Overview of returns to soft skills

What is known about the wage returns to soft skills, as observable and directly applicable behaviour important for high job performance (e.g. overcoming barriers in communications, building relationships with customers, work planning, cooperation with others, etc.), and closely related non-cognitive traits? This section provides a brief and highly selective overview of the empirical literature, focused only on the skills corresponding most closely to those considered by this article (see Note 1). For more elaborated overview of wage returns to soft skills, see Reference BalcarBalcar (2014).

Reference Borghans, ter Weel and WeinbergBorghans et al. (2006) focused on wage returns to a whole range of soft skills, whose presence was approximated by the importance of ‘people tasks’ for job performance. They also examined whether an individual’s sociability (measured during childhood and early adulthood) is associated with a wage premium in jobs where people tasks are important. They found that individuals working in jobs where people tasks are important face lower wages. The wage penalty connected with a standard deviation increase in the importance of people tasks was estimated at circa 5% in US and 4%–9% in the UK. Most sociability variables, though not all, indicated a very low wage premium in those jobs (circa 1%).

Reference Bacolod and BlumBacolod and Blum (2008) estimated wage returns to people skills, cognitive skills and motor skills. They found that wage returns to people skills nearly doubled during the years 1968–1990 (one standard deviation increase in people skills was associated with a 3.2% wage increase in 1968 and 6.0% increase in 1990), while returns to cognitive skills increased ‘only’ by 60% and returns to motor skills decreased by −50%. Detailed examination revealed that people skills have no value themselves, but only as a complement to other skills. The growth of wage returns to people skills was, in fact, caused by increasing returns to a combination of cognitive (or motor) and people skills during the years 1968–1990. These findings correspond to those of Reference WeinbergerWeinberger (2011) who identified a significant growth of employment only in jobs requiring both cognitive and people skills (in this case leadership).

Reference Borghans, ter Weel and WeinbergBorghans et al. (2008) applied a novel approach to the examination of wage returns to soft skills by focusing on an interpersonal interaction approximated by a trade-off between directness, which facilitates accurate communication, and care, which establishes a cooperative environment. They found a significant increase in returns to directness during the examined period, as a one standard deviation increase in directness relative to care raised wages by 9.6% in 1997 and 10.8% in 2001 in the UK, and by 3.8% in 1979 and 10.2% in 1998 in Germany. The wage premium for directness was higher in occupations where it is more important.

Reference Kuhn and WeinbergerKuhn and Weinberger (2005) focused on wage returns to leadership skills in the case of White US males. These skills were measured by the observable leadership activities of individual during study at high school and by their self-assessment. These leadership variables were regressed on annual and hourly earnings 9–13years later. Estimates showed an earnings premium for men who had led sporting teams or social organisations at high school and who self-assessed as having leadership skills. The authors also found that such males had a higher probability of subsequently occupying managerial roles.

The reviewed articles (Reference BalcarBalcar, 2014) show that changes in employers’ requirements for soft skills in employees, accompanied by corresponding development of their wage returns, together with changes in distribution of these skills between genders, led to a significant reduction of the gender wage gap. This conclusion is consistent with the results of Reference Bacolod and BlumBacolod and Blum (2008), Reference Black and Spitz-OenerBlack and Spitz-Oener (2007) and Reference Borghans, ter Weel and WeinbergBorghans et al. (2006).

