Change in Europe: How Technology and Education Transform the Job Structure, Oxford University Press.

Oldenski, L. (2014), “Offshoring and the Polarization of the U.S. Labor Market”, Industrial and Labor Relations Review.

Salvatori, A. (2015), “The Anatomy of Job Polarisation in the UK”, IZA Discussion Paper, No. 9193, Bonn.

Timmer, M. et al. (2015), “An Illustrated Guide to the World Input-output Database: The Case of Global Automotive Production”, Review of International Economics, Vol. 23, pp. 575-605.

Tuhin, R. (2015), “Impact of International Trade on Employment: Evidence from Australian Manufacturing Industries”, Research Paper No. 2/2015, Departement of Industry and Science of the Australian Government, Office of the Chief Economist.

World Bank (2016), World Development Report 2016: Digital Dividends, World Bank, Washington, DC, https://doi.org/10.1596/978-1-4648-0671-1.

WTO – World Trade Organization (2015), International Trade Statistics 2015.

Database references

EU KLEMS (2009, 2016), EU KLEMS Growth and Productivity Account, www.euklems.net/.

OECD (2017), OECD.Stat, Business enterprise R-D expenditure by industry; ISIC Rev. 3.1, http://stats.oecd.org/Index.aspx?DataSetCode=BERD_INDUSTRY.

OECD (2017), OECD.Stat, Business enterprise R-D expenditure by industry; ISIC Rev. 4, http://stats.oecd.org/Index.aspx?DataSetCode=BERD_INDUSTRY_ISIC4.

OECD (2017), OECD.Stat, Strictness of employment protection – temporary contracts, Version 1, http://stats.oecd.org/Index.aspx?DataSetCode=EPL_T.

OECD (2017), OECD.Stat, Average annual wages, http://stats.oecd.org/Index.aspx?DataSetCode=AV_AN_WAGE.

OECD (2017), OECD.Stat, Minimum wages at current prices in NCU, http://stats.oecd.org/Index.aspx?DataSet Code=MW_CURP.

OECD (2017), OECD.Stat, Trade union density, http://stats.oecd.org/Index.aspx?DataSetCode=UN_DEN.

OECD-WTO (2015), Trade in value added (TiVA) Database, http://stats.oecd.org/Index.aspx?DataSetCode= TIVA_2016_C1.

World Input Output Database (2013), World input output tables, www.wiod.org/release13.

World Input Output Database (2016), World input output tables, www.wiod.org/release16.

World KLEMS (2012), WORLD KLEMS Canadian productivity data, www.worldklems.net/data.htm

Annex 3.A1. Additional evidence on polarisation

Annex 3.A2. Estimates on selected countries, 2000-15

Table 3.A2.1. Unpacking polarisation, 2000-15, selected countries, manufacturing sector
Explaining polarisation using manufacturing sector data (ISIC two-digit) in the period 2000 to 2015 (selected OECD countries)
	(1) top	(2) bottom	(3) top	(4) bottom	(5) top	(6) bottom
	(1) top	(2) bottom	(3) top	(4) bottom	(5) top	(6) bottom	ICT	0.16*	0.18	0.15	0.22	0.16*	0.22
	(0.09)	(0.12)	(0.09)	(0.13)	(0.09)	(0.13)
R&D intensity			0.06	-0.04	0.06	-0.04
			(0.04)	(0.03)	(0.04)	(0.03)
Imp.pen^CHN					0.05	-0.01
					(0.03)	(0.04)
N	1 159	1 157	1 149	1 147	1 149	1 147
Standard errors in parentheses. *, , * statistically significant at 1%, 5% and 10% levels respectively.
“ICT” is the ratio of ICT capital services per hour worked. “R&D intensity” is the ratio of research and development expenditure over value added. “Imp.pen^CHN” is the ratio of Chinese imports over total domestic absorption. Trade in value added (TiVA) data is only available up to 2011, so it is not included in the above analysis. Countries included in the above analysis are: Austria, Belgium, Finland, France, Germany, Italy, the Netherlands, Spain and the United Kingdom. Standard errors are clustered at the industry level. All sets of analysis include dummies for country by year fixed effects, and also country by industry fixed effects. All the variables are converted to a logarithmic scale. Observations are weighted by industry share of total employment.
Source: OECD calculations based on the European Labour Force Survey; labour force surveys for Canada (LFS), Japan (LFS) and the United States (CPS MORG); the World Input-Output Database (WIOD); the EU KLEMS growth and productivity accounts; and the OECD Research and Development Statistics Database.
https://doi.org/10.1787/888933478230

