Center for Economic Analysis

The data for this paper comes from the 2014 National Survey of Business Competitiveness-Rural Establishment Innovation Survey (REIS), a survey of 10,913 US businesses (53,234 including null responses) collected by the USDA Economic Research Service. The REIS data includes 257 variables for each firm including 40 measures of innovation (e.g., patent activity, improved market share, reduced labor costs, etc.). Many papers have suggested that patents are a poor measure of innovation because not all innovation is patented (or even patentable) and not all patents are indicative of real innovation. In our data, we find that 6% of the firms produce 100% of the patents. Further, we find that patent activity is only weakly correlated with our other measures of innovation. While this might indicate that innovation should not be defined as patent activity, these factors alone do not preclude patent activity from being used a proxy variable for the innovation process. To test this hypothesis, we construct joint models for patent activity and another innovation measure to test for significant differences between the models. We model these innovation measures using a bivariate probit specification with fixed-effects for the census region and the 3-digit NAICS industry code. Within this framework, we find very few statistically significant differences between patent models and other more general measures of innovation. We isolate three variables which have a disproportionately large impact on the patent model: difficulty hiring skilled workers, the age of the firm, and having an employee ownership plan. Our findings could be interpreted to support firm-level research papers which use patents as proxies for innovation, with the caveat that any coefficients on difficulty hiring, firm age, and employee ownership are likely to be overstated in the patent models. Our findings could also be interpreted to support macroeconomic research papers which use patents as proxies for innovation, realizing that regional variation in difficulty hiring may have a slight downward bias on the innovation proxy.