Access the full text.
Sign up today, get DeepDyve free for 14 days.
Loading next page...
/lp/springer-journals/labour-market-polarisation-revisited-evidence-from-austrian-vacancy-HQo1sVAXfi
- Publisher
- Springer Journals
- Copyright
- Copyright © The Author(s) 2021
- ISSN
- 1614-3485
- eISSN
- 2510-5027
- DOI
- 10.1186/s12651-021-00290-4
- Publisher site
- See Article on Publisher Site
Abstract
Recent research suggests that new technologies are important drivers of empirically observed labour market polari- sation. Many analyses in the field of economics are conducted to evaluate the changing share of employment in low-skill, medium-skill and high-skill occupations over time. This occupation-based approach, however, may neglect the relevance of specific skills and skill bundles, which potentially can be used to explain the observable patterns of labour market polarisation. This paper adds to the literature in two ways: First, we present the results of an analysis of data on job vacancies rather than the currently employed and, second, we derive occupation-defining skills using network analysis tools. The analysis and tool usage allowed us to investigate polarisation patterns in Austrian vacancy data from 2007 to 2017 and identify changes in the skills demanded in job vacancies in Austria. In contrast to most previous research, we find no evidence for polarisation, but rather a trend towards upskilling. Keywords: Skill demand, Polarisation, Network analysis, Vacancies, ESCO JEL Classification: J24, J63, O15, O33 Nonetheless, the empirically observed polarisation of 1 Introduction labour markets (i.e. the increase in the share of low- and The effects of digital technologies on labour markets high-wage/-skill occupations in the US and the UK) is worldwide are currently being widely discussed. Several often explained by the capacity of technologies to substi- authors predict that these technologies will destroy jobs tute for specific work tasks (Autor 2015; Goos and Man - to various extents due to automation (Arntz et al. 2016; ning 2007; Goos et al. 2014). These analyses are often Frey and Osborne 2017), while others stress that they will based on occupation wages and employment shares lead to the emergence of new occupations and the crea- (Autor 2015, 2013; Autor and Dorn 2013; Goos et al. tion of new jobs (Bainbridge 2015) or the transformation 2009, 2014) and the distinction of routine and non-rou- of existing jobs (Berger and Frey 2015). Retrospective tine tasks as introduced by Autor et al. (2003). studies present less dystopian views than future-oriented However, several research groups recently indicated research regarding the spectre of job destruction due to that these studies need refinement (e.g. Alabdulkareem automation. For example, Graetz and Michaels (2018) et al. 2018; Caines et al. 2017; Salvatori 2018), noting that demonstrate that the increased usage of robots in 17 the traditional occupation-based approaches may neglect countries from 1993 to 2007 in industrial production the relevance of specific skills and skill bundles, which had a positive impact on productivity and that no over- can potentially be used to explain patterns of labour mar- all negative employment effects were observed; however, ket polarisation. This aspect is also highlighted by Dem - the employment share of low-skill labour decreased. ing and Kahn (2018) who use vacancy data and exploit the detailed job descriptions to demonstrate that the demand for skills is heterogeneous within occupations, *Correspondence: laura.zilian@uni-graz.at Graz Schumpeter Centre, University of Graz, Universitätsstraße 15/F, industries and across geographic locations. Hershbein Graz, Austria Full list of author information is available at the end of the article © The Author(s) 2021. This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/. 7 Page 2 of 17 L. S. Zilian et al. and Kahn (2018) also provide evidence for upskilling of level and skill specialisation needed to perform the tasks vacancies during recessions. in a job. The latter two studies place a focus on skills, but they To answer the second question, we use network anal- also indicate that researchers can analyse labour market ysis tools to connect the ESCO classification to the polarisation based on vacancy data as a viable alternative vacancy data. This allows us to determine the skill con - to conducting research based on employment data. Most tent of vacancies. This step is necessary, because no tex - scholars in this field discuss the polarisation of employ - tual description of the jobs is provided in the vacancy ment patterns for currently employed individuals. While database, which would allow the skills to be measured this discussion illustrates shifts that have already taken directly. By taking the network approach, we can identify place (since occupational employment can be consid- and quantify occupation-defining skills even though no ered as the equilibrium outcome of labour supply and direct link exists between ESCO and the Austrian occu- demand), an analysis of job vacancies may shed light pational classification system. Unfortunately, ESCO does on emerging trends, providing information about the not provide information about the complexity of a skill or (unmet) labour demand. We predict that the observable skill level, so we refer to the ILO skill-level categorisation polarisation patterns in employment will also be visible to approximate this missing information. in vacancy data, as technological change is assumed to The rest of the paper is structured as follows: In Sect. 2, affect labour market polarisation through its impact on we briefly review the related literature. The data sources labour demand. used for the analysis are described in Sect. 3. In Sect. 4, u Th s, we applied a promising approach to learn more we present evidence for the development of job vacancies about the technology-driven polarisation of labour in Austria from 2007 to 2017. The results regarding skills markets by analysing job vacancy data and the skills demanded in vacancies are presented in Sect. 5. In Sect. 6 demanded in vacancies. we compare our skill level ranking approach to a wage- This paper contributes to the existing literature in two based occupational ranking approach, and, in Sect. 7, we ways: First, we present the results of an evaluation of discuss the findings and present our conclusion. labour market polarisation based on Austrian vacancy data and, second, we examine the changes in skills that 2 Related literature are demanded to fill these job vacancies based on infor - Several studies suggest that methods used to analyse mation derived from the classification of European polarisation patterns need further refinement. For exam - Skills, Competences, Qualifications and Occupations ple, Caines et al. (2017) introduce a measure of task com- (ESCO). In particular, we address the following research plexity and demonstrate that, based on this measure, questions: routine intensity does not significantly predict wages and wage growth. Furthermore, Salvatori (2018) shows that the changing the skill mix acts as an important driver • Can we observe polarisation patterns in Austrian for occupational polarisation in the UK. The findings vacancy data by performing a skill ranking of vacan- of Alabdulkareem et al. (2018) support these results. cies rather than occupational groups? These authors use network analysis methods to dem - • Can we observe changes in skills demanded in the onstrate that workplace skills in the USA are polarised. Austrian vacancy data? Their results indicate that the polarisation of workplace skills is connected to the hollowing out of the mid- To answer the first question, we analyse job vacancy data dle class. In other research, network analyses were per- provided by the Public Employment Services Austria formed on big data extracted from publications, course (PES) using skill level categories. In contrast to previ- syllabi and job advertisements published between 2010 ous research in which wage-based occupational rankings and 2016. The analysis results show that unique human (Goos et al. 2014) or task-based approaches (Autor et al. skills, e.g. soft skills such as communication, presenta- 2003) were used, we apply the comprehensive skill level tion and teamwork, are needed to complement technical classification framework for occupations proposed by and engineering skills. Specifically Börner et al. (2018) the International Labour Organization (ILO 2012). This use a Multivariate Hawkes Process model and construct framework provides four broad skill level categories a directed network to analyse how soft and hard skills characterised by several dimensions related to the skill influence each other. The authors found that specific hard skills can predict specific soft skills and vice versa. One The skill levels are operationalised based on the ISCO major group, the level