key: cord-0569936-ku3opj1h
authors: Stephany, Fabian
title: Does it Pay Off to Learn a New Skill? Revealing the Economic Benefits of Cross-Skilling
date: 2020-10-22
journal: nan
DOI: nan
sha: 3421e608d478161146d795752b6bcd222ed4c106
doc_id: 569936
cord_uid: ku3opj1h

This work examines the economic benefits of learning a new skill from a different domain: cross-skilling. To assess this, a network of skills from the job profiles of 4,810 online freelancers is constructed. Based on this skill network, relationships between 3,525 different skills are revealed and marginal effects of learning a new skill can be calculated via workers' wages. The results indicate that the added economic value of learning a new skill strongly depends on the already existing skill bundle but that acquiring a skill from a different domain is often beneficial. Likewise, the data illustrate how to reveal valuable skills required for new and opaque technology domains, such as Artificial Intelligence. As technological and social transformation is reshuffling jobs' task profiles at a fast pace, the findings of this study help to clarify skill sets required for mastering new technologies and designing individual training pathways. This can help to increase employability and reduce labour market shortages.

This paper asks: Does it pay-off to learn something new? It examines the economic benefits of cross-skilling; the process of learning a new skill from a different skill domain. In doing so, this work leverages data of 4,810 online freelancers and their skill portfolios to create a network in which 3,525 skills are connected if they are jointly held by the same worker. The findings of this study show that learning a new skill can add between 20 to 250 percent on the average worker wage. The results suggest that acquiring a skill from a new profession could be of even higher economic value. However, the benefits of cross-skilling largely depend on the composition of the existing skill bundle. Similarly, this work exhibits how to contextualise skill requirements for newly emerging and still opaque technology domains, such as Artificial Intelligence (AI).

The work is motivated by the rapidly changing composition of occupations due to task automation (Frey & Osborne, 2017; Acemoglu & Autor, 2011) , resulting in the paradoxical situation of simultaneous unemployment and labour shortage (Autor, 2015) . A conventional policy response has been to align national education systems with changing labour market demand. This response is increasingly ineffectual as technological and social transformation outpaces national education systems (Collins & Halverson, 2018) . Workers have to some extent begun to assume greater personal responsibility for reskilling, via skill-based online training (Allen & Seaman, 2015; Lehdonvirta, Margaryan, & Davies, 2019) . However, often the economic benefits of reskilling strategies are unclear and precise skill requirements for mastering emerging technologies, such as AI or Big Data, remain opaque (De Mauro et al., 2018) .

This work aims to overcome re-skilling limitations by assessing the economic benefit of cross-skilling strategies and sketching valuable training pathways, in reference to existing individual skill sets. Furthermore, the empirical relationship of digital skill sets will help to establish a common taxonomy to be used by policy makers, education providers, and recruiters, so that job market mismatches can be reduced. The know-how of this real-time and market data-driven evaluation can be developed into a tool for job market entrants and targeted re-education campaigns. Globally, the value of such a "cross-skilling compass", as presented with a first interactive online prototype for this project 3 , could be highest in regions where traditional education infrastructure is lagging behind.

The remainder of this study is organised as follows: In the next section, a literature review embeds the work into discussions on automation of tasks and personalisation of training. Section two illustrates the approach of the work and highlights the importance of skill diversity. Section four presents the data collected and the methods. Section five summarises the results and section six concludes with policy implications and possible extensions of the work.

The periodic warning that automation and new technologies are going to terminate large numbers of jobs is a recurring theme in economic literature (Frey & Osborne, 2017; Brynjolfsson & McAfee, 2014; Acemoglu & Autor, 2011) . A popular early historical example is the Luddite movement of the early 19th century: A group of textile artisans in England protested the automation of their industry by seeking to destroy some of the machines. In contrast to recurring fears of mass unemployment, current literature shows that the (digital) technology revolution, rather automates tasks than vanishing entire occupations (Autor, 2015) .

