OECD Employment Outlook 2026

Geographic Disparities in Jobs and Incomes

Report

OECD Employment Outlook

7 July 2026

Available in:

English
français

Download PDF

4. Skills at work: New evidence on returns to skills and training

Copy link to 4. Skills at work: New evidence on returns to skills and training

Julie Lassébie

Patricia Navarro-Palau

Roland Tusz

Abstract

Labour market megatrends are reshaping labour demand and altering the value of different skills, contributing to a decline in returns to formal education and training. At the same time, participation in adult training is falling in many countries, and when undertaken, training is often very short in duration. These developments raise important questions about which types of skills and learning activities are most valued in today’s labour market. Using data from the 2022-2023 wave of the Survey of Adult Skills (PIAAC), this chapter analyses the links between wages and information processing skills, educational attainment and training participation. It explores how the links between wages and information processing skills and formal education interact with non‑cognitive skills and workplace practices, and how different training modalities and topics are rewarded in the labour market. The chapter concludes with policy implications to strengthen the reward of skills and enhance the labour‑market value of adult learning.

In Brief

Copy link to In Brief

Key findings

Returns to education and training are declining. Rapid technological change is reshaping labour demand and altering the value of different skills. Tasks and roles once performed mainly by higher education graduates are increasingly carried out by workers without higher education. At the same time, the growing emphasis on skills‑first approaches – hiring and workforce development practices that prioritise demonstrated competencies over formal credentials – suggests a shift in the relative importance of skills compared with formal education.

In parallel, recent evidence from the Survey of Adult Skills (PIAAC) indicates that participation in training is falling in many countries, and when training does occur, it is often very short, typically lasting less than one week or even less than a day. These changes raise important questions about which types of training – in terms of content and design – are most valued in today’s labour market.

Drawing on the most recent Survey of Adult Skills data, collected in 2022-2023, this chapter analyses how wage returns to skills and education have evolved over the past decade, how they vary across occupations, and how skills interact with different workplace practices. It also examines how returns to training differ across training types, topics, and characteristics, shedding light on the kinds of learning that generate the greatest labour market benefits.

The key takeaways from the chapter are as follows:

Information processing skills (proxied by numeracy or literacy) and formal education remain important predictors of employment and wages, but their influence has declined in the last decade. On average across OECD participating countries, a one‑standard‑deviation increase in years of education (around 3.1 years) is associated with a 1.2‑percentage‑point increase in the probability of being employed and a 15% increase in wages. A one‑standard‑deviation increase in numeracy (58 points in the Survey of Adult Skills) corresponds to a 0.65‑percentage‑point rise in employment probability and an 11% wage increase. However, associations of years of education and numeracy with the probability of being employed have weakened by about 1 percentage point (p.p.) relative to the first cycle of the Survey of Adult Skills (2011‑2018), and the association of wages with years of education by 2.2 p.p.
Occupational sorting is a major driver of wage differences: education and cognitive skills raise wages largely because they improve access to higher‑paying occupations and firms. Once occupation is taken into account, the estimated associations between years of education and wages, and between numeracy and wages, fall by around 7 and 4 p.p., respectively, to 8% and 7%. The links between wages and education and skills also vary substantially across occupations, with the strongest links observed in high‑skilled roles and declining steadily across the occupational spectrum.
Non‑cognitive skills and workplace practices also play an important role in wage outcomes. Higher levels of task discretion and greater co‑operation at work are associated with higher wages – by 1% and 6% on average across participating OECD countries – reflecting the growing importance of human‑centred work practices. Cognitive (information processing) and non‑cognitive skills appear strongly complementary: the wage association with numeracy is significantly larger for workers who enjoy greater autonomy and operate in co‑operative work environments.
The evidence presented above underscores the importance of skills and the potential relevance of skills‑first approaches, which prioritise demonstrated capabilities over formal credentials. Governments have a central role to play in supporting talent recognition by embracing policies that make skills visible such as Recognition of Prior Learning (RPL) or skills signalling, using, for example, instruments like skills passports.
Some forms of training are unambiguously associated with higher wages. Non‑formal training – planned, short, institution‑based learning that is not formally accredited – shows a clear positive association with wages. Adults who participated in non‑formal learning in the 12 months prior to the survey earn, on average across participating OECD countries, 6.2% higher wages than those who did not. Evidence also suggests that the association between wages and training has declined over time. The data do not allow to provide conclusive evidence on other types of learning.
The relationship between non‑formal training and wages appears to be partly driven by occupational selection. Once occupation controls are included, the link between participation in non-formal training and wages declines by around 2 p.p. This indicates that individuals in higher‑paying occupations are also more likely to participate in non‑formal training.
Training topics matter. Courses focussed on teamwork, leadership, and project management are associated with largest wages, with the training wage premium exceeding 8.5%. These large associations are consistent with the finding that managers have larger wage associations with non‑formal training participation, as these subjects are among the most common areas of training for this occupational group, and could be explained by selection into these training areas. However, overall, training in health and safety, which has a weak association with wages, is the most common.
The evidence suggests that flexibility in training delivery mode and timing does not diminish wage returns. Online and hybrid learning formats are associated with wage benefits at least as large as traditional in‑person training while offering greater flexibility and access. Similarly, the links between wages and non‑formal training do not differ significantly between very short courses (completed within a day) and longer courses (lasting more than a week), nor between training delivered in a single block and training spread over several days or months. Flexible learning opportunities – which may help in addressing barriers to participation in training – can be supported and promoted by policy tools such as micro‑credentials, RPL, and strengthened National Qualification Frameworks (NQFs).
Employer‑financed training is associated with high wage returns. This pattern suggests that employers may be better positioned to identify training that is closely aligned with job needs and likely to generate productivity gains – and thus higher wages – at least in the short run. It could also indicate that employers select their most promising employees to obtain higher returns from their training. Governments can encourage employer‑provided training – particularly among small and medium enterprises (SMEs) – through financial support and measures that strengthen firm capacity, including skills assessment and anticipation services, shared learning and training networks, and managerial training.

Introduction

Copy link to Introduction

Returns to education and training are declining. Evidence suggests that, after several decades of growth, the “university wage premium” – the additional earnings university graduates receive relative to non‑graduates – has recently stagnated and, in some economies such as the United States and the United Kingdom, has even begun to fall (Bengali, Valletta and Zhao, 2025[1]; Boero et al., 2024[2]). Evidence presented in Chapter 1 also suggests that the employment advantage of young college graduates has been on a shrinking trend for many years in several OECD countries.

Historically, skill‑biased technological change resulted in significant wage premia for university graduates. In particular, advances in information and communication technologies increased the productivity of highly educated workers with specialised skills, thereby boosting their wages relative to others (Atkinson, 2007[3]; Autor, Katz and Krueger, 1998[4]; Autor, Katz and Kearney, 2008[5]; Katz and Murphy, 1992[6]).

Today’s technological change, especially the rapid spread of artificial intelligence, is reshaping labour demand and the returns to different skills in new ways. Tasks and roles traditionally associated with university graduates are increasingly performed by workers without higher education. For example, computer use has become widespread across the workforce. Consistent with this shift, Bengali, Valletta and Zhao (2025[1]) provide evidence that the stagnation of the university wage premium in the United States is driven by labour‑demand factors, including rising substitutability between university and high‑school graduates.

Additionally, or alternatively, employers’ demand for skills that are either not developed, or may not be developed effectively in university programmes could have increased. Recent studies indicate that AI adoption amplifies the importance of non‑cognitive skills (Lane, Williams and Broecke, 2023[7]; Milanez, 2023[8]). However, there exists limited evidence on how these skills are best developed within education and training systems.

The increased popularity of skills-first approaches – hiring and workforce development practices that prioritise individuals’ demonstrated skills and competencies over traditional credentials such as degrees or job titles – may also signal a shift in the relative importance of skills compared with formal education (Bone, González Ehlinger and Stephany, 2025[9]; OECD, 2024[10]; OECD, 2025[11]). This trend, particularly pronounced in certain sectors, suggests that employers may be placing increasing value on workers’ skills rather than on the qualifications they hold, particularly for skills in emerging areas for which formal programmes are slow to adjust.

In parallel, recent evidence from the Survey of Adult Skills (PIAAC) shows that training participation is declining in many countries (OECD, 2025[12]). Participation in job‑related formal education, for instance, has fallen by more than 2 p.p. between survey cycles, with only 8% of adults enrolled in formal learning programmes in Cycle 2, collected in 2022-2023. Non‑formal job‑related learning has also declined, by 3 p.p. on average, with 37% of adults participating in Cycle 2. Furthermore, short or very short training activities – i.e. lasting less than one week or less than one day, respectively – are the norm. Short training formats can lower barriers to participation and support meaningful skills development over time. However, if they are not well designed, they may offer limited opportunities for deeper or more transformative reskilling. Given that lack of time is the most commonly reported barrier to training participation (OECD, 2025[12]), it is essential to assess whether short and flexible training formats can generate returns comparable to those of more traditional forms of training.

To shed light on these labour market developments, the chapter draws on the most recent data from the Survey of Adult Skills. It examines how returns to skills and education have evolved over the past decade and how they differ by occupation. It also explores how individuals’ education and skills interact with a range of workplace practices. Finally, the chapter analyses how returns to training vary across different types of learning – including formal, non‑formal, and informal learning – as well as across other dimensions such as training topics, which reflect the skills being developed, and training characteristics, such as the degree of flexibility and training duration.

Data from the Survey of Adult Skills is particularly well suited for this type of analysis, as it provides direct assessments of foundational skills together with extensive socio-demographic information, labour market outcomes, and detailed job characteristics. Importantly, the survey also contains rich information on participation in adult training – including the content of training activities, how they are implemented, and any barriers to participation. This breadth of information enables accurate identification of associations between education, skills, training, and wages, helps mitigate certain endogeneity concerns, and supports an exploration of heterogeneity across both job and training characteristics.

Furthermore, 31 countries and economies participated in Cycle 2 of the Survey of Adult Skills (2022-2023), with 27 of them also having taken part in Cycle 1 (2011-2018). This cross-country and over-time coverage makes it possible to examine how wage associations vary internationally and how the relationships between education, skills, and wages have changed in the past decade.

To better assess whether skills are becoming relatively more important in the labour market compared to education and to understand current returns to training, this chapter examines the association between education, skills, and training with wages. Generally, wage returns can be considered a lower‑bound estimate of the productivity gains associated with higher skill levels or participation in training and therefore provide useful insights into the economic value of skills and training. They are also an important incentive for workers to invest in training, although training may have benefits beyond wage increases. For this reason, wage returns should not be interpreted as returns on investment as they capture neither the costs of acquiring skills or participating in training nor the broader benefits that may accrue to adults, firms or society as a whole.

Wage returns are estimated by regressing the logarithm of hourly wages on training participation (for returns to training), numeracy proficiency (for returns to skills), years of education, work experience, and a rich set of worker, firm and job characteristics, along with country fixed effects. The chapter also investigates the relationship between the probability of being employed (versus unemployed) and education level and skills. It is important to note that the term “returns” is adopted here for brevity. The analysis relies on cross-sectional data from Cycles 1, gathered between 2011 and 2018, and 2 of the Survey of Adult Skills, meaning individuals cannot be followed over time. While the richness of these data allows addressing several endogeneity issues highlighted in the literature (potential endogeneity concerns and the approaches used to address them are discussed in Annex 4.A), the results presented in this chapter should be interpreted as associations rather than causal effects, since changes in individuals’ labour market outcomes after participating in training cannot be observed.

Differences in the macroeconomic context prevailing during data collection for the two cycles must be considered when interpreting the results. Most countries participating in Cycle 1 of the Survey of Adult Skills collected data in 2011‑2012, when economies were still recovering from the global financial crisis. Unemployment rates were high and wage growth was weak, with real wages stagnating or declining in many countries. By contrast, labour markets in 2022‑2023 were historically tight, characterised by high employment rates and widespread labour shortages, which may have improved employment prospects for less‑qualified or lower‑skilled adults.

The remainder of the chapter is organised as follows. The first section examines the associations between skills, education and wages and assesses how they have evolved since Cycle 1 of the Survey of Adult Skills, conducted a decade earlier. It also explores how these associations vary across occupations and countries, and whether they are amplified by the frequent use of certain non‑cognitive skills. The second section investigates associations between wages and job-related training, distinguishing between formal, non‑formal and informal learning, and documents how these associations differ by occupation, country and training topic. It further considers whether other training characteristics influence the magnitude of training returns. The third and final section discusses the main policy implications resulting from the chapter’s findings.

