OECD Employment Outlook 2020

Worker Security and the COVID-19 Crisis

Report

OECD Employment Outlook

7 July 2020

Available in:

English
français

Download PDF

4. What is happening to middle-skill workers?

Abstract

Driven by mega trends such as automation, ageing and globalisation, the share of middle-skilled jobs has been declining in the majority of OECD labour markets (a process also referred to as job polarisation). Middle-skill jobs are defined as occupations in the middle of the occupation-wage distribution. One little explored question is what is happening to the workers who have traditionally occupied these jobs? This chapter starts by examining whether the fall in the share of middle-skill employment is explained primarily by attrition or transitions. Attrition accounts for fewer younger workers entering these jobs compared to older workers retiring. Transitions explain changes in career patterns after a person has started working. The chapter then studies the characteristics of what would have been a “typical” middle-skill worker and uses this profile to examine how the jobs they hold have changed over time.

In Brief

The share of middle-skill jobs – defined as occupations whose average wages place them in the middle of the wage distribution – declined in OECD countries over the past two decades. From the mid‑1990s (1994‑1996) to the latest available period (2016‑2018), the share of total employment accounted for by middle-skill occupations – e.g. truck drivers and machine operators for men, cashiers and secretaries for women – declined over 11 percentage points. This contrasts with growth of 9 percentage points in high-skill occupations and of 3 percentage points in low skill occupations. The employment share of middle-skill occupations declined because the number of jobs in low- and particularly high‑skill occupations grew strongly, while the number of middle-skill jobs held broadly steady.

The key question addressed in this chapter is how this decline in employment shares of middle-skill occupations has taken place. Is it due to attrition or transitions? This difference amounts to whether the decline is borne by new entrants to the labour market, or mid-career workers. Attrition is primarily driven by new cohorts of workers entering the labour force in middle-skill occupations at lower rates than previous cohorts. Transitions are changes in the career patterns of different occupational groups, in which workers reallocate to other skill groups (including possible spells of non-employment) during their careers at different rates than in the past. Transitions could include earlier cohorts who started their working lives in middle-skill jobs being laid off mid-career, for example, while attrition could take the form of young workers entering the labour market in different occupations. Attrition is about labour market entry, while transitions account for mid-career changes.

The main findings of the chapter are as follows:

In contrast to popular perceptions (and anecdotal evidence concerning mass dismissals), this chapter shows, using data primarily from European OECD countries, that the share of middle‑skill employment has declined more because of attrition than transitions. Changing patterns of labour market entry appear to be key. For cohorts born before 1970, some 32.8% of workers were employed in middle-skill occupations when aged 25‑29. For cohorts born after 1970, this share decreased to 26.5%. The share of those in high-skill employment exhibited the reverse pattern.
Meanwhile, except during the financial crisis, workers separating from middle-skill jobs have tended to transition to other middle-skill jobs or to non-employment at similar rates as in the past.
The changing demographic composition of younger cohorts entering the labour force is a contributing factor to the decline in middle-skill employment. New entrants to the labour force are more likely to be women and have a tertiary education compared with 20 years ago. Workers without a tertiary education were the most likely to hold middle-skill jobs in the past. There are therefore fewer workers entering the labour force who would typically hold middle-skill jobs, which would be expected to cause a mechanical decline in the share of middle-skill employment. However, the analysis confirms that this explains only part of the contraction in middle-skill employment shares.
New cohorts of workers without tertiary education are less likely to start their careers in middle‑skill jobs. In fact, even those who would have been regarded as “typical” middle-skill workers in the past are now much less likely to start working in middle-skill jobs, and more likely to be in low‑skill employment. A formal decomposition of changes in employment shares confirms that attrition accounts for most of the decline in middle-skill employment.
Employment shares of women without a tertiary degree in low-skill occupations have increased dramatically. The largest percentage point increases for those holding low-skill jobs have been among men and women with a middle level of education (an upper‑secondary qualification rather than a tertiary degree).
In most countries, middle-educated workers are now also less likely to hold high-skill jobs than in the 1990s. A few countries appear to be exceptions, however. Over the past two decades, Sweden, Germany, Norway and Denmark have seen a significant rise in the propensity of middle-educated workers to be employed in high-skill occupations, for both men and women. Although not addressed explicitly in this chapter, these countries place an emphasis on vocational education and training, as well as maintaining a tradition of cooperative social dialogue (see also Chapter 5).

Introduction

At its peak, manufacturing employed millions of workers at wages solidly in the middle of the pay distribution, which helped support a strong middle class across industrialised nations (Helper, Krueger and Wial, 2012[1]; OECD, 2019[2]). However, since at least the 1970s, employment in OECD economies has been shifting from manufacturing to service industries. At the same time, the share of employment in middle-skill occupations within industries has steadily decreased. While these jobs declined, employment grew in high‑skill occupations such as human resources administrators and information technology support. Low‑skill service jobs such as janitors, home care workers and retail sales assistants flourished as well. Employment moved from assembling cars on the shop floor to stocking shelves on the sales floor.

Economists and policy makers termed this trend job polarisation. Defining the skill level of jobs by the average wage in an occupation (similarly using task content, or education), economists found that the employment shares of both higher- and lower-skill occupations have increased in many but not all countries, while shares of employment in middle-skill occupations have declined (Autor, Levy and Murnane, 2003[3]; Goos and Manning, 2007[4]; Goos, Manning and Salomons, 2009[5]). Subsequent research identified automation as the main cause of this job polarisation (OECD, 2017[6]; Autor and Dorn, 2013[7]).1 Increasing penetration of information technology and robotics have eroded jobs consisting of routine tasks. These routine jobs were traditionally situated in the middle of the skill distribution.

A common concern is that the decline of middle-skill employment may have resulted in distress, job insecurity and displacement for workers who held those jobs – see e.g. Autor (2010[8]), Cortes (2016[9]), and OECD (2017[10]). Yet this need not be the case. If job polarisation results in workers transitioning to higher-paid jobs, its effects may be more benign than previously thought. Moreover, the theory of job polarisation and the history of job destruction is ambiguous with respect to overall employment and wages in the long run (Autor, 2015[11]; Autor, 2015[12]; Acemoglu and Restrepo, 2018[13]). The important question is then how are middle-skill workers who need to transition to new jobs affected by job polarisation?

This chapter reviews where workers holding middle-skill jobs are going in the face of the shrinking share of such jobs. The first question it seeks to answer is how this adjustment has taken place. Are firms increasingly dismissing middle-skill workers, forcing mid-career workers to find new employment in other skill groups (“transitions”), or are older middle-skill workers gradually retiring and younger workers entering other, growing occupations (“attrition”)? In addition to the adjustment mechanism, this chapter asks what types of jobs workers who fit the traditional profile of a middle-skill worker are taking.

Understanding both the nature of adjustment and the final destinations of middle-skill workers will help inform policy makers on the types of policies that can aid this restructuring of OECD labour markets. For example, if job polarisation is driven mainly by workers losing middle-skill jobs, then labour market policy needs to focus on helping these workers make the transition to other occupations where employment opportunities are emerging (OECD, 2018[14]). By contrast, if patterns of labour market entry are the key factor shaping this process, then policies need to accompany young workers in starting their career and ensuring its sustainability over time (see e.g. Chapter 5).

The analysis begins by briefly documenting the near universal trend of job polarisation, and disentangling the dynamics of the shifting share of jobs across occupation groups (Section 4.1 and Section 4.2). This includes evidence for whether middle-skill employment shares have shrunk due to transitions of mid-career workers, or through attrition and differences in the labour market entry patterns of younger cohorts. In Section 4.3 the chapter turns to building the profile of the “typical” middle‑skill worker of the past and identifies the characteristics associated with middle‑skill work two decades ago. Section 4.4 uses the profile of middle-skill workers to shed light on where they are going. Specifically, the analysis shows the types of occupations that workers of different demographic groups are employed in compared with two decades ago.

4.1. How are middle‑skill jobs changing?

In order to better assess how middle-skilled jobs are changing, this section provides descriptive evidence on how the share of middle-skill jobs has changed over the preceding decade. The analysis in this section confirms that the share of employment in middle-skill jobs – defined by the average wage in an occupation (Box 4.1) – declined across OECD countries over the past decade.

Confirming earlier analyses, the share of employment in middle-skill occupations in OECD countries declined from the mid‑1990s (1994‑1996) to the mid‑2010s (2016‑2018). The analysis in this chapter relies on cross-sectional and panel survey data from across Europe and the United States (see Annex 4.A). Figure 4.1 depicts the share of employment in middle-skill jobs at both time periods for a broad range of OECD countries.2 Across countries, the share of middle‑skill employment declined by a little less than 11 percentage points. This compares to increasing shares of high-skill employment (9 percentage points) and low-skill employment (3 percentage points).