major finding of their network analysis is that “soft and of formal educational requirements for performing these tasks, the amount of informal on-the-job training and job-related experience required to perform hard skills influence each other recursively in a continu - the tasks in an occupation. See Table in Appendix for a description of the skill ous cycle” (Börner et al. 2018, p. 12636). These findings levels. Labour market polarisation revisited: evidence from Austrian vacancy data Page 3 of 17 7 indicate that soft skills and technical data science or engi- Moreover, because of the increasing technological capa- neering skills are tightly connected and, in fact, that it bilities to substitute for human labour, it is crucial to under- is difficult to disentangle them. This conclusion is quite stand the changing skills and skill mixes more thoroughly. similar to one that can be drawn from a review of the Acemoglu and Restrepo (2018) point out that, even though economics literature on task-based technological change, productivity gains have positive effects due to automation, namely, that jobs usually consist of a combination of which leads to rising labour demand under certain condi- many complementary abstract, manual, routine and non- tions, a shortage in worker’s skills could have significant routine tasks (Autor 2015). negative effects with far-reaching implications for inequality. Börner et al. (2018) present an argument regarding the They highlight the dangers associated with workers acquir - complementarity of skills that is supported by the results ing the wrong set of skills, and especially in the upcoming of Deming (2017) and Grundke et al. (2018). The authors era, when the use of technologies such as artificial intelli - highlight the increasing importance of specific skills and gence (AI) will become wide-spread. skill bundles for labour market outcomes, such as wages. Overall, the review of the relevant literature reveals the Deming (2017) stresses the fact that employees with a com- importance of conducting further research on polarisa- bination of social and mathematical skills receive higher tion patterns. Notably, a promising approach seems to rewards than those with other skills and skill combinations, be to shift the focus of the analysis from wages to skills but Grundke et al. (2018) show that workers with certain and to analyse vacancies rather than employment. By skills (i.e. higher levels of self-organisation and advanced taking this approach, we can—to some extent—to disen- numeracy) receive higher rewards in digital intensive indus- tangle labour demand from labour supply forces. Due to tries compared to less digitally intensive industries. Finally, the nature of our data, we cannot study within-occupa- Anderson (2017) illustrates that the applicability of skills tion change; instead, we address the issue of skill shifts matter: Even when the returns to a diverse set of skills are between occupations by considering overlapping skills higher than returns to more specialised skills, workers and then by investigating changes in the skill structure. whose skills can be applied in many different jobs (ubiqui - tous skills) receive fewer rewards than workers who have 3 Data combined skills that allow them to fill gaps in the labour For our analysis we use data from two different databases: market. the labour market database provided by the Public Employ- While all of these results indicate that the demand for ment Services Austria (PES) and the ESCO database pro- skills is changing, the actual shift of skill demand in the vided by the European Commission. In the following recent past is quantified in few empirical studies, for exam - section we discuss each database and its limitations. ple, Goos et al. (2014), but they place a focus on the cur- rently employed. 3.1 The labour market database However, only studying changes in employment shares The labour market database houses a great deal of infor - may neglect important aspects of the actual demand for mation, including information about open vacancies reg- skills, as shown by Hershbein and Kahn (2018) and Deming istered with the Public Employment Services Austria. To and Kahn (2018). On the one hand, employment serves as focus on more recent developments, we extracted data a proxy for skill demand and skill requirements simultane- on job vacancies (i.e. stock of vacancies) from 2007 to ously. This is due to the fact that occupational employment 2017; this limited our ability to compare our results with is the equilibrium outcome of labour demand and labour those from studies that cite pronounced polarisation supply, which is assumed to be inelastic in the short run. tendencies in the 1990s. However, since the time period This belief is challenged by Deming and Kahn (2018), who under consideration starts around a time where “[m]ost find substantial heterogeneity in skill requirements even of the old industries have been rejuvenated by the ICT within narrowly defined occupations, using job vacancies revolution and are all poised to innovate” (Perez 2013, p. provided by “Burning Glass Technologies”. On the other 13), our analysis results offer some insight into the labour hand, analyses based on employment at the occupational market effects of more recent technological trends, such level neither capture the heterogeneity of skill demand and as AI and advanced robotics. skill mix changes at the intensive margin (i.e. skills required The variable “stock of vacancies” reports the number of within occupations) nor measure overlapping skills (i.e. job vacancies per occupational group registered on the skills required by several occupations). The latter implies record date for the observation period set. The data are that shifts in skill mixes at the extensive margin (between occupations) cannot be adequately captured when employ- Note that these data do not provide any information on the content of the job openings (e.g. the text of the job ads, the job advertiser), but instead ment-based approaches are taken. reports the number of job posts for each occupational group coded according to the national classification scheme. 7 Page 4 of 17 L. S. Zilian et al. reported on a monthly basis and vary considerably dur- Table 1 Mapping of ISCO-08 major groups to skill levels according to International Labour Organization (2012). Source: ing the year due to variations in seasonal labour demand. ILO Hence, to obtain comparable yearly data, the public employment service calculates and reports the arithmetic ISCO-major group Skill level mean of the monthly values for each year. Consequently, 1 Managers 3 + 4 the reported figures can be interpreted as the seasonally 2 Professionals 4 adjusted monthly average for every year. As the vacan- 3 Technicians and Associate Professionals 3 cies provided by the Austrian labour market database are 4 Clerical Support Workers 2 classified according to the national classification scheme 5 Services and Sales Workers 2 (PES-classification), we convert them to an international 6 Skilled Agricultural/Forestry/Fishery Workers 2 classification scheme [the International Standard Clas - 7 Craft and Related Trades Workers 2 sification of Occupations (ISCO-08)], which is based on 8 Plant and Machine Operators, and Assemblers 2 a correspondence table provided by the public employ- 9 Elementary Occupations 1 ment service. The national classification scheme is more 0 Armed Forces Occupations 1 + 2 + 4 detailed than the lowest hierarchical level of the ISCO- Occupations of the submajor group 14 Hospitality, Retail and Other Service 08 (four-digit). For this reason, 5357 PES occupations Managers are at Skill Level 3 are subsumed under 380 ISCO-08 four-digit occupation Each of the three submajor groups is at a different skill level codes. 