4.1. Skills play a key role in shaping labour market outcomes beyond formal education

Copy link to 4.1. Skills play a key role in shaping labour market outcomes beyond formal education

4.1.1. Cognitive skills continue to be an important determinant of labour market outcomes

The question of how skills are rewarded in the labour market is a long-standing one in the economic literature. Because skills are difficult to observe directly, seminal studies have traditionally relied on estimating the relationship between education and earnings (Card, 1999[13]; Griliches, 1977[14]; Juhn, Murphy and Pierce, 1993[15]). However, school attainment is an imperfect measure of an individual’s actual human capital. More recent work extends traditional human-capital models by incorporating measures of occupation-specific tasks or skills proficiency. Ingram and Neumann (2006[16]) were among the first to propose an alternative measure of skills, based on the tasks that individuals perform in their occupations. Their findings show that between 1971 and 1997, returns to cognitive skills (mathematical and verbal ability) increased substantially, while rewards for routine and manual skills (including clerical skills, co‑ordination, fine motor abilities and strength) declined or remained modest. They also found that returns to years of education remained constant over time once cognitive skills are accounted for.

Cycle 1 of the Survey of Adult Skills, conducted in three rounds between 2011 and 2018 in 39 countries, enabled a direct and precise measurement of individuals’ skills and several studies have used these data to obtain reliable estimates of labour-market returns to skills. Regarding employment status, going up one (out of five) PIAAC proficiency levels in numeracy is associated with an average increase in the likelihood of being employed of about 8 p.p. on average across participating OECD countries (Hampf, Wiederhold and Woessmann, 2017[17]). Studies focussing on earnings (Hampf, Wiederhold and Woessmann, 2017[17]; Hanushek et al., 2015[18]) found statistically and economically significant wage returns to numeracy skills, averaging 18%. These returns vary considerably across countries, ranging from 12% in Finland to 28% in the United States.

A substantial body of literature has examined the relationship between returns to skills and inequality. In general, countries where skills are highly rewarded – such as the United States – tend to be characterised by higher wage inequality (Broecke, Quintini and Vandeweyer, 2018[19]; OECD, 2015[20]). At the individual level, socio‑economic disparities in employment prospects and wages linked to parental background, notably parental education and occupational status, are largely driven by differences in educational attainment and skills across groups. Differences in educational attainment and skills also partly explain disparities in employment prospects and wages between individuals who grew up in urban areas and those raised in rural areas (OECD, 2025[21]).

Cycle 2 of the Survey of Adult Skills, conducted in 2022-2023, provides an opportunity to revisit the question of how skills are rewarded in the labour market. It offers new estimates of adults’ proficiency in literacy, numeracy and adaptive problem solving – key information-processing skills that both individuals and societies need to thrive – as well as rich socio-demographic background, labour market outcomes, and job characteristics. Thirty-one countries and economies (mostly OECD Member countries) participated in the 2023 Survey of Adult Skills. As 27 of them also took part in previous rounds, the data additionally allow for examining how associations between education, skills and wages have evolved over the past decade.

Across participating countries, employment status (more specifically the probability of being employed rather than unemployed) is positively associated with both educational attainment and, independently of education, numeracy skills (Figure 4.1) – see Box 4.1 for the methodology. On average, for OECD participating countries, after controlling for individual characteristics and country-specific factors, a one‑standard-deviation increase in years of education (approximately 3.1 years) is associated with a 1.18 p.p. increase in the likelihood of being employed, while a one‑standard-deviation rise in average numeracy scores (58 points) corresponds to a 0.65 p.p. increase in employment probability.1 The positive and significant association with employment and wages also holds when literacy is used instead of numeracy.

Box 4.1. Empirical strategy to estimate associations of years of education and numeracy skill level with labour market outcomes

Copy link to Box 4.1. Empirical strategy to estimate associations of years of education and numeracy skill level with labour market outcomes

Association between years of education and numeracy skills with the probability of being employed

To examine the relationship between individuals’ years of education and their numeracy skill levels with the probability of being employed, a linear probability model is used, as follows:

$y_{i c} = α + β S k i l l s_{i} + δ {E d u c a t i o n}_{i} + γ x_{i} + ρ_{c} + ε_{i c}$

where the dependent variable $y_{i c}$ indicates the employment status of individual i in country c, being equal to 1 if individual i is employed and equal to 0 if individual i is unemployed. Inactive individuals are excluded from the sample. ${S k i l l s}_{i}$ is the average of all ten plausible values for numeracy (or literacy) for individual i normalised to have an average value of 0 and a standard deviation of 1.1 ${E d u c a t i o n}_{i}$ refers to the individual’s completed years of education normalised to have mean 0 and standard deviation equal to 1, $x_{i}$ is a set of individual control variables including age, sex, immigrant background, parental education, whether one lives with a partner or has children, work experience and work experience squared, and $ρ_{c}$ are country fixed effects. The coefficients of interest are $β$ and $δ$ , which capture the change in the probability of being employed associated with a one‑standard-deviation increase in average numeracy scores (58 points) and in years of education (approximately 3.1 years), respectively.

Association between years of education and numeracy skills with wages

To analyse the relationship between individuals’ years of education and their numeracy (or literacy) skill levels with wages, the following ordinary least squares specification is used:

$l n (w_{i j f o s c}) = α + β S k i l l s_{i} + δ {E d u c a t i o n}_{i} + γ x_{i} + θ u_{j} + δ z_{f} + μ_{o} + σ_{s} + ρ_{c} + ε_{i j f o s c}$

where $w_{i j f o s c}$ is the gross hourly wage of individual i employed in job j, firm f, occupation o, sector s and country c. ${S k i l l s}_{_i}$ is the average of all ten plausible values for numeracy (or literacy) for individual i normalised to have an average value of 0 and a standard deviation of 1. ${E d u c a t i o n}_{i}$ refers to the individual’s completed years of education normalised to have mean 0 and standard deviation equal to 1; $x_{i}$ is a set of individual control variables including age, sex, immigrant background, parental education, whether one lives with a partner or has children, work experience and work experience squared; $u_{j}$ is a set of job characteristics including whether the individual works part-time or with a temporary contract, such as a fixed term contract, seasonal contracts, a temporary employment agency contract, a zero-hour contract, a freelance, contractor and/or consultant contract, working as an apprenticeship or other training scheme or without a contract; $z_{f}$ is a set of firm characteristics including whether the individual works in a small and medium enterprise (SME), in a growing firm (i.e. the number of employees increased over the past 12 months), and in a private sector firm; $μ_{o}$ are occupation (ISCO 2 digits) fixed-effects; $σ_{s}$ are sector (ISIC 1 digit) fixed-effects, and $ρ_{c}$ are country fixed-effects. The coefficients of interest are $β$ and $δ$ , which reflect the percentage change in wages associated with a one‑standard-deviation increase in average numeracy scores (58 points) and in years of education (approximately 3.1 years), respectively.

1. See OECD (2025[22]) for more technical details on plausible values for individuals’ proficiency.

The relationship between years of education and employment status is positive and significant in half of the countries analysed. In the remaining countries, it is positive but not significant, except in Switzerland and Israel, where the coefficient is negative but close to 0 and not statistically significant. In the case of Switzerland, a strong Vocational Education and Training system, which leads to similar levels of employment than tertiary education, could explain the lack of association between years of education and employment status. Numeracy skills are also positively associated with employment status in most countries, above and beyond years of education, however, this relationship is statistically significant in only a quarter of the countries analysed. Interestingly, in six countries, numeracy shows a larger association with the probability of being employed than years of education.

Figure 4.1. Education and skills are positively associated with employment status
Copy link to Figure 4.1. Education and skills are positively associated with employment status

Percentage point change in probability of being employed (vs. unemployed) for a one‑standard-deviation increase in numeracy proficiency or years of education, by country

In addition to their link with employment status, both educational attainment and numeracy skills are also strongly associated with wages. While educational attainment has generally larger associations with wages, in several countries the association is stronger for numeracy skills (Figure 4.2). On average across participating OECD countries, after controlling for individual characteristics and country-specific factors, an additional 3.1 years of schooling is linked to a 15% increase in wages. There is substantial variation across countries participating in the Survey, with coefficients ranging from 7% in France to 32% in Singapore. The association between numeracy and wages is also strong and statistically significant: a one‑standard-deviation (58‑point) increase in average numeracy scores is associated with a 11% rise in wages on average for participating OECD countries. This association is stronger for more educated workers. As in the case of associations with years of education there is considerable cross-country variation. For example, the rise in wages is equal to 5% in Korea and the Slovak Republic and reaches 23% in Estonia. In six countries, the association between numeracy and wages is stronger than between years of education and wages. In Estonia, numeracy show stronger associations with both labour market outcomes – employment status and wages – than years of education.

Cross-country variation in the strength of the associations between years of education or numeracy and wages appears to be correlated with country inequality: generally, countries with higher levels of wage inequality2 tend to show larger associations between years of education or numeracy and wages.3 Greater wage‑setting flexibility at the firm level may allow employers to differentiate pay more sharply based on workers’ skills or qualifications.

Figure 4.2. Education and numeracy are positively, and independently, associated with higher wages across virtually all participating countries
Copy link to Figure 4.2. Education and numeracy are positively, and independently, associated with higher wages across virtually all participating countries

Change in hourly wages associated with a one‑standard-deviation increase in numeracy or years of education, percentage, by country

4.1.2. Education and numeracy skills are associated with higher wages primarily by enabling access to occupations with larger wages

A substantial share of the association between education or cognitive ability and wages operates through sorting: more capable adults tend to select into jobs and firms that best reward their skills. Using data from the Cycle 1 of the Survey of Adult Skills, Heisig and Solga (2017[23]) found that there is a strong association between skills and occupational status. Furthermore, since occupations differ in their skills and education requirements and in the extent to which they reward these requirements, skills tend to have stronger associations with earnings in occupations that rely on high-skill, non-routine and analytical tasks (Hanushek et al., 2015[18]; Hampf, Wiederhold and Woessmann, 2017[17]).

Using data from Cycle 2 of the Survey of Adult Skills, when controlling for occupational status, job and firm characteristics, the association between years of education or numeracy skills and wages weakens significantly in all countries analysed. Figure 4.3 illustrates this point by comparing the relationship between years of education or numeracy and wages when worker characteristics, occupation, job (temporary contract, part-time work) and firm characteristics (small and medium enterprise, private sector, growing firm) are included as control variables (solid colours). This compares individuals within the same occupations, types of jobs, and types of firms with the relationship observed when only worker characteristics are controlled for (entire bar with dashed borders), as in Figure 4.2. On average for participating OECD countries, a one‑standard-deviation increase in years of education is associated with a 15% increase in wages when only worker characteristics are controlled for, but an 8% increase in wages when including all controls. For numeracy, the change is from a 11% to a 7% increase in wages, after including all control variables. Occupational sorting emerges as the key driver: in every country analysed, controlling for occupation results in by far the greatest attenuation of the regression coefficients (regression tables available upon request).

Figure 4.3. Occupational sorting accounts for a substantial share of variation in wages
Copy link to Figure 4.3. Occupational sorting accounts for a substantial share of variation in wages

Change in hourly wages associated with a one‑standard-deviation increase in numeracy or years of education, percentage, by country, with and without controlling for occupations

Furthermore, the magnitude of the associations between education or numeracy and wages varies across occupations. Figure 4.4 shows wage associations with education and numeracy by occupation, with high-skilled occupations on the left of the figure and lower-skilled occupations appearing towards the right. The associations between years of education and wages are strongest in high-skilled occupations, occupations that require high levels of information processing skills, and generally decline moving down the occupational skill distribution. Returns to education are largest for managers and professionals (around 14% and statistically significant for an increase of one standard deviation), remain substantial and statistically significant for technicians (about 11%) and workers (between 6 and 10%), and drop to nearly zero for elementary occupations. By contrast, associations between numeracy and wages exhibit less heterogeneity across occupations, and the remaining variation is less systematically related to occupational skill requirements.4 Numeracy premia are highest for managers (around 10%), for technicians and associate professionals, and clerical support workers (8‑9%) but they remain positive even in elementary occupations (around 5%).

