Selection bias in innovation studies: A simple test

Does the use of single-office patent counts lead to selection bias in econometric estimates of invention production functions?

Reliable measurement of innovation is crucial for understanding the economic impact of R&D and innovation policies. Using patent counts exclusively from a single office can lead to selection biases that skew the results of econometric studies, which may trend to favor certain types of organizations or filing strategies. Such biases can misinform policy decisions and the academic discourses surrounding innovation economics, leading to inadequate strategies that do not account for the comprehensive landscape of innovation activities. Therefore, identifying and mitigating these biases is essential for accurate analysis of innovation output and the effectiveness of R&D expenditure.

Utilizing a novel dataset comprising all patents filed by Belgian firms and German universities, we applied a two-step methodology to detect selection bias due to reliance on single-office patent counts. Our empirical analysis involved 429 firm-year observations of Belgian firms and 134 university observations from German institutions. We employed Poisson regression models to analyze patent production functions and assess the econometric implications of using single-office counts.

We find that relying on a single-office count results in significant selection bias affecting econometric estimates. Notably, we detect selection bias in specifications where firm size, R&D intensity, and market exposure appear biased in their association with inventive output. We also find that estimates based on the single-office count differ significantly from those using a global count, highlighting up to a 40% misestimation of the true effect of variables such as R&D intensity and underscoring the need for comprehensive patent count methodologies.

• 429 firm-year observations from the Belgian R&D survey (2004, 2006, 2008)
• Belfirst database providing annual balance sheets and income statements
• Patstat database from the EPO (April 2009 version)

• The decision to file patents at a single office is not random, implying selection bias when using that count.
• The use of priority patent applications reflects a complete measure of an organization's innovative output, although this can be distorted by filing strategies.