3.2 The ESCO database routine physical tasks and the completion of basic educa- The ESCO database provides detailed descriptions of tion may be required (ISCED-97 Level 1). At Skill Level 2 occupations and associated skills, competences and (medium(-low)-skill occupations) workers typically carry knowledge. This information is highly relevant for the out physical and socio-cognitive tasks andthe comple- development of labour market policies and vocational tion of first stage secondary education (ISCED-97 Level education schemes. It includes 2942 occupations, 13485 2) up to vocation-specific education (ISCED-97 Level 4) skills/competences and 8136 qualifications. For instance, may be required. Occupations at Skill Level 3 (medium- the occupation “ICT network technicians” is character- high-skill occupations) are characterised by requirements ised by ten essential skills (e.g. adjust ICT system capac- to perform complex technical and practical tasks and ity, analyse network bandwidth requirements, “create require 1–3 years of higher education (ISCED-97 Level solutions to problems”, identify suppliers, use precision 5b). At Skill Level 4 (high-skill occupations), workers tools), three optional skills (e.g. migrate existing data), carry out complex problem-solving, decision-making, or four areas of essential knowledge (e.g. ICT networking creative tasks and 3–6 years of higher education (ISCED- hardware) and nine areas of optional knowledge (e.g. 97 Level 5a and higher) are required. Skill Level 3 and electronics principles). While some of these skills are Skill Level 4 can also be summarised as high-skill occu- highly specific (e.g. “adjust ICT system capacity”), others pations (International Labour Organization 2020). are quite general (e.g. “create solutions to problems”) and are applicable in many other occupations as well. Each 3.4 Survey data on employment and job vacancies ESCO occupation can be mapped to one ISCO-08 four- To contextualise our analyses with respect to previous digit occupation. This hierarchical structure provides the research on labour market polarisation and highlight evi- basis for our further analysis, as we use the ISCO-08 to dence of potential bias within our data source, we also link the ESCO descriptions to the vacancy data described included data from two publicly available data sources in Sect. 3.1. provided by Statistics Austria in our analyses: (i) the number of employees per ISCO two-digit occupation 3.3 The ILO skill levels based on the European Labour Force Survey (LFS) and The International Labour Organization (ILO 2012) cat - (ii) data on job vacancies collected by the Job Vacancy egorises ISCO-08 major and sub-major groups (ISCO-08 Survey (JVS). One main advantage of these data sources one- and two-digit codes) according to four different skill is their public availability, but they also have significant levels (Table 1): Workers in occupations at Skill Level 1 drawbacks regarding time and classification consistency. (low-skill occupations) typically perform simple and The Classification of European Skills, Competences, Qualifications and Occupations was developed by the European Commission in cooperation Note, that (extensive) on-the-job training and (prolonged) experience can with the CEDEFOP and various stakeholders from 2016 until 2018. substitute for formal education in some cases at Skill Levels 2, 3 and 4. Labour market polarisation revisited: evidence from Austrian vacancy data Page 5 of 17 7 Table 2 Shares of ISCO-08 major groups in percent in 2013 and 2017 ISCO-major groups 2013 2017 PES JVS Empl. PES JVS Empl. 1 Managers 1.1 2.96 4.95 1.52 2.88 4.96 2 Professionals 5.49 11.39 18.41 5.61 13.03 20.14 3 Technicians and Associate Professionals 14.4 18.88 21.82 13.95 19.27 22.16 4 Clerical Support Workers 5.02 6.71 9.70 5.50 6.19 9.36 5 Services and Sales Workers 25.03 31.83 20.14 24.17 24.62 21.02 a a 6 Skilled Agricultural/Forestry/Fishery Workers 0.53 0.78 – 0.37 0.76 – 7 Craft and Related Trades Workers 28.4 14.66 15.18 26.15 18.61 12.76 8 Plant and Machine Operators. and Assemblers 4.52 6.54 6.91 5.55 6.24 a a 9 Elementary Occupations 14.39 8.27 3.25 15.82 9.10 3.37 Due to sampling errors some data had to be excluded (indicated with ) Data are based on the Job Vacancy Survey (JVS) provided by Statistics Austria, open vacancies registered with the Public Employment Services Austria (PES) and data on employment (Empl.) collected using the European Labour Force Survey First, the JVS, which was implemented in 2009, used in absolute values, by examining the development of broader occupational groups prior to 2013. Data on the shares of vacancies (skills) by skill level. Since we look one-digit level of ISCO-08 are only available from 2013 at the relative growth of different skill shares, i.e. we onward. Second, until 2010, the LFS data were classified study changes in the job vacancy structure, this low- according to the predecessor classification of ISCO-08, skill bias of the data should not dramatically affect our namely, ISCO-88. While a conversion between the two results as long as the degree of underrepresentation is feasible, ISCO-88 is seriously outdated in some areas, remains relatively stable. However, when we compare most notably where technological change has affected the PES data with the JVS data (see Table 2), we can the nature of the occupations significantly. This and other see an increase in the coverage rate of the former; this limitations of the data used are discussed in the next means that the PES data have become more accurate section. over time as the spectrum of job vacancies increases. This aspect influenced the interpretation of our results, as the more comprehensive inclusion of high-skill 3.5 Limitations vacancies could have contributed to developments in One limitation concerns of this study concerns the rep- that segment. Thus, our results need to be interpreted resentative nature of the vacancies in the labour market cautiously. database. Only 40% to 60% of vacancies are advertised Another limitation concerns inconsistencies between via the public employment services. Furthermore, different classifications. We lose unique information unlike studies based on online vacancies which are in ESCO when we connect the ESCO skills, which are biased towards high-skill jobs (Deming and Kahn 2018; linked to 2942 ESCO occupations, to the more aggre- Hershbein and Kahn 2018), our study is based on the gated ISCO-08 four-digit occupations. These 427 occu - vacancies advertised at PES, which are biased towards pations are further reduced to 380 ISCO-08 four-digit low-skill jobs (Edelhofer and Käthe 2013). If we com- occupations due to the conversion between the PES pare the shares of each ISCO major group of the vacan- classification and ISCO-08. Moreover, to conduct anal - cies registered with the public employments services yses based on the skill level categorisation as described to the vacancies collected by the Job Vacancy Survey by Goos et al. (2014, 2009), we had to reclassify the data and the number of employees (Table 2), we see that according to ISCO-88 (see Sect. 6). This results in the Managers and Professionals are particularly underrep- reduction of 43 ISCO-08 two-digit submajor groups resented in our main data source and that both Craft to 28 ISCO-88 two-digit groups. In addition, nine and Related Trade Workers and Elementary Occupa- ISCO-88 two-digit groups had to be omitted from the tions are overrepresented. Thus, we recognise that our employment data set due to sampling errors. data only cover parts of the Austrian labour market and Finally, due to the chosen time period, the results especially high-skill vacancies are underrepresented. presented in this paper are not directly compara- This consideration is important when comparing our ble with evidence presented in the original studies on results with previous findings of studies that tested the labour market polarisation that find strong polarisation polarisation hypothesis. We try to circumvent this bias 7 Page 6 of 17 L. S. Zilian et al. Fig. 1 Percentage change of the shares of occupation in total vacancies by ISCO-08 1-digt from 2007 to 2017 (Source: PES) tendencies for the 1990s (Autor et al. 2003, 2008; Goos to 5% in 2007 and 6% in 2017. In our opinion, the magni- and Manning 2007; Goos et al. 2014), a time when rou- tude and increase of this share are negligible. tine-task replacing technologies started to permeate Although these issues are interesting in their own right, the economy. we do not address them explicitly in this paper; instead we focus on our main research goal, which is to study polarisation. To achieve this goal, we use the share of 4 Are vacancies in Austria polarised? each occupation ( v ) in total vacancies ( κ = v / v ) and i i i i i∈I In the short run, the number of job vacancies reflects calculate their growth factor over the whole observation business cycle fluctuations, i.e. during an economic period from 2007 ( t = 1 ) until 2017 ( t = 2 ) using to downturn the stock of job vacancies decreases and Eq. 1. Figure 1 shows the percentage changes of the during an economic upswing the stock of vacancies shares of vacancies, categorised by ISCO 1- digit major increases. The period of 2007 to 2017 was characterised groups. While the shares of Managers, Professionals, by the financial crisis of 2007 which was followed by the Technicians and Associate Professionals, Clerical Sup- Euro crisis. These business cycle trends are also visible in port Workers and Service and Sales Workers increased, our vacancy data (see Table 4 in Appendix). they decreased for Skilled Agricultural, Forestry and Another determinant of job posting behaviour is Fishery Workers, Craft and Related Trade Workers, Plant labour shortage. To determine whether the development and Machine Operators and Assembler and Elementary of vacancies could be traced back to changes in labour Occupations. These results do not allow us to identify shortages, we used a list of shortage occupations in Aus- polarisation