Has wage polarization due to Skill-Biased Technological Change persisted through 2008 – 2014? : a data analysis of nine European countries

(1)

Has wage polarization due to Skill-Biased

Technological Change persisted through

2008 – 2014?

A data analysis of nine European countries

Statement of Originality

This document is written by Student Aron Hammond who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Abstract

Many studies have been done focussing on Skill-Biased Technological Change in the twentieth century. This paper attempts to analyse whether the effects observed between 1980 and 2004 are also observable in more recent years. Following a broad literature review, data on wage inequality and ICT intensity between 2008 and 2014 will be presented and analysed for nine European countries. The analysis serves as a stepping stone to further research, suggesting that the models used to study STBC don't hold up.

Author: Aron Hammond

Student number: 10437215 Supervisor: Nicoleta Ciurilia

(2)

1

Introduction

Over the past decades, there have been some major computing revolutions. Starting with the advent of personal computers in the 80’s, to the internet in the 00’s and now smartphones and other smart devices/appliances. This trend will continue as progress is made in the field of artificial intelligence and Moore’s law keeps progressing. Moore’s law is the observation made by Moore that the number of transistors on a computer chip will double every eighteen months. Because of this, the price decrease of computing power is known for a long time in advance. (Schaller, 1997). It important to study the effects of technological innovation on the economy so that we are well prepared for this new age.

Computers are very good at performing certain tasks, and they are getting better. The tasks they are now best at, are those performed by people in the middle of the skill spectrum. Therefore, these workers' jobs are susceptible to automation (Autor, Levy and Murnane, 2003). This

substitution of labour by capital will leave a certain group of the working population either doing jobs they are overqualified for or unemployable because they are not qualified enough. This job polarization can cause wage inequality, because of ‘hollowing out of the middle' (Oesch and Menes, 2010).

Since the advent of computing in business around 1980, there have been many studies into the effect of technological change on the wage distribution. This research attempts to add to that by giving an overview of these studies and using the model of Michaels, Natraj and van Reenen (2014) to analyse the wage dynamics of 9 European countries from 2008 – 2014. The original paper named ‘Has ICT Polarized Skill Demand’, covered a substantial period. But in 2016 a new release of the EUKLEMS dataset was released, making even more recent data available.

The rest of this text will proceed as follows. In the next section, an overview of the theories around technological progress and wage inequality will be presented. This will start off with skill-biased technological change (SBTC) and why it didn't suffice. After that, alternative formulations of the theory will be presented. These include a theory based on general purpose technologies (Aghion and Howitt, 2002) and a model based on the task composition of jobs (Autor, Levy, and Murnane, 2003; Goos, Manning and Salomons, 2014; Michaels, Natraj and van Reenen, 2014). In the section after that, the data will presented and discussed with respect to the study and findings of Michaels, Natraj and van Reenen (2014).

Literature Review

Over the last decades, there have been many studies concerning the rise in wage inequality. The most cited reasons behind this development are globalization, labour market institutions and

(4)

3

the advent of computer technology. Globalization has led to greater international trade and the possibility to outsource jobs in the lower end of the wage structure to developing countries. The decreased demand for this labour in the developed countries could have contributed to rising wage inequality. Labour market institutions, on the other hand, are names as a force counteracting wage inequality.

Trade openness has been shown to have a significant effect, however, the aim of this research is to evaluate the link between computer technology and wage inequality. So, for the rest of this text, there will be no more mention of the other factors. While reading, one has to keep in mind that the effects described don't affect wage inequality in isolation.

The late 20th_{century has been ridden with technological innovation. This section will discuss} the possible ways in which these innovations could have affected wage inequality. The first theory that was formulated to explain the dynamics between technology and wage inequality is called skill-biased technological change (SBTC). Later, based on criticism by the academic community, the theory was revised into the task based approach. In the rest of the section, this paper will review some alternatives to SBTC. These are routine biased technical change and a technological explanation of wage inequality based on the diffusion of general purpose technologies.

Skill Biased Technological Change

The most widely named factor that can describe wage inequality caused by technology, is Skill-Biased Technological Change (SBTC). It mainly concerns the idea that new technologies affect the wages of high-skilled workers positively and those of low-skilled workers negatively. The theory has come about as an explanation for the fact that since the end of the 1970's, the supply of skilled workers has risen without suppressing the return of education (Acemoglu, 2002). This means that the estimated additional earnings that an extra year of education would earn you have not decreased, while that would have been expected based on conventional knowledge about supply and demand.