These differences are consistent with the idea that high-skilled occupations, characterised by abstract, analytical, or complex problem-solving tasks provide greater scope for workers to convert both formal education and information processing skills into higher productivity and wages. In lower-skilled occupations, weaker associations with wages – particularly for education – may reflect limited opportunities to deploy additional schooling, for instance because performance is driven by other types of proficiency, such as physical or technical precision, be the result of institutional wage‑setting practices that compress returns, or both. The flatter profile for numeracy suggests that cognitive skills, and particularly numeracy, are rewarded across a broader range of tasks, even where formal qualifications carry little wage premium.

Figure 4.4. Higher-skilled occupations reward numeracy and, especially, education
Copy link to Figure 4.4. Higher-skilled occupations reward numeracy and, especially, education

Change in hourly wages associated with a one‑standard-deviation increase in numeracy or years of education, percentage

4.1.3. The relationship between education and labour market outcomes has weakened over time

Recent empirical evidence suggests that the importance of cognitive skills as predictors of labour market success seems to have declined over the past decade, at least in some countries, even though the demand for college‑educated labour grew over the same period. In the United States, Castex and Kogan Dechter (2014[24]) show that cognitive skills had a 30%–50% larger effect on wages in the 1980s than in the 2000s. In Sweden, economic returns to cognitive skills decreased modestly during the 2000s (Edin et al., 2022[25]), and in Finland, Izadi and Tuhkuri (2024[26]) find a reduction in the value of cognitive skills between 2001 and 2015.

Results obtained using data from the Cycle 2 of the Survey of Adult Skills are consistent with these findings as they show that in participating OECD countries, the strength of the relationship between education, and even more so, skills, and employment prospects has weakened over time (Figure 4.5). On average across participating OECD countries, the association between years of education and an individual’s probability of being employed (rather than unemployed) declined by 0.7 p.p. between Cycle 1 and Cycle 2. The reduction was even larger for numeracy, as the relationship between numeracy scores and employment status decreased by 1.3 p.p.

One possible explanation for this reduction in the association between numeracy, and to a lower extent education, and probability of being employed between the two cycles of data collection is that during the second cycle of data collection (2022‑2023), labour markets in most OECD countries were tighter than during the first phase (2011‑2018), and that as a consequence, low-skilled individuals were more likely to find a job (OECD, 2024[27]). Another, complementary explanation is that structural factors such as the rapid diffusion of digital technologies and, more recently, artificial intelligence in particular, are reshaping labour demand and the demand for skills. Some tasks, such as computer use, previously performed by high-skilled individuals only can now be undertaken by workers with lower levels of cognitive skills, thereby reducing the former’s relative advantage in access to jobs. In the case of education, additionally, the continued expansion in tertiary educational attainment among younger cohorts (OECD, 2025[28]) may have moderated the employment premia associated with completing a tertiary degree, relative to the earlier period.

The decline in the relationship between years of education and employment status is particularly pronounced and significant at around 2 p.p. in Poland, Estonia, Ireland and the Slovak Republic and at about 1 p.p. in Singapore. One notable exception is Chile where the association increased by 1.6 p.p. In all other countries, the change is not statistically different from zero. The association between numeracy and employment status has decreased in nine countries: Ireland, the Slovak Republic, Sweden, Lithuania and Denmark where the decrease ranges from 2 to 4 p.p., and France, the United States, Hungary and Germany, where the decline is lower than 2 p.p. This association has only increased for one country, Singapore, where the association between numeracy and employment status has increased by 1.4 p.p.

Figure 4.5. In general, the associations between unemployment and education and skills have weakened over time
Copy link to Figure 4.5. In general, the associations between unemployment and education and skills have weakened over time

Percentage point change in association between numeracy or years of education and probability of being employed (vs. unemployed) between Cycle 1 (2011‑2018) and Cycle 2 (2022‑2023), by country

Declines in the strength of the relationship between education and wages are observed in a number of countries, while it is globally stable for skills and wages (Figure 4.6). On average across participating OECD countries, the association between years of education and wages has decreased by 2.2 p.p. between Cycle 1 (2011‑2018) and Cycle 2 (2022‑2023), while the relationship between numeracy and wages has declined by 0.4 p.p., statistically insignificant at standard levels. The decrease for educational attainment is driven by changes in France, Lithuania, Austria, Czechia, Hungary and Poland, with magnitudes ranging from 3.9 p.p. (France) to 12 p.p. (Poland). In the Netherlands and in Sweden, the strength of the association between education and wages has actually increased rather than decreased. A possible explanation could be changes in the composition of workers, for example, through an influx of low qualified workers earning lower wages, which could have increased wage associations with education. The decrease for numeracy skills is concentrated in three countries, the United States and the Slovak Republic and Singapore. For all other countries, the change is not significant or even significantly positive (France).

Figure 4.6. The relationship between education, skills and wages weakened in the last decade for some countries
Copy link to Figure 4.6. The relationship between education, skills and wages weakened in the last decade for some countries

Percentage point change in association between numeracy or years of education and hourly wages between Cycle 1 (2011‑2018) and Cycle 2 (2022‑2023), by country

4.1.4. Labour market success depends not only on what workers know, but also on the non-cognitive skills they use in their jobs

Recent research has emphasised the importance of non-cognitive skills as determinants of labour market outcomes (Deming, 2017[29]; Edin et al., 2022[25]; Izadi and Tuhkuri, 2024[26]). These skills have gained importance partly because they are difficult to replicate through automation technologies, and because the adoption of ICT technologies in the 1990s increased the value of teamwork, where social interaction and co‑ordination are essential (Bresnahan, 2002[30]). More recently there are indications that AI adoption is increasing the need for non-cognitive skills such as creative and social intelligence, reasoning, and critical thinking (Lane, Williams and Broecke, 2023[7]).

More specifically Deming (2022[31]) highlights the central role of collaborative and decision making practices in current labour markets. The literature on labour market returns to teamwork is still nascent and several studies have focussed on developing methodologies to identify individuals’ contribution to team productivity (Bonhomme, 2021[32]; Weidmann and Deming, 2021[33]). Empirical evidence on the link between teamwork, productivity, and salaries exists, but it is limited to specific occupations. For instance, Devereux (2021[34]) looks at data on co‑authorship of academic papers for economists and shows that co‑authors’ value‑added is more closely tied to one author’s salary than own value‑added. Arcidiacono et al. (2017[35]) use data from US professional basketball to show how individual performance strongly affects peers’ performance. Yet they also find that an individual’s wage is largely determined by his own performance with little weight given to his contribution to the performance of his peers. However, while informative, these findings are difficult to generalise given the highly specific nature of the occupations studied.

Evidence on decision making is also limited but growing. An important contribution is a recent paper by Caplin et al. (2023[36]), who develop a measure of decision making using a laboratory experiment and show that this measure strongly predicts income and income growth in representative samples of full-time workers in the United States and Denmark. The predictive power is particularly strong for individuals in management and other decision-intensive occupations.

The Survey of Adult Skills includes several variables related to the skills workers use in their jobs, including measures of co‑operation and task discretion that are closely related to the collaborative and decision making practices discussed in recent literature. Using data from Cycle 1 of the Survey of Adult Skills, Quintini (2014[37]) shows that the inclusion of skills use variables weakens the effect of both education and proficiency on wages by about a third on average, and that co‑operation and task discretion are positively and significantly correlated with wages in about half of the participating countries. Using Cycle 2 of the Survey of Adult Skills, a recent report (OECD, 2026[38]) shows that a more frequent deployment of skills in the workplace is associated with higher wages, after controlling for education and skills levels. The strongest association between skills use and wages is observed for influencing and information-processing skills such as reading, ICT and numeracy. Task discretion and co‑operation show moderate but positive and statistically significant associations with wages, and dexterity and physical skills are, on the contrary, negatively correlated with wages (all else being equal, including qualification, skills proficiency and occupation).

In fact, cognitive and non-cognitive skills should not be viewed in isolation but as mutually reinforcing dimensions. This complementarity, initially explored in the context of skill formation among children (Cunha et al., 2006[39]; Cunha, Heckman and Schennach, 2010[40]), has also been documented among working-age adults. For instance, in the United States, Weinberger (2014[41]) finds growing complementarity between cognitive and social skills between the 1980s and the 2000s. In the workplace, data from the Survey of Adult Skills show that in digital-intensive industries workers endowed with a high level of numeracy skills receive an additional wage premium if they also have high levels of self-organisation or managing and communication skills (Grundke et al., 2018[42]).

New analyses using Cycle 2 of the Survey of Adult Skills show that a higher degree of task discretion and co‑operation in one’s job is associated with higher wages (by 1% and 6%, respectively, on average across participating OECD countries), consistent with the growing importance of human-centred work practices. Including these variables does not significantly change the association between numeracy, years of education, and wages. In contrast to the declining returns to skills and education observed over the past decade, the associations between the degree of co‑operation and task discretion with wages are strong and have remained stable over time.

Furthermore, the association between numeracy and wages is more pronounced among workers who exercise higher degrees of task discretion and co‑operation in their jobs, suggesting strong complementarity between information processing and non-cognitive skills (Figure 4.7). The association between years of education and wages also tends to be stronger for workers who use high levels of task discretion and co‑operation, although this difference is not statistically significant.

Figure 4.7. The association between numeracy and wages is more pronounced among workers with higher degrees of task discretion and co‑operation
Copy link to Figure 4.7. The association between numeracy and wages is more pronounced among workers with higher degrees of task discretion and co‑operation

Association between numeracy or years of education and hourly wages, by degree of co‑operation and task discretion

4.2. Training and learning are associated with positive labour market outcomes, particularly in the case of non-formal training

Copy link to 4.2. Training and learning are associated with positive labour market outcomes, particularly in the case of non-formal training

4.2.1. Associations between participation in job-related training and wages differ by the type of learning

The theoretical literature on the relationship between on-the‑job training and wages dates back to Becker (1962[43]). Becker argued that, under conditions of perfect competition in the labour market, general training – i.e. training that develops skills transferable across all firms – should be financed by workers. Workers pay training by perceiving lower wages during the training period, since once trained, workers can move to another employer as the skills they have developed would be portable across jobs.

In the presence of labour-market imperfections – such as wage rigidities, search frictions or employers’ market power over wage setting – firms may partially finance general training because they can be more confident of retaining workers after training (Acemoglu and Pischke, 1998[44]; Acemoglu and Pischke, 1999[45]; Stevens, 1994[46]). Firms may also finance general training when they need specific combinations of general skills that are valuable only to a limited set of employers (for example, a combination of knowledge of history and advanced digital skills, which are both general skills) (Lazear, 2009[47]). In all these cases, wages would lie above workers’ productivity during training and below it afterwards. As a result, wages either rise or remain constant following training, depending on the productivity differentials between the two periods.

For job-specific training, neither the firm nor the worker have a strong incentive to pay the full cost of training. To avoid under-provision of specific training, a sharing mechanism is needed. In practice, such sharing typically results in wages being set below the worker’s productivity after training, but above the wage received before training (Hashimoto, 1981[48]).

Since these seminal works, the empirical literature has sought to quantify the wage returns to on-the‑job training. Evidence in this area, coming from different countries and periods and produced using diverse methodologies, points to robust, and in most cases statistically significant, positive effects of work-related training on wages (Duncan and Hoffman, 1979[49]; Veum, 1995[50]; Parent, 2003[51]; Bassanini et al., 2007[52]; Albert, Garcìa-Serrano and Hernanz, 2010[53]; Almeida and Faria, 2014[54]; Biesebroeck, 2008[55]). Most of these studies acknowledge that simply comparing trained and untrained workers’ wages is misleading, as individuals who undertake training tend to have higher abilities and/or motivation. They therefore try to correct for this bias by comparing similar workers or track workers over time to address this selection issue. More recent studies, leveraging random or quasi-random assignment into training that enables a causal interpretation of estimates, also usually find substantial positive wage effects (Adhvaryu, Kala and Nyshadham, 2023[56]; De Grip and Sauermann, 2012[57]; Denzler, Ruhose and Wolter, 2025[58]; Leuven and Oosterbeek, 2008[59]; Osman and Speer, 2025[60]).

Using data from Cycle 1 of the Survey of Adult Skills, Fialho, Quintini and Vandeweyer (2019[61]) highlight the importance of distinguishing between formal, non-formal, and informal learning (see Box 4.2 for a definition of each learning type). After accounting for a number of socio-demographic and job characteristics and controlling for selection into employment, they find that participation in formal learning is associated with a small decline in wages. In contrast, participation in non-formal and informal learning is associated with higher wages, by 11% and 3.5% respectively. The study also suggests that employers benefit more from the training they provide than workers do, as the rise in value added per hour worked exceeds the increase in wages. Furthermore, the authors find that more costly training – potentially of better quality – is associated with greater productivity gains.