In this process, technological and social transformation change the skill composition of professions (Acemoglu & Autor, 2011) . The work that is thereby eliminated has different skill requirements than the newly created jobs, resulting in the paradoxical situation of simultaneous unemployment and labour shortage (Autor, 2015) . As the pace of technological and social change accelerates, the skills gap grows rapidly (Milano, 2019) . History suggests that the skills gap, even more so than the elimination of jobs per se, causes heightened economic inequality (Card & DiNardo, 2002) and retards firm growth (Krueger & Kumar, 2004) during times of technological and social transformation.

More fundamentally, the very notion of occupations is increasingly problematic in large sectors of the economy. The contemporary notion of occupations arose from the industrial revolution, as mass production required large numbers of workers with uniform bundles of skills (Featherman & Hauser, 1979) . But today's knowledge workers strive to build unique specialisms and combinations of skills that differentiate them from other workers (Hendarman & Tjakraatmadja, 2012) . Even low-end service workers can end up developing extremely heterogeneous skill sets, because they cobble together incomes from idiosyncratic combinations of gigs ranging from coffee serving to Uber driving (Fuller, Kerr, & Kreitzberg, 2019) . Tracking labour demand in terms of occupations assumed to consist of uniform bundles of skills therefore fails to produce the kind of information that individual, corporate, and national decision makers need to successfully overcome the skills gap.

A conventional policy response to closing the skill gap has been to align national education systems with changing labour market demand. This response is increasingly ineffectual as technological and social transformation outpaces national education systems (Collins & Halverson, 2018) . Large employers are likewise struggling to keep their workforces' skills up to date (Illanes et al., 2018) . Workers have to some extent begun to assume greater personal responsibility for reskilling, via online courses, distance education tools, and entrepreneurial approaches to work (Allen & Seaman, 2015) . This trend is amplified as the COVID-19 pandemic tightens economic budgets and forces workers into individual and remote reskilling.

Similar to the reshuffling of task compositions, digital technologies have enabled a process that has become a defining paradigm of the digital economy: rebundling (McManus et al., 2018) . First, in the early days of the Internet, download platforms, at times operating illegally, allowed music lovers to access songs individually without having to acquire the artist's entire album. The single item (song) was unbundled from the original bundle (album). Later, at a second stage, streaming platforms, like Spotify, reversed the trick by allowing the (re)bundling of previously unrelated items. Users could listen to songs from different artists for one single price. The mastery of this strategy has made digital entertainment companies superstars firms (Eriksson et al., 2019) .

In music (Dabager et al., 2014) , broadcasting (Hoehn & Lancefield, 2003) or gaming (McManus et al., 2018) , things that have been unbundled rarely remain that way. The economic benefit of individualised rebundling is too strong. Similarly, this paradigm has affected the way we learn new skills. At first, in the debundling phase, digital technologies allowed education providers to provide topical online courses (Wulf et al., 2014) . At a later stage, platforms like Coursera or DataCamp performed the rebundling and offered a whole set of topical courses for a single price (Bates, 2019) . The acquisition of individual skills (programming in Python) has been detached from its original domain of training (studying informatics). However, despite advances in personalised reskilling, a sizable skill gap persists on the labour market.

Current approaches to addressing the skills gap are based on predicting demand for entire occupations or at best for abstract skills such as social skills or creativity. But in many occupations, technological and social transformation is leading the concrete skills that make up the occupation to change regularly and decisively. Nursing has been transformed by successive generations of electronic health record systems, clinical decision support systems, and diagnostic technologies (Adams et al., 2000) . Web application development has rapidly rotated from perl to PHP to Ruby to Python to other development platforms (Purer, 2009 ). Firms and workers who fail to reskill while there is still demand for their skills risk dropping out of the market entirely once demand tips.

Most recent research shows that just-in-time skills development, motivated by the demands of the work at hand, or by perceived market shifts, has emerged, as formal training courses are unaffordable for workers who can't take time off paid work (Kester et al., 2006) . In addition, cultural aspects in traditional STEM education, for example, still hinders female participation, despite efforts to alter it (Kahn & Ginther, 2017) . Instead research shows that independent professionals, including women, prefer informal, digital, social learning resources like Stack Overflow and tutorial videos to develop new skills (Yin et al., 2018) .