trends on the 1-digit level. tria in 2016 [as defined by the CEDEFOP (2016)] and Finally, to illustrate to the extent to which vacan- calculated the share of shortage occupations among all cies in different skill level groups have gained (or lost) vacancies in our observation period. This share amounts Labour market polarisation revisited: evidence from Austrian vacancy data Page 7 of 17 7 Fig. 2 Growth factor of fraction of vacancies by skill level from 2007 to 2017 after correcting for outliers outside the 1.5 IQR. Every violin describes the distribution of the data for each skill level category. The growth factor is calculated for the shares of each occupation in total vacancies and plotted against its respective skill level, where n refers to the number of ISCO-08 four-digit occupations falling into the respective skill level category. The skill levels are assigned to each ISCO-08 four-digit occupation according to the ILO classification (Source: PES, ILO) relative importance, we categorise the vacancies into between 2007 and 2017. Thus, we do not observe a polar - the four skill level groups provided by the ILO classifi - isation of job vacancies (in the sense of a hollowing out cation of the ISCO major groups described in Sect. 3.3. of the middle-skill jobs), but rather a trend towards the This allows us to examine the distribution of growth growing importance of medium-high-skill and high-skill factors within skill level categories. We assign each vacancies. vacancy a skill level according to the ISCO major group to which it belongs. This yields 106 high-, 74 medium- 5 From jobs to skills: a network approach high, 174 medium-low and 28 low-skill vacancies. One major drawback of the occupation-based approach is that its application only gives rudimentary insights into the skills demanded in vacancies. While it provides an κ − κ it it 1 2 g = overview of the demand in different skill level categories, κ (1) it it does not provide information about the demand for skills at the extensive margin, i.e. skills that are in demand In Fig. 2, the growth factor ( g on the y-axis) of each in various occupations at different skill levels. For this occupation share is plotted against the skill levels (x-axis) reason, we present an alternative approach to study after correcting for outliers outside the 1.5 IQR. Every skills demanded in vacancies. Since the vacancy data do violin describes the distribution of the data for each skill not contain detailed information about the actual skills level category. According to the polarisation hypothesis demanded, we have to rely on this alternative approach one would expect that the fraction of low-skill and high- to infer the skills demanded. Using network analysis tools skill vacancies has increased, while that of medium skills we can easily identify the need for skills in vacancies. has decreased. However, we see in Fig. 2 that the median growth factor of the fraction of vacancies increases with 5.1 Methods the skill level. In fact, on average, the fraction of low and 5.1.1 F rom ESCO to ISCO using network analysis medium-low-skill vacancies decreased, while the fraction We rely on concepts and tools used in network analysis of medium-high-skill and high-skill vacancies increased in our study. According to network theory, each graph 7 Page 8 of 17 L. S. Zilian et al. bipartite network isco(c, s) = esco(j, s) · isco(c, j) . This weighted network consists of 427 occupational groups and 8258 essential skills linked by 22,805 edges, with a median of 41 and an average of 54 skills per occupation. This approach yields ISCO-08 four-digit occupational groups which are characterised by many different skills; i.e. they are quite diverse. Consequently, we need to identify the most relevant skills, when performing these occupations. Hence, we use the concept of revealed comparative advantage (RCA) to identify skills that are over-expressed in ISCO-08 occupational groups. Fig. 3 Bipartite network of jobs (uppercase letters) and skills (lowercase letters) (Source: personal illustration) isco(c, s)/ isco(c, s) s∈S rca (c, s) = isco (2) isco(c, s)/ isco(c, s) c∈C c∈C ,s∈S or network consists of two different components. Nodes (also called vertices) can describe arbitrary objects, in We compare the relative importance of skill s to an occu- our case, skills and occupations. These nodes are con - pational group (numerator in Eq. 2) to the relative impor- nected by links, the so-called edges. Here, we define tance of a skill on aggregate (denominator in Eq. 2). that an occupation and a skill are connected if the skill Next, we construct a network/matrix of essential skills is required to perform the occupation in question. If that E(c, s) = 1 with E ∈{0, 1} if rca > 1 , where an occu- isco skill is not relevant, no direct link is established between pational group c relies on a skill more than expected them. This leads to a network within which an occupa - on aggregate. This helps us to identify key occupational tion can be connected to several skills, and also a skill can features and controls for ubiquitous skills (e.g. create be connected to several occupations. However, since the solutions to problems). This process yields a network edges are always between an occupation and a skill, an consisting of 8258 skills and 427 occupations linked by occupation will never be directly connected to another 22,675 edges. occupation. The same is true for skills. This class of net - works, which contains two distinct categories of nodes 5.1.2 D eriving skills demanded in vacancies that are never directly connected, is called a bipartite net- We derive the demand for skills based on the vacancy work. Figure 3 provides an example of a bipartite network data extracted from the labour market database. After where jobs (uppercase letters) are linked to skills (lower- mapping the vacancies from the national classification to case letters). ISCO-08 using the official conversion of PES, we use the We use the methodology described by Alabdulkareem matrix of the monthly average of vacancies in each occu- et al. (2018) and construct a bipartite network of skills s pational group for every year AV(y, c) and the matrix of and occupations j, where we base the existence of an edge essential skills E(c, s) to calculate the average demand for between s and j on ESCO esco(j, s). each occupation-defining skill s per year in Eq. (3). Each essential skill, s ∈ S , is matched to occupations, S(y, s) = AV(y, c) · E(c, s). (3) j ∈ J , using esco(j, s) ∈{0, 1} , whereby esco(j, s) = 1 indi- cates that s is essential to j and esco(j, s) = 0 indicates Hence, we connect the skills which were identified that s is not required for occupation j. This network con - as occupation-specific from the RCA to the vacan - sists of 2937 occupations and 8258 skills linked by 46,062 cies provided by the public employment services via edges their ISCO-08 code. This yields a weighted matrix with The hierarchical structure of ISCO-08 four-digit occu - 380 occupations and 8258 skills. Note that E(c, s) is pational groups, c, and ESCO occupations, j yields the Note that not all of the 436 ISCO-08 occupations contain ESCO occupa- tions. These are typically occupational groups with no economic activity in the EU such as “ fire wood collector”. RCA has been used for different research contexts including the analy - Note that the network of skills and occupations could also be used to deter- sis of exports of nations Caldarelli et al. (2012), Hidalgo et al. (2007), the mine the skill complementarity of occupations as in Alabdulkareem et al. emergence of technology-based sectors Colombelli et al. (2014) or the role (2018). However, this would go beyond the scope of the paper at hand. of technological relatedness for technological change Boschma et al. (2014). Labour market polarisation revisited: evidence from Austrian vacancy data Page 9 of 17 7 time-constant, while the weights of the occupations A = v s i (5) AV(c, y) vary over time due to changes in the posted i=1 vacancies. Next, in Eq. (4), we normalise matrix S by the sum of u Th s, we obtain a continuous skill value for each skill in the monthly averages of vacancies in each year. the range of 0 to 1 ( A ∈[0, 1]). −1 ψ(v, y) = S(y, s) · AV (c, y) (4) 5.2 Results c∈C 5.2.1 Sk ill demand of vacancies in Austria 2007–2017 Each element of the matrix ψ represents the weight of We now deepen the analysis of skills demanded in vacan- occupation-defining skills in relation to all posted vacan - cies by looking at the skill values obtained in the previous cies in a given year. This allows us to interpret the ele - section. To do so, we use the occupation-defining essen - ments of matrix ψ as the relative importance of an tial skills identified by RCA and described in Sect. 5.1.2 occupation-defining skill in a given year. Therefore, by and calculate the growth factor of the relative impor- focusing on ψ , we can