Why do low-educated workers invest less in further training?

Abstract

Several studies document that low-educated workers participate less often in further training than high-educated workers. This article investigates two possible explanations: low-educated workers invest less in training because of (1) the lower economic returns to these investments or (2) their lower willingness to participate in training. Controlling for unobserved heterogeneity, we find that the economic returns to training for low-educated workers are positive and not significantly different from those for high-educated workers. However, low-educated workers are significantly less willing to participate in training. We show that this lesser willingness to train is driven by economic preferences, and personality traits.

Keywords:

• returns to training
• preferences
• noncognitive skills

JEL Classification::

• J24
• J31

Acknowledgements

We thank Eric Bonsang, Bart Golsteyn, Renske Jongsma, Olivier Marie, an anonymous referee, and participants of the LOPSI conference in Milan, the Education in Adulthood and the Labour Market workshop in Nuremberg, the ESPE Conference in Essen, the International Workshop on Applied Economics of Education in Catanzaro, and ROA/AEII seminar series in Maastricht for their useful comments on earlier versions of this article.

Notes

¹ Labour market institutions matter in this respect. For example, Peraita (2005) shows that the high wage compression and education system in Spain results in unequal provision of firm-sponsored training between the high- and low-educated workers.

² Bassanini and Ok (2007) discuss such heterogeneity in returns to training in a number of European countries.

³ Restricting the sample to people aged 18–54 yields similar results.

⁴ Data on the workload associated with the training are not consistently available over the years.

⁵ Evidence of a growing gap in the training participation of low- and high-educated workers is also found in Germany (Riphahn and Trübswetter, 2007).

⁶ Our analyses control for the number of hours worked.

⁷ The DNB Household Survey was previously used, for example, to study the effect of personality on earnings (Nyhus and Pons, 2005).

⁸ The three most important competencies were the same, irrespective of education level, with people skills being the most frequently reported competency, followed by occupation-specific skills and communication skills.

⁹ Manski (1990) shows that such intentional data have a predictive value for actual behaviour.

¹⁰ One could argue that the relation between economic preferences and personality traits, on the one hand, and the willingness to train is not necessarily linear. However, Mueller and Plug (2006), who investigate the effect of personality on earnings, show that personality traits can be used linearly.

¹¹ The concept of future orientation is highly related to an individual's time preference and is also seen as a determinant of human capital investments (Killingsworth, 1982).

¹² Using factor scores rather than a summary index does not affect our findings.

¹³ Locus of control is not assumed to be time constant.

¹⁴ The full list of statements is reported in Appendix D.

¹⁵ See http://ipip.ori.org/New_IPIP-50-item-scale.htm.

¹⁶ We use principal component factor analysis with promax rotation to allow for correlation across factors. We imposed a five-factor solution in anticipation of the underlying factor structure. The 50 items and their factor loadings are reported in Appendix E. The full results are available from the authors upon request.

¹⁷ Using the hourly wage (computed from the monthly wage and the actual number of weekly working hours) rather than the monthly wage does not affect our results.

¹⁸ Education is taken as time constant and equal to its value the first time an individual is observed in the data. Henceforth, education is dropped in the fixed effects estimation.

¹⁹ The Ordinary Least Square (OLS) version of the model also includes gender and year dummies.

²⁰ The wage return to training for intermediate-educated workers equals . The return for high-educated workers equals .

²¹ In our data 75% of the individuals are in paid employment.

²² Alternatively, the lower return could be due to measurement error on the training variable, resulting in downward bias of the estimated coefficient in the fixed effects specification. However, this is not likely to play a role here, since, except for the last three waves of the panel, the data were gathered by very experienced interviewers. If misreporting were an issue, one would expect training to be reported differently by experienced panel respondents and new panel members, since they differ in experience with the questionnaire. However, tests have shown that this is not the case.

²³ This is possible since the training information is available for persons with a job as well as for persons without a job.

²⁴ We also implemented two Instrumental Variable (IV) approaches. First, analogous to Budría and Pereira (2007) we used a dummy for holding a second job as an instrument for training participation. In their view, this proxies for an individual's work commitment and motivation, and they show the statistical validity of this instrument. In our data, second job holding is also positively related to training, but tests reject the validity of this instrument (F-test < 10). Second, we exploited a change in the Dutch tax law in 1998 to instrument training participation. The additional tax deduction that was then introduced implies that 120% of the training costs can be deducted from profits (Leuven and Oosterbeek, 2004). In our analyses the generic change in the tax deduction regime is not found to be a valid instrument (F-test < 10). One possible reason is that the additional deduction is not very large. In the absence of suitable instruments, we estimated nearest-neighbour matching as well as kernel-based matching estimators and found that, as is the case in the OLS model in Table 1, low-educated workers who participated in training earn a 5% higher wage than those who did not participate.

²⁵ We estimated an OLS model to test the relation between the monthly wage (in log) and work-related training in the past 2 years in the ROA Life Long Learning Survey (the model controls for gender, age and education). We found positive returns to training for low-educated workers, and these returns are not significantly different from the returns for the intermediate- and high-educated workers.

²⁶ An F-test rejects the hypothesis that the Big Five personality traits are jointly equal to zero.

²⁷ To the extent that the fixed effects (μ_i ) capture differences in abilities such as preferences and personality traits, an additional analysis on the OSA Labour Supply Panel has revealed that low-educated workers who participated in training have higher ability levels than those who did not.