Supporting Findings

The theory is first demonstrated by Krueger (1991) by showing that the wages of workers that used computers on the job were higher than those of workers who did not use computers at work. In the same study, he found that this wage differential favoured the highly educated. A regression shows that without a computer use dummy, the return on education increased by 1 point from 1984 to 1989. With the dummy, this increase was 0,6 points. From this result, Krueger (1991) concludes that 40% of the increased return on education can be explained by computer use. In a later paper, Autor, Katz, and Krueger (1997) corroborate that finding (figure 1).

(5)

4

Other studies (Acemoglu, 2002; Autor, Katz and Krueger, 1997) have looked at longer periods, also showing this effect. Both study approximately the same period, (1940's to 1990's). To analyse if the relation between technological change and the increase in wage inequality could stem from a change in the supply or demand of skilled workers, they use a simple supply and demand framework. This framework then forms a basis for predicting the movement of the dependent variable. The papers have different implementations of such a framework, but they are essentially the same. The production factors in this simple economy are high-skilled and low-skilled workers, defined by college educated and high-school educated workers respectively. The supply of these factors is taken as given and is inelastic and the market is competitive. Also, two technology factors are added. One that complements the skilled workers and one that complements the unskilled. A relationship between the change in the skill-premium (ie. the ratio between wages of high over low-skilled workers) and the technology factor can be deduced. That relationship shows that depending on the elasticity of substitution between the two factors, an increase in the relative productivity of skilled workers (a relative increase of the skill-complimentary technology factor) can lead to an increase in the skill premium. The threshold for this is an elasticity higher than 1. Estimates of the elasticity of substitution indicate that it is somewhere between 1 and 2 (Acemoglu, 2002; Autor, Katz and Krueger, 1997). Based on this simple supply and demand model and the observation that both the skill premium and the supply of skilled labour have increased, the researchers conclude that the period after 1960 must have been characterized by SBTC.

There are three main ways in which changes in supply and demand of skilled versus unskilled labour and technological change could lead to wage inequality. These are by a decline in the growth of the supply of skilled labour, while demand and technological progress grow at a steady rate. This is called the steady demand hypothesis. The second is that inequality could have stemmed from a sudden acceleration in the demand for skilled labour, driven by innovation (Acemoglu, 2002; Autor, Katz and Krueger, 1997; Juhn, Murphy and Pierce, 1993). The third is only put forward by Acemoglu (2002) and argues that technological change is an endogenous factor and that it will be skill-biased as long as the market size for the skill-intensive product is large enough. Both papers conclude that while it is not unambiguous and there are more factors in play, the acceleration of the demand for skilled labour is the most likely candidate explanation of the increased inequality observed after 1970. That demand, in turn, is demonstrated to be linked to computer use (Krueger, 1991; Autor, Katz and Krueger, 1997).

In their review of the theory, Card and DiNardo (2002) distinguish between two versions to SBTC, both considered in the same supply and demand framework as described above. The first is the "computer-use-skill-complementarity" version, and the other is the "rising-skill-price"

(6)

5

hypothesis. The first assumes that groups who are more likely to use computers have skills that are complementary with computers and therefore experience a rise in productivity due to progress in computer technology (Autor, Katz and Krueger, 1997). The second assumes that technological change has increased the productivity of skilled workers in general, leading to the higher wage differentials (Juhn, Murphy and Pierce, 1991).

While most of the empirical data they discuss doesn't match up with the predictions of the theory, there are some trends that fit well. Namely, the movement of the skill premium for both men and women are in line with the rising-skill-price hypothesis. In the 1970's the college premium for males declined by 9%, with the mention that female earnings relative to that of men didn't change much over that period (Levy and Murnane, 1992). The college premium for males rose from by 25% from 1979 to 1987. For females, the college premium went up by 21% during this period (Card and DiNardo, 2002). Assuming the rising-skill-price hypotheses and because the returns to education were similar to begin with, one would expect SBTC to increase the college premium for both men and women. Another observation that is supported by this interpretation of the SBTC theory, is the rise in returns to education for women. The return to experience for high-school as well as college educated women rose after 1970 (Card and DiNardo, 2002). The rising-skill-price hypothesis expects this wage gap to widen as a result of technological change because experience is a proxy for skill.

The other version of the theory, computer-skill-complementarity, also has some empirical support. The rising college-high school wage gap for both men and women during the 1980's is supported by the fact that computer use among college-educated workers and college-equivalents (some college experience) was almost 10% higher in 1984. Also, the fact that returns to experience have been flat over the years for high-school educated men is in line with the computer-use-skill-complementarity version of SBTC because there was no real difference between the computer use of older and younger high-school educated men (Card and DiNardo, 2002).