Newest findings show that independent IT professionals today develop new skills incrementally, adding closely related skills to their existing portfolio (Lehdonvirta, Margaryan, & Davies, 2019) . Their work examines the skill development of freelancers on online labour platforms. Indeed, online freelance platforms might have become early "laboratories" for the de-and rebundling of incremental skills. It could be argued that, for work, freelance platforms, such as UpWork 4 , have become what Spotify is for music: They allow freelancers to jointly sell previously detached skill components for one hourly price. The role of the Data Scientist is a prime example of how the rebundling of skills from different domains, i.e., visualisation, programming, and statistics, is an economically profitable offer. The work by Anderson (2017) confirms that diverse rebundles of skills from different domains are profitable in general.

In this situation of rapidly changing market dynamics, systematic oversight is key. However, individuals often lack foresight into which skills are rising, which skills are most valuable and which skills their existing portfolio is complementary to. They get locked into path dependencies that may result in dead ends that prevent them from re-skilling into new areas (Escobari, Seyal, & Meaney, 2019) .

In light of the rapid reshuffling of occupational profiles and the failed attempts to develop farsighted re-skilling strategies, this work proposes an economic evaluation of cross-skilling pathways with, at least, the following four goals: 1) Develop an endogenous categorisation of skills. 2) Evaluate the economic benefit of learning a new skill.

3) Reveal the skill context of the domain of Artificial Intelligence. 4) Sketch valuable cross-skilling trajectories based on individual skill bundles.

In pursuing these goals, the study can rely on previous data-driven approaches to assess skill and human capital evaluation. Traditionally, measures of human capital rely on the count years of experience, training, or education or divide workers categories, e.g., of laborers and management (Willis, 1986) . However, a growing body of literature suggests that years of training and broad worker categories fail to address the importance of skill specialisation, diversity, and recombination in knowledge generation (Hong & Page, 2004; Lazear, 2004; Woolley et al., 2010; Ren & Argote, 2011; Aggarwal & Woolley, 2013) . In addition, the rise of the knowledge economy (Powell & Snellman, 2004) has sparked new interest in a more nuanced measure of skill composition. In this context, several papers have taken skill diversity and individual cognitive abilities into account for estimating their effect on wages (Bowles et al., 2001; Heckman et al., 2006; Borghans et al., 2008; Altonji, 2010; Autor & Handel, 2013) .

A central conclusion of past contributions on skill diversity is that the relationship between wages and skills does not only depend on a worker's individual skills but also, how they are combined. The question of skill synergies arises (Allinson & Hayes, 1996) . For some skills (e.g., programming in JavaScript and visualisation techniques) it can be argued that skill synergies emerge. The bundle of skills is more valuable than the sum of its parts. It could be argued that skill synergies are constrained to an occupational domain, e.g., programming in python and translating Russian should have little skill synergies. Certainly, the value of additional skills depends on the skill portfolio that the worker already possesses (Altonji, 2010) . However, this work precisely investigates how limited synergy effects of skill bundling are and if cross-skilling, the acquisition of a new skill outside of the existing skill portfolio, might indeed be profitable.

The data for this analysis stems from the freelancing platform UpWork 5 , which falls under the category of online labour markets (OLM). These platforms are websites that mediate between buyers and sellers of remotely deliverable cognitive work (Horton, 2010) . The sellers of work on OLMs are either people in regular employment earning additional income by "moonlighting" via the Internet as freelancers or they are self-employed independent contractors. The buyers of work range from individuals and early-stage startups to Fortune 500 companies (Corporaal & Lehdonvirta, 2017) . OLMs can be further subdivided into microtask platforms, e.g., Amazon Mechanical Turk, where payment is on a piece rate basis or freelancing platforms, such as UpWork, where payment is on an hourly or milestone basis (Lehdonvirta, 2018) .