Box 4.2. Types and incidence of adult learning in Cycle 2 of the Survey Adult Skills

Copy link to Box 4.2. Types and incidence of adult learning in Cycle 2 of the Survey Adult Skills

In this chapter adult learning comprises three types of learning:

Formal training refers to accredited education or training that is planned, delivered within an institutional setting, lasts at least one semester, and is officially recognised by the relevant authorities. In this chapter, formal training is defined as learning undertaken by adults aged 25 to 65 who: i) were studying towards a qualification at the time of the survey; ii) had completed a qualification in the 12 months preceding the survey interview; or iii) had interrupted studies towards a formal qualification in the 12 months prior to the survey interview.
Non‑formal training involves planned, institution‑based education or training, but it is either shorter than one semester or not formally accredited. In this chapter, non‑formal training is measured as having participated in organised learning activities that do not involve studying towards a qualification in the 12 months preceding the survey interview. These include courses, webinars, workshops, lectures, or private lessons aimed at acquiring skills for a current or future job.
Informal learning is intentional but takes place outside institutional settings; it is less structured than formal or non‑formal learning and can occur in any environment. In this chapter, informal learning is measured by the frequency with which adults engaged in activities such as learning from others, learning by doing, or learning about new products and services provided by their employer in the 12 months prior to the survey interview. Adults who engaged in any of these activities at least once per week are considered to have engaged in informal learning.

To exclude younger adults still in initial education and to concentrate on adults undertaking formal and non-formal learning that is more closely linked to jobs, the analysis focusses on job-related adult learning for the population aged 25‑65, disregarding learning for other reasons.

On average, around 8%, 37% and 65% of adults in OECD countries participating in Cycle 2 of the Survey of Adult Skills took part in formal, non‑formal and informal learning, respectively (Figure 4.8). Participation in formal training is relatively low across all countries – below 20% everywhere – whereas participation in non‑formal and informal learning varies much more widely. For example, more than 53% of adults in Norway engage in non‑formal training, compared with fewer than 10% in Korea. By definition, informal learning – having engaged in relevant learning activities at least once a week in the 12 months prior to the interview – tends to be more widespread. Still, countries differ substantially: about 25% of adults report engaging in informal learning in Poland, compared with 80% in Portugal. Despite these differences, participation rates across the three learning types are generally correlated. Countries with relatively high levels of one type of learning tend also to show higher levels of the other two.

Figure 4.8. Participation in adult learning varies by type of learning and country
Copy link to Figure 4.8. Participation in adult learning varies by type of learning and country

Share (%) of adults who participated in adult learning, by type of learning

Participation in learning also varies by individuals’ occupation (Figure 4.9). In general, adults in higher‑skilled occupations – such as managers and professionals – are more likely to engage in learning activities. By contrast, those in lower‑skilled roles, including elementary occupations, tend to have the lowest participation rates.1

Figure 4.9. Employees in higher-skilled occupations participate more in adult learning
Copy link to Figure 4.9. Employees in higher-skilled occupations participate more in adult learning

Share (%) of employees who participated in adult learning, by occupation

The positive association between non-formal training and wages as well as the negative association between formal training and wages identified in Fialho, Quintini and Vandeweyer (2019[61]) also emerge when using data from Cycle 2 of the Survey of Adult Skills, collected in 2022-2023 (Figure 4.10). Adults who participated in non-formal learning in the previous 12 months in participating OECD countries earn, on average, wages that are about 6.2% higher than those of workers who did not participate. As in Fialho, Quintini and Vandeweyer (2019[61]), these estimates control for an extensive set of worker, job and firm characteristics, including occupation. However, despite the extensive list of controls, unobserved variables, such as attitudes towards learning or motivation, which are associated with both wages and participation in training, could bias the observed relationship. To address this concern, following Leuven and Oosterbeek (2008[59]), the comparison group was restricted to workers who wished to participate in non-formal training but were unable to do so for unexpected reasons, thereby offering a closer counterfactual. Results using this control group (shown in Annex Table 4.B.3) are qualitatively and quantitatively similar to the main results. While this approach does not address all endogeneity concerns, this result suggests that omitted variables correlated both with non-formal training participation and wages do not explain the association.

Box 4.3. Empirical strategy to estimate associations of training and learning participation with wages

Copy link to Box 4.3. Empirical strategy to estimate associations of training and learning participation with wages

To analyse the relationship between training and learning participation with wages, the following ordinary least squares specification is used:1

$l n (w_{i j f o s c}) = α + β T r a i n i n g_{i} + γ x_{i} + θ u_{j} + δ z_{f} + μ_{o} + σ_{s} + ρ_{c} + ε_{i j f o s c}$

where $w_{i j f o s c}$ is the gross hourly wage in purchasing power parity-adjusted 2022 USD of individual i employed in job j, firm f, occupation o, sector s and country c. $T r a i n i n g_{i}$ includes three dummy variables indicating whether an individual has taken part into a learning activity during the year preceding the survey interview. In the context of this chapter, the three indicators are defined as follows:

$F o r m a l t r a i n i n g_{i}$ is equal to 1 if individual i: i) was studying towards a qualification at the time of the interview; ii) had completed a qualification in the 12 months preceding the interview; or iii) had interrupted studies towards a formal qualification in the 12 months preceding the interview.
$N o n - f o r m a l t r a i n i n g_{i}$ is equal to 1 if individual i participated in organised learning activities that did not involve studying towards a formal qualification in the 12 months preceding the interview. These include courses, webinars, workshops, lectures, or private lessons aimed at acquiring skills for a current or future job.
$I n f o r m a l t r a i n i n g_{i}$ is equal to 1 if individual i engaged in learning from others, learning by doing or learning about new products and services offered by their employer at least once per week in the 12 months preceding the interview.

$x_{i}$ is a set of individual control variables including numeracy score and completed years of education as defined in Box 4.1, age, sex, immigrant background, parental education, whether one lives with a partner or has children, work experience and work experience squared; $u_{j}$ is a set of job characteristics including whether the individual works part-time or with a temporary contract, such as a fixed term contract, seasonal contracts, a temporary employment agency contract, a zero-hour contract, a freelance, contractor and/or consultant contract, working as an apprenticeship or other training scheme or without a contract; $z_{f}$ is a set of firm characteristics including whether the individual works in a SME, in a growing firm (i.e. the number of employees increased over the past 12 months), and in a private sector firm; $μ_{o}$ are occupation (ISCO 2 digits) fixed-effects; $σ_{s}$ are sector (ISIC 1 digit) fixed-effects; and $ρ_{c}$ are country fixed-effects. The coefficient of interest is $β$ , which captures the percentage change in wages associated with participating in formal or non-formal training or engaging in informal learning.

1. The Survey of Adult Skills is cross‑sectional, meaning that individuals are observed only at one point in time. As a result, the data contain no information on respondents’ wages prior to their participation in training. Thus, it is not possible to compare pre‑ and post‑training wage levels for the same individuals.

Participation in non‑formal learning is associated with higher wages in nearly all countries participating in Cycle 2 of the Survey of Adult Skills, although the magnitude of the relationship differs (Figure 4.10). As with wage associations with skills, the link between non‑formal training participation and wages tends to be greater in countries with higher levels of wage inequality and for workers with higher levels of education.5 This pattern may reflect greater flexibility in wage‑setting practices and may also capture differences in the characteristics of adults who engage in training. While job mobility is one of the main channels through which adults achieve wage gains (Topel and Ward, 1992[62]), this mechanism cannot be examined using Survey of Adult Skills data.6

Wages for adults who engaged in informal learning activities at least once per week in the last 12 months – such as learning from others, learning by doing or learning about new products and services offered by their employer – in participating OECD countries are, on average, not significantly different from those of workers who did not undertake these activities or did so less frequently. A high engagement in informal learning is linked to significant wage differences in only a limited number of countries (Figure 4.10). In Singapore, the Slovak Republic and Germany, more engagement in informal learning activities is associated with higher wages, whereas in Denmark it is associated with lower wages. The positive relationship observed in most of these countries appears to be driven largely by strong associations between informal learning and wages among individuals working in health‑related occupations – such as health professionals and health associate professionals – and managers in hospitality, retail and other services. In these countries and occupational groups, associations between informal learning and wages were smaller in magnitude and generally not statistically significant in the Cycle 1 of the Survey of Adult Skills, suggesting that the more pronounced associations seen in Cycle 2 may be partly related to post‑COVID‑19 pandemic wage developments in these occupations and industries. On the other hand, the negative association observed in Denmark may be driven by adults with lower wages self-selecting into engaging in informal learning. However, these results should be interpreted with caution: the Survey of Adult Skills does not collect information on individuals’ willingness and barriers to engage in informal learning, making it impossible to rule out endogeneity arising from omitted variable bias or reverse causality.

Finally, participation in formal training is associated with lower wages (Figure 4.10), although the robustness of this finding cannot be established. This negative association holds in most countries with a few exceptions – such as Chile, Latvia and Estonia. Unlike in Fialho, Quintini and Vandeweyer (2019[61]), this result is driven by workers who have already completed formal training, rather than by those earning lower wages while they are actively participating in training. It is also worth noting that the negative association between formal training and wages is statistically significant only for individuals (i) having started a new job within the past year, (ii) being hired under a contract other than an indefinite contract, or (iii) working part-time and that results seem driven by younger adults and, to some extent, migrants.7 Workers under open-ended contracts, working full-time, and holding their position for more than a year do not experience such negative association of formal training with wages. Thus, this association could be capturing lower earnings received by younger workers, just entering the labour market or an occupation. As with informal learning, however, the interpretation of these patterns is limited by data constraints: the Survey of Adult Skills does not collect information on individuals’ willingness to participate in, or barriers to accessing, formal training. Consequently, it is not possible to assess the extent to which these associations may reflect endogeneity arising from omitted variable bias or reverse causality.

Figure 4.10. On average in OECD countries, there is a clear and positive association between non-formal training and wages
Copy link to Figure 4.10. On average in OECD countries, there is a clear and positive association between non-formal training and wages

Percentage change in hourly wages associated with participation in adult learning in the previous 12 months across countries, by training type

Although these findings point in the same direction as those in Fialho, Quintini and Vandeweyer (2019[61]), their magnitudes are smaller, to the point that associations of wages with informal learning are not significant anymore. Average associations of wages with training participation for countries that participated in both cycles of the Survey fell by 4.6, 5.6 and 3.7 p.p. for non-formal training, informal learning formal training, respectively. These results could suggest that returns to training have declined over time across all three types of learning. However, regarding non-formal training, this interpretation remains uncertain, as changes in how the question on non‑formal training participation was administered between the two survey cycles could account for some of the observed differences.8

In addition, the relationship between non‑formal training and wages appears to be partly driven by selection into occupations. Once occupational controls are included, the estimated wage premium declines by about 2 p.p. This suggests that individuals working in higher‑paying occupations are also more likely to participate in non‑formal training.

4.2.2. The relationship between training participation and wages is heterogeneous across occupations

The association of training participation and wages also varies substantially across occupations, as shown in Figure 4.11. Participation in non-formal training is associated with higher wages across all occupational groups, with managers and craft and related trades workers experiencing the largest wage premia. At the other end of the spectrum, professionals and workers in elementary occupations exhibit slightly lower associations between non-formal training and wages. Taken together, the results suggest that non‑formal training appears to be a relevant form of skill development for most occupations.

Figure 4.11. Managers and skilled trades have the strongest association between non-formal training and wages
Copy link to Figure 4.11. Managers and skilled trades have the strongest association between non-formal training and wages

Percentage change in hourly wages associated with participation in non-formal training in the previous 12 months across occupational categories

4.2.3. Labour market outcomes depend on what, how, and who pays for training

Recent empirical work on the returns to training distinguishes between training for cognitive and noncognitive skills and seems to support the idea that training for noncognitive skills, such as communication, time management, decision making, and effective teamwork, is particularly effective. Adhvaryu, Kala and Nyshadham (2023[56]) randomly allocated workers in five garment factories in India into training for soft skills and found that trained workers were 13.5% more productive. Productivity gains were largest among workers who initially had low levels of leadership skills and were most pronounced when trainees worked alongside other coworkers in operation teams. This work is among the very few, if not the only, studies examining the causal impact of on-the‑job training for soft skills for employed individuals.