Paper

Rise in skill premium explained by

computer use

Krueger (1991)

40%

Autor, Katz, and Kruger (1997)

30 – 50%

Figure 1

Contradictory Findings

While the original formulation of SBTC as a cause of wage inequality was accepted by many in the academic community, there have also been papers showing its shortcomings (Card and DiNardo, 2002; Sanders and Ter Weel, 2000). This section will discuss some empirical findings regarding the change in some specific wage gaps that are contradictory to what one would expect

(7)

6

based on models of SBTC. First, some educational wage gaps will be considered. After that will come gender, race, and experience.

College high school wage gap

The rise in returns to college for men and women has followed roughly the same path over the 1975-1999 period (Card and DiNardo, 2002). Levy and Murnane (1992) give a more detailed description of relative wages from 1970-1987. In the 1970's the college premium for males declined by 9%. Female earnings relative to that of men didn't change much over that period. The college premium for males rose from by 25% from 1979 to 1987. For females, the college premium went up by 21% during this period. This, together with the fact that men are more likely to use computers that women, is interesting in the light of the two versions of SBTC. The

computer-use-skill-complementarity version suggests that because the relative computer use of college educated over high-school educated workers was higher for men than for women (Card and DiNardo, 2002), the college-high school wage gap should have risen more for men than for women. Since the gap actually moves the same, the theory is inconsistent with the data. On the other hand, the fact that the college-high school wage gap moves the same for both genders is in accordance with the

predictions based on the rising-skill-price hypothesis. Because the gaps are similar, SBTC should have a similar effect on the level of change.

Age and education

In the 1970's, the returns to education were higher for older men. However, as from 1980, the college-high school wage gap for younger men rose steeply while that of the older group did not rise much. The same is observed for women, yet to a lesser extent (Card and DiNardo, 2002). The rising-skill-price hypothesis has little use here since it predicts all wage differentials to widen as a result of technical change. The computer-skill-complementarity version of SBTC has more potential in providing an explanation for the observations. Computer use, measured in 1989, was higher for the younger groups than for the older. Therefore, computer related technology growth could have led to gains in productivity in the former groups. However, the gap in computer use flattened over the 1984-1997 period while the returns to education widened between the age groups. This poses a problem for the theory because neither version can fully explain these trends.

Returns to technical degrees

Evidence for SBTC has been only based on overall wage differentials between college and high school educated workers. But, the advent of technology could also imply a subsequent rise in demand for people with specifically technical skills. Salaries for workers with a technical degree

(8)

7

dropped in the 1980-1986 period, which overall is characterized by a steep increase in wage

inequality (Card and DiNardo, 2002). These observations are inconsistent with both versions of SBTC because (1) computer use among people with technical degrees is relatively high and (2) the returns to those degrees were the highest. So, both versions predict a rise in returns to technical degrees as a consequence of technological change.

Gender

The relative earnings of men over women were stable during the 1970's. Then, during the 1980-1992 period, the gap closed by approximately 15%. After that, the gender-gap was again relatively stable (Card and DiNardo, 2002). The simple interpretation of the rising-skill-price

hypothesis, namely that it widens wage gaps, is inconsistent with the closing gap between 1980 and 1992. However, the fact that computer use among women was higher in that period gives some support to the computer-use-complementarity version of SBTC. But, when the level of education is taken into account the theory falls short again because college educated men were more likely to use a computer than college educated women (Card and DiNardo, 2002).

Race

The dynamics of wage inequality between black and white workers are different from those of gender and the overall wage inequality. In the overall relatively stable 1970’s, the white-black wage gap fell from 28% to 18% for men and from 18% to 4% for women. Then, when overall wage inequality grew sharply and the gender gap was closing, racial wage inequality was relatively stable. This continued throughout the 1990’s (Card and DiNardo, 2002; Neal, 2002).

Although the racial wage gap moved differently, the same arguments apply to it with respect to the validity of the SBTC theories as apply to the gender gap. The difference in computer use between whites and blacks is roughly the same as between men and women (Card and DiNardo, 2002). Therefore, one would have expected the racial wage gap to widen in the 1980’s. Both

versions of SBTC fall short, there must have been other factors counteracting the potential effects of technological change on the racial wage gap.

Experience

For the experience factor in wage, Card and DiNardo (2002) follow the model described by Mincer (1974) that assumes log wages as a separable function of education and potential labour market experience. Labour market experience in this model is age minus education minus six. The distinction between returns experience and the interaction between age and education is that

(9)

8

return to experience can be seen as a return to age separated by education and gender, while the latter is a return to education separated by age and gender.

