Measuring the persistence in innovation in Spanish manufacturing firms: empirical evidence using discrete-time duration models

Abstract

This paper measures the level of persistence in innovation using a large representative sample of Spanish manufacturing firms for the period 1990–2008. We determine survival in innovation activities using discrete-time duration models, which control for some of the existing problems in the continuous-time duration models used in previous studies (namely, unobserved heterogeneity and the proportional hazards assumption). This paper examines the relationship between the firm-specific characteristics of technological regimes and the persistence measured by innovative spells at the firm level. The results show that high technological opportunities, patents, cumulativeness of learning based on previous experience and accumulated R&D, as well as the use of generic knowledge provided by universities enhance persistence in innovative activity.

Keywords:

• persistence in innovation
• discrete-time duration models
• panel data
• technological regimes

Acknowledgements

We are grateful to two anonymous referees as well as the editors, for comments. We also thank participants at Workshop for the Special Issue of Economics of Innovation and New Technology ‘Innovation Persistence’ that took place on 8 March at the Department of Economics ‘Cognetti de Martiis’ in Torino (Italy).

Notes

1. Higher opportunities can favor the entry of new innovators, thus reducing the stability in the ranking of innovators (measured by the rank correlation between the hierarchies of firms patenting in two different periods). However, high opportunity conditions in conjunction with high appropriability may also be associated with a Schumpeter Mark II pattern (Breschi, Malerba, and Orsenigo 2000, 394).

2. Examples include Doiron and Gørgeus (2008), Joyce (2005) and Fougère, Kramarz, and Magnac (2000).

3. Our results are compatible with those obtained in previous studies based in duration analysis (Geroski, Reenen, and Walters 1997; Cabagnols 2003; Jang and Chen 2011). Geroski, Reenen, and Walters (1997) estimate that the corrected mean spell length over 5178 spells is 1.92 years. Although they use a large panel of patent data (3304 firms from 1969 to 1988), other studies that consider shorter panels also obtain comparable results. Jang and Chen (2011) show that the average spell length is 2.18 years using a sample of 125 manufacturing firms in Taiwan and considering only 160 patent spells over the period 1990–2001.

4. High and medium-high-technology industries include chemicals and pharmaceuticals; machinery and equipment; computer products, electronics and optical: electrical materials and accessories; vehicles and accessories; and other transport equipment. Low and medium-low technology industries include meat products; food and tobacco; beverage; textiles and clothing; leather, fur and footwear; timber; paper; printing; plastic and rubber products; non-metal mineral products; basic metal products; fabricated metal products; furniture; and other manufacturing.

5. Patents measure the persistence of innovative leadership rather than of innovative behavior because not all inventions are patented (Duguet and Monjon 2004)

6. A more detailed explanation of the suitability of using the number and duration of previous spells can be found in Fritjers (2002). In broad terms, there is no risk of endogeneity because the dependent variable and proxies have a different nature. The dependent variable (hazard rate) is a conditional function of density in t-time, while the number and the duration of previous spells refer to cumulative and unconditional values measured before starting the current spell. Additionally, from an econometric point of view, considering state dependence provides more efficient estimations because it allows us to control for the effect of multiple spells (Joyce 2005).

7. See Collier (2005) for the field of labor economics and Obashi (2010) and Hess and Persson (2011) in international economics.

8. These papers refer to external trade relationships, but their conclusions about the advantages of this methodology can easily be extrapolated to the duration analysis of innovation.

9. Because the introduction of this variable is not possible in a Cox model, we only report the two first specifications.

10. Additional work regarding the influence of the number and the duration of previous spells is developed. The joint influence of lagged and occurrence dependence is examined by introducing interaction terms defined by the number of previous spells multiplied by the duration of previous spell (see Table A1 for the descriptive statistics and Table A4 for the estimation results). The results confirm that the longer the duration of previous spells – even with more than one spell – the higher the persistence in innovation (interaction 4). We are very grateful to C. Antonelli for suggesting we perform this robustness analysis.