analyse and compare the increas- tance of skills in job vacancies ( g in Eq. 6) to analyse the ing or declining importance of each occupation-defining change observed between 2007 and 2017. skill in the posted vacancies. However, we cannot capture ψ − ψ t t 1 2 any shifts in the skill-content within vacancies. g = ψ (6) 5.1.3 Skill levels In Fig. 4, each violin represents the distribution of To analyse the polarisation of skills demanded in vacan- growth factors within predefined skill value ranges. Due cies, we need to assign a skill level to each skill. Since the to the fact that many skills are linked to several occupa- ESCO does not provide information on the complexity tions, which fall into different skill levels, we obtain a of a skill, we approximate this missing information by more detailed picture of the skill content of vacancies using the ILO skill levels of the one- and two-digit ISCO- in Austria by performing this analysis than we obtained 08 occupations (summarised in Sect. 3.3). First, we re- with the analysis in Sect. 4. From these violin plots, scale the skill levels to obtain values ranging from 0 to we gain two main insights. First, the dispersion of the 1, whereby Skill Level 1 (low) corresponds to the value 0 growth factor increases with the skill value. This implies and Skill Levels 2 (medium-low), 3 (medium-high) and 4 that the development of the relative importance of occu- (high) correspond to the values 0.33, 0.66 and 1, respec- pation-defining skills in the top-skill ranges was more tively. Next, each ISCO-08 four-digit occupation is cate- heterogeneous than it was for those in the bottom-skill gorised according to its skill level group (see also Sect. 4). ranges. Second, on average we observe a positive rela- Based on the skill level values (v) obtained for each occu- tionship between the skill value and the growth level of pation (c), we infer unique skill values for each skill s in the relative importance of the skills. This can be seen E(c, s) . Since 48% of all skills are linked to more than one when we compare the arithmetic mean (red square) in ISCO-08 four-digit occupation and 70% percent of these Fig. 4 across the skill value ranges. If we then compare occupations belong to more than one skill level category, the means of the growth factors, we also see that the rela- we need to calculate the average skill value of each skill tive importance of occupation-defining skills in the bot - (s). We assume that all occupation-defining skills are tom skill range decreased the most, while those in the top equally important for each ISCO-08 occupation to which skill range increased the most. Thus, based on these data, they are linked. Similarly, the ISCO-08 occupations to we do not see a polarisation pattern but rather a general which a skill is linked to by E(c, s) are equally important trend: It became more important over this time period for determining the skill level values of the respective for applicants to have top skills in order to fill the vacan - skill. Hence, we take the arithmetic mean ( A ) of the skill cies posted vacancies posted with the public employment level values (v) of those ISCO-08 four-digit occupations services. to which the respective skill is linked to by E(c, s) = 1 , given in Eq. 5. 5.2.2 Quantile regression u Th s far, we only examined descriptive evidence which suggests a positive relationship between the skill level and the growth factor of the relative importance of skills. However, due to the continuous nature of the derived skill values of the occupation-specific skills, we can also In addition, we calculated the median skill level of each skill and obtained explore the relationship between changes in the relative similar growth factors. The results are provided in Appendix . 7 Page 10 of 17 L. S. Zilian et al. Fig. 4 Growth factors of the relative importance of occupation-specific skills by skill value ranging from 2007 to 2017. Each violin represents the distribution of growth factors within predefined skill value ranges. The growth factor is calculated for the weight of occupation-defining skills in relation to all vacancies in a given year. The skill value ranges are calculated using network analysis methods and are based on ILO skill levels and ESCO (Source: PES, ESCO, ILO) importance of skills and their skill value more thoroughly. skill value s of each skill), and τ denotes the quantiles 0.2, To do so, we use a quantile regression approach which 0.3, 0.4, 0.5, 0.6, 0.7 and 0.8. Fitting Eq. 7 yields estimates was first introduced by Koenker and Bassett (1978). for the seven conditional quantiles of the growth fac- Quantile regression is used to estimate quantiles of the tor given the skill level (see Table 5). The coefficients are independent variable distribution and provide a more interpreted as the effect of a unit change in the depend - complete picture of the distribution than OLS, which is ent variable on quantiles of the distribution of the inde- used to estimate the mean. As our main interest lies in pendent variable. To make the regression results more studying the phenomenon of polarisation, we can use accessible, they can also be examined graphically. quantile regression to determine what happens at the In Fig. 5, each black dot represents the slope coefficient tails of the distribution. More precisely, we can study the for the quantile indicated on the x-axis and the lines with relationship between the skill level and the growth fac- the 95% confidence envelope connect them, whereas the tor for skills at different quantiles of the change between red line corresponds to the least square estimates and the two observation periods. To achieve this purpose, its confidence interval. The graphs clearly show that we estimate a conditional quantile function (CQF) of the the skill-value effect on the growth factor of the relative growth factor of relative importance of a skill ( g ) as a importance of skills is significantly positive. However, −1 function of skill value (s), Q (τ|s) = F (τ|s) where the skill-value effect is much lower at the lower quan - ψ g τ ∈[0, 1] denotes the quantiles (Hao and Naiman 2007). tiles than at the higher quantiles. But while the size of The econometric quantile regression model can then be the effect increases with the skill value in absolute terms, expressed as: (τ ) (τ ) (τ ) y = β + β x + ǫ , (7) i i 0 1 i The descriptive statistics and regression coefficients are provided in Appen - where y is the dependent variable (i.e. the growth factor dix. As most of the quantiles lie beyond the least square estimates, an OLS g of each skill), x is the independent variable (i.e. the ψ i regression does not adequately describe the data at hand. Labour market polarisation revisited: evidence from Austrian vacancy data Page 11 of 17 7 Fig. 5 Quantile regression of growth factors of the relative importance of skills. Each dot represents the slope coefficients for the quantile indicated on the x-axis. The lines with the 95% confidence envelop connect them. The solid red line corresponds to the least square estimates and its confidence interval (dashed red lines) (Source: PES, ESCO, ILO) the relative increase associated with higher skill values wage-based skill level classification described by Goos becomes smaller for higher quantiles of the growth fac- et al. (2014, 2009) and, on the other hand, the ILO skill tor. Put differently, the skill value effect becomes weaker level classification of occupations. in the upper half of the distribution of the growth factors. To follow the approach of Goos et al. (2014), we cat- Consequently, the results from the quantile regression egorise the PES-vacancies and the Austrian employment confirm the positive relationship between the skill value data described in Sect. 3.4 according to the four lowest- and the growth factor of the relative importance of a skill. paying, nine middling and eight highest-paying occupa- In addition, these results reveal that the size of the slope tions provided by Goos et al. (2014). Moreover, using coefficient increases as the skill value increases, albeit at a the approach described in Sect. 5.1.3, we assign the skill diminishing rate. levels defined by these authors to the derived occupation- Overall, our evidence does not indicate that a polarisa- defining skills. Likewise, we classify the employment data tion of skills demanded in vacancies is occurring; instead, according to the skill levels of the ILO classification based it highlights the shift that is taking place to assign on the ISCO-08 two-digit submajor occupational groups increasing relative importance to higher skills. described in Sect. 3.3. In order to be able to compare our categories more effectively with the three catego - 6 Are our results driven by our indicators? ries described by Goos et al. (2014, 2009), we reduce the Our approach deviates from that taken in most of the number of ILO skill level categories from four to three labour market polarisation literature and, when apply- by grouping the medium-high and high skill levels under ing this approach, we did not find polarisation patterns. one category (high skill). We need to perform this step u Th s, the question arises whether our results are driven with the vacancy and employment data, but not with the by the approach we have chosen. More specifically, are these results driven by the analysis of the growth factor of skill shares, by the occupation-based derivation of skill Note that Goos et al. (2014) use the ISCO-88 classification, hence, we values and, consequently, the skill levels, or by our data use the conversion table provided by the ILO to reclassify our employment, source? To address this question, we compare changes vacancy and skill data according to ISCO-88. To analyse the skill levels, we decided to use the ISCO-08 classification of skill levels as this classification in the structure of employment, vacancies and skills. To is highly similar to the classification of ISCO-88 skill levels (International perform this comparison, we use, on the one hand, the Labour Organization 2020). 