The fall in the share of middle-skill employment across countries accompanied an increase in high- and low-skill employment shares. Two decades earlier, middle-skill employment comprised slightly more than 42% of employment in OECD countries compared to about 35% and 24% for high- and low-skill employment, respectively. In twenty years the share of middle-skill employment fell to be closer to low‑skill employment than high‑skill, with the shares averaging 32%, 43% and 27% for middle‑, high‑ and low‑skill employment respectively.

Figure 4.1. Employment shares of middle-skill occupations declined sharply

Employment in middle-skill occupations as a share of total employment, average of years 1994‑1996 and 2016‑2018

Box 4.1. Defining occupations in terms of “skill levels”

The term middle-skill seems simple, but it typically means different things depending on the data available, the research question, or the country involved. The original research from Autor, Levy and Murnane (2003[3]), who documented the declining share of middle-skill jobs, defined “skill” by the underlying tasks performed in an occupation (“routine” or “non-routine”) for the United States. Since this original research, further work on job polarisation employs an ever expanding conception of “skill”, including task, wage, and education while sometimes finding contradictory evidence for job polarisation (Hofer, Titelbach and Vogtenhuber, 2017[15]; Tåhlin, 2019[16]; Oesch and Piccitto, 2019[17]).

This chapter will use “skill” or “occupation” as short-hand for a wage‑based ranking of occupations. For continuity, and in order to build on previous OECD work, this chapter will use the classification of occupations employed in OECD (2017[6]). The occupation groups follow the classification defined in Goos, Manning and Salomons, (2014[18]), who define International Standard Classification of Occupations (ISCO) occupations by their average wage. The authors use data from the European Community Household Panel (ECHP, the predecessor of EU-Statistics on Income and Living Conditions EU‑SILC), which has wage information, to define the occupations by their average wage at the same occupation level available in the EU-LFS. Workers whose occupation had an average wage in the middle of the occupation‑wage distribution would be classified as middle-skilled regardless of their formal education, training, or labour market experience. This chapter uses the same EU-LFS data, and therefore employs the same country-invariant classification:

High-skill, or high-occupation. ISCO‑88 one-digit occupations 1‑3.
Middle-skill, or middle-occupation. ISCO‑88 one-digit occupations 4, 7, 8.
Low-skill, low-occupation. ISCO‑88 one-digit occupations 5, 9.

The chapter uses a similar occupation-based grouping for the United States. OECD (2017[10]) mapped the ISCO‑88 classification to Standard Occupational Classification (SOC) 2000 codes. This chapter then merged this mapping to Dorn (2009[19]) to provide harmonised Census codes from the 1990s to the present. The U.S. classification scheme shares many characteristics with classifications which group occupations based on tasks or education (Acemoglu and Autor, 2011[20]).

At its simplest, this chapter is concerned with how educational attainment maps into wage-based rankings of occupational outcomes, and how this relationship has changed. When referring to measures of education the following terms will be employed:

High-education. This refers to a person who has at least a tertiary degree corresponding to International Standard Classification of Education (ISCED) level 5 and above.
Middle-education. For persons with an upper-secondary degree or a post-secondary non‑tertiary degree corresponding to ISCED levels 3‑4.
Low-education. This refers to all persons without an upper-secondary degree: ISCED level 2 and below.

4.2. What drives the fall in the share of middle‑skill jobs?

To answer where middle-skill workers are going, it is important to understand how the share of middle‑skill jobs declined. The share of middle-skill employment can decline for several reasons, which may evolve over time. A central question addressed in this chapter is whether the share of middle‑skill employment adjusted due to gradual adjustment through attrition, and/or more abrupt adjustment through transitions.

The two different paths both lead to a diminished middle-skill employment share, but they point to different policy responses. With attrition, workers enter into different occupation groups early in their careers. This is primarily driven by new cohorts of workers entering the labour force by starting in low- and high-skill occupations at higher rates than previous cohorts. In the case of adjustment via transitions, workers reallocate to other skill groups (with possibly spells of non-employment) due to increased separations in middle-skill employment.

4.2.1. With the exception of the global financial crisis, middle-skill separation rates were stable

The level of middle-skill employment held steady in most OECD countries until the 2008‑09 financial crisis, however employment growth was more robust in low- and high-skill employment. Table 4.1 shows average rates of hires and separations across European OECD countries for four time periods.3 The time periods roughly align to the 1990s, 2000s pre-crisis, the crisis and immediate aftermath, and post-crisis. Before the financial crisis, on average across countries, middle-skill hiring and separation rates were about equal, while hiring rates clearly exceeded separation rates for low- and high-skill occupations. This implies that employment in absolute numbers was growing in low- and high-skill occupations before the crisis, while remaining more or less constant for middle-skill employment. The higher employment growth rate in low- and in particular high-skill employment led to a decreasing share of middle-skill employment.

During the financial crisis, the level of middle-skill employment declined, while high and low‑skilled occupations continued to add jobs. For all occupation groups across countries, hiring rates have gradually declined over the past twenty years.4 For low- and high-skill occupations, separation rates remained remarkably consistent, and below hiring rates. However, for middle-skill jobs, the crisis resulted in a sharp increase in separation rates. During the recovery from the crisis, separation rates for all groups returned to their (lower) pre-crisis levels. For the dynamics of employment adjustment across occupation groups, the crisis accelerated the declining share of middle-skill employment by actually destroying jobs, rather than employment simply growing more slowly than other occupation groups as happened in the 15 years preceding the crisis.

	Low-skill		Middle-skill		High-skill
	Hires (%)	Separations (%)	Hires (%)	Separations (%)	Hires (%)	Separations (%)
1995‑2000	23.6	21.5	16.4	15.9	13.8	10.8
2001‑2007	22.8	20.7	15.8	15.8	12.4	10.6
2008‑2012	21.5	20.4	14.4	18.2	11.4	10.4
2016‑2018	22.3	21.0	16.4	15.2	13.6	11.0

Notes: Countries included and time periods: Austria, Belgium, the Czech Republic, Estonia, Denmark, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Luxembourg, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland and the United Kingdom are included in all time periods averaging available years. Latvia, Poland and the Slovak Republic are included starting in 2001.

Source: European Labour Force Survey (EU-LFS).

Statlink

Looking more closely at individual countries reveals the large variation in separation rates between and within countries. Countries that experienced large decreases in middle-skill employment saw large increases in middle-skill separation rates during the crisis driving the overall increase in separations (Figure 4.2). These countries included Greece, Luxembourg and Spain. However, almost all countries in the sample have returned to, or experienced lower separation rates than their pre-crisis rates.

Employment adjustment across skill groupings is a tale of two time periods. Before 2007, labour markets experienced a declining share of middle-skill jobs caused by higher rates of growth in low- and high-skill occupations. The quantity of middle-skill employment remained mostly constant. During the crisis, low‑ and high‑skill employment continued to grow, but at slower rates than pre-crisis. The quantity and share of middle-skill employment declined sharply because of a hike in separation rates. During the recovery, separation and hiring rates largely returned to their pre-crisis averages. In short, with the exception of the global financial crisis, separations have declined gradually, and middle-skill employment has held steady or declined slightly over the past twenty years.

With the exception of the financial crisis, labour market destinations for workers separating from middle-skill jobs also remained stable

Although middle-skill separations have remained mostly stable (except for the financial crisis period), one question that naturally arises is whether the destinations for workers separating from middle-skill jobs changed in any meaningful way. This has two important implications. First, a changing composition of the destinations (occupation groups for new jobs, or non-employment) of workers separating from middle-skill jobs is a signal that the change in middle-skill employment shares is happening through transitions (rather than attrition) of employed middle-skill workers. Conversely, a mostly stable distribution of labour market outcomes for workers separating from middle-skill jobs combined with mostly stable separation rates (as above) points to a diminished role of the transition channel.

Figure 4.2. Changes in separation rates were large during the global financial crisis, but have mostly returned to pre-crisis levels

Hiring and separation rates for middle-skill occupations, 2001‑2018

Second, the destinations of workers previously in middle-skill jobs provide evidence for the normative implications of the decline in middle-skill employment. If these workers are increasingly moving into high‑skill jobs, policy makers may worry less about the implications of shrinking middle-skill employment. By contrast, increasing transitions into low-skill employment or non-employment may signal to policy makers greater distress among workers who previously held middle-skill employment.