Between 2017 and 2020, the global market for online labour has grown approximately 50% (Kässi & Lehdonvirta, 2018) . In light of the COVID-19 pandemic and it's significant economic repercussions across industries (Stephany et al., 2020a) , OLMs continue to increase in popularity due to a general trend of work at distance (Stephany et al.. 2020b) . UpWork is usually perceived as the globally most popular freelance platform (Kässi & Lehdonvirta, 2018) . This study utilises OLM data, as platforms like UpWork have become early "laboratories" of the rebundling of skill sets. Their data allow us to monitor skill rebundling in a global workforce by near real-time reporting location, asking wages, previous income, gender attributes (forenames), and up-to-date skill bundles on a granular level.

For the methodological approach of this paper, the work by Anderson (2017) is referential. Anderson constructs a human capital network of skills from online freelancers and shows that workers with diverse skills earn higher wages. The limitation of Anderson's work is that a skill specific evaluation in the context of cross-skilling is not addressed. This work aims at adding this cross-skilling perspective and exemplary sketches economically valuable cross-skilling pathways in times of shifting occupational profiles.

Similar to Anderson (2017) , this work uses the rich toolbox of network analysis for the characterisation of skill relationships. Given a sample of 4,810 freelancers with multidimensional skill portfolios, a network is constructed in which 3.525 skills are nodes and two skills are connected by a link if a worker has both. Links are weighted according to how often the two skills co-occur. First, this skill network provides us with an endogenous categorisation of skills based on their relationship in application and the context dependency of human capital. In a second step, the wage proposals 6 of workers allow a statistical assessment of skills. Via calculating regression coefficients, the economic value of the 30 most popular individual skills 7 can be derived:

age β ountry β og(earned) β kill w i = 0 + β 1 * c i + 2 * l + 3 * s i,j + e i (1) ε 1, .., and j ε 1, .., 1 i . n . 2

The linear regression model (1) uses all workers ( ) as ε 1, .., i

. n observations and considers their country of origin and amount of money earned as characteristics when estimating the worker's asking wage. In addition, each of the 30 most popular skills are considered as an explanatory feature in the linear regression. Lastly, the method addresses the issue of cross-skilling. Based on the existing skill portfolio of a worker, skill coefficients are again calculated. This time only a subset of workers is considered:

(2) age β ountry β og(earned) β kill , w k = 0 + β 1 * c k + 2 * l k + 3 * s k,j + e i ε a, .., and j ε 1, .., 1 k . m < n . 2

Here the accounts between a and m fall into a specific occupational domain, e.g., translation and writing, as the majority of their skills are located in this skill cluster. Hence the coefficient of the skill characteristic ( ) only refers β 3 to the additional wage this skill contributes within a smaller subset of workers with the skill bundle . This cross-referencing allows ε a, .., k

. m < n us to indicate the potential additional marginal value of acquiring a new skill if added to a specific skill portfolio.

As the first part of the analysis a skill network is constructed, shown in Figure 1 . The network uses the information of 4,810 workers and 3,525 unique skills. In this network, unique skills are represented as nodes. They are connected if simultaneously advertised by the same worker. The edges between two nodes grow in strength the more workers combine a pair of skills. The (unweighted) degree centrality of each node is represented by its size. Based on the relationship between skills via workers, a Louvain clustering method is applied 8 that minimises the number of edges crossing each other. Seven distinct clusters emerge, as highlighted in different colours. By highlighting the ten most prominent skills -in terms of degree centrality -of each cluster, we can see conceptual consistency within 7 Once a larger set of skills is considered at the same time, the model validity is scrutinised by potential multicollinearity. 8 The skill clusters differ in size (of skills and workers employing them) but their conceptual consistency underlines the effectiveness of this endogenous clustering approach. In comparison to human classification of skills in the context of online freelance markets, similarities and differences occur. Kässi and Lehdonvirta (2018) , for example, classify skills in six different domains that do not include 3D, Graphic, and Audio Design individually. The endogenous clustering approach, however, clearly indicates that these skill groups form individual clusters of their own but they are similar to each other and though group together at the left hand side of the skill network.