Literature using randomised experiments concerning soft skills development for recent graduates has focussed in developing countries and reports mixed results. Groh et al. (2016[63]) find no employment impact of a training on interpersonal skills in either the short or medium term. By contrast, Brudevold-Newman and Ubfal (2024[64]) show that short-run employment prospects improve for individuals who receive training in effective communication, networking, interpersonal (teamwork, collaboration, trust, empathy, negotiation) and intrapersonal (self-awareness, personal initiative, and perseverance) skills. Acevedo et al. (2020[65]) finds positive effects, but only for women. Osman and Speer (2025[60]) examine how soft‑skills and technical‑skills training interact and show that individuals who receive both types of training earn 20‑27% higher incomes than those who participate in only soft‑skills or only technical‑skills programmes.

Another strand of the literature evaluates the effects of Active Labour Market Policies (ALMPs). Although a full review of this extensive body of work is beyond the scope of this chapter,9 one recent paper analysing an intervention focussing on soft skills deserves particular attention. Schlosser and Shanan (2025[66]) study an ALMP that focusses on enhancing soft skills of welfare recipients in Israel and find that the programme increased participants’ employment rates and decreased income‑support recipiency. The impacts persist five to six years after the programme’s implementation.

Results using Survey of Adult Skills data indicate that the associations of non-formal training with wages differ considerably by training topic (Figure 4.12). Consistent with the broader literature showing positive effects of non‑cognitive skill development on labour‑market outcomes, participation in training aimed at developing teamwork and leadership, and project management skills shows the largest association with wages, although the robustness of this result cannot be established given that the topic of training cannot be observed for adults who did not participate in training. Given evidence that such non-cognitive skills may be relatively easy to develop even later in life (Adhvaryu, Kala and Nyshadham, 2023[56]; Cunha et al., 2006[39]; Cunha, Heckman and Schennach, 2010[40]; Osman and Speer, 2025[60]), encouraging training in these areas could be particularly beneficial. The strong wage associations observed for these training topics also align with the finding that managers appear to benefit the most from non‑formal training, as these areas, jointly with computer and software skills, are among the most common training subjects for this occupational group. This suggests that the stronger wage associations may partly reflect the higher participation of workers in better-paid occupations in such training. Trainings on reading and writing, foreign languages, computer and software and communication and presentation skills are associated with higher wages of about 5.5 to 8.5%. These types of training appear to be relatively more common among technicians and associate professionals, as well as among service and sales workers, compared with other occupational groups.

Finally, training on numeracy, health and safety, operating machinery or equipment, handling customers or clients, and job-related training on creative or musical skills and sports shows relatively weak associations with wages. Health and safety‑related training is the most common type of training completed by adults and is particularly common among skilled agricultural, forestry and fishery workers, plant and machine operators and assemblers, and workers in elementary occupations, accounting for more than one‑quarter of all training they undertake. Similarly, training on operating machinery or equipment is especially prevalent among plant and machine operators and assemblers and among workers in elementary occupations, again representing over a quarter of their completed training. These types of training are often mandatory for performing specific tasks in these occupations, which likely contributes to their lower observed wage differentials.

Figure 4.12. Training on teamwork, leadership and project management is associated with larger wages
Copy link to Figure 4.12. Training on teamwork, leadership and project management is associated with larger wages

Percentage change in hourly wages associated with participation in non-formal training in the previous 12 months by topic

Training characteristics can also influence both the effectiveness of training and its association with wages. Although training flexibility is increasingly prominent in the current policy agenda – recently highlighted by an OECD publication underscoring its potential to expand participation and improve inclusiveness (OECD, 2023[67]) – limited evidence exists on whether flexible training modalities are as effective as traditional formats in developing workers’ skills. Adult learning opportunities can offer flexibility across four key dimensions: time, place, mode, and content; and the small existing literature focusses primarily on delivery mode and its implications for skill acquisition. Evidence shows that participants in blended learning, where in‑person instruction is combined with online components, report higher satisfaction and engagement, but also face challenges such as technological barriers and increased cognitive load (Bower et al., 2015[68]). Blended learning may lead to improved test performance, though part of this improvement reflects higher dropout rates among lower-performing learners (Deschacht and Goeman, 2015[69]).

Results from Cycle 2 of the Survey of Adult Skills indicate stronger wage associations for distance/online learning and blended/hybrid learning than for fully in-person training (Figure 4.13). These patterns partly reflect that low-return training topics, such as health and safety, operating machinery or equipment and handling customers or clients, are predominantly offered in person (they account for about 40% of in‑person training). Nonetheless, the findings for distance learning are robust to including controls for training topic. This suggests that this training format does not yield lower wage benefits than traditional training. Similarly, wage associations for non‑formal training do not differ significantly between shorter courses (completed within a day) and longer ones (lasting over a week), nor between training delivered in one or a few consecutive days and recurring courses spread over several days or months. Overall, these results indicate that flexibility in delivery mode and timing is not associated with lower wages.

The relationship between training participation and wages also varies depending on who finances the training. The limited existing evidence comparing employer-sponsored training with training paid for by individuals suggest that, at least in the short run, formal training leads to higher wages when employers – rather than workers – bear the cost (Bishop, 1997[70]). This may indicate that employers are better able to identify effective formal training programmes. Alternatively, the results may reflect non-random selection by employers, who may tend to sponsor training for their most promising workers or for those they wish to retain or promote (Dietz and Zwick, 2020[71]).

Findings from the Survey of Adult Skills for non-formal training are similar: training financed by employers shows a stronger association with wages than training paid for by individuals, by other funders, or offered free of charge (Figure 4.13). Likewise, given that more than 85% of employer‑funded training occurs during working hours, training undertaken during working hours is also more strongly associated with higher wages than training completed outside of working hours. These associations partly reflect differences in the topic of the training and in the characteristics of participating workers. Employer‑funded non‑formal training is more commonly targeted to job‑specific needs or organisational processes. Topics such as security, computer and software, teamwork and leadership, project management, operating machinery or equipment and handling customers or clients account for over 70% of employer-funded training. Importantly, however, employer‑funded training continues to display a stronger association with wages even after controlling for the topic of training. This pattern suggests that employers may be more effective at identifying training that is particularly relevant or likely to yield higher productivity and, consequently, returns. However, employers may select high‑potential employees to reap greater returns from their training, as found in Dietz and Zwick (2020[71]), and higher wages could be explained, at least partly, by this selection, instead of training participation.

Furthermore, non‑formal training that leads to a certificate is not associated with higher wages than comparable training without certification. More than 40% of workers who completed certificate‑granting training did so in topics with relatively low associations with wages – such as health and safety, customer handling, or machinery and equipment operation. These courses are often mandatory, which may limit their impact on wages. In addition, many non-formal training certificates lack recognition or portability across employers, which likely contributes to the absence of a wage premium for certificate‑based non-formal training.

Figure 4.13. Flexible learning and employer-funded training are associated with higher wages
Copy link to Figure 4.13. Flexible learning and employer-funded training are associated with higher wages

Percentage change in hourly wages associated with participation in non-formal training by characteristic of the training

4.3. Fostering skills development and high-return training requires renewed public policies

Copy link to 4.3. Fostering skills development and high-return training requires renewed public policies

4.3.1. Signalling, rewarding and recognising skills are becoming increasingly central

Results presented in this chapter show that information processing and non-cognitive skills seem to have a direct and significant effect on wages beyond education, underscoring the value of skills. These results support the case for skills-first approaches, which refer to ways of designing hiring, education, and public policy that prioritise what people know and can do – their skills – over formal credentials such as degrees, job titles, or years of experience. These approaches should not replace traditional credential-based systems, but rather complement them.

The skills-first readiness and adoption index developed by the OECD and SUSS-IALS in 2025 provides a structured way of analysing measures to foster skills signalling, rewarding and recognition (OECD/SUSS-IAL, 2025[72]). This framework suggests that governments have a central role to play in supporting talent recognition by embracing policies that make skills visible such as Recognition of Prior Learning (RPL) or skills signalling. Government initiatives to advance skills-first approaches should, wherever possible, align with existing skills-first practices already adopted by employers.

Through RPL, individuals can have knowledge and skills acquired through non‑formal or informal learning formally validated and certified, allowing these competences to be recognised and rewarded in the labour market. France has a long-standing RPL tradition, with its system in place for more than 20 years. The French RPL framework is closely linked to the National Directory of Professional Certifications (RNCP), ensuring transparency and comparability of qualifications and guaranteeing that RPL awards carry the same status as those obtained through traditional education and training pathways.

Other skills‑signalling policies, such as skills passports, also help make individuals’ competences more visible, thereby facilitating the hiring of adults with relevant skills and supporting their recognition and reward in the labour market. One example is Japan’s Job‑Card System, which documents individuals’ skills, work history and career aspirations. While its primary purpose is to support individuals in understanding and planning their career development, the Job‑Card can also be used to demonstrate acquired skills to prospective employers (OECD, 2026[73]).

4.3.2. Enhancing training flexibility can raise participation and deliver benefits comparable to traditional formats

Many adults face multiple, multifaceted, and interrelated constraints, including limited time due to work or family responsibilities, as well as training opportunities offered at inconvenient times or locations. By reducing these constraints, more flexible training arrangements can make participation feasible for a wider range of learners. Flexible training refers to adult learning provision that adapts to learners’ individual circumstances by allowing greater choice over how, when, where, and what they learn (OECD, 2023[67]).

The evidence presented in this chapter indicates that short training, training delivered at least partly online, or training offered over multiple sessions are not associated with lower returns than less flexible learning formats, indicating that it may have a similar labour market value than longer and less flexible training. If flexible training delivers at least comparable outcomes to in‑person or longer programmes – while also boosting participation – flexible training could be promoted. At the same time, in-person training, should be maintained to ensure that adults with limited digital skills and those living in areas with poor connectivity can participate.

Policymakers can use different measures to increase the flexibility of adult learning provision. One common approach is modularising adult learning programmes – that is, breaking longer programmes into a number of shorter modules that can be completed individually. These modules often lead to micro-credentials, which are short, targeted learning certifications that recognise specific skills or competencies and certify the successful completion of individual learning modules that stand alone or that can be combined to build toward larger qualifications. The short duration of these learning opportunities enhances their accessibility compared with traditional pathways, making them especially appealing to time‑constrained individuals. They also allow for personalisation across delivery mode, location, and content (OECD, 2025[12]). Denmark’s well‑established modular approach to adult learning shows how flexible, stackable learning pathways can expand opportunities for adults. By documenting every module, learners and providers are enabled to easily track progress and combine modules across institutions. This supports flexible participation and contributes to a high share of adults achieving further qualifications (OECD, 2023[67]). Higher education students can also benefit from the modularisation of programmes through time and cost savings, for example by receiving credit towards a qualification for prior relevant education or training.

RPL can support flexible adult learning pathways in different ways. It may allow learners to skip selected subjects and reduce duplication of learning, leading to shorter training duration and alleviating time constraints. When exemptions from admissions processes exist, it may also reduce application costs, thereby reducing cost-related barriers to training (OECD, 2023[74]). In Norway, RPL procedures can be used to access post‑secondary and tertiary education, broadening opportunities by allowing adults aged 25 and over to have their formal, non‑formal and informal learning recognised for entry (OECD, 2023[67]).

Other initiatives, such as National Qualifications Frameworks (NQFs), credit transfer systems, and regularly updated national qualification standards, play a key role in supporting modularisation and, consequently, RPL. NQFs are instruments by which countries organise, recognise, and assign value to qualifications and place them within a hierarchy based, ideally, on learning outcomes (Cedefop, 2017[75]; European Training Foundation, 2017[76]). They facilitate modularisation and RPL by helping institutions locate prior learning within the qualifications’ hierarchy and foster learning flexibility by supporting learner mobility across institutions, programmes, and levels. Strong NQFs encompass and link different types of learning (formal, non-formal and informal) at different levels (basic skills and qualifications, vocational education and training, higher education). However, many countries operate separate frameworks for academic and vocational education and training (Martin and Godonoga, 2020[77]). NQFs should be complemented with and linked to credit transfer systems that allow learning outcomes or credits earned in one programme or institution to be recognised and used in another. In addition, the qualifications that are organised within NQFs, national qualification standards, should be up to date. These standards define the core competences or learning outcomes required to obtain a formal qualification, describing what learners are expected to know, understand, and be able to do upon completion, and are crucial to establish a shared reference for competences and to foster RPL across providers (Cedefop, 2017[78]).