Measurement of soft skills

It is very difficult to measure soft skills (e.g. cooperation, consumer orientation, leadership, etc.) as there is no test, according to author’s knowledge, which objectively assesses the individual’s workplace behaviour as it is an interactive process depending on context. Observation, which is a suitable method for this task, is too expensive for application on big samples. Empirical literature focused on wage returns to soft skills uses two different measurement methodologies (Reference BalcarBalcar, 2014). The direct one is based on questioning individuals on their past behaviour approximating given skills (e.g. Reference Kuhn and WeinbergerKuhn and Weinberger, 2005; Reference WeinbergerWeinberger, 2011); the indirect one approximates an individual’s soft skills by job tasks (e.g. Reference Bacolod and BlumBacolod and Blum, 2008; Reference Black and Spitz-OenerBlack and Spitz-Oener, 2007; Reference Borghans, ter Weel and WeinbergBorghans et al., 2006, Reference Borghans, ter Weel and Weinberg2008), which are identified by trained experts or workers themselves. ^{Footnote 2} The indirect approach raises a question whether it measures soft skills reliably. The approximation of individual’s skills by job tasks is supported by empirical studies finding a statistically significant match of soft skills with job tasks (Reference Borghans, ter Weel and WeinbergBorghans et al., 2006, Reference Borghans, ter Weel and Weinberg2008; Reference WeinbergerWeinberger, 2011) or at least job type (Reference Kuhn and WeinbergerKuhn and Weinberger (2005) match leadership skills with employment in managerial jobs). Such attempts at approximation correspond to the fact that soft skills represent a decisive criterion for hiring, while qualifications are used for a preselection of job applicants (Reference Balcar, Homolová and KarásekBalcar et al., 2011). The result is that only individuals with soft skills at the required level (or as close to it as possible) are hired, as their subsequent development is slow and expensive (see ‘Introduction’). This also solves the potential problem with reverse causality between soft skills and performed job. Evidence in this paragraph suggests that an indirect approach to soft skills measurement through job characteristics provides relevant data.

The indirect approach to the approximation of soft skills is used also in this article, but it goes further than the reviewed papers by shifting focus from the necessity of particular soft skills for job performance (usually approximated by tasks performed) to levels of their mastery. This focus enables more accurate estimation of returns to soft skills, avoiding the problem that skills at a low or average level of mastery may still be highly important for job performance. Results of the model focusing on the importance of soft skills would be then significantly different in comparison with a model focusing on their level.

Wage returns to hard and soft skills

The original version of Reference MincerMincer’s (1974) wage equation, explaining differences in individuals’ wages by knowledge and skills acquired through education and work experience, was used as a base model in the analysis. It was hypothesised that its relevancy and predictive power would be increased by the inclusion of a soft skills variable (approximated by job requirements for a soft skills level), enabling both the estimation of the returns to particular soft skills and the elimination of their influence on education and work experience variables, whose regression coefficients would thus provide information on returns only to hard skills. Subsequent model modifications would shed light on the importance of soft and hard skills for wage determination, the role of the education system in soft skills accumulation and gender differences in returns to soft skills (see equation (1); each parenthesis represents significant model modification). ^{Footnote 8} All models in the article use gross monthly wage as the dependent variable. ^{Footnote 9} It can be noted that the results, to the author’s knowledge, provide the first estimation of wage returns to soft skills for the Czech Republic

(1) $\ln w=f\left[\begin{array}{c}\left(\mathrm{education},\mathrm{work}\mathrm{experience},\mathrm{location}\right),\left(\mathrm{soft}\mathrm{skills}\right),\left(\mathrm{gender}\right),\\ \left(\mathrm{physical}\mathrm{characteristics},\mathrm{cognitive}\mathrm{and}\mathrm{noncognitive}\mathrm{traits},\mathrm{family},\mathrm{background}\right),\\ \left(\mathrm{job},\mathrm{employer}\right)\end{array}\right]$

It proved unrealistic to include as regressors all 15 soft skills defined by the NSP and to estimate wage returns to each of them because of high mutual correlation among these skills (the Pearson correlation coefficient reached values of 0.33–0.49 only in six cases, 0.5–0.69 in 27 cases and 0.7–0.86 in 72 cases). Factor analysis revealed that the set of soft skills could be represented by communication skills, which would explain 74.26% of the variability in the original soft skills variables, but the mean level of all 15 soft skills provided even better results (slightly over 75%). Therefore, the latter was chosen as the variable approximating individual’s soft skills in the wage models. It is noteworthy that Cronbach’s alpha (at the value of 0.959) suggests that soft skills defined by the NSP represent a reliable measurement of the general level of individuals’ soft skills.

