Does Innovation Destroy Employment in the Services Sector? Evidence from a Developing Country

Abstract

The employment effect of innovation in the heterogeneous universe of services is investigated using firm-level data provided by the 2004–9 Uruguayan services innovation surveys. The empirical analysis shows that the effect of product innovation on employment is positive, while process innovation appears to have no effect. Process innovation activities tend to replace low-skilled jobs with jobs of a higher level of qualification. Product innovation allows for efficiency gains in the production of new services with unskilled labor, and no gains with skilled labor. The results found for knowledge-intensive business services and small firms, with some exceptions, are similar to those found for the whole sample.

KEY WORDS:

• employment
• firm size
• innovation
• innovation surveys
• knowledge intensity
• services sector

Acknowledgments

Comments and suggestions by Gustavo Crespi, David Kaplan, Alessandro Maffioli, Jacques Mairesse, Pierre Mohnen, Ezequiel Tacsir, Marco Vivarelli, Pluvia Zuñiga, and participants at the first and second discussion workshop of the Inter-American Development Bank (IDB) Employment Generation, Firm Size and Innovation in Latin America research project (Washington, DC and Costa Rica), at MEIDE 2011, at the 2011 Economic Meeting of the Central Bank of Uruguay, at the 2011 IDB economic seminars (Montevideo), and at Globelics 2011 are gratefully acknowledged. Paola Cazulo provided excellent research assistance. The views expressed in this article are the authors’ and are not necessarily shared by the OECD or its member countries.

Notes

1. The services sector accounts for more than 60 percent of both GDP and employment in Latin American and Caribbean countries and more than 70 percent in developed countries (World Development Indicators 2011, World Bank).

2. For the manufacturing sector, the evidence is ample. For Latin America, see, for example, Aboal et al. (2015) and de Elejalde et al. (2014).

3. Implicitly, we are assuming that process innovation only affects efficiency in the production of old products; that is, the term -(lnθ₁₂-lnθ₁₁) in Equation (1). However, process innovation could also potentially affect β. In the Robustness Analysis section, we will try to see if this is indeed the case.

By definition, all the sales of the previous period are old in the current period. Therefore, it is not possible to compute the growth rate of sales of new services.

Specifically, the error term is as follows: ν = – π₁ – βπ₂(Y₂₂/Y₁₁) – (ω₁₂– ω₁₁), where π₁ is the rate of increase of the prices of old services and π₂ is the difference of the prices of new services with respect to the prices of the old services.

4. Harrison et al. (2008) transform the original model that was in real terms to include the sales in nominal terms. This generates an additional problem: The unobserved disturbance includes prices of the new products that are correlated with g₂. In any case, the bias here is an attenuation bias.

5. If this variable is a good proxy for the rate of increase of prices of old services, then the error term ν will not include the change in prices of old services. Still, the prices of the new services will be in the error term.

6. Note that the left side is a proxy for the rate of productivity growth with opposite sign.

7. The authors show that ignoring the presence of - π₁ in the error term of Equation (3) generates an attenuation bias in OLS estimation of α₁.

8. Firms with missing information on sales or employment were also excluded, as were the first and ninety-ninth percentiles of variables l and g to avoid outliers and negative values of the variable g₂. By construction (see Appendix B), g₂ in very exceptional cases can be negative (four cases in our sample).

9. The matching with the EAS was not without loss. The 2004–6 SIS was performed with the same sampling frame as the EAS 2005. This implies that a significant number of firms that were surveyed in the 2004–6 SIS did not participate in the 2004 EAS. A similar problem arises when matching the 2007–9 SIS (which is a subsample of the 2009 EAS) and the 2007 EAS. Both facts explain the loss of observations when matching the surveys. In addition, the questions relating employment and sales in the two surveys are not exactly the same. In the case of employment, firms in the SIS are asked for the number of people employed on average in a year, including professionals, technicians without a dependent relationship, owners and business associates working in the firm, and unpaid family workers; in the EAS survey, the same firms are asked for total employment including dependent workers, owners and business associates working in the firm, and unpaid family workers. This can cause some discrepancies between sources.

We do not present results for the subsample of small firms since the sample size is reduced to only thirty-three firms. This happens because we are only including firms present in both surveys, which are in general of big size since they are of mandatory inclusion in the SIS.

Additional information

Funding

This research was possible thanks to the financial support of the IDB. The usual declaimer applies.