Levy and Murnane (1992) show that the returns to experience (age) for males increased for all levels of education in the 1970's. Card and DiNardo (2002) show that from the 1980's onwards young (less potential experience) high-school educated males did not show a change in the returns to experience. Young college-educated males with the same potential experience, however, showed an increase in the returns to experience in the 1980's and 1990's. For the college educated males in the middle range of experience the data shows a slight decline in the returns to experience over the same period. These observations, together with the fact that age-related wage differentials for high-school educated men were relatively constant indicates a decrease in college-high high-school wage premium for older relative to younger men. For women, the return to experience increased for all levels of education and experience. This is consistent with the widening of the college-high school wage gap of older versus newer workers (Card and DiNardo, 2002).

This relates to the problems and puzzles regarding the two versions of STBC as follows. Using the rising-skill-price hypothesis one would expect the experience profiles to steepen. For women, this is in line with the data. But for men, there was not a real systematic change in the returns to experience for the high-school educated. These returns even flattened for the college-educated men.

The other version of the theory, computer-skill-complementarity, suggest that the experience groups were computer use is higher, the profiles should steepen more. This only coincides with the movement, or rather stability, of the returns to experience of high-school educated men. The reason being that computer use in that group was relatively flat over the entire period. The movements in the other groups, however, are not easily explained by their respective computer use.

Methodological arguments

Krueger (1991) was the first researcher that used the percentage of people that using a computer on the job as an explanatory variable for the rising wage inequality during the 1980’s. One critique of this approach is that the data is not consistent( Card and DiNardo (2002)). They note that before the data used by Krueger (1991) starts in 1984, there were already specialized word

processors in use and people were already working in offices that contained a mainframe computer. They call upon Bresnahan (1999), who found that by 1971, one-third of American employees worked in an office with mainframe computer access. This makes is hard to compare the changes in

(10)

9

Furthermore, a lot comes down to the proxy used for technological change. Some studies use R&D, some use innovation and others use the use of or investment in computer equipment. This may be problematic because this shows the researchers assume technological change to be skill-biased by nature. This is because these proxies measure only the rate and not the direction (Sanders and Ter Weel, 2000).

Routine-Biased Technical Change

For all the work that has been done on SBTC, the only measured results are correlations. These numbers indicate that the change in computer use in the studied economies moves similar to the change in wage inequality over the same period. In response to this critique on the original theory of SBTC, academics tried to expand the theory to explain some observations that were inconsistent with the theory and provide more insight into the causal relationship between computer technology and the labour market (Autor, Levy and Murnane, 2003; Goos, Manning and Salomons; 2014).

The rise in employment in the lower and high end of the wage structure at the expense of the middle could not be explained by the simple supply and demand models used in the early formulation of the theory. This is because there was no middle-skilled production factor in the formulas. To resolve this, Autor, Levy, and Murnane (2003) introduce a task-based model that uses data from the Dictionary of Occupational Titles (DOT), which classifies job-task requirements for approximately 450 types of occupations. A task is considered ‘routine' if it can be performed by a computer following an explicit, deterministic, set of instructions. Task are considered ‘non-routine' if the rules for performing them are not well enough understood to specify them in the form that a computer can execute (Autor, Levy and Murnane, 2003). By analysing the task composition of jobs, and the investment in computers in those sectors, the researchers found that technological change has shifted demand from routine tasks to non-routine tasks. The assumption here is that computers substitute for routine tasks and complement non-routine tasks. College-educated workers have a comparative advantage in the latter, resulting in wage inequality based on skill level.

The model proposed is based on three assumptions. The first, introduced above, is that computer capital better substitutes routine human labour that its non-routine counterpart. Second, routine and non-routine tasks are imperfect substitutes. And third, increased intensity of routine input complements non-routine input. Given these assumptions, Autor, Levy, and Murnane (2003) derive several equations. The details of these equations are not relevant to understand the implications. These propositions follow from the model:

(11)

10

1. Following a decline in the price of computing, industries that are intensive in routine task input will adopt computers at a faster rate.

2. A decline in the price of computing will increase the demand for non-routine tasks input because routine task input complements non-routine tasks.

Empirical analysis

The empirical trends regarding task-composition seem to support the implications of the models. From 1960 to 1998, the share of employment requiring non-routine mental task rose noticeably. On the flipside, the share of employment requiring manual or cognitive routine tasks decreased over the same period. There is also a shift away from non-routine manual task, which is most pronounced in the 1960's and slows down in subsequent decades (Autor, Levy and Murnane, 2003). For this last observation, the decline in non-routine manual jobs, there is no explanation as to why this might occur even though it is in contrast with the propositions derived from the model. To test the link of these demand shifts with computerization, the researchers perform four different regressions. They split the data into four decades, from the 1960's to the 1990's. The dependent variable is the change in task input of a specific task type in a specific industry. There is one independent variable, namely the change in the number of people using a computer in the same industry. Based on the conceptual model described above, one would expect the two variables to be highly correlated for the decades starting from 1970, but not for the 1960's. This is because the microchip was introduced in the 1970's, making computing accessible and affordable for all industries and occupations.