4.3.3. Quality and information are essential to increase returns to training

While making training more flexible, it is important to ensure that it remains of high quality. To this end, appropriate quality assurance procedures should be put in place, ensuring high‑quality implementation and supporting continuous improvement.

Furthermore, as adult learning systems become more flexible, they generally become increasingly complex for individuals to navigate. Policymakers should consider expanding and improving existing career guidance services that can help individuals make better choices about education, training, and employment. (OECD, 2023[67]).

4.3.4. Capacity-building measures can be used to encourage employer-provided training

Employer-provided training is highly related to employees’ jobs and generally shows high associations with wages. Yet, small and medium enterprises (SMEs) face significant barriers to training their workforce. In addition to financial constraints, they often lack information on training opportunities or available support mechanisms and on the benefits of investing in training, and they are usually less able to identify their skills needs than larger companies. Public intervention is therefore interesting to facilitate and encourage employer-provided training in SMEs (OECD, 2021[79]).

Financial constraints often limit employers’ ability to invest in training, particularly among SMEs. To address these barriers, direct subsidies are a common policy tool to reduce the costs borne by employers when training their workforce (OECD, 2021[79]). An alternative approach is the use of sectoral training funds, through which firms pool resources to finance training that meets the specific needs of their sector, as is the case in Flanders (Belgium).

Capacity-building measures such as skills assessment and anticipation services, learning and training networks to share planning and costs, and training managers to develop a learning culture within the firm, are promising instruments to support and expand employer-provided training in SMEs. Skills assessment and anticipation services help firms identify skills gaps in relation to current and future needs. In France, the Advance Management of Skills (Gestion prévisionnelle des emplois et des compétences – GPEC) helps organisations anticipate and adapt jobs, workforce levels and skills to evolving economic, technological, social and legal environments. It is a structured method to analyse future skill needs, manage career development, prevent recruitment difficulties, address ageing‑workforce issues, and support organisational change (Ministère du Travail et des solidarités, 2024[80]). These services may be provided by public employment services, by employer associations, by learning networks, or by private firms. In addition, collaboration among companies allows them to pool resources, achieve economies of scale, and create a critical mass in training demand, lowering per-worker costs. Learning and training networks can also directly support companies through guidance on available financial instruments or offering directly skills assessments services, for instance (OECD, 2021[79]). Ireland’s Skillnet Business Networks are collaborative clusters of private‑sector businesses that come together by sector or region to address shared skill needs. There are 70 Skillnet Business Networks through which their members can identify skill gaps and provide upskilling programmes (Skillnet Ireland, 2026[81]).

Managers’ and entrepreneurs’ attitudes towards learning within the firm play a critical role in shaping investments in skills. Strengthening their competences and fostering an understanding of human capital as a productive investment can therefore support SMEs’ workforce upskilling and skills development. Coaching, mentoring, and peer learning among managers are interesting programmes that can facilitate knowledge sharing and strengthen managers’ and hence SMEs’ capacity and learning culture. Approaches that combine peer learning with individualised support – such as subsidised consulting or coaching services – appear particularly effective. These programmes should be delivered in a flexible manner, for example through modular courses and online formats (OECD, 2021[79]) A notable example is Enterprise Ireland’s Mentor Network, a government‑supported programme that connects SMEs with experienced entrepreneurs and industry experts. Through 5‑10 one‑to‑one mentoring sessions delivered over 6‑12 months – typically online – SME managers can discuss key operational and strategic challenges with their mentors (Enterprise Ireland, 2026[82]).

More broadly, firms could be encouraged to adopt High Performance Work Practices (HPWPs), organisational practices that promote employee autonomy, collaboration, and effective use of skills in the workplace. Existing evidence suggests that HPWPs are conducive to increased opportunities for on-the‑job training (OECD, 2021[83]). The findings in this chapter also indicate that cognitive skills, such as numeracy, are strongly rewarded in the labour market but even more so in collaborative work environments and when employees have some autonomy over task sequencing. Adaptive organisational structures that give workers discretion to respond to context-specific conditions create an environment in which cognitive skills can be fully leveraged and translated into higher performance. In practice, the promotion of HPWPs typically involves measures such as guidance, knowledge sharing, and the dissemination of best practices (OECD, 2021[79]).

4.4. Concluding remarks

Copy link to 4.4. Concluding remarks

This chapter shows that both cognitive skills and formal education remain an important determinant of labour market outcomes – both employment status and wages. However, their importance has declined over time. Occupational sorting plays a key role, as education and cognitive skills are associated with higher wages primarily because they facilitate access to high-paying occupations. Furthermore, the strength of the association between education, skills, and wages varies substantially across occupations and is especially pronounced in high-skilled occupations. The chapter also highlights the important role of non-cognitive skills and work practices in shaping labour market success, as greater task discretion and co‑operation at work are associated with higher wages. Finally, the relationship between numeracy and wages is stronger among workers with higher levels of task discretion and co‑operation, pointing to a strong complementarity between cognitive and non-cognitive skills.

Results presented in this chapter also show that participation in non-formal training is consistently associated with higher wages, whereas the relationship between wages and other forms of learning – namely formal training and informal learning – is less clear. As with education and skills, the association between non-formal training and wages appears to be partly driven by the fact that high-paying occupations are also the ones where individuals participate more in training and learning activities. Beyond the type of learning activity, several training characteristics also matter. Training aimed at developing teamwork, leadership, and project management skills is associated with the largest wage premia. Online learning – whether delivered fully online or in hybrid format – also shows strong associations with wages, suggesting that flexible training formats can deliver wage benefits comparable to traditional modes while potentially increasing participation. Non-formal training financed by employers is also strongly associated with wages.

While this chapter employs several strategies to address potential endogeneity, the findings remain non causal and only show associations with wages. More research is therefore needed to establish the causal impact of skills, education and training on labour market outcomes. In particular, studies that use random allocation of adults into training programmes help expand the causal evidence base on training returns. Strengthening this evidence is essential for designing more effective skills policies.

Nevertheless, the findings in this chapter suggest that policies which improve the visibility of skills – such as RPL and other skills‑signalling tools – can have positive effects on wages by providing clearer information on what individuals are able to do, in addition to allowing to shorten the time necessary to obtain qualifications and potentially reduce costs of education and training. Since associations with wages of online, short, or otherwise flexible forms of training are not lower than those of traditional formats, enhancing the flexibility of adult learning should be a priority where it increases participation. Key strategies include modularising longer programmes, using RPL to shorten training pathways or facilitate access, and strengthening NQFs, all backed by robust quality assurance and supported by high-quality career guidance. Given the strong wage associations of employer‑provided training, such training should be further encouraged – particularly among SMEs – through measures that strengthen firm capacity, including skills assessment and anticipation services, shared learning and training networks, and managerial training.

Effective skills policies also depend on robust systems for monitoring, evaluating, and learning to ensure that public investments generate value in the labour market. Governments are increasingly expected to demonstrate what their interventions achieve, not only in terms of implementation but also in terms of outcomes. Comprehensive results-based monitoring and evaluation frameworks can help determine whether a project, programme, or policy has achieved its intended medium- and long-term objectives, and can provide valuable evidence on the labour‑market impacts of training.

Such frameworks can also help explain how progress has been made, identify challenges encountered along the way and highlight areas for improvement. Programmes that prove ineffective should be adapted based on evaluation findings or, when necessary, discontinued so that resources can be redirected toward initiatives that deliver greater impact. However, modifying or terminating established programmes can be politically challenging. Launching interventions on a smaller scale – such as through pilots or staggered implementation across regions or sectors – can help address this challenge. These approaches allow for testing, learning and refining before scaling programmes up, making it easier to adjust, or discontinue programmes based on clear evidence of their effectiveness.

References

[45] Acemoglu, D. and J. Pischke (1999), “Beyond Becker: Training in Imperfect Labour Markets”, The Economic Journal, Vol. 109/453, pp. 112-142, https://doi.org/10.1111/1468-0297.00405.

[44] Acemoglu, D. and J. Pischke (1998), “Why Do Firms Train? Theory and Evidence”, The Quarterly Journal of Economics, Vol. 113/1, pp. 79-119, https://doi.org/10.1162/003355398555531.

[65] Acevedo, P. et al. (2020), “How vocational education made women better off but left men behind”, Labour Economics, Vol. 65, p. 101824, https://doi.org/10.1016/j.labeco.2020.101824.

[56] Adhvaryu, A., N. Kala and A. Nyshadham (2023), “Returns to On-the-Job Soft Skills Training”, Journal of Political Economy, Vol. 131/8, pp. 2165-2208, https://doi.org/10.1086/724320.

[53] Albert, C., C. Garcìa-Serrano and V. Hernanz (2010), “On‐the‐job training in Europe: Determinants and wage returns”, International Labour Review, Vol. 149/3, pp. 315-341, https://doi.org/10.1111/j.1564-913x.2010.00089.x.

[54] Almeida, R. and M. Faria (2014), “The wage returns to on-the-job training: evidence from matched employer-employee data”, IZA Journal of Labor & Development, Vol. 3/1, https://doi.org/10.1186/2193-9020-3-19.

[35] Arcidiacono, P., J. Kinsler and J. Price (2017), “Productivity Spillovers in Team Production: Evidence from Professional Basketball”, Journal of Labor Economics, Vol. 35/1, pp. 191-225, https://doi.org/10.1086/687529.

[3] Atkinson, A. (2007), “The distribution of earnings in OECD countries”, International Labour Review, Vol. 146/1-2, pp. 41-60, https://doi.org/10.1111/j.1564-913x.2007.00004.x.

[5] Autor, D., L. Katz and M. Kearney (2008), “Trends in U.S. Wage Inequality: Revising the Revisionists”, Review of Economics and Statistics, Vol. 90/2, pp. 300-323, https://doi.org/10.1162/rest.90.2.300.

[4] Autor, D., L. Katz and A. Krueger (1998), “Computing Inequality: Have Computers Changed the Labor Market?”, The Quarterly Journal of Economics, Vol. 113/4, pp. 1169-1213, https://doi.org/10.1162/003355398555874.

[52] Bassanini, A. et al. (2007), “Workplace Training in Europe”, in Education and Training in Europe, Oxford Academic, https://doi.org/10.1093/acprof:oso/9780199210978.001.0001.

[43] Becker, G. (1962), “Investment in Human Capital: A Theoretical Analysis”, Journal of Political Economy, Vol. 70/5, pp. 9-49, http://www.jstor.org/stable/1829103.

[1] Bengali, L., R. Valletta and C. Zhao (2025), “Explaining Stagnation in the College Wage Premium”, Federal Reserve Bank of San Francisco, Working Paper Series, Vol. 2025/01, pp. 01-59, https://doi.org/10.24148/wp2025-01.

[55] Biesebroeck, J. (2008), Wage and Productivity Premiums in Sub-Saharan Africa, National Bureau of Economic Research, Inc.

[70] Bishop, J. (1997), “What we know about employer-provided training: a review of the literature”, in Research in Labor Economics, Emerald Publishing.

[2] Boero, G. et al. (2024), “The college wage premium in the UK: decline and fall?”, Oxford Economic Papers, https://doi.org/10.1093/oep/gpae014.

[9] Bone, M., E. González Ehlinger and F. Stephany (2025), “Skills or degree? The rise of skill-based hiring for AI and green jobs”, Technological Forecasting and Social Change, Vol. 214, p. 124042, https://doi.org/10.1016/j.techfore.2025.124042.

[32] Bonhomme, S. (2021), “Teams: Heterogeneity, Sorting, and Complementarity”, in Chernozhukov V, H. (ed.), Advances in Economics and Econometrics: Twelfth World Congress, Econometric Society Monographs, Cambridge University Press.

[68] Bower, M. et al. (2015), “Design and implementation factors in blended synchronous learning environments: Outcomes from a cross-case analysis”, Computers & Education, Vol. 86, pp. 1-17, https://doi.org/10.1016/j.compedu.2015.03.006.

[30] Bresnahan, T. (2002), “Information Technology, Workplace Organization, and the Demand for Skilled Labor: Firm-Level Evidence”, The Quarterly Journal of Economics, Vol. 117/1, pp. 339-376.