Table 3.A2.2. Unpacking polarisation, 2000-15, selected countries, services sector
Explaining polarisation using non-manufacturing sector data (ISIC one-digit) in the period 2000 to 2015 (selected OECD countries)
	(1) top	(2) bottom	(3) top	(4) bottom	(5) top	(6) bottom
	(1) top	(2) bottom	(3) top	(4) bottom	(5) top	(6) bottom	ICT	0.11**	-0.08	0.10	-0.11	0.10*	-0.12
	(0.05)	(0.14)	(0.06)	(0.15)	(0.05)	(0.12)
R&D intensity			0.03	0.02	0.03	0.02
			(0.02)	(0.02)	(0.02)	(0.02)
Imp.pen^CHN					0.00	0.01
					(0.02)	(0.03)
N	630	629	560	559	550	549
Standard errors in parentheses. *, , * statistically significant at 1%, 5% and 10% levels respectively.
“ICT” is the ratio of ICT capital services per hour worked. “R&D intensity” is the ratio of research and development expenditure over value added. “Imp.pen^CHN” is the ratio of Chinese imports over total domestic absorption. Trade in value added (TiVA) data is only available up to 2011, so it is not included in the above analysis. Countries included in the above analysis are: Austria, Belgium, Finland, France, Germany, Italy, the Netherlands, Spain and the United Kingdom. Standard errors are clustered at the industry level. All sets of analysis include dummies for country by year fixed effects, and also country by industry fixed effects. All the variables are converted to a logarithmic scale. Observations are weighted by industry share of total employment.
Source: OECD calculations based on the European Labour Force Survey; labour force surveys for Canada (LFS), Japan (LFS) and the United States (CPS MORG); the World Input-Output Database (WIOD); the EU KLEMS growth and productivity accounts; and the OECD Research and Development Statistics Database.
https://doi.org/10.1787/888933478240

Notes

← 1. The statistical data for Israel are supplied by and under the responsibility of the relevant Israeli authorities. The use of such data by the OECD is without prejudice to the status of the Golan Heights, East Jerusalem and Israeli settlements in the West Bank under the terms of international law.

← 2. The analysis covers Austria, Belgium, Canada, the Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Japan, the Netherlands, Norway, Portugal, the Slovak Republic, Slovenia, Spain, Sweden, the United Kingdom and the United States. Country coverage was limited by data availability. Limitations in the EU KLEMS growth and productivity accounts (used to form a measure of ICT intensity) represented the main constraint. ICT intensity data was unavailable for Chile, Iceland, Israel, Mexico, New Zealand, Norway, Switzerland, and Turkey. The analysis of Australia and Korea was limited by the availability of employment level data presented across both occupations and industries.

← 3. Pay has generally been found to be a good proxy for skill levels, at least as captured by education in 3-digits occupations (Acemoglu and Autor, 2011; Green and Sand, 2015). Since the analysis in this chapter uses broad occupational categories (at the 1-digit level), the results are particularly unlikely to be affected by the specific metric used to rank them. In line with previous literature, self-employment is excluded from this analysis and from the data used in the remainder of the chapter.

← 4. The overall pattern of polarisation shown in Figure 3.1 for Central Europe as a whole is the result of a process of occupational upgrading in Hungary and the Czech Republic (where bottom occupations have declined more than all other groups), and of a clearer process of job polarisation in Slovenia and the Slovak Republic.

← 5. China and India together account for the largest technically automatable employment potential in the G20, with more than 700 millionfull-time equivalents between them (McKinsey Global Institute, 2017).

← 6. It is plausible that similar effects might have occurred even in industries that have experienced a relatively modest increase in import penetration, if innovation is pursued by firms to improve efficiency in order to prevent significant breakthroughs by foreign competitors.

← 7. Another way in which trade and technology interacts is that trade facilitate the transfer of technology across different countries (Acharya and Keller, 2009).

← 8. More generally, the type of developments, rate adoption, and modality of use of technology are note entirely exogenous, as they will be driven by choices firms and workers make taking into account a number of factors – including policies. An analysis of these mechanisms is beyond the scope of this chapter, but a better understand of the extent to which policies can influence how technology is used in the labour market is an important topic for future research.

← 9. The choice to focus on exports, as opposed to total production, is driven by data availability in the TiVA dataset.

← 10. It may also result in a disproportionate growth in high-skill occupations (i.e. top polarisation) in advanced economies, which may tend to specialise in in high-skill production and offshore low-skill tasks to less developed countries with lower labour costs (a pattern that would be consistent with the prima-facie evidence shown in Figure 3.1).

← 11. Those effects are also likely to differ substantially between countries in different segments of the GVC, andmost crucially, between those that are already well integrated in GVCs and those on the verge of entering GVCs.