As expected, a strong relationship is found between an increase in computer use and the employment of non-routine analytical and interactive tasks. For example, during the 1990 – 1998 period, the test shows that an increase of 10 percent in computer use corresponds with a 1.2 centile annualized increase in the employment of labour performing non-routine analytical task. The other post-computer decades are also highly significant. And, like expected, the regression for the 1960’s doesn’t give any significant results for the relationship between computerization and a change in task input. This supports the hypothesis that computerization played a role in the shifting demand for tasks between 1970 and 1998 (Autor, Levy and Murnane, 2003).

This research gives more insight into the relationship between computerization and the observed changes in the labour market during the period of last three decades of the twentieth century. It shows that computer technology affects the demand for specific sorts of task and their corresponding level of skill. Computer technology substitutes for routine labour (manual and cognitive), which is associated with low-skill and it complements non-routine labour in which high-skilled workers have comparative advantage. The developed model and insights were also used in

(12)

11

later studies, that look more at the implication for the wage structure following such demand shifts (Goos, Manning, and Salomons, 2014; Michaels, Natraj and van Reenen, 2014).

General Purpose Technologies

The view that technological growth is biased towards skill doesn't suffice. Aghion and Howitt (2002) put forward views that are not mentioned in the conventional literature on STBC. They start off by describing some questions the theory raises for which it has no easy answer. After that, they explain other ways that wage inequality between skill groups can have come about. Finally, they make the argument that technological progress is not biased towards skill but adaptability. People who quickly adapt a new technology have a short term advantage over others.

A general-purpose technology is a fundamental technological breakthrough that

dramatically affects the entire economic system (Aghion and Hewitt, 2002). The GPT that has been responsible for the wage dynamics in the last decades of the twentieth century has been ICT and namely the widespread adaptation of computers and the advent of the internet. The way in which GPTs spread though the economy can be used to explain the rise in returns to skill. Unlike the theories of SBTC, this explanation also accounts for the stagnant production that is otherwise expected as a result of the increased productivity due to technological progress.

One of the alternative explanations of the rising skill premium comes from Acemoglu (1998). He states that a change in direction of technological change could be responsible and calls his theory the market size effect. To understand this, suppose that there is a market in which the final output is produced by two intermediate products. One is operated by highly educated workers and the other by high-school graduates. Technological progress will increase productivity of either of these inputs, based on the effort that is put in the innovation of the two. A decision has to be made as to where to focus this effort. When the relative supply of college workers rises, so will the skill premium.

However, this short run effect can possibly be offset by the net result of two opposing effects (depending on the elasticity between the two intermediate products). The negative effect is that because the skill premium is lower, the relative cost per unit that can be saved by improving the technology that goes into the skill-intensive product is lower. The positive effect is that because the cost of the skill-intensive product is lower, the amount produced will be higher. So, the cost saved per unit will be earned over a larger volume. The level of innovation in the economy is assumed constant. It is just the direction towards which the effort is focused that changes (Acemoglu, 1998). In their book, Aghion and Howitt (1998, ch.9) describe a model that exemplifies the market size effect. One distinction is, however, that they take into account that R&D (innovation) is a skill-intensive activity. So, an increase in skilled workers will accelerate the technological progress and thus accelerate the rise in wage inequality.

(13)

12

Papers arguing for SBTC all take human capital as a factor of production of goods. This, in combination with perfectly competitive markets, would dictate that the relative supply of skills cannot go together with an increase in the returns to skill since factor-demand schedules are downward-sloping. Since this did happen, there must have been an increase in the productivity on the margin of the skills of college-educated workers, relative to those of high-school workers (for any given level of employment). When you take into account Schumpeterian growth theory, which is built on the idea that human capital can also be used for the production of ideas, human capital becomes more involved (Aghion and Howitt, 2002). For instance, human capital is responsible for R&D (the production of innovation). Moreover, human capital can be viewed as more than just a production factor, but also a factor for generating and implementing technological change (Nelson and Phelps, 1966). Other sources of technological progress are learning by doing and absorption and diffusion of a recent innovation (Aghion and Howitt, 2002). People with a higher skill level have a comparative advantage in all of these.