Across European OECD countries, the destinations for workers separating from middle-skill jobs have remained stable on average. Figure 4.3 shows the four mutually exclusive and collectively exhaustive destinations for workers who separated from middle-skill jobs one year earlier: low-, middle-, high-skilled employment and non-employment. The propensities are calculated one year after separating, which allows for short non-employment durations. The analysis compares the average of years 2005‑2007, 2008‑2012, and 2015‑2017. Although separations grew during the financial crisis, this chapter is most concerned with the long-term decline in the share of middle-skill employment, and omitting the crisis period attenuates movements due to cyclical variation.

Non-employment and transitions to other middle-skill jobs are the most likely destinations, and have remained so post-crisis. The countries are ordered by the percentage point change in the share of employment in middle-skill jobs one year later pre- and post-crisis – i.e. the difference between dark blue diamonds and bars in Panel B. Non-employment was the most likely destination pre-crisis with 51.4% of workers separating from middle-skill jobs to non-employment one year later followed by a different middle‑skill job with 35.1%. Post-crisis the likelihood of ending up in non-employment fell to 47.6%, while the probability of working in another middle-skill job increased slightly to 35.6%.

Figure 4.3. Outcomes for workers separating from middle-skill jobs remain stable

Skill-group of employment and non-employment outcomes for workers who separated from a middle-skill job one year prior

Once again, the largest changes in the patterns of workers separating from middle-skill jobs occurred during the crisis. The share of workers separating from middle-skill jobs into non-employment grew to 59.4% during 2008‑2012. The shares separating into the other occupation groups all declined. In particular, the share separating into another middle-skill job one year later declined by over 7 percentage points to 27.8%.

The propensity to find low- or high-skill jobs within one year after separation increased uniformly, but modestly from pre- to post-crisis periods. On average, the low-skill propensity increased from 7.1% to 8.4% and the high-skill post-separation employment propensity increased from 7.4% to 8.2%. All but six countries in the sample saw an increase in the low-skill propensity with Sweden, Poland and the United Kingdom experiencing the largest percentage point increases. Transitions into high-skill employment were similarly broad-based. Austria, Sweden and the United Kingdom saw some of the largest percentage point increases. Of course, the composition of the labour force likely changed over this time period, a dimension which will be explored further in Section 4.4.

Younger workers were disproportionately affected by the financial crisis and concurrent spike in separation rates for middle-skill jobs

In addition to how employment is adjusting across skill groupings, the question of who is involved in the adjustment is just as important. The adjustment out of middle-skill work was mostly born by the young. Table 4.2 presents the same hiring and separation rates as Table 4.1 but limited to workers in middle‑skill occupations and further divided into workers who are less than 30 years old and those that are 30 years old and older. The rate of hires and separations is much higher for workers younger than 30 than those that are older. This is expected as younger workers have more volatile employment histories due to increased job hopping, higher rates of temporary contracts, and generally trying to find their way in the labour market.

	Younger workers		Older workers
	Hires (%)	Separations (%)	Hires (%)	Separations (%)
1995‑2000	31.5	32.3	10.2	9.7
2001‑2007	31.5	32.2	10.4	10.5
2008‑2012	29.1	36.2	9.8	12.6
2016‑2018	34.4	33.7	11.3	10.2

Notes: Younger workers are those aged 16‑29, older workers are aged 30‑64. Countries included and time periods: Austria, Belgium, the Czech Republic, Estonia, Denmark, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Luxembourg, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland and the United Kingdom are included in all time periods averaging available years. Latvia, Poland and the Slovak Republic are included starting in 2001.

Source: European Labour Force Survey (EU-LFS).

Statlink

The crisis led to decreased middle-skill employment in all age groups including prime age and older workers. The impact was much stronger for younger workers. During the crisis hires declined and separations increased for both older and younger workers in middle‑skill occupations. The magnitudes differed greatly with younger workers experiencing a net employment loss of 7 percentage points annually compared to 3 percentage points for older workers. In sum, both older and younger age group workers experienced serious employment declines in middle-skilled occupations during the crisis, but younger workers incurred a disproportionate burden.

The results in this section compare well with the previous literature, which finds that polarisation is a result of differential hiring rates, as well as layoffs. Earlier research from the United States found that the decline in middle-skill employment was concentrated during the past two recessions (Jaimovich and Siu, 2014[21]). The work in this chapter suggests that the global financial crisis acted as an accelerant for job polarisation on average across European OECD countries as well. In these countries, adjustment took place gradually until the crisis. Separation rates then spiked for workers in middle-skill jobs during the crisis before mostly returning to pre-crisis levels. The results further generalize the finding – again, previously found in the United States – that job polarisation is the result of higher flows into low- and high-skill occupations, and lower flows into middle-skill occupations, with the pattern most pronounced among the young (Smith, 2013[22]).

4.2.2. Differential labour market entry of younger workers explains most of the change in middle-skill employment shares

The remainder of this section will examine workers by birth cohort to see how their employment shares across different occupation groups change over their working-ages. By examining workers by birth cohort, the analysis decomposes employment shares in different occupational groups by labour market entry and labour market history. Figure 4.4 shows employment shares in low-, middle-, and high-skilled occupations as well as non-employment by birth cohorts.5

The analysis uses eight different European birth cohorts of six contiguous years with the number and size of the cohorts determined by the 24 years of available data.6 Each panel of the figure displays the share for a given skill group and non-employment. For ease of interpretation, the figure averages the birth cohorts’ shares into cohorts born before and after 1970.7 Each line represents one of the two average birth cohorts. The figures present the shares such that they allow comparison for cohorts at the same age.

The entry shares for the two average cohorts shows the greatest divergence. Workers aged 25‑29 worked in middle-skill occupations at a rate of 32.8% before 1970. For cohorts born after 1970, the share decreased to 26.5%. However, the employment trajectories of cohorts as they aged diverged only slightly. Birth cohorts before 1970 saw a slight decreasing propensity to work in middle-skill employment over the life cycle. For the post 1970 birth cohorts, the trajectory was mostly flat.

The shares in high-skill employment exhibited the reverse pattern. Workers aged 25‑29 before 1970 entered high-skill employment at a rate of 22.9%. For birth cohorts after 1970 the rate was 30.7%. Both average birth cohorts saw a slight upward trajectory for the shares employed in high-skill jobs over the life cycle, but birth cohorts born after 1970 were and are employed in high-skill jobs at a higher rate at every age group.

Employment shares in low-skill jobs and non-employment did not diverge appreciably across average birth cohorts. Workers in the pre‑1970 cohort entered low‑skill employment at a rate of 16.3% as 25‑29 year olds, while workers born post‑1970 entered low-skill employment at a rate of 18.1%. Entry shares for non‑employment also diverged only slightly. The pre‑1970 birth cohorts were slightly more likely to be non‑employed at age 25‑29. For the pre‑1970 birth cohorts the rate was 28%, while for post‑1970 birth cohorts averaged 24.8%. After the youth entry share, the trajectories and propensities for the pre- and post‑1970 average birth cohorts were remarkably similar. However, the post‑1970 birth cohorts were always more likely to be both employed, and working in low-skill employment, although only slightly.

Although each successive cohort entered skill groups at different rates, their career trajectories followed markedly similar paths. Put differently, when examining Figure 4.4, the lines showing different employment shares at different ages for each cohort are largely parallel. After entering the labour market in their youth, each cohort has progressed in the labour market similarly.

The exception is the life-cycle patterns for employment shares in middle-skill and high-skill occupations. The employment shares do not seem to be decreasing for the post‑1970 cohorts as they did for the pre‑1970 cohorts. Both have become flatter suggesting perhaps that young workers entering middle-skill occupations are less likely now to upgrade towards high-skill occupations. This seems consistent with findings (Chapter 5) on vocational graduates, which appear to have a flatter career pattern now with respect to the past. It also consistent with recent findings on the decline in internal labour markets (Maurin and Signorelli, 2019[23]).

Figure 4.4. Propensities to work in different skill groups have been driven by differential entry of younger cohorts

Shares of different occupation-skill groups at different ages by birth cohort

The analysis in this section shows that, with the exception of the years after the Global Financial crisis, the rate of separations, and the destinations of workers separating from middle-skill jobs have stayed constant. Younger workers are more likely to enter low- and high-skill employment and are less likely to take middle‑skill jobs highlighting the greater importance of attrition for the decline in the share of middle-skill employment. Annex 4.B provides a formal decomposition of the forces of attrition, transition and cohort size on the change in middle-skill employment shares, which confirms that attrition is the dominant channel accounting for the decline in middle-skill occupations.