7 Page 12 of 17 L. S. Zilian et al. Table 3 Initial shares and percentage point change 2007–2017 (by classification). Source: PES, ESCO, Statistics Austria Employment Vacancies Skills Share 2007 Percentage point Share 2007 Percentage point Share 2007 Percentage (in percent) change 2007–2017 (in percent) change 2007-2017 (in percent) point change 2007–2017 Goos et al. Skill Levels High 39.4 4.2 17.2 29.9 29.5 14.7 Medium 34.0 − 16.7 53.2 − 16.1 45.7 − 4.7 Low 22.3 9.3 26.8 15.9 24.4 20.5 ILO Skill Levels High 43.6 8.3 15.6 31.9 14 22.5 Medium 51.1 − 3.3 68.1 − 7.4 43.1 7.8 Low 5.3 − 36.2 16.2 0.5 42.9 1.8 1.5cmILO Skill Levels High 20.7 21.2 4.5 55.7 10.8 24.9 Medium High 22.9 − 3.4 11.1 22.3 19.6 14.3 Medium Low 51.1 − 3.3 68.1 − 7.4 61.4 2.8 Low 5.3 − 36.2 16.2 0.5 8.2 1.4 Skills refers to the relative importance of an occupation-defining skill in a given year The data are categorised according to the wage-based skill level classification of Goos et al. (2014) (Goos et al. Skill Levels) and the skill levels of the ILO classification based on the ISCO-08 two-digit submajor groups (ILO Skill Levels). For comparability, the ILO skill levels are reduced to three skill levels (high, medium and low) skill data, since these are assigned continuous skill values wage-based skill level categories derived from 21 ISCO (see Sect. 5.1.3). two-digit occupations are not detailed. As described in To determine whether polarisation can be observed, Sect. 5.1.2, we have applied our approach in attempt to we calculate the percentage point changes of the shares overcome this drawback by combining information on of employment, vacancies and skills between 2007 and individual skills from ESCO and skill levels based on 2017 within the different categories (Table 3). While ISCO. By taking this approach, we demonstrate that the polarisation patterns can be detected within the catego- same skill can be identified as occupation-defining for ries of Goos et al. (2014, 2009), the polarisation is not a number of different occupations that are assigned to evident when using the ILO categorisations. Instead, one different skill level categories. This finding is interesting can observe a tendency towards upskilling when the ILO in its own right, but more importantly, it highlights the categorisation is used. These findings suggest that these importance of skill shifts at the extensive margin. These patterns are determined by the skill level categorisation shifts have been neglected in approaches that use broadly used; interestingly, this observation was also made by defined occupational groups. Therefore, we believe Fernández-Macías (2012). In particular, he analyses the that the derived skill values are more suitable to analyse polarisation in the given data set. same data as used in Goos et al. (2009) and shows that the polarisation patterns are less pronounced when using a different grouping of employment shares by occupa -7 Discussion and conclusion tions. Moreover, Fernández-Macías (2012) presents an In this paper, we contribute to the literature on the well- alternative approach to uncovering polarisation patterns studied phenomenon of labour market polarisation. Most in European counties using job quality tiers based on studies in the field of economics have placed a focus on wage quintiles and education. Instead of a unified polari - the change of employment shares in low-skill, medium- sation trend, he finds heterogeneous patterns of changes skill and high-skill occupations, wage-based occupa- in the employment structure between 1995 and 2007. tion classifications, or task-based classifications (routine One drawback of Goos et al. (2009), which is also or non-routine). While all these studies indicate that a underlined by Fernández-Macías (2012), is that the polarisation of employment has been taking place, they Labour market polarisation revisited: evidence from Austrian vacancy data Page 13 of 17 7 disguise the heterogeneity of different skills within occu - rate. Overall, our results indicate that labour demand, pations (Deming and Kahn 2018). Moreover, a network as represented by vacancies, is characterised by upskill- analysis revealed the fact that the network of workplace ing rather than polarisation. This finding could (partly) skills itself is polarised (Alabdulkareem et al. 2018). be explained by the chosen time period: While the origi- We extend the existing research by studying three dif- nal studies on labour market polarisation identified the ferent aspects: First, like Hershbein and Kahn (2018) strongest polarisation tendencies in the 1990s, we can and Deming and Kahn (2018), we use vacancy data as an examine observations for occupations that have been indicator of (unmet) labour demand. Second, we employ subject to routine-task displacement since the 1980s. network analysis tools proposed by Alabdulkareem et al. Our findings deviate from most empirical results pre - (2018) to identify occupation-defining skills. This identi - sented in the labour market polarisation literature. For fication step allows us to evaluate the changing demand this reason, we calculate changes in the structure of for those skills. Third, instead of relying on wage-based employment, vacancies and skills using the wage-based occupation rankings to approximate skill levels, we turn skill level classification of Goos et al. (2009) and Goos to the skill-level ranking of occupations provided by the et al. (2014) and compare them to the ILO skill level ILO. This ranking is based on the nature of the tasks classification of occupations. This additional analysis performed and the minimum education requirements. provides results that show that the observation of polari- Fernández-Macías (2012) also highlighted the necessity sation is sensitive to the skill level classification used. of turning to an alternative occupation ranking pointing However, as our disentanglement of occupation-defining out inconsistencies regarding wage-based polarisation skills shows, the same skill can be occupation-defining analyses. for several occupations that fall into different skill level Given that technological change has been identified as categories. Consequently, approaches that involve the use the main driver of polarisation, we assume that polarisa- of broadly defined occupational groups but do not take tion is mainly demand-driven and, therefore, this change into account skill shifts at the extensive margin, may pro- should be clearly visible in vacancy data. However, our duce misleading results. Therefore, we believe that the approach did not allow us to identify a polarisation of derived skill values are more suitable to analyse shifts in vacancies in Austria; instead, we detected a trend of labour demand in the case of our data set, which provides increasing relative importance of medium-high-skills the number of vacancies per ISCO four-digit occupation. and high-skills regarding vacancies between 2007 and Nevertheless, to assess the benefits of the derived skill 2017. Furthermore, if we compare the number of skills values, further research should be conducted to try to to the number of job vacancies, we find that the demand apply the approach presented in this paper to comparable for skills in the higher skill value ranges (i.e. skill values data sets for job vacancies. As we used international clas- greater than 0.4) grew on average more than the demand sification systems – ESCO, ISCO and ILO skill levels— for skills in the lower skill ranges. These results are con - our method can easily be used to analyse vacancy data in sistent with the findings of Hershbein and Kahn (2018), other European countries. providing evidence for upskilling during recessions. Even though our method seems promising, some In