In light of this idea, a body of literature has appeared based on the Nelson-Phelps

framework of Schumpeterian theory. In short, it implies that one can show that all that is required is an increase in the value of skilled labour in the technology for producing ideas. This is because the production of ideas uses skilled labour more intensively. An example is an increase in productivity in R&D (which is intensive) will raise the skill premium, even if the R&D activity generates

skill-unbiased innovation (Aghion and Howitt, 2002). Dinopoulos and Segerstrom (1999) use this idea to

argue that trade could be more heavily responsible for wage inequality than normally thought. Because the liberalization of trade grows the potential market for an innovator, it increases the returns of R&D and consequently the amount of R&D that will be undertaken. Thus, because R&D is skill intensive, trade liberalization will increase the skill premium and wage inequality.

When a GPT is introduced, it diffuses and spurs a wave of secondary innovations that are based on the original technology. An example is the development of a word processor after the introduction of the desktop computer. This diffusion increases the demand for skill in the sector that adopts the GPT because adaptation and experimentation require of new technology require skill (Aghion and Howitt, 2002). This explains the rising skill premium as a result of the introduction of a GPT. To come to the conclusion that adaptability (not skill) is rewarded with a premium following the introduction of a GPT, and in this case ICT, Aghion and Howitt (2002) make the assumption that in the long run, technology substitutes for skill. But, the point is that in the short run, some pick up the technology faster than others and are rewarded for that adaptability and that is where the observed wage inequality results from.

(14)

13

There remains one puzzle concerning the GPT based approach to wage inequality. This being that since the spread of a new GPT takes place over a long period of time, one wouldn't expect it to have quantitatively significant macroeconomic consequences (Aghion and Howitt, 2002). The initial disruption would, of course, be significant. However, the following diffusion and its associated costs are spread out over decades, at a uniform rate. The paper sums up several factors that suggest a more fitting view, namely that diffusion is gradual, but not uniform. One factor could be the existence of strategic complementarities between certain sectors that create lock-in effects. Also, exogenous factors could be at play that give a head start to some sectors before others decide to adopt the GPT. Examples of these exogenous factors are a continuous increase in real labour costs, trade liberalization, an intensification in competition or a sharp increase in skilled labour supply.

Data analysis

As concluded by many researchers, there exist a relationship between technological change and rising wage inequality. However, the periods examined by those researchers concentrate around the 1980's. The prevalent technological change at the time was computerization of the workplace. The way that computerization affects the wage distribution is by substituting for routine cognitive and routine manual labour. The falling price of computing power makes it more attractive to use computers for those tasks, instead of human labour which is now relatively more expensive. In the following section, some data regarding ICT investment and the distribution of wages in 9 European countries will be shown and discussed.

Data

For the analysis of western European countries and the relationship between ICT investment and wage-bill shares of three skill groups, Michaels, Natraj and van Reenen (2014) use data from the EUKLEMS database. In their analysis, they look at the period between 1980 and 2004. Because more recent data is now available, this section will look at the period between 2008 and 2014. The data for wage-bill shares per skill group are limited to these 6 years because there was a methodological change from the previous releases. This makes the releases not directly comparable (Jäger, 2016). For that reason, the analysis will be limited to 2008 – 2014.

The dataset consists of a collection of economic variables for ten major European

economies. These include Austria, Belgium, Germany, Finland, France, Italy, the Netherlands, Spain, Sweden and the UK. The data are collected from each country its national statistical institute (NSI) and are disaggregated into industries, by their two-digit ISIC codes. For this analysis, the relevant economic variables are ICT capital stock and wage bill shares per skill group. Because of the unavailability of wage bill shares, Belgium is not covered in this study.

(15)

14

The variables of interest are namely the wage bill shares of three educational groups, and the intensity of the investment in ICT. The educational groups are university graduates, workers with some education and workers with no formal education. These three groups shall be referenced as high, middle and low educated workers respectively. Because the EUKLEMS database reports wage bill shares disaggregated to education, age, and sex groups, I have aggregated these back to the level of education. The ICT investment measure comes directly from the database and is represented by the ratio of ICT capital compensation over total value added of the industry or aggregate economy.

Descriptive statistics

The 2016 release of the EUKLEMS dataset covers 9 European economies, with data over 7 years. To analyse all these economies would be beyond the scope of this text. Therefore, some summarizing statistics will be presented. Based on those statistics, the analysis will zoom into two economies and do a more detailed comparison.