4.3. Who were middle‑skill workers?

Looking at only the transitions of workers currently in middle-skill jobs misses the labour market decisions of new cohorts entering the labour market. As the previous section showed, this is likely the most salient mechanism for the shifting shares of skill groups. As the share of middle-skill occupations declines, the probability of any given worker finding a middle‑skill job also declines and the probability she finds a job in either high‑ or low‑skill occupations increases. To account for the full picture of “where middle‑skill workers are going”, one must therefore also consider what those workers who in the past would have been a middle-skill worker are doing today. In short, one needs to account for the possibly shifting demographics8 of middle-skill workers.

To build a clearer picture of a “middle-skill worker”, it is essential to define the profile of the “typical” middle‑skill worker. Using variables which best identify a middle‑skill worker in the period when job polarisation began, this chapter sketches the profile of a middle-skill worker from twenty years prior. Section 4.4 will examine workers with this profile to compare how their labour market outcomes compare to workers with the same profile twenty years prior. This approach helps remedy the problem of non‑observability of counterfactual outcomes for cohorts who enter the labour market in times of more or fewer middle-skill jobs, and who may therefore look vastly different to someone holding that job in the past.

The rest of this section is concerned with identifying the characteristics that best describe a middle-skill worker from times past. The following analysis paints a picture of a typical middle-skill worker which remains relatively constant over time and does not change with shifts in the labour market.

4.3.1. Middle‑skill workers are predominantly workers without a tertiary degree

To determine where middle-skill workers are going, it is necessary to find a set of characteristics that best predict work in middle-skill occupations. These characteristics should ideally be independent of labour market conditions and outcomes.9

Education was the strongest indicator of middle-skill work

Education was the single best predictor of being a middle-skill worker twenty years prior. Figure 4.5 shows the share of middle-skill workers divided into four categories: men and women separately who had at least an upper‑secondary degree but no tertiary degree, and men and women with less than an upper‑secondary degree. The shares are averages for each country of the years 1994‑1996 (years vary depending on availability, see figure notes). Across the OECD countries for which data are available, slightly more than 90% of middle-skill workers lacked a tertiary degree with the share increasing to over 95% in ten of the countries. The Czech Republic and Austria had the highest share while Estonia and Belgium had the lowest shares with 83% and 86.8%, respectively.

Among those without a tertiary degree, a majority of middle-skill workers did have at least an upper‑secondary non-tertiary degree. On average in the OECD, 56.8% of middle-skill workers possessed an upper‑secondary education. The highest shares were in the Slovak Republic, Poland and the Czech Republic, while the lowest shares of middle-skill workers holding at least an upper‑secondary diploma were found in Portugal, Spain and Greece.10 Over a third of middle-skill workers possessed less than an upper‑secondary diploma implying that middle-skill jobs were accessible to even those with little education.11

Figure 4.5. Middle-skill workers were workers without a tertiary degree

Share of middle-skill workers by gender and education, 1994‑1996 (average)

Middle-skill workers were more likely to be male

Middle-skill employment was also dominated by men. Among middle‑skill workers without a tertiary degree, a little less than two thirds were men on average across the OECD. Among countries for which data are available, the highest shares were found in Spain and Luxembourg, which had male shares of middle-skill workers without a tertiary degree of over 70%. The lowest shares were found in the United States and Slovenia. Despite having the lowest shares of men, men still exceeded 50% of middle-skill workers in each of these countries.

Further enforcing the gender disparities in middle-skill work, the share of men in middle-skill work without an upper‑secondary degree exceeded the share of women with one. Men without an upper‑secondary degree made up 26% of middle-skill workers on average. Women who held at least an upper‑secondary degree, but less than a tertiary degree, comprised only 18.6% of middle-skill workers.

4.3.2. Both male and female middle-skill workers were most likely to work in manufacturing

Although substantial gender differences existed among middle-skill workers, both men and women were most likely to work in manufacturing. Figure 4.6 shows the share of middle-skill workers for men and women, respectively, in the three industries they were most likely to work in. For both men and women in middle-skill work, manufacturing represented the modal industry, employing 37.7% of men, and 35.3% of women. Slovenia and Italy employed the highest share of male middle-skill workers in manufacturing. For women, Slovenia and Portugal employed the highest share of middle‑skill women in manufacturing, with 58.3% and 55.7%, respectively.

Figure 4.6. Male and female middle-skill workers were concentrated in manufacturing

Share of middle-skill workers by industry for men (Panel A) and for women (Panel B), 1994‑1996 (average)

The gender differences for middle-skill workers are most apparent in industries other than manufacturing. For women, the next most probable industry was wholesale and retail trade which averaged 11.7% of women in middle-skill employment followed by public administration with 9.4%. The United States and the Netherlands employed the largest share of middle-skill women in wholesale and retail trade with 21.6% and 19.1%, respectively. For public administration, Belgium with 17.8%, and Greece with 16.9%, employed the highest shares.

Middle-skilled men were more likely to be employed in construction, as well as in transportation and storage. After manufacturing, construction employed the largest share of middle-skill male workers across the OECD followed by transportation and storage with 16.8% and 13.3%, respectively. The highest share of workers in construction were found in Luxembourg and Austria, with 24.3% and 21% respectively. For transportation and storage, the highest shares of middle-skilled male workers were found in Finland and Latvia with a little over 16% in both countries.

Occupations differed greatly for middle-skill workers by gender

For men, the industry distribution of middle-skill workers was reflected in their most common occupations. Table 4.3 shows the three most likely detailed occupations across OECD countries for middle-skill workers from two decades prior by gender.12 The most likely occupations were drivers, building finishers, and machinery mechanics and repairers. The three most likely occupations are indicative of the three most likely industries: transportation & storage, construction, and manufacturing.

Women	Men
Secretaries	Truck, Delivery Drivers
Cashiers	Machine Operators
Bookkeepers and Accounting Clerks	Building Finishers (floors, roofing, insulation)

Notes: First row is the most prevalent occupation for each gender, the last row is the third most prevalent. For countries with no data in 1994, "mid‑1990s" is the three earliest years of data. The earliest years are: 1995 (Austria), 1996 (Netherlands, Norway, Slovenia, Switzerland), 1997 (Estonia, Finland, Hungary, Sweden), 1998 (Czech Republic, Latvia, Slovak Republic), 2002 (Poland).

Source: European labour force survey (EU-LFS), The German Socio-Economic Panel (SOEP) for Germany, and the Current Population Survey (CPS) for the United States.

The occupations most likely to be held by middle-skill women did not follow as clearly from the most likely industries to employ middle-skill women. Middle‑skill women were most likely to be employed as secretaries, cashiers and book keepers or auditing clerks. The latter two reflect two of the modal industries most likely to employ female middle-skill workers: wholesale and retail trade, and public administration. The most likely occupation, secretaries, are employed across industries and therefore make up the modal occupation without any explicit tie to the modal industry, manufacturing.13

4.4. Where are middle‑skill workers going?

This section ties together the results of the previous two sections. First, it shows that changes in the composition of the employed population are not the main cause of the declining share of middle-skill employment shares, an issue complicating the results of Section 4.2. It also establishes that groups likely to have been middle-skill workers in the past (Section 4.3) experienced a decreased tendency to work in middle-skill jobs. The rest of the section shows that groups who were previously likely to work in middle‑skill jobs, especially workers with an upper-secondary degree but no tertiary degree, are now more likely to work in low-skill employment.

4.4.1. The decline in middle-skill employment is not primarily due to changing demographics

The working-age population is more highly educated today than twenty years prior. This shift alone may account for the decline in middle-skill employment. This would complicate the results in Section 4.2 because the types of workers holding middle-skill jobs today have a different demographic profile compared to middle-skill workers two decades prior. To answer the question of whether or not shifts in the demographic composition of the working‑age population are causing the shares of middle-skill employment to shrink, the analysis turns to a shift-share analysis.14 The shift-share analysis decomposes the change in the share of middle-skill employment into shifts induced by changes in the composition of the workforce and changes in propensity to be employed in middle-skill employment within groups. In other words, it shows what the share of middle-skill employment would have been if the skill composition of the work‑force had not changed in each country over 20 years (composition effect), as well as if the propensity to work in middle-skill jobs among individuals of each skill groups had not changed over the same period of time (propensity).15

The analysis in this section is most similar to Cortes, Jaimovich and Siu (2017[24]) who perform a similar analysis for the United States. Researchers have performed similar analyses for Germany (Bachmann, Cim and Green, 2018[25]), Finland (Maczulskij and Kauhanen, 2017[26]), and the United Kingdom (Salvatori, 2015[27]).

A simple example helps to explain the shift-share and argue for its importance. From Section 4.3 it is apparent that workers with a tertiary degree are less likely to work in middle-skill jobs compared to workers with less education. If the share of the labour force with a tertiary degree increases over time, the share of employment in middle-skill work will likely decline. In this case, middle-skill workers are not “moving.” Workers across the education distribution are possibly employed in different skill groups at the same rates as before, but the shift in composition makes it look like middle-skill employment has declined.

