The TIS Lab directs research in four main areas:
- Machine Learning
- Renewable Energy
Review of Economics and Statistics
Forthcoming ( 2021 )
The United States patent system is unique in that it requires an applicant to cite documents they know to be relevant to the examination of their patent application. Some research has suggested that applicants strategically withhold 21-33% of relevant citations from patent examiners, suggesting that more than one in ten patents are fraudulently obtained. We challenge this view. We examine the institutional details of how courts identify strategic withholding and find that such claims are inconsistent with both legal standards and standard operating procedures. We compile a more up-to-date and detailed set of data to reassess the empirical basis for the claim and find no evidence that applicants withhold citations from examiners.
The RAND Journal of Economics
Vol. 51, No. 1: pp. 109–132 ( 2020 )
Many studies rely on patent citations to measure intellectual heritage and impact. In one article, we use a vector space model of patent similarity, and new data from the USPTO, to show that the nature of patent citations has changed dramatically in recent years. Today, a small minority of patent applications are generating a large majority of patent citations, and the mean technological similarity between citing and cited patents has fallen dramatically. We replicate several well-known studies in industrial organization and innovation economics and demonstrate how overly generalized assumptions about the nature of patent citations have misled the field.
Review of Economics and Statistics
Vol. 102, No. 3: pp. 569–582 ( 2020 )
Research in economics and organizational behavior suggests that a “tolerance for failure” may be required to motivate individuals to select exploratory courses of action. We examine how individuals select between options to explore or exploit when confronted by prolonged periods of negative feedback. We run randomized, online experiments in a two-dimensional maze game to shed light on behavioral strategies with respect to exploration & exploitation. Our methods extend beyond analytical, two-period, Bayesian models of decision-making, to account for stochastic learning and behavior by individuals in longer-running, dynamic contexts.
Journal of Economic Behavior & Organization
Vol. 150: pp. 162-181 ( 2018 )
Executives play an important role in leading firm innovation. We examine how executives and middle managers interact, therein shaping the rate and scope of innovation by the firm. Empirically, we exploit a natural experiment in Delaware (USA) where court rulings increased takeover protection. Difference-in- differences estimates show that increased takeover protection reduced the rate of innovation by firms, and that it also reduced the scope of innovation across several key dimensions (technological, temporal, organizational, and international). We examine how these effects interact with the size of the firm, and the substitutive relationship between competitive pressures from the takeover market and product market.
IEEE – 18th Intl. Conference on Machine Learning and Applications
DOI 10.1109/ICMLA.2019.00120 ( 2019 )
Recent work has benchmarked the feasibility of using new machine learning models to develop more accurate measures of technological invention and innovation. Such measure construction relies on the use of natural language processing and text analysis techniques, especially text embedding methods. We evaluate the performance of different vector space models to perform this task. We address the real-world problem of modeling patent-to-patent similarity and compare TFIDF (and related extensions), topic models (e.g., latent semantic indexing), and neural models (e.g., paragraph vectors).
Social Science Research Network
The TIS lab analyzes high-dimensional models of technological relatedness. With support from Google and the Google Cloud Computing platform, we use Big Data methods to map out the similarity space of all patents granted by the USPTO. The core of the project is a 450 TB vector space model for over 14 trillion patent-to-patent (pairwise) comparisons. We also work on methods to visualize the space, with an aim to develop data-driven tools for decision-making on issues of Intellectual Property (IP).
Vol. 25: pp. 652-677 ( 2016 )
We estimate the firm‐level returns to retaining employees using difference‐in‐differences analysis and a natural experiment where the enforcement of employee noncompete agreements was inadvertently reversed in Michigan. We find that noncompete enforcement boosted the short‐term value of publicly traded companies by approximately 9%. The effect is increasing in local competition and growth opportunities, and offset by patenting.
How Anticipated Employee Mobility Affects Acquisition Likelihood: Evidence from a Natural Experiment
Vol. 36: pp. 686-708 ( 2015 )
Abstract: This study draws on strategic factor market theory and argues that acquirers’ decisions regarding whether to bid for a firm reflect their expectations about employee departure from the firm post‐acquisition, suggesting a negative relationship between the anticipated employee departure from a firm and the likelihood of the firm becoming an acquisition target. Using a natural experiment and a difference‐in‐differences approach, we find causal evidence that constraints on employee mobility raise the likelihood of a firm becoming an acquisition target. The causal effect is stronger when a firm employs more knowledge workers in its workforce and when it faces greater in‐state competition; by contrast, the effect is weaker when a firm is protected by a stronger intellectual property regime that mitigates the consequences of employee mobility.
The Changing Frontier: Rethinking Science and Innovation Policy
Adam Jaffe and Ben Jones, editors – Chapter 7: pp. 199 – 232 ( 2015 )
Abstract: We document three facts related to innovation and entrepreneurship in renewable energy. Using data from the US Patent and Trademark Office, we first show that patenting in renewable energy remains highly concentrated in a few large energy firms. In 2009, the top 20% firms accounted for over 40% of renewable energy patents in our data. Second, we compare patenting by venture capital-backed startups and incumbent firms. Using a variety of measures, we find that VC-backed startups are engaged in more novel and more highly cited innovations, compared to incumbent firms. Incumbent firms also have a higher share of patents that are completely un-cited or self-cited, suggesting that incumbents are more likely to engage in incremental innovation compared to VC-backed startups. Third, we document a rising share of patenting by startups that coincided with the surge in venture capital finance for renewable energy technologies in the early 2000s. We also point to structural factors about renewable energy that have led the availability of venture capital finance for renewable energy to fall dramatically in recent years, with potential implications for the rate and trajectory of innovation in this sector.
NREL/TP-6A20-50624 ( 2011 )
Abstract: Low-carbon energy innovation is essential to combat climate change, promote economic competitiveness, and achieve energy security. Using U.S. patent data and additional patent-relevant data collected from the Internet, we map the landscape of low-carbon energy innovation in the United States since 1975. We isolate 10,603 renewable and 10,442 traditional energy patents and develop a database that characterizes proxy measures for technical and commercial impact, as measured by patent citations and Web presence, respectively. Regression models and multivariate simulations are used to compare the social, institutional, and geographic drivers of breakthrough clean energy innovation. Results indicate statistically significant effects of social, institutional, and geographic variables on technical and commercial impacts of patents and unique innovation trends between different energy technologies. We observe important differences between patent citations and Web presence of licensed and unlicensed patents, indicating the potential utility of using screened Web hits as a measure of commercial importance. We offer hypotheses for these revealed differences and suggest a research agenda with which to test these hypotheses. These preliminary findings indicate that leveraging empirical insights to better target research expenditures would augment the speed and scale of innovation and deployment of clean energy technologies.