An estimation of Mincer’s model without soft skills variable (Model 1 in Table 1) provided standard and expected results. An additional year of schooling ^{Footnote 10} was associated with a wage premium at the level of 6.51%. Returns to different educational levels, compared to primary education, International Standard Classification of Education 2A/European Qualification Framework 2 (ISCED 2A/EQF 2), revealed that students of secondary general or technical schools, ISCED 3A/EQF 4, gained double returns to their educational level (24.80%) compared with their friends at secondary vocational schools, ISCED 3C/EQF 3 (12.61%), and they could double these returns again by reaching a masters or doctoral degree, ISCED 5A and 6/EQF 7 and 8 (55.14%); see Model 9 in Table 3. ^{Footnote 11} Tenure was also a statistically significant wage predictor, whose returns reached the value of 2.19% per year. A year of other work experience was connected approximately with 1.47% wage premium, which confirms a higher relevance of experience acquired in current employment.

Education and work experience develop both hard and soft skills. Inclusion of a soft skills variable (Model 2 in Table 2), which approximates the level of their development, enabled separation of returns to these different kinds of skills (therefore, the variables education, tenure and other work experience in all models except Models 1 and 9 represent only hard skills). The returns to one standard deviation of soft skills (i.e. 0.8459 of a behavioural level at 6-point scale defined by the NSP) are equal to 8.51%. The inclusion of soft skills significantly influenced returns to education, which suggests that the educational system plays an important role in the accumulation of not only hard skills but also soft skills. The return to a year of schooling decreased from 6.51% (returns to both hard and soft skills) to 4.23% (returns only to hard skills), which means a decrease of regression coefficient by −35.02%. An analysis of changes in regression coefficients for different levels of education brings interesting results (compare the results of Models 9 and 10 in Table 3). The relative decrease in wage returns was similar for all educational levels (decrease of regression coefficients from −30.03% to −34.07%), except secondary vocational schools (regression coefficient decreased by −23.47%). These findings suggest that the proportion of soft skills to hard skills development is relatively stable across the whole Czech education system, except secondary vocational schools, which are more focused on development of hard skills. Inclusion of a soft skills variable had only a minor impact on returns to tenure, that is, the change of returns from 2.19% to 1.98% (regression coefficient decreased by −9.59%), and minimal impact on returns to other work experience (the change of regression coefficient was smaller than 5.00%). These results support the existence of specific soft skills, whose value is bounded only or mainly with the place of their accumulation (e.g. a firm-specific form of communication with clients), although the theoretical literature assumes all soft skills to be generic and thus transferable. They also suggest that work experience for both current and previous employers leads mainly to the accumulation of hard skills.

Table 2. Returns to hard and soft skills, Czech Republic, 2012.

Source: Author.

Robust standard errors in parentheses.

a Health limitation of work performance, Difference between individual’s height and average gender height, Body mass index (BMI) categories.

b Grades from math at age 15, Strong need to excel and be better than others, Persistence in following difficult goals, Self-esteem, Locus of control, Highest life priority, Feeling of personal responsibility for ensuring an adequate income, Feeling of personal responsibility for ensuring everyday housework and taking care of children, Loss of individual’s income would lead to significant decrease of living standard, Preference of job security, Preference of job flexibility, Preference of individual’s self-fulfilment, Preference of less demanding and stressful work, Preference of good interpersonal relations at the workplace.

c Marital status, Number of children in 5 age categories.

d Number of siblings, Mother tongue.

e Occupation according to 1-digit International Standard Classification of Occupations (ISCO classification), Prevailing economic activity according to Statistical Classification of Economic Activities (NACE classification), Workload, Difference in number of hours really devoted to a work and official workload, Absenteeism, Field of education and job match, Work performance dependence on co-workers, Not strictly specified working process and freedom to create/try new working processes, Rate of subjectivity in wage-system, Relation with boss, Way of getting the job.

f Number of employees, Ownership, Natural person dummy, Age of firm/institution.

g Region according to Nomenclature of Territorial Units for Statistics (NUTS 3), Residence town size.

*** p<0.01; **p<0.05; *p<0.1.

Table 3. Returns to interactions of hard and soft skills, Czech Republic, 2012.