A key part of the shift-share is the division of workers into distinct groups. The analysis divides workers into mutually exclusive and collectively exhaustive groups based on education and gender. The analysis in the previous section found that they are the best predictors of middle-skill employment. The association between these predictors and middle-skill employment is the deciding factor for their inclusion. They are similar to the demographic characteristics used in Cortes, Jaimovich and Siu (2017[24]) who follow the same methodology for the United States.16

The results of the shift-share show that both changes in composition and changes in the propensity to work in middle-skill jobs contributed to the decline in the share of middle-skill employment. Across OECD countries in the sample, the share of middle‑skill employment declined by 4.2 percentage points.17 That decline can be decomposed into the part due to composition changes, 2.1 percentage points, and decreased propensity to work in middle-skill jobs within groups, 2.3 percentage points. The decrease in middle‑skill employment due to compositional changes is not surprising given the increased share of the population with a tertiary degree.18

Decreases in the propensity to work in middle-skill employment exceed compositional effects in the majority of countries in the sample (Figure 4.7). Luxembourg, Slovenia, and Norway saw the largest declines in middle-skill employment due to decreased propensity in absolute numbers. The countries with the largest declines in propensity as a share of the total decline in middle-skill shares were Estonia, Spain, Ireland and Belgium.

Figure 4.7. Shrinking share of middle-skill employment due (slightly more) to diminished propensity to work more than changing composition

Percentage point change in share of middle-skill employment due to composition and propensity, mid‑1990s to mid‑2010s

4.4.2. Workers without a tertiary degree are more likely to be employed in low-skill occupations

The preceding shift-share analysis confirmed that propensity to work in middle‑skill jobs contributed to the decline in the share of middle-skill employment. Changes in the composition of the labour force – greater educational achievement and women’s increased participation – are a contributing factor. The analysis did not show where workers are increasingly likely to work. To see where they are working, this analysis digs deeper into the changing propensities of where workers likely to be middle-skill are employed.

Middle-educated men have seen only a modest drop in middle-skill employment shares

Across OECD countries for which data are available, workers without a tertiary degree have become less likely to work in middle-skill occupations. This is not entirely surprising given the decline in middle-skill employment overall. However, it was not a given that all groups would experience a drop in their propensity to work in middle-skill occupations. Middle‑educated men have been the least affected with the share of the working‑age population in middle-skill occupations dropping a little over 2 percentage points (Figure 4.8). Low-educated men saw their share decrease by 7 percentage points. Women with low and middle-education experienced a decline of a little over 4 percentage points each.

Figure 4.8. The propensity to work in middle-skill occupations fell for all workers without a tertiary degree

Percentage point change in the shares in middle-skill jobs by gender-education groups, mid‑1990s to mid‑2010s

Middle‑educated workers are much more likely to be in low‑skill employment

As the share of middle-occupation employment has declined, workers without a tertiary degree are increasingly likely to work in low-skill occupations. Figure 4.9 shows the percentage point increase in the propensity to work in low‑skill occupations for women without a tertiary degree. The figure further divides employed women into those without an upper‑secondary degree, and those with at least an upper‑secondary degree, but no tertiary degree. The shift is most pronounced for middle-educated women. The propensity of employed middle-educated women to work in low‑skill occupations increased from 18.8% to 27% twenty years later. For low‑educated women the increase was more muted. Low-educated women saw their propensity to work in low-skill occupations grow from 17.8% in the mid‑1990s to 21.2% in the mid‑2010s.19

Figure 4.9. Women without a tertiary degree more likely to work in low-skill occupations

Women’s shares in low-skill occupations by education, mid‑1990s and mid-2010s

Middle-educated women in Finland, Portugal and Spain experienced the largest increase in the propensity to work in low‑skill occupations. The percentage point increase in all three countries exceeded 17%. Low‑educated women experienced the largest increase in the share working in low‑skill occupations in Spain, Portugal and Estonia.

The propensity for employed men without a tertiary degree to work in low‑skill occupations also increased across OECD countries. Figure 4.10 shows the change in the share in low-skill occupations for male, middle‑ and low‑educated workers. Though still large, men saw a more muted shift towards low-skill occupations compared to women. Low-educated men’s propensity to work in low-skill occupations increased from 12.1% to 14% over the previous twenty years across OECD countries. For middle‑educated men, the increase was larger, rising from 11.2% to 15.1%.

Figure 4.10. Men without a tertiary degree are also increasingly in low-skill occupations

Men’s shares in low-skill occupations by education, mid‑1990s and mid‑2010s

When examining individual countries, it is clear middle-educated men experienced a larger increase in the share of low-skill employment compared to low-educated men. Low‑educated men in the Netherlands, Sweden and Hungary experienced the largest increase in the propensity to work in low-skill occupations. Their propensities increased by 8.6, 7.8 and 6.3 percentage points, respectively. For middle‑educated men, the greatest increase occurred in Portugal, the United Kingdom and Spain, where increases topped 8 percentage points in each country.

Middle‑educated workers of both genders are less likely to work in high‑skill occupations, with some notable exceptions

There is not a corresponding increase in the propensity to work in high‑skill occupations. Across OECD countries, the propensity to work in high-skill occupations declined by 0.5 percentage points for middle‑educated men, and 0.4 percentage points for middle‑educated women (Figure 4.11). For low‑educated men, the change was a decline of 0.7 percentage. Low-educated women in OECD countries saw no change in the propensity to work in high-skill jobs.20

Figure 4.11. Middle-educated men and women are only slightly more likely to work in high-skill occupations

Shares of middle-educated workers in high-skill occupations by gender, mid‑1990s and mid‑2010s

Although propensities to work in high-skill occupations declined, there were some countries that saw substantial increases in the share of workers in high-skill occupations. In Denmark, Germany, Norway and Sweden the share of middle-educated women in high-skill occupations increased by more than 5 percentage points. Denmark, Germany, Norway and Sweden (together with Estonia) also had the largest increases in the share of middle-educated men moving into high-skill occupations over the preceding twenty years. Interestingly, many of these countries have a vocational education and training system (VET) at the intermediate education level that seem to work particularly well (see Chapter 5).

Changing entry propensities for middle- and low-skill employment are almost entirely due to workers without a tertiary education

To bring the analysis to a close, the chapter returns to the cohort analysis introduced at the end of Section 4.2. Figure 4.12 shows the life-cycle labour market employment shares for pre- and post‑1970 birth cohorts again, but this time further broken out by educational attainment. Breaking out the life-cycle employment shares by education further targets workers most likely to have been middle-skilled in the past (Section 4.3).

Figure 4.12. Propensities to work in different skill groups mostly driven by differential entry of younger cohorts without a tertiary degree

Shares in different occupation-skill groups at different ages by birth cohort and education

Changing entry propensities for middle- and low-skill employment are almost entirely due to workers without a tertiary education. This finding reinforces the main results from the beginning of this section. There is little change in either entry propensities or life-cycle trajectories for high-skill employment and non-employment by educational attainment. Similarly, workers with a tertiary education show no discernible differences pre- and post-1970 with regard to employment in middle- and low-skill employment. The major change is that workers without a tertiary degree are less likely to work in middle-skill employment and more likely to work in low-skill employment. The trajectories to the eye are quite similar with entry patterns showing the greatest difference between pre- and post-1970 birth cohorts without a tertiary education.

4.5. Concluding remarks

The share of middle-skill jobs in OECD labour markets has declined over the past three decades. Middle‑skill jobs once made up a large share of overall employment, but automation and offshoring have reduced the share of middle-skill employment relative to low‑skill and high-skill occupations, a trend that has been termed job polarisation. What is happening to workers who could have previously expected to be employed in middle-skill occupations is an enduring question for governments in OECD countries.

In the past, employment in middle‑skill occupations provided many workers with a good standard of living. The decline in employment opportunities in these occupations has meant that workers who previously would have held these jobs are increasingly employed in low-skill occupations.

The shift in the share of middle-skill jobs towards low- and high-skill employment has mostly taken place through attrition, at least in European countries. That is to say through successive cohorts of younger workers being less likely to enter the labour force in middle-skill jobs, and more likely to start in low-skilled, and to a lesser extent, high-skilled jobs. Their subsequent job trajectories over the life cycle are a contributing but secondary factor. Workers without a tertiary degree are sliding down the job ladder. Compared with twenty years ago, workers without a tertiary degree are less likely to work in middle‑skill occupations. This has been matched almost exactly by an increase in low‑skill employment for this group. The analysis in this chapter, however, relies on cross-sectional and panel survey data from across Europe and the United States. Further analysis would be required to fully generalise these findings to the whole OECD.