Skill clusters also differ significantly with regard to the asking wages of workers, as shown in Figure 2 , ranging from a median of 20 USD per hour, asked by workers in Admin and Support to 30 USD/hour in Software and Tech. to 100 USD/hour in Legal. However, the wage spread within and across skill domains is sizable. In Software and Technology, asking wages range from 3.75 to 495 USD per hour. 

The rich variety of skill combinations allows us to assess the value of adding a new skill to a worker's skill portfolio. Via linear regression models, we can calculate beta coefficients for the 30 most popular skills for six domains 9 , as shown in Figure 3 .

The added value of learning one of the most popular 30 skills varies significantly. On average, for ten of the 30 skills the beta coefficient is positive, for four skills (Voice Talent, Python, Audio Editing, and Copywriting), results are statistically significant (p>0.05). But even within the group of significantly profitable skills, the spread is large. Being knowledgeable in Voice Talent adds 240% to the average worker's asking wage, while Copywriting skills contribute 66% 10 . Negative coefficients indicate that workers with these skills ask for significantly lower wages than the average online freelancer.

With the endogenous classification of skill groups at hand, we can perform an evaluation of learning a new skill based on the already existing skill bundle of a worker. For this purpose, two analyses are performed. First, the value of skill diversity is assessed. Skill domains are attributed to each worker based on her skill bundle. While each worker has a dominant skill category (the relative majority of all of her skills are in this domain), some workers also add skills from other domains to their portfolio. In Figure 4 , wage distributions of different skill diversities are shown across the major six skill groups. In general, it can be noticed that across skill domains, workers with more diverse skill bundles have higher wages. In particular, for workers with very diverse bundles, i.e., adding skills from three domains other than the major skill category, wages on the 4th quartile (lines) of the distribution are shifted upwards. Figure 4 : Across skill domains, asking wages increase when skills from other domains are added to the workers' portfolio. Workers in the top earning quartiles (lines) ask for significantly higher wages when demanding skills from three domains other than their defining skill bundle (Observation sizes in Legal are too small).

In a second step, the linear regression explaining workers' asking wages is performed for the complete set of workers and the six major skill domain subsets. Figure 3 summarises the coefficients of the 21 most popular skills in the eight scenarios. Figure 5 : Learning a new skill, like Copywriting or Python, pays off in general, but even more when added to skill bundles like translation and writing or admin and support.

In contrast to the added economic value for the complete set of workers, we see that some skills, like programming in Python, increase the worker asking wage over proportionally in the skill context of Admin & Support. Similarly, knowing how to do Copywriting contributes significantly more when added to Graphic Design. Other skills, like Audio Editing, however, do contribute to an average worker's wage, but fail to make a difference conditional to special skill bundles. Figure 5 illustrates three of these cross-skilling trajectories. Figure 6 : The cross-skilling trajectories for skills like Python, JavaScript, or Copywriting, can be examined individually and compared with each other.

As an example, it is of little surprise that, on average, knowing how to program with Python, allegedly THE data science super skill (Grus 2019) , adds more to a worker wage than knowing how to work with JavaScript. However, once we add these skills to a bundle in the domain of 3D Design, the picture turns upside down. Designers that know Python can't add wage but workers that are skilled in JavaScript can increase their wage by more than 200% on average. Similarly, as shown in the lower panel of Figure 5 , the skill of Copywriting adds about 60% to the average worker wage. In the domain of Graphic Design, on the other hand, knowing how to do Copywriting has an impressive additional value of more than 600% of the average wage in this domain.

These skill trajectories are an illustration of what online labour market data allow us to say about cross-skilling trajectories in general. The data enable us to evaluate the economic benefit of individual skills based on the existing skill bundle of a person and to sketch sustainable cross-skilling pathways.

A further obstacle in developing effective and timely reskilling pathways is the lack of skill contextualisation of newly emerging technologies. The rapid expansion of such emerging digital technologies, like AI, is creating a huge demand for labour skilled in the development and application of these domains. However, the fast change of skill profiles in new technology environments makes it difficult for companies to find adequately trained experts. At the same time, it is unclear which types of skills constitute newly emerging areas of digital technologies (De Mauro et al. 2018) .