addition to performing a descriptive analysis based limitations regarding the data and also the methodol- on pre-defined skill value ranges, we exploit the continu - ogy must be discussed. First, the PES vacancy data are ous nature of the derived skill values by applying a quan- biased towards low-skill vacancies. However, during the tile regression analysis. The quantile regression results observation period, the coverage rate of the vacancy data reveal a statistically significant relationship between skill registered at the PES increased. If we compare the struc- values and the 0.2 to 0.8-quantiles of the growth factor ture of the vacancy data, we conclude that the low-skill of skill importance. Furthermore, the size of the effect bias becomes less dominant over the period. u Th s, our increases as the skill values rise but at a diminishing results may partially be driven by this improvement in 7 Page 14 of 17 L. S. Zilian et al. the coverage of high-skill vacancies. This aspect requires approach presented in this paper provides evidence for further investigation. Second, the skill structure within the shift of relative importance towards high skills in occupations changes over time but it is not possible for Austrian labour demand, further research is needed to us to capture this change with the data used. Hence, we gain additional insights into the evolution of labour mar- may actually have underestimated the extent of upskill- kets. Such research would include the complementary ing that took place from 2007 to 2017 if, if upskill- analysis of unemployment data to capture possible struc- ing also occurred within occupations. Another aspect tural changes in labour supply. This would also allow us that we cannot adequately address with the available to study the extent of skill mismatch in labour markets. data, is related to qualitative changes in the skill struc- The network approach used by Alabdulkareem et al. ture; namely, some skills become obsolete while others (2018), which we partially applied in this work, seems emerge. Third, no table is yet available to convert ESCO promising in this respect, if ESCO is applied to European occupations to PES occupations. Thus, we have to con - labour market data in the future. Furthermore, although vert ESCO and PES occupations to ISCO four-digit occu- our initial assumption was that technological change pations, and the latter is is a less detailed classification is the main driver of labour market polarisation, we did than both of the former. As a result, highly specific skills not attempt to conduct a causal analysis of the impact of that are important and thus define for ESCO occupations technological change on labour demand. However, our (e.g. “supervise camp operations” for the ESCO occupa- results indicate that future research on labour market tion “camping ground manager”) are identified as occu - polarisation needs to extend the focus beyond employ- pation-defining skills for much broader ISCO four-digit ment shifts and use classifications other than wage-based occupations (e.g. 1439 “Services managers not elsewhere skill level classifications. Specifically, this research should classified”). As a result, we cannot meaningfully name be extended to incorporate alternative skill level meas- the actual skills that have gained in relative importance, ures and other labour market indicators such as vacan- although this would be feasible from a technical point cies and unemployment. of view if a full integration of ESCO were possible at the national level. Therefore, we lose a lot of unique informa - tion due to data aggregation. Appendix Finally, this study allowed us to identify several inter- See Tables 4, 5, 6; Fig. 6 esting future research directions. While the data-driven Table 5 Austria: quantile regression results. Source: PES, ESCO, Table 4 Monthly average stock of vacancies in Austria 2007– ILO 2017. Source: PES Quantiles (Intercept) Skill value Year Vacancies ∗∗∗ ∗∗∗ 0.2 0.529 0.349 ( 0.014) ( 0.023) 2007 37,461 ∗∗∗ ∗∗∗ 0.3 0.545 0.550 ( 0.012) ( 0.024) 2008 36,776 ∗∗∗ ∗∗∗ 0.4 0.582 0.709 2009 26,736 ( 0.011) ( 0.024) ∗∗∗ ∗∗∗ 0.5 0.616 0.864 2010 30,593 ( 0.018) ( 0.032) ∗∗∗ ∗∗∗ 2011 31,798 0.6 0.721 0.917 ( 0.017) ( 0.027) 2012 28,858 ∗∗∗ ∗∗∗ 0.7 0.867 0.937 ( 0.023) ( 0.035) 2013 25,833 ∗∗∗ ∗∗∗ 0.8 1.085 0.876 ( 0.024) ( 0.038) 2014 25,675 Standard errors are in parentheses 2015 28,484 ***p < .01 2016 39,034 2017 54,477 Labour market polarisation revisited: evidence from Austrian vacancy data Page 15 of 17 7 Table 6 Skill levels according to the International Labour Organization (ILO 2012). Skill Level Description Level 1 Occupations at Skill Level 1 typically involve the performance of simple and routine physical or manual tasks. They may require the use of hand-held tools, such as shovels, or of simple electrical equipment, such as vacuum cleaners. They involve tasks such as cleaning; digging; lifting and carrying materials by hand; sorting, storing or assembling goods by hand (sometimes in the context of mechanized opera- tions); operating non-motorized vehicles; and picking fruit and vegetables. Many occupations at Skill Level 1 may require physical strength and/or endurance. For some jobs basic skills in literacy and numeracy may be required. If required these skills would not be a major part of the work. For competent performance in some occupations at Skill Level 1, completion of primary education or the first stage of basic education (ISCED-97 Level 1) may be required. A short period of on-the-job training may be required for some jobs. Occupations classified at Skill Level 1 include office cleaners, freight handlers, garden labourers and kitchen assistants Level 2 Occupations at Skill Level 2 typically involve the performance of tasks such as operating machinery and electronic equipment; driving vehi- cles; maintenance and repair of electrical and mechanical equipment; and manipulation, ordering and storage of information. For almost all occupations at Skill Level 2 the ability to read information such as safety instructions, to make written records of work completed, and to accurately perform simple arithmetical calculations is essential. Many occupations at this skill level require relatively advanced literacy and numeracy skills and good interpersonal communication skills. In some occupations these skills are required for a major part of the work. Many occupations at this skill level require a high level of manual dexterity. The knowledge and skills required for competent perfor- mance in occupations at Skill Level 2 are generally obtained through completion of the first stage of secondary education (ISCED-97 Level 2). Some occupations require the completion of the second stage of secondary education (ISCED-97 Level 3), which may include a signifi- cant component of specialized vocational education and on-the-job training. Some occupations require completion of vocation-specific education undertaken after completion of secondary education (ISCED-97 Level 4). In some cases experience and on-the-job training may substitute for the formal education. Occupations classified at Skill Level 2 include butchers, bus drivers, secretaries, accounts clerks, sewing machinists, dressmakers, shop sales assistants, police officers, hairdressers, building electricians and motor vehicle mechanics Level 3 Occupations at Skill Level 3 typically involve the performance of complex technical and practical tasks that require an extensive body of factual, technical and procedural knowledge in a specialized field. Examples of specific tasks performed include: ensuring compliance with health, safety and related regulations; preparing detailed estimates of quantities and costs of materials and labour required for specific projects; coordinating, supervising, controlling and scheduling the activities of other workers; and performing technical functions in support of professionals. Occupations at this skill level generally require a high level of literacy and numeracy and well-developed interpersonal communication skills. These skills may include the ability to understand complex written material, prepare factual reports and communicate verbally in difficult circumstances. The knowledge and skills required for competent performance in occupations at Skill Level 3 are usually obtained as the result of study at a higher educational institution for a period of 1–3 years following completion of secondary education (ISCED-97 Level 5b). In some cases extensive relevant work experience and prolonged on-the-job training may substitute for the formal education. Occupations classified at Skill Level 3 include shop managers, medical laboratory technicians, legal secretaries, commercial sales representatives, diagnostic medical radiographers, computer support technicians, and broadcasting and recording technicians Level 4 Occupations at Skill Level 4 typically involve the performance