[19] Broecke, S., G. Quintini and M. Vandeweyer (2018), “Wage Inequality and Cognitive Skills: Reopening the Debate”, in Education, Skills, and Technical Change: Implications for Future US GDP Growth, University of Chicago Press, https://www.nber.org/books-and-chapters/education-skills-and-technical-change-implications-future-us-gdp-growth/wage-inequality-and-cognitive-skills-reopening-debate.

[64] Brudevold-Newman, A. and D. Ubfal (2024), “Soft-skills, networking, and workforce entry: Impacts of a training program for recent graduates in Rwanda”, Labour Economics, Vol. 91, p. 102650, https://doi.org/10.1016/j.labeco.2024.102650.

[36] Caplin, A. et al. (2023), Economic Decision-Making Skill Predicts Income in Two Countries, National Bureau of Economic Research, Cambridge, MA, https://doi.org/10.3386/w31674.

[13] Card, D. (1999), “The Causal Effect of Education on Earnings”, in Handbook of Labor Economics, Elsevier, https://doi.org/10.1016/s1573-4463(99)03011-4.

[85] Card, D., J. Kluve and A. Weber (2010), “Active Labour Market Policy Evaluations: A Meta‐Analysis”, The Economic Journal, Vol. 120/548, pp. F452-F477, https://doi.org/10.1111/j.1468-0297.2010.02387.x.

[24] Castex, G. and E. Kogan Dechter (2014), “The Changing Roles of Education and Ability in Wage Determination”, Journal of Labor Economics, Vol. 32/4, pp. 685-710, https://doi.org/10.1086/676018.

[78] Cedefop (2017), “Defining, writing and applying learning outcomes: a European handbook”.

[75] Cedefop (2017), “Global inventory of regional and national qualifications frameworks”.

[39] Cunha, F. et al. (2006), “Chapter 12 Interpreting the Evidence on Life Cycle Skill Formation”, in Handbook of the Economics of Education, Handbook of the Economics of Education Volume 1, Elsevier, https://doi.org/10.1016/s1574-0692(06)01012-9.

[40] Cunha, F., J. Heckman and S. Schennach (2010), “Estimating the Technology of Cognitive and Noncognitive Skill Formation”, Econometrica, Vol. 78/3, pp. 883-931, https://doi.org/10.3982/ecta6551.

[57] De Grip, A. and J. Sauermann (2012), “The Effects of Training on Own and Co‐worker Productivity: Evidence from a Field Experiment”, The Economic Journal, Vol. 122/560, pp. 376-399, https://doi.org/10.1111/j.1468-0297.2012.02500.x.

[31] Deming, D. (2022), “Four Facts about Human Capital”, Journal of Economic Perspectives, Vol. 36/3, pp. 75-102, https://doi.org/10.1257/jep.36.3.75.

[29] Deming, D. (2017), “The Growing Importance of Social Skills in the Labor Market*”, The Quarterly Journal of Economics, Vol. 132/4, pp. 1593-1640, https://doi.org/10.1093/qje/qjx022.

[58] Denzler, S., J. Ruhose and S. Wolter (2025), “Labour market effects of work-related continuous education in Switzerland – evidence from administrative data”, Economics of Education Review, Vol. 107, p. 102683, https://doi.org/10.1016/j.econedurev.2025.102683.

[69] Deschacht, N. and K. Goeman (2015), “The effect of blended learning on course persistence and performance of adult learners: A difference-in-differences analysis”, Computers & Education, Vol. 87, pp. 83-89, https://doi.org/10.1016/j.compedu.2015.03.020.

[34] Devereux, K. (2021), “Returns to Teamwork and Professional Networks: Evidence from Economic Research”, Working Papers 202101, School of Economics, University College Dublin..

[71] Dietz, D. and T. Zwick (2020), “The retention effect of training: Portability, visibility, and credibility<sup>1</sup>”, The International Journal of Human Resource Management, Vol. 33/4, pp. 710-741, https://doi.org/10.1080/09585192.2020.1737835.

[49] Duncan, G. and S. Hoffman (1979), “On-The-Job Training and Earnings Differences by Race and Sex”, The Review of Economics and Statistics, Vol. 61/4, p. 594, https://doi.org/10.2307/1935790.

[25] Edin, P. et al. (2022), “The Rising Return to Noncognitive Skill”, American Economic Journal: Applied Economics, Vol. 14/2, pp. 78-100, https://doi.org/10.1257/app.20190199.

[82] Enterprise Ireland (2026), Mentor Network.

[76] European Training Foundation (2017), “Getting organised for better qualifications: A toolkit”.

[61] Fialho, P., G. Quintini and M. Vandeweyer (2019), “Returns to different forms of job related training: Factoring in informal learning”, OECD Social, Employment and Migration Working Papers, No. 231, OECD Publishing, Paris, https://doi.org/10.1787/b21807e9-en.

[14] Griliches, Z. (1977), “Estimating the Returns to Schooling: Some Econometric Problems”, Econometrica, Vol. 45/1, p. 1, https://doi.org/10.2307/1913285.

[63] Groh, M. et al. (2016), “The impact of soft skills training on female youth employment: evidence from a randomized experiment in Jordan”, IZA Journal of Labor & Development, Vol. 5/1, https://doi.org/10.1186/s40175-016-0055-9.

[42] Grundke, R. et al. (2018), “Which skills for the digital era?: Returns to skills analysis”, OECD Science, Technology and Industry Working Papers, No. 2018/09, OECD Publishing, Paris, https://doi.org/10.1787/9a9479b5-en.

[17] Hampf, F., S. Wiederhold and L. Woessmann (2017), “Skills, earnings, and employment: exploring causality in the estimation of returns to skills”, Large-scale Assessments in Education, Vol. 5/1, https://doi.org/10.1186/s40536-017-0045-7.

[18] Hanushek, E. et al. (2015), “Returns to skills around the world: Evidence from PIAAC”, European Economic Review, Vol. 73, pp. 103-130, https://doi.org/10.1016/j.euroecorev.2014.10.006.

[48] Hashimoto, M. (1981), “Firm-Specific Human Capital as a Shared Investment”, The American Economic Review, Vol. 71/3, pp. 475-82.

[23] Heisig, J. and H. Solga (2017), “How returns to skills depend on formal qualifications: Evidence from PIAAC”, OECD Education Working Papers, No. 163, OECD Publishing, Paris, https://doi.org/10.1787/332a43d7-en.

[16] Ingram, B. and G. Neumann (2006), “The returns to skill”, Labour Economics, Vol. 13/1, pp. 35-59, https://doi.org/10.1016/j.labeco.2004.04.005.

[26] Izadi, R. and J. Tuhkuri (2024), “Evolving Returns to Personality”, Journal of Labor Economics, https://doi.org/10.1086/733226.

[15] Juhn, C., K. Murphy and B. Pierce (1993), “Wage Inequality and the Rise in Returns to Skill”, Journal of Political Economy.

[6] Katz, L. and K. Murphy (1992), “Changes in Relative Wages, 1963-1987: Supply and Demand Factors”, The Quarterly Journal of Economics, Vol. 107/1, pp. 35-78, https://doi.org/10.2307/2118323.

[7] Lane, M., M. Williams and S. Broecke (2023), “The impact of AI on the workplace: Main findings from the OECD AI surveys of employers and workers”, OECD Social, Employment and Migration Working Papers, No. 288, OECD Publishing, Paris, https://doi.org/10.1787/ea0a0fe1-en.

[47] Lazear, E. (2009), “Firm‐Specific Human Capital: A Skill‐Weights Approach”, Journal of Political Economy, Vol. 117/5, pp. 914-940, https://doi.org/10.1086/648671.

[59] Leuven, E. and H. Oosterbeek (2008), “An alternative approach to estimate the wage returns to private‐sector training”, Journal of Applied Econometrics, Vol. 23/4, pp. 423-434, https://doi.org/10.1002/jae.1005.

[77] Martin, M. and A. Godonoga (2020), “Policies for Flexible Learning Pathways in Higher Education: Taking Stock of Good Practices Internationally”.

[8] Milanez, A. (2023), “The impact of AI on the workplace: Evidence from OECD case studies of AI implementation”, OECD Social, Employment and Migration Working Papers, No. 289, OECD Publishing, Paris, https://doi.org/10.1787/2247ce58-en.

[80] Ministère du Travail et des solidarités (2024), La gestion prévisionnelle de l’emploi et des compétences (GPEC), https://travail-emploi.gouv.fr/la-gestion-previsionnelle-de-lemploi-et-des-competences-gpec#anchor-navigation-504 (accessed on 16 February 2026).

[73] OECD (2026), A Skills-First Labour Market, Getting Skills Right, OECD Publishing, Paris, https://doi.org/10.1787/2e1b85f0-en.

[38] OECD (2026), How Workers Use, or Don’t Use, their Skills in the Workplace, Getting Skills Right, OECD Publishing, Paris, https://doi.org/10.1787/0e7c6dc9-en.

[86] OECD (2026), OECD Data Explorer: Earnings: DSD_EARNINGS (DEC_I), https://data-explorer.oecd.org/s/44g (accessed on 13 February 2026).

[28] OECD (2025), Education at a Glance 2025: OECD Indicators, OECD Publishing, Paris, https://doi.org/10.1787/1c0d9c79-en.

[11] OECD (2025), Empowering the Workforce in the Context of a Skills-First Approach, OECD Skills Studies, OECD Publishing, Paris, https://doi.org/10.1787/345b6528-en.

[21] OECD (2025), OECD Skills Outlook 2025: Building the Skills of the 21st Century for All, OECD Publishing, Paris, https://doi.org/10.1787/26163cd3-en.

[22] OECD (2025), Survey of Adult Skills 2023 Data Analysis Manual, OECD Publishing, Paris, https://doi.org/10.1787/25a87a9d-en.

[12] OECD (2025), Trends in Adult Learning: New Data from the 2023 Survey of Adult Skills, Getting Skills Right, OECD Publishing, Paris, https://doi.org/10.1787/ec0624a6-en.

[10] OECD (2024), Bridging Talent Shortages in Tech: Skills-first Hiring, Micro-credentials and Inclusive Outreach, Getting Skills Right, OECD Publishing, Paris, https://doi.org/10.1787/f35da44f-en.

[27] OECD (2024), Do Adults Have the Skills They Need to Thrive in a Changing World?: Survey of Adult Skills 2023, OECD Skills Studies, OECD Publishing, Paris, https://doi.org/10.1787/b263dc5d-en.

[84] OECD (2024), Survey of Adult Skills – Reader’s Companion: 2023, OECD Skills Studies, OECD Publishing, Paris, https://doi.org/10.1787/3639d1e2-en.

[67] OECD (2023), “Flexible adult learning provision: What it is, why it matters, and how to make it work”, OECD, Paris.

[74] OECD (2023), “Recognition of prior learning: A practical guide for policy makers”, OECD, Paris.

[79] OECD (2021), Incentives for SMEs to Invest in Skills: Lessons from European Good Practices, Getting Skills Right, OECD Publishing, Paris, https://doi.org/10.1787/1eb16dc7-en.

[83] OECD (2021), Training in Enterprises: New Evidence from 100 Case Studies, Getting Skills Right, OECD Publishing, Paris, https://doi.org/10.1787/7d63d210-en.

[20] OECD (2015), OECD Employment Outlook 2015, OECD Publishing, Paris, https://doi.org/10.1787/empl_outlook-2015-en.

[72] OECD/SUSS-IAL (2025), Skills-First Readiness and Adoption Index: Methodological Note, OECD, Paris.

[60] Osman, A. and J. Speer (2025), “Are Soft Skills Enough? Experimental Evidence on Skill Complementarity for College Graduates”, ILR Review, Vol. 78/4, pp. 692-711, https://doi.org/10.1177/00197939251316849.

[51] Parent, D. (2003), “Employer-supported training in Canada and its impact on mobility and wages”, Empirical Economics, Vol. 28/3, pp. 431-459, https://doi.org/10.1007/s001810200140.

[37] Quintini, G. (2014), “Skills at Work: How Skills and their Use Matter in the Labour Market”, OECD Social, Employment and Migration Working Papers, No. 158, OECD Publishing, Paris, https://doi.org/10.1787/5jz44fdfjm7j-en.

[66] Schlosser, A. and Y. Shanan (2025), “Fostering Soft Skills in Active Labor Market Programs”, Journal of Human Resources, pp. 0723-13013R1, https://doi.org/10.3368/jhr.0723-13013r1.

[81] Skillnet Ireland (2026), Building Competitiveness Through Skills., https://www.skillnetireland.ie/ (accessed on 16 February 2026).