Companies are not able to satisfy their rapidly growing demand for talents in information and communication technologies (ICT). Projections show that positions for digital technology talents are the fastest growing job segment in the United States with estimated 2,720,000 openings by the end of 2020 (Miller and Hughes 2017) , leading to a significant excess demand for ICT professionals. The results of qualification mismatches are lower labour and economic productivity (McGowan & Andrews 2015) . This can be particularly harmful for economies in regions with low levels of growth, where expert labour in ICT is already scarce.

Often, the description of skills and responsibilities of experts in domains such as Big Data or AI is fuzzy and firms tend to apply subjective interpretations. As De Mauro et al. (2018) illustrate in detail, the emergence of the expert role of the Data Scientist is a typical example for a simplistic job description that downsizes the complexity and variety of skills required to retrieve information and transform it into economically valuable insights. Research indicates that there is a clear gap regarding the formal taxonomy of skills and educational needs in new technology domains like AI (Miller & Hughes 2017; Song & Zhu 2016) .

The presented approach of creating skill networks can help to reveal precise skill sets that are related to AI. Figure 6 shows a subset of nodes from the skill network in Figure 1 . Here, only skills from profiles are considered that have appeared under the search term "Artificial Intelligence". Similar to Figure 1 , skills are connected if jointly advertised by the same worker. Skills are strongly clustered around the domains of software and technology and admin and support. In absolute terms, AI appears most frequently in the skill context of software and technology; 28 skills related to AI belong to this domain. However, taken the size of the different skill clusters into account, legal is the skill domain most strongly populated by AI skills; seven out of 46 legal skills (15.2%) are connected to AI.

Similarly to the analysis in the previous subsection, for all of the 80 skills that are related to the search term "Artificial Intelligence" wage coefficients are calculated. Figure 7 shows all positive (both significant and insignificant) coefficients.

In the domain context of AI, skills in the field sales, web and software services, as well as, learning Python add to the average worker's asking wage.

The most profitable skills to learn in the context of AI are situated in the field of ales (Sales Letter and Sales Writing). Workers with one of these skills earned on average up to eleven times more than the average worker. Likewise, skills in Software as a Service -Saas (six times) or Python (four times) are profitable skills in the AI domain.

As technological change accelerates, task automation shifts occupational skills requirements, challenging the global workforce to constantly reskill. To avoid skill gaps and systematic labour market mismatches, approaches to reskilling need to step-up, as traditional education policies are too slow for the fast-changing pace of technological and social change. In addition, situations like the COVID-19 lockdown further accelerate digitalisation trends while limiting economic resources of companies to up-skill their employees and constraining workers to learn remotely from home.

In light of this grand challenge, this work explores the foundations of new modes of re-skilling via sketching cross-skilling pathways based on online labour market data. Online labour markets have become early laboratories for the de-and rebundling of skills from previously unrelated domains. The statistical analysis of diverse skill portfolios and wages of online workers allows an evaluation of the economic benefit of learning a new skill. Furthermore, the endogenous categorisation of skills via skill networks gives us insights into the value of learning a new skill depending on the already existing skill portfolio of each worker. We see that some skills are, in terms of additional wage, more valuable than others. On average, performing Voice Overs tribles worker's wages, while knowing how to program in Python enhances wages by 120%. These figures are independent of a worker's previous earnings and location, which likewise influence asking wages.

In addition to the economic evaluation of individual skills, this work assesses the added economic value of learning a new skill in addition to a defined skill portfolio, i.e., cross-skilling. The conditioning on skill domains is a relevant perspective, as individual examples show. Compared to the average worker, Copywriting skills are, in terms of added wage, about ten times as valuable when added to a skill bundle in 3D Design. Sketching personalised cross-skilling trajectories is mandatory for future educational formats, as skill portfolios become more fragmented and re-skilling opportunities more granular.