Source: Author.

ISCED: International Standard Classification of Education; EQF: European Qualification Framework.

Robust standard errors in parentheses.

Work experience comprehends Tenure, Tenure squared, Other work experience, Other work experience squared.

a–g See notes a–g under Table 2 for a specification of other control variables.

*** p<0.01; **p<0.05; *p<0.1.

A comparison of the standardised beta coefficients of hard skills acquired through a year of schooling (0.2388) and soft skills (0.2301) in Model 2 suggests that the importance of each as a wage predictor is nearly the same. This inference can be illustrated by the wage returns to one standard deviation of each kind of skills, which were 8.84% for hard skills and 8.51% for soft skills. Also, the contribution of these skills to explaining variance in gross monthly wage can be mentioned. The coefficient of determination (R²) of Model 2 without Location variables (not shown here) reached the level of 0.2282. The Owen decomposition of R² revealed that the shares explained by hard skills acquired by schooling (40.10%) and by soft skills (39.22%) were equal. The importance of soft skills for explaining variance in the gross monthly wage can be illustrated also by a simple comparison of the R² of models with (0.251) and without (0.216) the soft skills variable.

The importance of both hard and soft skills for wage determination raises the question of gender differences in their returns as there is evidence that returns to education (i.e. hard and soft skills together) differ significantly between men and women in the Czech Republic (Reference ChaseChase, 1997; Reference FlanaganFlanagan, 1998). Model 3 in Table 2 shows the estimation of Model 2 extended by a gender variable and its interactions with years of schooling (approximating hard skills) and soft skills. It revealed that there was a statistically significant gender difference in returns to hard skills at the level of 2.68% for an additional year of schooling for women (it doubled their returns to hard skills in comparison with men). Returns to soft skills, on the other hand, were 1.46% per one standard deviation higher in the case of men, but the difference was statistically insignificant. This means that returns to soft skills are gender neutral. This evidence suggests that the gender differences in returns to education reported by Reference FlanaganFlanagan (1998) and Reference ChaseChase (1997) were connected with differences in returns to hard skills, not soft skills. ^{Footnote 12}

It should be noted that regression coefficients of hard and soft skills, as well as their gender differences, can be significantly affected by (a) omitted personal variables relevant to their accumulation and (b) segregation of men and women into different occupations and economic sectors. The problem of segregation can be illustrated by the fact that re-estimation of Model 3 without employees working in managerial occupations, ISCO 1 (not shown here), led to a decrease in the coefficient of gender and soft skills interaction by −41.04% as returns to soft skills are substantially higher in these occupations and they are performed more often by men than women (4.25% of men and 2.57% of women in the sample). Also, re-estimation of Model 2 extended by prevailing economic activity variables according to Statistical Classification of Economic Activities (NACE classification), (Eurostat, 2016) and their interactions with soft skills (not shown here) confirmed differences in returns to soft skills across economic sectors (compared with Manufacturing with 8.39% returns to one standard deviation of soft skills). Statistically significant sector premia for soft skills were identified in accommodation and food service activities, NACE I (16.95% per one standard deviation of soft skills); financial and insurance activities, NACE K (9.07%); professional, scientific and technical activities, NACE M (12.27%, statistically significant at 0.1 level); and arts, entertainment and recreation, NACE R (9.90%). On the other hand, agriculture, forestry and fishing, NACE A, and transportation and storage, NACE H, were identified as sectors with significantly lower returns to soft skills (compared to Manufacturing, NACE C). ^{Footnote 13} New model specifications, augmented by (a) personal characteristics and (b) job and employers characteristics, were estimated in order to overcome the described deficiencies of Models 1–3. However, this changed their focus from an assessment of general productivity of hard and soft skills (e.g. useful for a decision on development of soft skills in the frame of education system) to productivity in specific conditions (e.g. useful for decision on further development of these skills of already employed individuals).