Although the answer to the question of what is happening to middle-skill workers appears bleaker than expected, some countries are performing well in mitigating the adverse effects of job polarisation. Over the past two decades in Sweden, Germany, Norway and Denmark, the rise in the employment shares of middle-educated workers in high-skill occupations was almost as sizeable as the rise in their employment in low-skill occupations. The relative success of these countries shows that good jobs for formerly middle‑skill workers is not necessarily due to insurmountable structural forces. Automation and globalisation have reduced the number of middle-skill employment opportunities for workers without a tertiary degree. However, countries that have improved employment opportunities for middle-skill workers share a set of common policies. They have strong institutions and practices around social dialogue, as well as an emphasis on vocational education and training (Chapter 5). The explicit application of these policies to the dynamics studied in this chapter will be left for future research.

References

[20] Acemoglu, D. and D. Autor (2011), “Skills, Tasks and Technologies: Implications for Employment and Earnings”, in Ashenfelter, O. and D. Card (eds.), Handbook of Labor Economics, http://dx.doi.org/10.1016/S0169-7218(11)02410-5.

[13] Acemoglu, D. and P. Restrepo (2018), “Artificial Intelligence, Automation and Work”, NBER Working Paper Series, No. 24196, NBER, Cambridge, MA.

[11] Autor, D. (2015), Polanyi’s Paradox and the Shape of Employment Growth.

[12] Autor, D. (2015), “Why are there still so many jobs? The history and future of Workplace automation”, Journal of Economic Perspectives, Vol. 29/3, pp. 3-30.

[8] Autor, D. (2010), The Polarization of Job Opportunities in the U.S. Labor Market Implications for Employment and Earnings, The Hamilton Project, Washington, D.C., https://www.hamiltonproject.org/papers/the_polarization_of_job_opportunities_in_the_u.s._labor_market_implica (accessed on 13 January 2020).

[7] Autor, D. and D. Dorn (2013), “The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market”, American Economic Review, Vol. 103/5, pp. 1553-1597.

[30] Autor, D. and D. Dorn (2009), “This Job is “Getting Old”: Measuring Changes in Job Opportunities using Occupational Age Structure”, American Economic Review Papers and Proceedings, Vol. 99/2, pp. 45-51.

[3] Autor, D., F. Levy and R. Murnane (2003), “The Skill Content of Recent Technological Change: An Empirical Exploration”, The Quarterly Journal of Economics, Vol. 118/4, pp. 1279-1334, https://economics.mit.edu/files/11574 (accessed on 8 December 2017).

[25] Bachmann, R., M. Cim and C. Green (2018), “Long-run Patterns of Labour Market Polarisation: Evidence from German Micro Data”, Ruhr Economic Papers, No. 748, RWI.

[32] Cazes, S. and M. Tonin (2010), “Employment protection legislation and job stability: A European cross-country analysis”, International Labour Review, Vol. 149/3, pp. 261-285, http://dx.doi.org/10.1111/j.1564-913X.2010.00087.x.

[9] Cortes, G. (2016), “Where Have the Middle-Wage Workers Gone? A Study of Polarization Using Panel Data”, Journal of Labor Economics, Vol. 34/1, pp. 63-105, http://dx.doi.org/10.1086/682289.

[24] Cortes, M., N. Jaimovich and H. Siu (2017), “Disappearing Routine Jobs: Who, How, and Why?”, Journal of Monetary Economics, Vol. 91, pp. 69-87.

[19] Dorn, D. (2009), “Data Appendix”, in Essays on Inequality, Spatial Interaction, and the Demand for Skills, Dissertation University of St. Gallen no. 3613, September., https://www.ddorn.net/data/Dorn_Thesis_Appendix.pdf (accessed on 17 April 2019).

[4] Goos, M. and A. Manning (2007), “Lousy and Lovely Jobs: The Rising Polarization of Work in Britain”, The Review of Economics and Statistics, Vol. 89/1, pp. 118-133, http://www.mitpressjournals.org/doi/pdf/10.1162/rest.89.1.118 (accessed on 4 August 2017).

[18] Goos, M., A. Manning and A. Salomons (2014), “Explaining Job Polarization: Routine-Biased Technological Change and Offshoring”, American Economic Review, Vol. 104/8, pp. 2509-2526, http://dx.doi.org/10.1257/aer.104.8.2509.

[5] Goos, M., A. Manning and A. Salomons (2009), “Job Polarization in Europe”, American Economic Review, Vol. 99/2, pp. 58-63, http://dx.doi.org/10.1257/aer.99.2.58.

[1] Helper, S., T. Krueger and H. Wial (2012), Why Does Manufacturing Matter? Which Manufacturing Matters? A Policy Framework, Metropolitan Policy Program at the Brookings Institution.

[15] Hofer, H., G. Titelbach and S. Vogtenhuber (2017), “Polarisierung am österreichischen Arbeitsmarkt?”, Wirtschaft und Gesellschaft, Vol. 43/3, pp. 379-404.

[29] Hyatt, H. and J. Spletzer (2013), “The recent decline in employment dynamics”, IZA Journal of Labor Economics, Vol. 2/1, p. 5, http://dx.doi.org/10.1186/2193-8997-2-5.

[21] Jaimovich, N. and H. Siu (2014), “The Trend is the Cycle: Job Polarization and Jobless Recoveries”, NBER Working Paper Series, No. 18334, NBER, http://www.nber.org/papers/w18334.

[26] Maczulskij, T. and M. Kauhanen (2017), “Where do workers from declining routine jobs go and does migration matter?”, Työpapereita Working Papers, No. 314, Labour Institute for Economic Research.

[23] Maurin, E. and S. Signorelli (2019), The Decline in Internal Labor Markets and Technological Change, Working Paper.

[2] OECD (2019), Under Pressure: The Squeezed Middle Class, OECD Publishing, Paris, https://dx.doi.org/10.1787/689afed1-en.

[14] OECD (2018), OECD Employment Outlook 2018, OECD Publishing, Paris, https://dx.doi.org/10.1787/empl_outlook-2018-en.

[6] OECD (2017), “How technology and globalisation are transforming the labour market”, in Annexes of Chapter 3 of the OECD Employment Outlook 2017, OECD Publishing, Paris, https://www.oecd.org/els/emp/Employment-Outlook-2017-Annexes-Chapter-3.pdf.

[10] OECD (2017), OECD Employment Outlook 2017, OECD Publishing, Paris, https://dx.doi.org/10.1787/empl_outlook-2017-en.

[31] OECD (2009), OECD Employment Outlook 2009: Tackling the Jobs Crisis, OECD Publishing, Paris, https://dx.doi.org/10.1787/empl_outlook-2009-en.

[17] Oesch, D. and G. Piccitto (2019), “The Polarization Myth: Occupational Upgrading in Germany, Spain, Sweden, and the UK, 1992-2015”, Work and Occupations, Vol. 46/4, pp. 441-469, http://dx.doi.org/10.1177/0730888419860880.

[27] Salvatori, A. (2015), “The anatomy of job polarisation in the UK”, IZA Discussion Paper, No. 9193, IZA, https://www.iza.org/publications/dp/9193.

[28] Schmitt, J., H. Shierholz and L. Mishel (2013), “Don’t Blame the Robots”, EPI-CEPR Working Paper, Economic Policy Institute, Washington, DC, https://www.epi.org/publication/technology-inequality-dont-blame-the-robots/.

[22] Smith, C. (2013), “The Dynamics of Labor Market Polarization”, Finance and Economics Discussion Series, No. 2013-57, Federal Reserve Board, https://www.federalreserve.gov/pubs/feds/2013/201357/201357abs.html.

[16] Tåhlin, M. (2019), “Polariseringsmyten. Försvinner verkligen de medelkvalificerade jobben?”.

Annex 4.A. Data Sources

Unless otherwise noted, all samples use the working-age population defined as individuals ages 16‑64. Some samples further restrict the age range to younger workers (ages 16‑24), prime-age (25‑54) and older workers (55‑64).