[46] Stevens, M. (1994), “A Theoretical Model of On-the-Job Training with Imperfect Competition”, Oxford Economic Papers, Vol. 46/4, pp. 537-62, http://www.jstor.org/stable/2663510.

[62] Topel, R. and M. Ward (1992), “Job Mobility and the Careers of Young Men”, The Quarterly Journal of Economics, Vol. 107/2, pp. 439-479, https://doi.org/10.2307/2118478.

[50] Veum, J. (1995), “Sources of Training and Their Impact on Wages”, Industrial and Labor Relations Review, Vol. 48/4, p. 812, https://doi.org/10.2307/2524358.

[33] Weidmann, B. and D. Deming (2021), “Team Players: How Social Skills Improve Team Performance”, Econometrica, Vol. 89/6, pp. 2637-2657, https://doi.org/10.3982/ecta18461.

[41] Weinberger, C. (2014), “The Increasing Complementarity between Cognitive and Social Skills”, The Review of Economics and Statistics, Vol. 96/5, pp. 849-861, https://doi.org/10.1162/rest_a_00449.

Annex 4.A. Methodological annex

Copy link to Annex 4.A. Methodological annex

Discussion of potential endogeneity issues and strategies adopted to address them

Copy link to Discussion of potential endogeneity issues and strategies adopted to address them

A number of potential endogeneity issues affect the estimation of the effect of skills on employment status and wages. The table below lists these concerns and the different strategies adopted in this chapter to address them.

Analysis	Type of endogeneity	Description	Strategy to mitigate the problem
Returns to skills and to training – employment status and wages	Measurement error	Because skills are inherently difficult to measure, imprecise measurement is likely to generate attenuation bias, biasing estimates of the relationship between skills and employment outcomes towards zero.	The fact that the Survey of Adult Skills measures skills in several domains diminishes the risk of measurement error. The main analyses use numeracy proficiency, but robustness checks were conducted using literacy proficiency.
Returns to skills and training – employment status and wages	Reverse causation	This happens if employment patterns directly affect test scores over the lifecycle. The literature has shown that employment breaks could lead to skills depreciation. Besides, some jobs may lead to skills reinforcement (better use of skills and/ or participation into training).	Main regressions with wages as a dependent variable control for occupational status. Robustness checks were conducted adding controls for employment breaks, skills use and training participation.
Returns to skills and training – employment status and wages	Omitted variable bias	Individual characteristics such as family background, health, or personality traits could influence earnings and skills and lead to an overestimation of the link between skills and wages if not included in the regression.	Main regressions control for family background (parental education). Robustness checks were conducted adding controls for health status and personality traits.
Returns to skills and training – wages	Sample selection	Wages observed for employed workers only. If workers whose skills are obsolete or that do not participate in training become unemployed, the relationship between skills or training and wages would be underestimated.	If anything, the relationship between skills or training participation is underestimated and the coefficients can be interpreted as a lower bound of the true effect.
Returns to training	Sample selection	There might be selection into training based on unobserved characteristics such as ability, motivation, or employer’s practices (e.g. if they select into training individuals they would like to retain or promote or else if they select for training newly hired workers with insufficient skills).	Regressions control for skills (literacy). One robustness check was conducted by restricting the “control” group to individuals that wanted to undertake training but did not for an unexpected reason as in Leuven and Oosterbeek (2008[59]).

Yet, it is important to note that these solutions suffer from their own limitations. The list of additional covariates is limited and may not allow to control for all the sources of endogeneity. The method consisting in restricting the control group to individuals that wanted to undertake training but did not for an unexpected reason cannot be used to estimate interaction effects (i.e. association between different training characteristics – topics, training flexibility, employer-sponsored training, training usefulness and certificated training – and wages) because it suffers from low power. Moreover, it would not neutralise all possible sources of selection bias if the unexpected reason inducing workers to give up training opportunities affects also their earnings potential, for example if they drop out from employment for the same reason. Hence the relationships presented in this chapter should not be interpreted as causal estimates.

Annex 4.B. Robustness checks

Copy link to Annex 4.B. Robustness checks

	(1) Baseline	(2) Break in employment	(3) Skill use	(4) Non-formal training	(5) All controls
Skills and education
Numeracy (one‑standard-deviation)	0.0678***	0.0677***	0.0670***	0.0660***	0.0652***
	(0.00279)	(0.00278)	(0.00279)	(0.00279)	(0.00278)
Years of education (one‑standard-deviation)	0.0867***	0.0863***	0.0847***	0.0847***	0.0824***
	(0.00331)	(0.00329)	(0.00330)	(0.00331)	(0.00329)

Additional controls
Break in employment		-0.0696***			-0.0683***
		(0.00545)			(0.00542)
High co‑operation at work			0.00769**		0.00482
			(0.00379)		(0.00377)
High task discretion at work			0.0640***		0.0608***
			(0.00421)		(0.00420)
Non-formal training participation				0.0553***	0.0534***
				(0.00381)	(0.00379)

N	62 973	62 952	62 852	62 973	62 831
R-squared	0.605	0.606	0.608	0.607	0.611

Note: The table presents four specifications including various robustness checks relative to the baseline model (1). Model 2 controls for adults who reported a break in employment of at least three months over the past five years. Model 3 controls for the extent to which adults exercise task discretion and co‑operation at work. Model 4 controls for those who participated in non-formal adult learning activities. Model 5 includes all controls. In addition to the variables presented, all estimates control for worker characteristics (work experience, sex, age, marital status, whether one has children, parental education, and whether one is a native speaker), job and firm characteristics (contract type, firm size and growth, private sector), and includes fixed effects for country, occupational category, and industrial sector. Standard errors in parentheses. * p<.1; ** p<.05; *** p<.01.

	(1) Baseline specification	(2) Health	(3) Personality	(4) All controls
Skills and education
Numeracy (one‑standard-deviation)	0.0678***	0.0662***	0.0688***	0.0675***
	(0.00279)	(0.00277)	(0.00266)	(0.00266)
Years of education (one‑standard-deviation)	0.0867***	0.0838***	0.0859***	0.0842***
	(0.00331)	(0.00330)	(0.00303)	(0.00303)
Additional controls: Self-reported health
Excellent		0.0586***		0.0407***
		(0.00592)		(0.00582)
Very good		0.0334***		0.0269***
		(0.00429)		(0.00441)
Fair		-0.0409***		-0.0328***
		(0.00558)		(0.00590)
Poor		-0.0313**		-0.0175
		(0.0136)		(0.0147)

Additional controls: Personality traits
Agreeableness			-0.00640***	-0.00774***
			(0.00198)	(0.00198)
Conscientiousness			0.0104***	0.00934***
			(0.00212)	(0.00212)
Emotional stability			0.0192***	0.0138***
			(0.00206)	(0.00212)
Extraversion			0.0230***	0.0206***
			(0.00202)	(0.00203)
Open-mindedness			-0.00231	-0.00217
			(0.00197)	(0.00197)

N	62 973	62 950	58 018	58 003
R-squared	0.605	0.607	0.614	0.616

Note: The table presents three specifications including various robustness checks relative to the baseline model (1). Model 2 controls for adults’ self-reported health (ref. category: Good). Model 3 controls for individuals’ personality traits. Model 4 includes all controls. In addition to the variables presented, all estimates control for worker characteristics (work experience, sex, age, marital status, whether one has children, parental education, and whether one is a native speaker), job and firm characteristics (contract type, firm size and growth, private sector), and includes fixed effects for country, occupational category, and industrial sector. Standard errors in parentheses. * p<.1; ** p<.05; *** p<.01.

	(1) Participation relative to all non-participants (baseline)	(2) Relative to those facing unexpected barriers to participation	(3) Relative to all other non-participants
Non-formal job-related training within past year	0.0550***	0.0584*	0.0547***
	(0.00383)	(0.0349)	(0.00384)
Other adult learning controls
Formal training within past year	-0.0406***	-0.0388***	-0.0409***
	(0.00724)	(0.00914)	(0.00723)
Informal learning at least once a week	0.0137***	0.00812	0.0137***
	(0.00403)	(0.00594)	(0.00404)
Skills and education
Numeracy (one‑standard-deviation)	0.0668***	0.0782***	0.0668***
	(0.00280)	(0.00390)	(0.00280)
Years of education (one‑standard-deviation)	0.0844***	0.110***	0.0845***
	(0.00332)	(0.00449)	(0.00333)
N	62 669	29 782	62 442
R-squared	0.607	0.592	0.608

Note: Each model presented shows the estimated change in wages associated with participation in non-formal job-related adult learning. Model 1 shows the effect relative to all adults who did not participate; Model 2 presents results relative only to those adults who reported that they wished to participate in non-formal training, but were unable to do so due to unexpected barriers; Model 3 presents results relative to all other non-participants (i.e. those who did not report any unexpected barriers). In addition to the variables presented, all estimates control for worker characteristics (work experience, sex, age, marital status, whether one has children, parental education, and whether one is a native speaker), job and firm characteristics (contract type, firm size and growth, private sector), and includes fixed effects for country, occupational category, and industrial sector. Standard errors in parentheses. * p<.1; ** p<.05; *** p<.01.

Testing the null hypothesis that the estimated penalty for workers who wished to participate in non-formal training but were unable to do so for unexpected reasons is equal to the estimated penalty for workers who did not participate in non-formal training for other reasons, both relative to workers who completed non-formal training, leads to a p-value of 0.9127. This means that we cannot reject the null hypothesis that the two penalties are equal and suggests that the estimated association of participating in non-formal training with wages when using as a control group adults who wanted to participate but could not due to unexpected reasons is qualitatively and quantitatively similar to the associations using the full sample.

Notes

Copy link to Notes

← 1. Respondents to the Survey of Adult Skills are scored according to their assessed performance in three skill domains. For numeracy (and literacy), there are six levels – from Below Level 1 to Level 5 – and the gap between each level is 50 points. See OECD (2024[84]) for more information.

← 2. Measured as the ratio of the 90th to the 10th percentile of the wage distribution.

← 3. Using data on earnings from the Survey of Adult Skills, the correlation between the estimated association of years of education and earnings at the country level with countries’ inequality measure is 0.72. This correlation is 0.46 when using the association between numeracy and earnings. This correlation is about 0.50 for both associations when using data from OECD (2026[86]), which excludes information for non-OECD Members, as Singapore and Croatia.

← 4. Heterogeneity in the associations of wages with literacy across occupations is less pronounced than with numeracy.

← 5. Wage inequality is measured as the ratio of wages at the 90th and 10th percentiles using wages information from the Survey of Adult Skills.

← 6. While the Survey of Adult Skills allows identification of whether an individual changed jobs and whether they completed training within the past year, it does not provide information on the temporal order of these events.

← 7. The stronger negative association between wages and formal training participation for migrants is statistically significant only at the 10% level.

← 8. In the Cycle 1 of the Survey of Adult Skills, conducted between 2011 and 2018, respondents were asked sequentially about their participation in each type of training activity. In Cycle 2 (2022-2023), however, they reported all training activities undertaken in the past 12 months in a single question. Although the underlying content of the questions remained identical across cycles, this change in administration may have influenced how some respondents reported their participation. For further details, see Box 1.3 in OECD (2025[12]).

← 9. See for instance Card, Kluve and Weber (2010[85]) for a meta‑analysis of ALMP evaluations.

Featured topics

Agriculture and fisheries

Climate change

Development

Digital

Economy

Education and skills

Employment

Environment

Finance and investment

Governance

Health

Industry, business and entrepreneurship

Regional, rural and urban development

Science, technology and innovation

Society

Taxation

Trade

Energy

Nuclear energy

Transport

Featured topics

Agriculture and fisheries

Climate change

Development

Digital

Economy

Education and skills

Employment

Environment

Finance and investment

Governance

Health

Industry, business and entrepreneurship

Regional, rural and urban development

Science, technology and innovation

Society

Taxation

Trade

Energy

Nuclear energy

Transport

Countries A - C

Countries D - I

Countries J - M

Countries N - R

Countries S - T

Countries U - Z

Regional and global engagement

Countries

Countries A - C

Countries D - I

Countries J - M

Countries N - R

Countries S - T

Countries U - Z

Regional and global engagement

Publications

Publications

Featured publications

Data

Data

Featured data

News & events

News & events

Featured events

About OECD

About

Engage with us

Work with us

Featured topics

Agriculture and fisheries

Climate change

Development

Digital

Economy

Education and skills

Employment

Environment

Finance and investment