Model 4 in Table 1 represents a wage model with detailed information on individual cognitive and non-cognitive traits, preferences and other characteristics, which can influence wages both directly and indirectly through their impact on acquired level of hard and soft skills. The magnitude of the indirect effect will be reflected by a decrease in the corresponding regression coefficients. Inclusion of personal, family and background characteristics (compared to the Model 2) led to a decrease of returns to a year of schooling approximating hard skills acquisition from 4.23% to 2.93% (a decrease in the regression coefficient by −30.73%) and in returns to one standard deviation of soft skills from 8.51% to 6.70% (a decrease in the regression coefficient by −21.27%), suggesting the significant role of the newly controlled variables on hard and soft skills development. However, introduction of the new variables (compare results of Models 5 and 3) had no effect on gender differences in returns to a year of schooling, but led to decreased gender differences in returns to soft skills (the coefficient remained statistically insignificant). The comparison of standardised beta coefficients in Model 4 suggested that development of soft skills (0.1811) can be slightly more profitable than development of hard skills through schooling (0.1655). It can be also noted that the predictive power of Model 2 (R²=0.251) increased substantially by adding variables on personal, family and background characteristics (R² of Model 4 reached the value of 0.429).

Levels of education, soft skills, cognitive and non-cognitive abilities, preferences and other personal characteristics usually have a crucial effect on both choice of a desirable job made by an individual and selection of a suitable worker made by an employer. There may be some (and maybe substantial) indirect effect of these factors on wages through their influence on job characteristics. This raised the question of whether hard and soft skills have any direct effect on wages when job and employer characteristics are controlled, or whether they are entirely sterile and their effect on wages is intermediated only by acquired job. This question has very practical consequences as it shows whether investment in the further accumulation of hard and soft skills of employed individuals has any influence on productivity. Some positive effects are expected because, for example, more skilled managers (ISCO 1) in Manufacturing (NACE C) can lead their teams more efficiently or can perform more demanding managerial positions, which both should lead to higher wages. Model 6, which augmented Model 4 with job and employer variables, was used for a verification of this assumption.

The inclusion of job and employer characteristics (Model 6 in Table 1) decreased both returns to hard skills acquired by a year of schooling to the level of 2.48% (a decrease of regression coefficient by −15.36% in comparison with Model 4) and returns to one standard deviation of soft skills to the level of 3.37% (a decrease of the regression coefficient by −49.75%). This shows that both kinds of skills are productive even controlling for a wide range of personal, job and employer characteristics. Gender differences in returns to hard skills decreased to the level of 1.90% per year of schooling (by −21.81%) and remained statistically significant (compare Models 7 and 5). On the other hand, there were practically no gender differences in returns to soft skills as the corresponding beta coefficient was statistically insignificant and its value was close to zero. The comparison of standardised beta coefficient in Model 6 confirmed, as in previous models, that hard skills acquired by schooling (0.1410) and soft skills (0.0910) belong, together with tenure and work experience, to the most important wage predictors. It can be noted that job and employer variables significantly increased the predictive power of the model as R² increased from 0.429 to 0.517.

Although the article discussed the returns to hard and soft skills separately, job applicants and employees need them both to succeed in the labour market (Reference Balcar, Janíčková and FilipováBalcar et al., 2014). The complementarity of these skills was tested by re-estimation of Model 2 with an interaction term of soft skills and years of schooling (Model 8 in Table 3). The results showed a decrease of statistical significance of both years of schooling (statistically significant at the level of 0.1) and soft skills (statistically insignificant). However, the interaction term was also found statistically insignificant (beta coefficient 0.0011). As these results can be caused by non-linear productive effects of soft and hard skills interaction, the model was re-estimated also with education levels instead of years of schooling (Model 11 in Table 3). This specification brought expected results. Variables for particular education levels were found statistically insignificant, soft skills (surprisingly) remained statistically significant and positive, and the interactions of soft skills and educational levels (approximating hard skills) were found positive and statistically significant except university education. Further extension of the model by individual, employer and job characteristics (Model 12 in Table 3) led to a substantial increase in beta coefficients and statistical significance of the interaction of soft and hard skills. These models thus suggest that the productive effect of hard and soft skills consists in their combination.