Figure 2 shows the change in wage bill shares for the three educational categories for each country from 2008 to 2014. For all 9 economies, the wage bill share awarded to the highly educated rose. Because of the way the wage bill shares are constructed, any change in one segment is offset by an equal opposite change in the other two segments. In the Netherlands, Finland, Spain and France, the rise in the top of the distribution was due to a decline in the bottom. This is expected, based on the basic theory of SBTC. However, the more recent articulations of the theory explain the divergence of wages to a hollowing out of the middle (Michaels, Natraj and van Reenen, 2014). In other countries, like Austria, the increase in wage bill share of highly educated workers was offset by a decline of that of the middle. This is in line with the theory of routine biased technical change, that predicts computerization will substitute for routine jobs done by workers with intermediate

education (Autor, Levy and Murnane, 2003). In the remaining economies, the movement is less unilateral.

EDU

NLD

FIN

AUT

ESP

ITA

SWE

DEU

FRA

GBR

HIGH

0,0146 0,0651 0,0833 0,0944 0,0289 0,0733 0,0057 0,0739 0,1077

MEDIUM -0,0013 -0,0042 -0,0850 -0,0138 0,0210 -0,0652 0,0053 -0,0043 -0,1474

LOW

-0,0133 -0,0609 0,0017 -0,0806 -0,0499 -0,0081 -0,0110 -0,0696 0,0397

Figure 2: Percentage changes of the wage share of each educational group between 2008 and 2014

Because the hypothesis to be corroborated is that computerization causes wages to diverge, the observed changes in the wage distribution have to be compared to the level of computerization in the respective economy. For this measure, Michaels, Natraj and van Reenen (2014) use a variable provided in the EUKLEMS database. This variable is the contribution of ICT capital services to total

(16)

15

value added. This variable, from now on referred to as ICT intensity, will be compared to the numbers in figure 2. In figure 3, you can see the change in ICT intensity from 2008 to 2014.

Figure 3: Percentage change of the contribution of ICT to value added for the total economy

The expectation is that those countries that show an increase in ICT intensity show a decline in wage bill shares of middle educated workers, and vice-versa. Furthermore, the change should be offset by an opposite change in the sum of the top and the bottom. Countries that fit this expectation are Austria, Italy, Germany and the UK. Furthermore, while the direction of these changes is as

expected, the magnitude of the change in ICT intensity doesn't seem to translate into an equivalent change in the wage distribution for these countries. The movement of the wage structure in the other countries cannot easily be explained by the theories covered in this research.

NLD FIN AUT ESP ITA SWE DEU FRA GBR

HIGH 0,45 -0,64 -0,64 -0,83 0,77 -0,62 -0,04 -0,14 -0,40

MIDDLE 0,36 -0,67 0,59 -0,48 -0,90 0,01 0,06 0,11 -0,18

LOW -0,34 0,98 0,38 0,85 0,73 0,28 -0,09 -0,03 0,31

Figure 4: Correlation of yearly changes of the wage distribution and yearly changes of ICT intensity

To get some more insight into the relationship between ICT intensity and wage bill shares, figure 4 shows a table of the correlations of each skill group their change in wage share and the change in ICT intensity. Like stated above, the expectation is that the wage bill share to middle educated workers decreases, as ICT intensity increases. Therefore, negative correlation values for the middle in combination with positive values for the top and bottom, give support to the hypothesis that the theory of routine biased technical change still holds for the period of 2008 – 2014. Italy is the one country that fits this pattern. Overall, the middle of the wage distribution shows a negative trendline when plotted against the change in ICT intensity. The high end of the distribution shows a flat trend line in the same context. For the low end of the distribution, the trend line shows that it is positively correlated with a change in ICT intensity (see figures 5-7).

Overall, the data doesn't seem to be very supportive of the presence routine biased technological change in Europe from 2008 – 2014. However, this overview was based on the aggregate economy of each country. Where there was a significant relationship found for earlier periods, the researchers used industry aggregates. Also, they incorporated non-ICT capital contribution to value added in their model.

∆ICT/VA

NLD

FIN

AUT

ESP

ITA

SWE

DEU

FRA

GBR

(17)

16

Figure 5: Trendline ICT investment and highly educated wage share 2008 - 2014

Figure 6: Trendline ICT investment and middle educated wage share 2008 - 2014

Figure 7: Trendline ICT investment and low educated wage share 2008 - 2014 0.000 0.020 0.040 0.060 0.080 0.100 0.120 -0.400 -0.300 -0.200 -0.100 0.000 0.100 Δwa ge -b ill s ha re h igh e du ca tio n ΔICT/VA

High education Linear (High education)

-0.160 -0.140 -0.120 -0.100 -0.080 -0.060 -0.040 -0.020 0.000 0.020 0.040 -0.400 -0.300 -0.200 -0.100 0.000 0.100 Δwa ge -b ill s ha re m id dl e ed uca tio n ΔICT/VA

