Full citation
Crépon, B., Duguet, E., & Mairesse, J. (1998). "Research, innovation, and productivity: An econometric analysis at the firm level". NBER Working Paper No. 6696, National Bureau of Economic Research.
Summary
This paper investigates how research investment drives innovation, and how innovation translates into higher productivity at the firm level. Rather than looking at any single link in isolation, the authors build a complete chain: firm decisions about R&D → innovation outputs (measured by patents and new product sales) → firm productivity.
The study uses data from French manufacturing firms and introduces three key innovations. First, it measures innovation output directly—patents and the share of sales from new products—rather than just counting R&D spending. Second, it accounts for the fact that only a minority of firms do R&D and hold patents, which creates bias if you ignore firms with zero activity. Third, it addresses the reality that R&D decisions are not independent of innovation success and market conditions; the authors handle this "endogeneity" problem using advanced econometric methods designed for limited dependent variables and simultaneity.
The main findings confirm what many had suspected but had not rigorously shown. Larger firms, firms with higher market share, and diversified firms are more likely to do R&D. Among R&D performers, effort scales proportionally with firm size. R&D effort strongly predicts both the number of patents and the proportion of sales from new products. Finally, and most importantly, firms with higher innovation output enjoy measurably higher labour productivity, even after accounting for capital intensity and the skill composition of the workforce.
Key topics by page
| Page(s) | Topic |
|---|---|
| 1–2 | Motivation and three novel methodological contributions to R&D and innovation analysis |
| 3–4 | Model structure: research decisions, innovation output equations, and productivity equation |
| 4–5 | Generalised tobit specification for R&D participation and intensity |
| 5–6 | Patent count data model (negative binomial); ordered probit for innovative sales intervals |
| 6–7 | Productivity equation and augmented Cobb–Douglas specification |
| 7–9 | Asymptotic least squares (ALS) estimation method for handling simultaneity and selection |
| 10–11 | Descriptive statistics: prevalence of R&D, patents, and skill composition |
| 11–13 | Basic model results: size, market share, diversification effects on R&D; R&D elasticity with patents and sales |
| 13–14 | Extended model: demand pull and technology push indicators; skill composition effects on productivity |
| 14–15 | Comparison of ALS estimates with OLS, 2SLS, and maximum likelihood; evidence of interaction between selectivity and simultaneity biases |
| 25–29 | Data construction details: accounting data, R&D capital perpetual inventory method (15% depreciation), patent matching, innovation survey design |
Quotes
"We find that using the more widespread methods, and the more usual data and model specification, may lead to sensibly different estimates. We find in particular that simultaneity tends to interact with selectivity, and that both sources of biases must be taken into account together." (p. 2)
"Our model thus includes three relationships: the research relation linking research to its determinants, the innovation equation relating research to innovation output measures, and the productivity equation relating innovation output to productivity." (p. 3)
Key insights
- On firm size and R&D: Probability of doing R&D rises with firm size (employees), market share, and diversification. However, among firms that do R&D, the intensity of research spending is proportional to size—not increasing with it—meaning larger firms do not necessarily spend more per employee.
- On R&D and innovation output: The elasticity of patents with respect to R&D capital intensity is approximately one (a 10% increase in R&D intensity yields roughly a 10% increase in patent output). By contrast, innovative sales show lower elasticity (~0.43), suggesting that not all innovations are patented or all patents are commercialised.
- On innovation and productivity: Firms with more patents or higher shares of innovative sales show measurably higher labour productivity. When skill composition is controlled for, the estimated elasticity of productivity with respect to patents falls from 0.13 to 0.09, indicating that knowledge capital and skilled labour are correlated but distinct contributors to productivity.
- On econometric method: The paper demonstrates that ignoring either selectivity (few firms have patents) or endogeneity (R&D responds to market opportunities) alone produces biased estimates. The two sources of bias interact, and both must be corrected together using appropriate methods.
- On demand vs. technology push: Market demand appears to influence both the likelihood and intensity of R&D more strongly than technological opportunity, though both matter.