The impact of educational mismatch on returns to R&D: evidence from manufacturing in OECD countries

ABSTRACT

This paper investigates the effect of educational mismatch of R&D workers on firm's returns to innovation. R&D labour mismatch emerges when R&D workers have competencies different from those required by their occupation providing a contribution to innovation lower than in the case of perfect educational matching. By estimating a knowledge production function on data for 13 manufacturing industries from 16 OECD countries between 2003 and 2011, we find that R&D labour mismatch may cause returns to R&D investment to be between 10 and 15% lower than estimated in the literature. These results are robust to controlling for institutional factors, simultaneity feedbacks and other mis-specification issues. The detrimental effect of the misallocation of R&D labour is found to be stronger in those sectors where R&D activities have greater potential (returns), i.e. high-tech sectors.

KEYWORDS:

• Educational mismatch
• returns to education in R&D
• patents
• R&D
• OECD manufacturing

Acknowledgments

We are grateful to the editor, three reviewers, Fabrizio Pompei, Alessandro Sterlacchini, Michela Vecchi, seminar participants at the University of Perugia for helpful comments. We are also grateful to Olof Ejermo and the Centre for Innovation and Research and Competence in the Learning Economy (Circle) for the hospitality and granting the access to Statistics Sweden's data.

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Francesco Venturini http://orcid.org/0000-0002-7295-268X

Notes

1 ORU is the acronym for Over-education, Required education, Under-education.

2 We wish to thank a reviewer for stressing the importance of knowledge spillovers in our analysis.

3 It is implicitly assumed that one individual researcher cannot influence the direction of technological change.

4 This methodology has been adopted to study, among others, the effect of labour market institutions (Bassanini, Nunziata, and Venn 2009), product market regulations (Bourles et al. 2013), patent protection (Aghion, Howitt, and Prantl 2015).

5 Country list: Country list: Austria (AT), Australia (AU), Belgium (BE), Germany (DE), Denmark (DK), Spain (ES), Finland (FI), Greece (GR), Hungary (HU), Ireland (IE), Italy (IT), South Korea (KR), Netherlands (NL), United Kingdom (UK), United States (US). Industry list (ISIC Rev. 3 classes): Food, beverages, tobacco (15–16), Textiles, leather and related products (17–19), Wood and products of wood and cork (20), Pulp, paper and paper products, printing and publishing (21–22), Coke, refined petroleum products and nuclear fuel (23), Chemicals and chemicals products (24), Rubber and plastic products (25), Other non-metallic mineral products (26), Basic metals and fabricated metal products (27–28), Machinery and equipment n.e.c. (29), Electrical and optical equipment (30–33), Transport equipment (34–35) and Manufacturing n.e.c. (36–37).

6 Patents are assigned to applicants proportionally to the number of firms involved in innovation activities. For instance, one patent applied by two firms is assigned for 0.50 to each patentee, 0.33 is assigned to each patentee if there are three applicants, etc.

7 The Swedish Educational Terminology (SUN2000, linked to the International Standard Classification of Education, ISCED97) categorizes the levels of education into six classes. We have converted the levels into years of schooling as follows: Lower secondary education (up to 9 years): 7 years. Lower secondary education (9–10): 9. Upper secondary education (less than 2): 10. Upper secondary education (2): 11. Upper secondary education (3): 12. Post-secondary education (less than 2): 13. Post-secondary education (2): 14. Post-secondary education (3): 15. Post-secondary education (4): 16. Post-secondary education (5): 17. Other postgraduate programme: 18. Licentiate programme (half of PhD): 19. Doctoral programme (4 years completed): 21.

8 The identification of R&D workers is based on the Swedish classification of occupations (SSYK, built on ISCO), which categorizes occupations according to their skill requirements. SSYK is organized into 10 major categories (2 digit classification): 1. legislators, senior official and managers; 2. professionals; 3. technicians and associate professionals; 4. clerks; 5. service workers and shop sales; 6. skilled agricultural and fishery workers; 7. craft and related trade workers; 8. plant and machine operators and assemblers; 9. elementary occupations; 10. armed forces.

9 Summary statistics on institutional variables by countries and industries are reported in the Appendix. Table 2 reports descriptive statistics for innovation variables at the industry level. Electrical and optical equipment industry (ISIC Rev. 3 category 30-33) arises for the greatest patenting propensity (6 patent application per million worker annually), followed by Chemicals (24) and Machinery and equipment (29). These same sectors stand out also for a very high research effort, for both cost components of R&D.

10 Since the sector of employment of each worker is expressed at one-digit level, these data cannot be exploited in the regression analysis. Moreover, data are unavailable for all countries covered by our analysis and, hence, Austria, Australia, Greece and Hungary are left out from Table 3.

11 Similar findings arise if, in col. 5, we omit the ratio to output of other research expenses ORDY (omitted for brevity).

12 Parameters of the regulatory variables are semi-elasticities and hence have to be divided by 100 to be comparable with the coefficients (elasticities) of PRDY and other logged variables.

13 Our main findings do not change even when we include other measures of FDI restriction such as discriminatory screening or approval mechanisms, limitations to the employment of foreigners as key personnel, other operational restrictions, or finally a combination of all these dimensions of the regulation (unreported).

14 By construction, our measure of external financial dependence, defined as interaction between the financial development indicator and asset intangibility (at 2003), is correlated with PRDY (0.50). To assess to what extent our results are sensitive to this condition, we have re-estimated the model using the industry share of R&D personnel expenses from the benchmark country (Sweden at 2003). As further robustness checks, we have also measured asset intangibility with the share of ICT capital on total capital or on value added. Moreover, we have adopted alternative indicators of external financial dependence, namely (i) the ratio between the value of stock market capitalization and GDP; and (ii) the proportion of net margins to total earnings of the banking sector. All these unreported regressions yield findings similar to those reported in Table 6.

15 The patent stock is built from the series on patent counts adopting the perpetual inventory method and assuming an annual depreciation rate of 15%.

16 Evidence consistent with the hypothesis of diminishing technological opportunities can be found in Venturini (2012a) and Bloom et al. (2017).

17 These weights follow the scheme proposed by Lichtenberg and Van Pottelsberghe de la Potterie (1998). See Venturini (2015) for a discussion of pros and cons of the weighting procedures adopted to measure cross-country (cross-industry) spillovers.

18 Our set of external instruments genuinely captures the variation in the interaction variable as these regressors are insignificant when regressed against PRDY.