← 12. For Canada, see Green and Sand (2015); for Germany, Antonczyk et al. (2010) and Dustmann et al. (2009); and for the United Kingdom, Salvatori (2015).

← 13. Consistently with this conjecture, there is evidence that clerical workers – the stereotypical victim of computer automation – have seen declining employment shares but strong wage performance in the United States, Canada and the United Kingdom (Autor and Dorn, 2013; Green and Sand, 2015; Salvatori, 2015).

← 14. “Real estate and business services” is also the second fastest growing sector in Japan, which is excluded from Figure 3.8 due to a structural break in the data. The importance of the business services sector highlighted in this analysis is interesting in light of the findings in Cortes and Salvatori (2016). Using British data, they show that changes in occupational specialisation at the firm level are closely linked to aggregate polarisation. They also document that such firm-level changes are entirely driven by firms providing goods and services to other firms, which are disproportionally concentrated in the business services sector.

← 15. The decomposition can be expressed as follows: , where Polari,c captures within-industry polarisation of industry i in country c, and Si,c is the employment share of the industry i relative to total employment in all considered industries in country c.

← 16. For a broader discussion of the potential role of changes that occur on the supply side of the labour market, see Oesch (2013) and Salvatori (2015). Mazzolari and Ragusa (2013) emphasise the role of growing demand for personal services by high-paid high-skill workers in explaining the increasing share of low-skill jobs.

← 17. This approach is preferable to using an overall measure of polarisation (i.e. the ratio of top plus bottom occupations relative to the middle), as the effects of the megatrends of interest may be very different at the top and bottom of the occupational distribution.

← 18. The R&D measure, however, differs from the ICT penetration proxy in some important ways. In particular, R&D expenditures are more likely to capture investment in cutting edge innovation rather than the pace of adoption of an already-available technology. Moreover, R&D investments are risky and might not actually lead to significant innovation. Furthermore, when innovations are achieved, their adoption on a scale sufficient to affect the labour market might require a significant amount of time.

← 19. Data on the foreign component of value added in exports, however, is only available in 1995, 2000, 2005 and 2008-11 in the TiVA dataset. In order to increase sample size, the analysis uses linear interpolation to fill the gaps.

← 20. An alternative would be to use a measure of forward participation, captured by the share of an industry’s exports that is part of foreign exports. In this case, the domestic industry is assumed to be at the beginning of the value chain. In their robustness checks, Breemersch et al. (2017) produce a set of estimates based on thisalternative proxy. They find no significant relationship between GVC forward participation and polarisation.

← 21. This follows the approach in Breemersch et al. (2017) and is in line with recent work on the same topic (see for example Autor et al., 2016).

← 22. Domestic absorption is equal to the domestic consumption of an industry’s goods. It is therefore equal to the country’s home production of a given industry’s goods plus imports of those same goods minus exports.

← 23. This is achieved by including interaction terms between the variables of interest and the strength of institutions.

← 24. The measure of EPL used in the analysis is the stringency of regulation for permanent contracts. The bite of the minimum wage is captured by the Kaitz index, the ratio of the nominal legal minimum wage to the average wage of the working population. As these variables do not vary across sectors, the identification of their effect comes from variation in the data over and above the country-specific time trends, which are included in all regressions.

← 25. It is also worth noting that an analysis using industry-level data such as this one is not well-suited to account for some of the factors that previous studies have singled out as likely drivers of the strong performance of low-skill service occupations in recent decades. These include the increase in the demand for such services by high-skill workers (Mazzolari and Ragusa, 2013) and complementarities in consumption between goods (whose prices are driven down by new technology) and services (Autor and Dorn, 2013).

← 26. For this reason, the specification used in this case controls forcountry, year and industry specific fixed effects separately, rather than by interacting them. This is less demanding of the data. The results do not change significantly.

← 27. For example, the effect for Northern Europe is given by the sum of the 0.11 coefficient in the first column (WE) and the 0.20 coefficient in the third column (NE-WE). For a 10% increase in ICT, this implies an increase in top polarisation by 3%.

← 28. The coefficient on Chinese import penetration, which never attracts significant coefficients is not shown for conciseness.

← 29. Beyond general skills, the design of education and training programmes should pay close attention to the needs of the labour market. For ICT specialist skills, for instance, basic programming is no longer enough. Advanced engineering and experience with machine learning are increasingly important. In addition, ICT specialists also need domain-specific knowledge, given the potential applications of ICT in business, health, education and industry (OECD, 2016b).

← 30. The index is compiled from the Survey of Adult Skills and accounts for the way respondents: relate new ideas to real life; like learning new things; relate to existing knowledge when coming across something new; get to the bottom of difficult things; figure out how different ideas fit together; and look for additional information.