Middle education Linear (Middle education)

-0.100 -0.080 -0.060 -0.040 -0.020 0.000 0.020 0.040 0.060 -0.400 -0.300 -0.200 -0.100 0.000 0.100 Δwa ge -b ill s ha re lo w e du ca tio n ΔICT/VA

(18)

17

Conclusion

Because of the amount of research that finds significant results for the presence of a bias of technological change towards facilitating relatively more high- and low-skilled work, at the expense of the middle, one has to accept that this is probably an existing relation. However, previous research has focussed mainly on the 1980's. In that decade, there was a rapid growth of computer use. When applied to other eras of technological change, these significant effects are not

unambiguously found. Furthermore, in more recent years, computers and other ICT equipment are already widespread throughout the economy. The question whether the bias of technological change is still present in the current decade remains unanswered. This study, however, tentatively suggests that this may not be the case. Proper qualitative research has to be done to draw any proper conclusions. One possible endevour that would benefit the field, would be to to an industry level analysis of the same data discussed in this analysis.

(19)

18

Sources

Acemoglu, D. (1998). Why do new technologies complement skills? Directed

technical change and wage inequality. Quarterly journal of economics, 1055-1089.

Acemoglu, D. (2002). Technical change, inequality, and the labor market. Journal of

economic literature, 40(1), 7-72.

Aghion, P., Howitt, P., Brant-Collett, M., & García-Peñalosa, C. (1998). Endogenous

growth theory. MIT press.

Aghion, P., & Howitt, P. (2002). Wage inequality and the new economy. Oxford

Review of Economic Policy, 18(3), 306-323.

Autor, D.H., Katz, L. F., & Krueger, A. B. (1997). Computing inequality: have

computers changed the labor market? (No. w5956). National Bureau of Economic Research.

Autor, D.H., Levy, F., & Murnane, R. J. (1992). US earnings levels and earnings

inequality: A review of recent trends and proposed explanations. Journal of economic

literature, 30(3), 1333-1381.

Bresnahan, T. F. (1999). Computerisation and wage dispersion: an analytical

reinterpretation. The Economic Journal, 109(456), 390-415.

Card, D., & DiNardo, J. E. (2002). Skill biased technological change and rising wage

inequality: some problems and puzzles (No. w8769). National Bureau of Economic Research.

Dinopoulos, E., & Segerstrom, P. (1999). A Schumpeterian model of protection and

relative wages. American Economic Review, 450-472.

Goldin, C., & Katz, L. F. (1999). The returns to skill across the twentieth century

United States. NBER Working paper, 7126, 27-28.

Goos, M., Manning, A., & Salomons, A. (2014). Explaining job polarization:

Routine-biased technological change and offshoring. The American Economic Review, 104(8),

2509-2526.

Greenwood, J., & Yorukoglu, M. (1997, June). 1974. In Carnegie-Rochester

Conference Series on Public Policy (Vol. 46, pp. 49-95). North-Holland.

Juhn, Chinhui, Kevin M. Murphy, and Brooks Pierce. "Accounting for the slowdown in

black-white wage convergence." Workers and their wages (1991): 107-43.

Katz, L. F. (1999). Changes in the wage structure and earnings inequality. Handbook

of labor economics, 3, 1463-1555.

Krueger, A. B. (1991). How computers have changed the wage structure: Evidence

from microdata, 1984-1989 (No. w3858). National Bureau of Economic Research.

(20)

19

Levy, F., & Murnane, R. J. (2003). The skill content of recent technological change: An

empirical exploration. The Quarterly journal of economics, 118(4), 1279-1333.

Mincer, J. (1974). Schooling, Experience, and Earnings. Human Behavior & Social

Institutions No. 2.

Michaels, G., Natraj, A., & Van Reenen, J. (2014). Has ICT polarized skill demand?

Evidence from eleven countries over twenty-five years. Review of Economics and

Statistics, 96(1), 60-77.

Neal, Derek. The measured black-white wage gap among women is too small. No.

w9133. National Bureau of Economic Research, 2002.

Nelson, R. R., & Phelps, E. S. (1966). Investment in humans, technological diffusion,

and economic growth. The American economic review, 56(1/2), 69-75.

Oesch, D., & Menes, J. R. (2010). Upgrading or polarization? Occupational change in Britain, Germany, Spain and Switzerland, 1990–2008. Socio-Economic Review, mwq029.

Sanders, M., & Ter Weel, B. (2000). Skill-biased technical change: Theoretical concepts, empirical problems and a survey of the evidence.

Has wage polarization due to Skill-Biased Technological Change persisted through 2008 – 2014? : a data analysis of nine European countries