This work is a first exploration of a quantitative and market data based assessment of cross-skilling. It comes with limitations and opens space for future investigations. The strong point of this work is the potential for individual recommendation based on existing skill bundles. Ideally, researchers and policy makers could use this blueprint to develop new tools for a granular and near real-time assessment of individual development potentials, recommending which skill to learn next 11 . As an improvement of this method, the presented findings should be enriched with on-site labour market wages from a broader set of occupations. Likewise, a monitoring of cross-skilling trajectories over time could enable more nuanced and far-sighted statements about the emergence of new skills and the future of learning something new. 

Skills, tasks and technologies: Implications for employment and earnings

Skill-mix changes and work intensification in nursing

Do you see what I see? The effect of members' cognitive styles on team processes and errors in task execution

Grade Level: Tracking Online Education in the United States

The cognitive style index: A measure of intuition-analysis for organizational research

Multiple skills, multiple types of education, and the labor market: A research agenda

Skill networks and measures of complex human capital

Why are there still so many jobs? The history and future of workplace automation

Putting tasks to the test: Human capital, job tasks, and wages

What's right and what's wrong about Coursera-style MOOCs

The economics and psychology of personality traits

The determinants of earnings: A behavioral approach

The second machine age: Work, progress, and prosperity in a time of brilliant technologies

Skill-biased technological change and rising wage inequality: Some problems and puzzles

Rethinking education in the age of technology: The digital revolution and schooling in America

Platform Sourcing: How Fortune 500 Firms are Adopting Online Freelancing Platforms

An empirical analysis of digital music bundling strategies

Human resources for Big Data professions: A systematic classification of job roles and required skill sets

Spotify teardown: Inside the black box of streaming music

Realism about Reskilling: Upgrading the Career Prospects of America's Low-Wage Workers. Workforce of the Future Initiative

On the measurement of occupation in social surveys

The future of employment: How susceptible are jobs to computerisation?

The Gig Economy: Leasing Skills to Pay the Bills

Data science from scratch: first principles with python

The effects of cognitive and noncognitive abilities on labor market outcomes and social behavior

Relationship among soft skills, hard skills, and innovativeness of knowledge workers in the knowledge economy era

Broadcasting and sport

Groups of diverse problem solvers can outperform groups of high-ability problem solvers

Online Labor Markets

Retraining and reskilling workers in the age of automation

ForceAtlas2, a continuous graph layout algorithm for handy network visualization designed for the Gephi software

Women and STEM (No. w23525)

Online labour index: Measuring the online gig economy for policy and research

Just-in-time, schematic supportive information presentation during cognitive skill acquisition

Skill-specific rather than general education: A reason for US-Europe growth differences?

Balanced skills and entrepreneurship

Skills formation and skills matching in online platform work: policies and practices for promoting crowdworkers' continuous learning (CrowdLearn)

Flexibility in the gig economy: Managing time on three online piecework platforms

Labour market mismatch and labour productivity

Steering incentives and bundling practices in the telecommunications industry

The digital skills gap is widening fast. Here's how to bridge it

The quant crunch: How the demand for data science skills is disrupting the job market

Digital Innovation Management: Reinventing innovation management research in a digital world

The knowledge economy

PHP vs. Python vs. Ruby-The web scripting language shootout

Transactive memory systems 1985-2010: An integrative framework of key dimensions, antecedents, and consequences

Big data and data science: what should we teach

Distancing bonus or downscaling loss? The changing livelihood of US online workers in times of COVID-19

The CoRisk-Index: A data-mining approach to identify industry-specific risk assessments related to COVID-19 in real-time

Back to the Future-Changing Job Profiles in the Digital Age

Everything You Always Wanted to Know About AI -Nowcasting Digital Skills With Wikipedia

Wage determinants: A survey and reinterpretation of human capital earnings functions. Handbook of labor economics

Evidence for a collective intelligence factor in the performance of human groups

Massive open online courses

Learning to mine aligned code and natural language pairs from stack overflow

Research commentary-the new organizing logic of digital innovation: an agenda for information systems research

 Table 1 : Network metrics of the 35 skills with the highest degree centrality in each of the seven skill cluster