European Labour Force Survey (EU-LFS). The EU-LFS is the largest European household sample survey covering labour force participation of people aged 15 years and older as well as people outside the labour force. The national statistical institutes design and execute their own respective labour force surveys. Eurostat harmonises and distributes the microdata in consultation with the respective national statistical institutes. All OECD members who are members of the European Union or are EFTA countries are included in this study with the exception of Iceland.
German Socio-Economic Panel (GSOEP). The GSOEP is a longitudinal survey of private households in Germany from 1984 to 2016. The database is produced by the Deutsches Institut für Wirtschaftsforschung (DIW). The survey covers labour force topics, as well as household composition, health, and satisfaction.
Current Population Survey (CPS) including supplements. The CPS is a monthly household survey conducted by the U.S. Census Bureau on behalf of the Bureau of Labor Statistics. The survey provides labour force information on household members age 15 years and older. In addition to the main questionnaire, each month one fourth of respondents rotate out of the survey and provide information on earnings.
European Union Statistics on Income and Living Conditions (EU-SILC). The EU-SILC is a European household sample survey covering income, poverty, social exclusion, living conditions and labour market outcomes. The EU-SILC began in 2004 and it now covers all EU countries, Iceland, Norway and Switzerland. The survey contains both a cross-sectional survey and a longitudinal element which observes individuals over-time. The longitudinal portion is employed in this chapter to track labour market transitions.

Annex 4.B. Attrition Transition Decomposition

The graphical analysis found in Figure 4.4 of the difference between transitions and attrition for the causes of the decline in middle-skill employment is useful, but incomplete. Annex Figure 4.B.1 shows the results of a more rigorous analytical decomposition of the contribution of attrition, transitions and cohort sizes to the decline in middle-skill employment shares. Attrition is meant as the contribution of younger cohorts showing different propensities to enter different occupation groups and non-employment when they are young (25‑29 or 30‑34, for example) while older cohorts exit the labour market. Transitions are meant to capture how propensities change as cohorts age, and whether these trajectories have meaningfully changed for subsequent birth cohorts. Finally, cohort size captures the fact that the analysis looks at the change in middle-skill employment over time, and birth cohorts appear at different ages at different points in time. Thus, the share of middle-skill employment can change – holding attrition and transitions constant – if birth cohort sizes vary.

Annex Figure 4.B.1. Attrition accounts for the majority of the decline in the share of middle-skill employment

Decomposition of the change in the share of middle-skill employment by attrition, transitions and cohort size, 1995‑2018

The more formal analysis shows that attrition is the primary mechanism for declining middle-skill employment shares. The figure shows the share of the change in middle-skill employment shares due to attrition, transitions and cohort size. Reported total changes differ from those shown in Figure 4.1 due to different time-periods. The first main finding is that the attrition component is sharply negative for almost all countries, and in all countries except five – Estonia, Switzerland, Norway, Hungary and Sweden – the contribution from attrition is greater than that from transitions. This indicates that the lower propensity to enter middle-skill employment for younger workers is driving the change. Second, the transition component is heterogeneous across countries. Few countries show a sharp negative contribution from transitions – Estonia, Switzerland, and Norway – while others show even a positive contribution. In countries where the transition component is positive, this implies that if entry rates were equal across age cohorts, the share of middle-skill employment would actually increase over-time. On average, attrition accounts for a little under two thirds of the change in middle-skill employment with the rest split almost evenly among transitions and cohort size.

Derivation of attrition and transition decomposition

What follows is the derivation of the decomposition into the two components (attrition and transition) as well as the age-cohort population weights. The derivation covers the general case of when the time period of the decomposition perfectly aligns with the minimum and maximum of the age groups. However, the decomposition generalizes to arbitrary time spans and the cases when cohorts are not perfectly observable. The analysis focuses on the cohort decomposition for middle-skill jobs (but the same reasoning/calculations hold for other types of jobs and non-employment).

There are three dimensions: age class j, cohort c and time t. Note that there is an unambiguous correspondence between any couple of these indexes and the third one. For example, class j at time t completely identifies cohort c. But at the same time cohort c and class j perfectly identifies time t, etc. To fix ideas, a unit of time in the derivation of the same span as the span of age classes. In Figure 4.4 age classes cover 5 years.

The derivation first proceeds with definitions. Let $M_{i, t}^{j}$ be the number of people of age class j in middle-skill jobs at time t (for cohort i) and $P_{i, t}^{j}$ the number of people of age class j at time t. We have, assuming that cohort c is in age class 1 at time t:

$S_{t} = \frac{\sum_{j = 1}^{k} M_{c - j + 1, t}^{j}}{\sum_{j = 1}^{k} P_{c - j + 1, t}^{j}} = \frac{\sum_{j = 1}^{k} (\frac{M_{c - j + 1, t}^{j}}{P_{c - j + 1, t}^{j}}) P_{c - j + 1, t}^{j}}{\sum_{j = 1}^{k} P_{c - j + 1, t}^{j}} = \sum_{j = 1}^{k} (M_{c - j + 1, t}^{j} / P_{c - j + 1, t}^{j}) w_{t}^{j} = \sum_{j = 1}^{k} s_{c - j + 1, t}^{j} w_{t}^{j}$

where $S_{t}$ stands for shares at time $t$ , $s_{i, t}^{j} = \frac{M_{i, t}^{j}}{P_{i, t}^{j}}$ are age-class-specific share, w are population weights (the share of population with age class j in total population within the range defined by the selected set of age classes, e.g. 25‑54) and k is the number of age classes.

The decomposition begins with a shift-share analysis between t and t+k-1 (with age-classes as the only grouping variable). This yields:

$S_{t + k - 1} - S_{t} = \underset{S h i f t [t; t + k - 1]}{\underset{⏟}{\sum_{j = 1}^{k} {w_{t}^{j} (s}_{c + k - j, t + k - 1}^{j} - s_{c - j + 1, t}^{j})}} + \underset{S h a r e [t; t + k - 1]}{\underset{⏟}{\sum_{j = 1}^{k} s_{c + k - j, t + k - 1}^{j} {(w}_{t + k - 1}^{j} - w_{t}^{j})}}$

The Share component simply gives the share of the change due to cohort size. The Shift component will yield the attrition and transition components. The idea is to compare each age-by-time specific share with the corresponding age-by-time specific share in the benchmark cohort.

We have:

$S h i f t [t; t + k - 1] = \sum_{j = 1}^{k} {w_{t}^{j} s}_{c + k - j, t + k - 1}^{j} - \sum_{j = 1}^{k} {w_{t}^{j} s}_{c - j + 1, t}^{j}$

$= \sum_{j = 1}^{k} w_{t}^{j} s_{c + k - j, t + k - 1}^{j} - \sum_{j = 1}^{k} w_{t}^{j} s_{c, t + j - 1}^{j} + \sum_{j = 1}^{k} w_{t}^{j} s_{c, t + j - 1}^{j} - \sum_{j = 1}^{k} w_{t}^{j} s_{c - j + 1, t}^{j}$

$= \sum_{j = 1}^{k} w_{t}^{j} (s_{c + k - j, t + k - 1}^{j} - s_{c, t + j - 1}^{j}) - \sum_{j = 1}^{k} w_{t}^{j} ({s_{c, t + j - 1}^{j} - s}_{c - j + 1, t}^{j})$

Before proceeding, define $\prod_{s = 1}^{j} p^{s}$ as the transition path for the benchmark cohort, which gives the trajectory of the benchmark cohort from its initial entry share. This implies that $s_{c, t + j - 1}^{j} = s_{c, t}^{1} (\prod_{s = 1}^{j} p^{s})$ . Similarly, define $\prod_{s = j}^{k} q^{s}$ as the inverse transition path of the benchmark cohort, which maps the benchmark cohort back from its terminal share at age group k implying $s_{c, t + k - 1}^{k} (\prod_{s = j}^{k} q^{s}) = s_{c, t + j - 1}^{j}$ . For clarity, in this example the benchmark cohort is cohort c, and enters the labour market (j=1) at time t.

The derivation into the transition and attrition components proceeds with the following two steps. First, substitute the two identities defined in the previous paragraph into the second and third terms of the most recent equation and then add and subtract ${\sum_{j = 1}^{k} s}_{c + k - j, t + k - 1}^{j} (\prod_{s = 1}^{j} p^{s}) w_{t}^{j}$ and ${\sum_{j = 1}^{k} s}_{c - j + 1, t + k - j}^{k} (\prod_{s = j}^{k} q^{s}) w_{t}^{j}$ , respectively. This yields

$S h i f t [t; t + k - 1] = T W T r [t; t + k - 1] + T W A t [t; t + k - 1]$

Where TWTr represents the (weighted) transition component across dates and TWAt represents the (weighted) attrition component across dates. Operationally we have:

$T W T r [t; t + k - 1] =$

$= \sum_{j = 1}^{k} (s_{c + k - j, t + k - 1}^{j} - s_{c + k - j, t + k - j}^{1} (\prod_{s = 1}^{j} p^{s})) + \sum_{j = 1}^{k} w_{t}^{j} (s_{c - j + 1, t + k - j}^{k} (\prod_{s = j}^{k} q^{s}) - s_{c - j + 1, t}^{j})$

The first term fixes the benchmark trajectory to each post-benchmark cohort and compares their hypothetical share (using the benchmark trajectory) to their actual share in the t+1 time period. The second term does the same with the pre-benchmark cohorts using the inverse transition path and comparing the hypothetical and actual share in period t. The result is the contribution of transitions. For attrition we have:

$T W A t [t; t + k - 1] =$

$= \sum_{j = 1}^{k} w_{t}^{j} (\prod_{s = 1}^{j} p^{s}) {(s}_{c + k - j, t + k - j}^{1} - s_{c, t}^{1}) + \sum_{j = 1}^{k} w_{t}^{j} (\prod_{s = j}^{k} q^{s}) (s_{c, t + k - 1}^{k} - s_{c - j + 1, t + k - j}^{k})$

The intuition is similar. The first term holds the transition component constant allowing the comparison of the entry share between the benchmark cohort, and each subsequent cohort. The second term does the same, except allowing the comparison of the benchmark cohort in its terminal age group (j=k) with the share in the terminal age group for all earlier cohorts. The result is the attrition component.