of tasks that require complex problem-solving, decision-making and creativity based on an extensive body of theoretical and factual knowledge in a specialized field. The tasks performed typically include analysis and research to extend the body of human knowledge in a particular field, diagnosis and treatment of disease, imparting knowledge to others, and design of structures or machinery and of processes for construction and production. Occupations at this skill level generally require extended levels of literacy and numeracy, sometimes at a very high level, and excellent interpersonal communication skills. These skills usually include the ability to understand complex written material and communicate complex ideas in media such as books, images, performances, reports and oral presentations. The knowledge and skills required for competent performance in occupations at Skill Level 4 are usually obtained as the result of study at a higher educational institution for a period of 3–6 years leading to the award of a first degree or higher qualification (ISCED-97 Level 5a or higher). In some cases extensive experience and on-the-job training may substitute for the formal education, or may be required in addition to formal education. In many cases appropriate formal qualifications are an essen- tial requirement for entry to the occupation. Occupations classified at Skill Level 4 include sales and marketing managers, civil engineers, secondary school teachers, medical practitioners, musicians, operating theatre nurses and computer systems analysts 7 Page 16 of 17 L. S. Zilian et al. Fig. 6 Growth factor of the relative importance of occupation-specific skills by (median) skill value range from 2007 to 2017. Each violin represents the distribution of growth factors within predefined skill value ranges.The growth factor is calculated for the weight of occupation-defining skills in relation to all vacanciesin a given year. The skill value ranges are calculated using network analysis methods and are based on ILO skill levels and ESCO (Source: PES, ESCO, ILO) Acknowledgements References We are very grateful to Prof. Alexandra Spitz-Oener, Prof. Heinz D. Kurz, Dr. Acemoglu, D., Restrepo, P.: Artificial intelligence, automation, and work. In: Marlies Schütz and two anonymous referees for their valuable comments. the economics of artificial intelligence: an agenda, National Bureau of Economic Research, pp 197–236 (2018) Authors’ contributions Alabdulkareem, A., Frank, M.R., Sun, L., AlShebli, B., Hidalgo, C., Rahwan, I.: All authors read and approved the final manuscript. Unpacking the polarization of workplace skills. Sci. Adv. 4(7), eaao6030 (2018) Funding Anderson, K.A.: Skill networks and measures of complex human capital. Proc. This work was supported by the Austrian Science Fund (FWF) (Project Num- Natl. Acad. Sci. 114(48), 12720–12724 (2017) ber: P 30434-G27). Arntz, M., Gregory, T., Zierahn, U.: The risk of automation for jobs in oecd coun- tries: a comparative analysis. OECD social, employment, and migration Availability of data and materials working papers (189):0\_1 (2016) The data that support the findings of this study are available from Public Autor, D.H.: The “task approach’’ to labor markets: an overview. J. Lab. Mark. Res. Employment Services Austria but restrictions apply to the availability of these 46(3), 185–199 (2013). https ://doi.org/10.1007/s1265 1-013-0128-z data, which were used under license for the current study; thus they are not Autor, D.: Why are there still so many jobs? the history and future of workplace publicly available. Data are, however, available from the authors upon reason- automation. J. Econ. Perspect. 29(3), 3–30 (2015) able request and with the permission of Public Employment Services Austria. Autor, D.H., Dorn, D.: The growth of low-skill service jobs and the polarization of the us labor market. Am. Econ. Rev. 103(5), 1553–1597 (2013) Declarations Autor, D.H., Levy, F., Murnane, R.J.: The skill content of recent technological change: an empirical exploration. Quart. J. Econ. 118(4), 1279–1333 Competing interests (2003) The authors declare that they have no competing interests. Autor, D.H., Katz, L.F., Kearney, M.S.: Trends in us wage inequality: revising the revisionists. Rev. Econ. Stat. 90(2), 300–323 (2008) Author details Bainbridge, S.: In the future, what will people do? Dolphin, T (ed). Technology, Graz Schumpeter Centre, University of Graz, Universitätsstraße 15/F, Graz, globalisation and the future of work in Europe. pp 80–86 (2015) Austria. Institute of Systems Sciences, Innovation and Sustainability Research, Berger, T., Frey, C.B.: Bridging the skills gap. Dolphin, T (ed). Technology, Glo- University of Graz, Merangasse 18, Graz, Austria. Vienna University of Eco- balisation and the future of work in Europe essays on employment in a nomics and Business, Welthandelsplatz 1, Vienna, Austria. digitised economy London. pp 75–79 (2015) Börner, K., Scrivner, O., Gallant, M., Ma, S., Liu, X., Chewning, K., Wu, L., Evans, Received: 12 December 2019 Accepted: 2 March 2021 J.A.: Skill discrepancies between research, education, and jobs reveal the Labour market polarisation revisited: evidence from Austrian vacancy data Page 17 of 17 7 critical need to supply soft skills for the data economy. Proc. Natl. Acad. Goos, M., Manning, A., Salomons, A.: Explaining job polarization: routine-biased Sci. 115(50), 12,630-12,637 (2018) technological change and offshoring. Am. Econ. Rev. 104(8), 2509–2526 Boschma, R., Balland, P.A., Kogler, D.F.: Relatedness and technological change (2014) in cities: the rise and fall of technological knowledge in us metropolitan Graetz, G., Michaels, G.: Robots at work. Rev. Econ. Stat. 100(5), 753–768 (2018) areas from 1981 to 2010. Ind. Corp. Change 24(1), 223–250 (2014) Grundke, R., Marcolin, L., Squicciarini, M., et al.: Which skills for the digital era?: Caines, C., Hoffmann, F., Kambourov, G.: Complex-task biased technological returns to skills analysis. OECD science, technology and industry working change and the labor market. Rev. Econ. Dyn. 25, 298–319 (2017) papers 2018(9), 0\_1-37 (2018) Caldarelli, G., Cristelli, M., Gabrielli, A., Pietronero, L., Scala, A., Tacchella, A.: A Hao, L., Naiman, D.: Quantile regression. SAGE Publications, Inc, Thousand Oaks network analysis of countries’ export flows: firm grounds for the building (2007) blocks of the economy. PLoS ONE 7(10), e47,278 (2012) Hershbein, B., Kahn, L.B.: Do recessions accelerate routine-biased technologi- CEDEFOP (2016) Austria: Mismatch priority occupations. https ://skill spano cal change? Evidence from vacancy postings. Am. Econ. Rev. 108(7), rama.cedef op.europ a.eu/en/analy tical _highl ights /austr ia-misma tch-prior 1737–72 (2018) ity-occup ation s#_short age_occup ation s. Hidalgo, C.A., Klinger, B., Barabási, A.L., Hausmann, R.: The product space condi- Colombelli, A., Krafft, J., Quatraro, F.: The emergence of new technology- tions the development of nations. Science 317(5837), 482–487 (2007) based sectors in European regions: a proximity-based analysis of International Labour Organization: international standard classification of nanotechnology. Res. Policy 43(10), 1681–1696 (2014) occupations (ISCO-08) (2012) Deming, D.J.: The growing importance of social skills in the labor market. International Labour Organization: international standard classification of Quart. J. Econ. 132(4), 1593–1640 (2017) occupations (ISCO) (2020). https ://ilost at.ilo.org/resou rces/metho ds/class Deming, D., Kahn, L.B.: Skill requirements across firms and labor markets: ifica tion-occup ation / evidence from job postings for professionals. J. Lab. Econ. 36(S1), Koenker, R., Bassett, G.: Regression quantiles. Econometrica 46(1), 33–50 (1978) S337–S369 (2018) Perez, C.: Unleashing a golden age after the financial collapse: drawing les- Edelhofer, E., Käthe, K.: Oene ff -Stellen-Erhebung 2009 bis 2012. In: Statis- sons from history. Environ. Innov. Soc. Transit. 6, 9–23 (2013). https ://doi. tische Nachrichten 11/2013, Statistik Austria (2013) org/10.1016/j.eist.2012.12.004 Fernández-Macías, E.: Job polarization in europe? changes in the employ- Salvatori, A.: The anatomy of job polarisation in the UK. J. Lab. Mark. Res. 52(1), ment structure and job quality, 1995–2007. Work Occup. 39(2), 8 (2018) 157–182 (2012) Frey, C.B., Osborne, M.A.: The future of employment: how susceptible are jobs Publisher’s Note to computerisation? Technol. Forecasting Soc. Change 114, 254–280 Springer Nature remains neutral with regard to jurisdictional claims in pub- (2017) lished maps and institutional affiliations. Goos, M., Manning, A.: Lousy and lovely jobs: the rising polarization of work in Britain. Rev. Econ. Stat. 89(1), 118–133 (2007) Goos, M., Manning, A., Salomons, A.: Job polarization in Europe. Am. Econ. Rev. 99(2), 58–63 (2009)
Journal
Journal for Labour Market Research – Springer Journals
Published: Mar 17, 2021