Annex 4.C. Additional Figures

Annex Figure 4.C.1. Education shares of the working-age population

Annex Figure 4.C.2. Alternative shift-share analysis

Annex Figure 4.C.3. Shares of low-education, prime-age workers in middle-skill employment

Annex Figure 4.C.4. Share of middle-educated, prime-age workers in middle-skill employment

Annex Figure 4.C.5. Share of low-educated, prime-age workers in low-skill employment

Annex Figure 4.C.6. Share of middle-educated, prime-age workers in low-skill employment

Annex Figure 4.C.7. Share of low-educated, prime-age workers in non-employment

Annex Figure 4.C.8. Share of middle-educated, prime-age workers in non-employment employment

Notes

← 1. The causes of job polarisation are not uniformly agreed upon. Researchers noted that the U‑shaped pattern underpinning job polarisation does not hold across time periods. In addition, changes in occupation shares do not explain skill patterns as the theory intended (Schmitt, Shierholz and Mishel, 2013[28]). This chapter notes the existing research and makes no original claims about why the shares of middle-skill occupations are declining. Rather, it focuses on the dynamics of how the share has declined, and where workers who would have held those jobs in the past are doing now.

← 2. European employment data beyond 2010 was mapped from ISCO-08 to ISCO-88 using a many-to-many mapping technique. This mapping technique is described in Annex 3.A4 (OECD, 2017[6]). For the United States, uniform Census codes come from Dorn (2009[19]).

← 3. Employment flows for each group $i$ , are derived following OECD (2009[31]): $∆ {E m p}_{i} = {H i r e s}_{i} - {S e p}_{i}$ , Where $∆ {E m p}_{i}$ is the net employment change in group $i$ between years $t$ and $t - 1$ . ${H i r e s}_{i}$ is the gross hires for group $i$ at time $t$ , which is determined by the number of workers with tenure less than one year. ${S e p}_{i}$ is the number of gross separations and is pinned down by the identity. In practice gross hires and separations are presented as a rate by dividing by average employment in years $t$ and $t - 1$ .

← 4. This is a longer term trend, which is well documented in the United States (Hyatt and Spletzer, 2013[29]), with varying trends in other OECD countries (Cazes and Tonin, 2010[32]).

← 5. Unlike the previous section, this analysis uses labour force surveys, and does not follow the same worker over time.

← 6. The size of each birth cohort is limited by sample size in the survey data. Ideally one would use each year as the size of a birth cohort, but that would leave too small a sample with labour force surveys. The number of age ranges is similarly limited by sample size, but also the maximum span of the surveys to 20 years of data.

← 7. This allows for roughly equal years before and after 1970 with birth years between 1946 and 1993. It also roughly aligns with “baby boom” and “post baby boom” generations.

← 8. This includes education.

← 9. For example, age, and for the most part gender are fixed and pre-determined at birth, and do not change with labour market conditions. Education and region of residence are more problematic and partly reflect local labour market conditions. Both also involve high switching costs, and may reflect exogenous factors such as historical family ties (place of residence) and labour market conditions many years prior to observation (education). Given their importance in predicting middle-skill workers, both are included in the set of possible predictors. Variables conditional on employment, such as industry and occupation, are not included as predictors. Skill groups are determined by occupation, and its inclusion will make any inferences about occupation tautological. Due to the strong correlation between occupations and industry, industry is not included for similar reasons to occupation. Industry and occupation will be explored to complete the picture of the typical middle-skill worker from 20 years prior, but they will be excluded from defining a middle-skill worker.

← 10. This cross-country variation partially reflects the relative abundance of low- and middle-educational attainment in the population.

← 11. As noted, education is included as one of the main predictors, but its inclusion is potentially problematic. Education is an endogenous choice for young workers or young people deciding whether to enter the labour force or pursue further schooling. If a given young person is, for example, confronted with satisfactory job opportunities once she obtains an upper-secondary degree, she may decide to forgo further education. That same person entering a labour market at a different, hypothetical, point in time may instead pursue further schooling in the face of a poor job market. This choice for people at the margin of pursuing further education complicates the interpretation of the outcomes of education cohorts at different points in time. Fixing this problem is not trivial, and all results should be interpreted with this in mind.

← 12. These are ISCO-88 3-digit occupations and the uniform Census codes from Dorn (2009[19]).

← 13. While the decline in manufacturing shifted the industry distribution of middle-skilled jobs, perhaps surprisingly, the modal occupations remained relatively consistent compared to 20 years prior.

← 14. See Annex 4.C for a depiction of the changing education shares of the working-age population.

← 15. The shift share decomposes the change in the share of middle-skill employed according to:

${\bar{π}}_{1}^{j} - {\bar{π}}_{0}^{j} = \sum_{g} ∆ w_{g 1} π_{g 0}^{j} + \sum_{g} w_{g 0} ∆ π_{g 1}^{j}$ . The term ${\bar{π}}_{t}^{j}$ is the share of the population in skill group $j$ at time $t$ . The term $w_{g t}$ is the share of the population in demographic group $g$ at time $t$ , and $π_{g t}^{j}$ is the share of group $g$ in skill group $j$ at time $t$ . The left-hand side of the equation is the change in the share of the population in skill group $j$ . The two terms on the right-hand side of the equation are (from left to right) the composition and propensity effects, respectively.

← 16. Other analyses include age as an important factor in middle-skill employment (Autor and Dorn, 2009[30]). The analysis also undertook the shift-share using age as a factor. All results are qualitatively similar. The change in propensities are stronger for younger workers, however. Results for the shift-share including age, and propensities for prime-age education by sex groups to be employed in different skill groups is available in Annex 4.C.

← 17. This is significantly more modest than the percentage point decrease presented at the beginning of Section 4.1. In the first section, the shares are defined as shares of the employed as originally constructed in the literature (Autor, Levy and Murnane, 2003[3]). The formulation here constructs the shares as a share of the working-age population, which allows the shift-share to account for shifts out of employment.

← 18. For some perspective, across OECD countries in this chapter, the share of the population aged 16‑64 without an upper-secondary degree decreased from 38.5% to 24.5%, while the population with at least a tertiary degree increased from 15.9% to 28.9%. The share with at least an upper-secondary degree but without a tertiary degree increased from 44.7% to 45.8%.

← 19. The increase in propensity for middle-educated women to work in low‑skill employment was partly the result of increased rates of employment. Low-educated women saw no meaningful increase in their employment to population ratio.

← 20. Not in the chart.

Featured topics

Agriculture and fisheries

Climate change

Development

Digital

Economy

Education and skills

Employment

Environment

Finance and investment

Governance

Health

Industry, business and entrepreneurship

Regional, rural and urban development

Science, technology and innovation

Society

Taxation

Trade

Energy

Nuclear energy

Transport

Featured topics

Agriculture and fisheries

Climate change

Development

Digital

Economy

Education and skills

Employment

Environment

Finance and investment

Governance

Health

Industry, business and entrepreneurship

Regional, rural and urban development

Science, technology and innovation

Society

Taxation

Trade

Energy

Nuclear energy

Transport

Countries A - C

Countries D - I

Countries J - M

Countries N - R

Countries S - T

Countries U - Z

Regional and global engagement

Countries

Countries A - C

Countries D - I

Countries J - M

Countries N - R

Countries S - T

Countries U - Z

Regional and global engagement

Publications

Publications

Featured publications

Data

Data

Featured data

News & Events

News & Events

Featured Events

About OECD

About

Engage with us

Work with us

Featured topics

Agriculture and fisheries

Climate change

Development

Digital

Economy

Education and skills

Employment

Environment

Finance and investment