Social Science and Technology Seminar Fall 2019
Seminars locations and times are TBD.
You can also subscribe to our mailing list.
Riitta Katila, Professor, Management Science & Engineering, Stanford Technology Ventures Program
Tim Bresnahan, Professor, Department of Economics
Woody Powell, Professor, Department of Education
Chuck Eesley, Associate Professor, Management Science & Engineering, Stanford Technology Ventures Program
October 16, 2019
David Obstfeld (Cal State Fullerton) - TALK POSTPONED. There will be no talk today!
October 30, 2019
Scott Stern (MIT) - presentation in the Gunn SIEPR Bldg, 2nd Floor, Room 224 (Johnson Conference Room)
Predicting Patent Quality: A Machine Learning Approach (with Lindsey Raymond)
Not all (patented) innovations are created equal. Since the vast majority of impact is associated with a small fraction of superstar innovations, accounting for the skewed nature of innovation quality is a central challenge in the study of innovation. However, traditional measures of realized impact (e.g., the citations ultimately received by a patent) conflate the initial intrinsic innovation quality with the dynamic influence of the institutional, strategic and economic environment. This paper develops a novel method for estimating patent quality on a prospective rather than retrospective basis. Our approach combines three key insights. First, rather than attempting to maximize predictive accuracy, estimating intrinsic patent quality should focus on technological information (such as patent claims or text) rather than environmental factors (such as the identity or geography of the inventor or assignee). By delineating clearly between technological and non-technological factors, we are able to assess the relative importance of intrinsic patent quality and the influence of the environment. Second, rather than attempting to maximize the predictability of the number of citations, patent quality estimation should focus on predicting the rank of a patent within the patent impact distribution. Because the distribution of patent quality is highly skewed, predicting the most impactful innovations can meaningfully increase the predictable fraction of patent impact. Finally, rather than focusing on the influence of specific patent characteristics, patent quality estimation should account for interdependencies and interaction effects that reflect the combinatorial nature of the innovation process. Machine learning methods, including context-dependent natural language processing and neural networks, are well suited to capture these combinatorial effects. We apply these ideas by estimating the 10-year citations realized by patents applied for in the United States between 2002 and 2006, documenting several key findings. First, across a range of approaches, the ability to predict patent quality increases monotonically over the patent distribution, with the most valuable being the most predictable. For example, more than 50% of patents in the top 1% of the realized citation distribution are in the top 10% of our estimated patent quality distribution. Second, we demonstrate that, though statistically significant, the influence of geography or identity of the inventor or assignee results in only a modest improvement in predicting patent impact: the vast majority of predictable differences in patent quality are embedded in technological information contained in the patent. Finally, the ability to predict high-impact patents (e.g., those within the top 1% of the citation distribution) combined with the fact that value of patents is highly skewed (e.g., the top 1% of patents may be associated with more than 40% of value) implies that a very significant fraction of patent value may be able to predicted a the time of patent grant. Our results suggest that technological factors may be the primary driver of the impact of breakthrough innovations, while environmental factors may loom larger in shaping the impact of incremental innovation.
November 6, 2019
Alfonso Gambardella (Bocconi) - presentation in the Gunn SIEPR Bldg, 3rd Floor, Room 320 (Doll Conference Room)
Small Changes with Big Impact: Experimental Evidence of a Scientific Approach to Decision-Making of Entrepreneurial Firms (joint work with Arnaldo Camuffo & Chiara Spina)
Identifying the most promising business ideas is key to the introduction of novel firms, but predicting their success can be difficult. We argue that if entrepreneurs adopt a scientific approach by formulating problems clearly, developing theories about the implications of their actions, and testing these theories, they make better decisions. Our theory predicts that the scientific approach corrects the problem of overestimation and underestimation of the returns from business ideas. This has implications for important entrepreneurial choices, such as discontinuing a business idea and pivoting, as well as for performance. Using a field experiment with 251 nascent entrepreneurs attending a pre-acceleration program, we examine the effect of a scientific approach to decision-making. In the field experiment, we teach the treated group to formulate the problem scientifically and to develop and test theories about their actions, while the control group follows a standard training approach. We collect 18 data points on the decision-making and performance of all entrepreneurs for 14 months. Results show that treated entrepreneurs are more likely to close their start-up. We also find that scientific entrepreneurs are more likely to pivot a small number of times, suggesting that the scientific approach makes them more precise in pivoting to more valuable ideas. Finally, we find that the scientific approach increases revenue, suggesting that a more accurate assessment of ideas helps entrepreneurs to make better decisions and eventually leads to better performance. This study shows that the scientific approach is a critical link between decision-making and performance of nascent entrepreneurs.
November 13, 2019
Rahul Kapoor (University of Pennsylvania) - presentation in the Gunn SIEPR Bldg, 3rd Floor, Room 320 (Doll Conference Room)
Two Faces of Value Creation in Platform Ecosystems: Leveraging Complementarities and Managing Interdependencies
A given innovation often does not create value on its own. Rather it is connected with other elements in the ecosystem for its value creation. We draw on this premise in a platform-based ecosystem in which participating firms innovate around a platform. We introduce the notion of connectedness to refer to the extent to which a given innovation connects with platform components and other complements in the ecosystem, and explore the implications for its commercialization success. While higher connectedness may allow the innovation to leverage complementary functionalities, it may subject the innovation to technological interdependencies that may limit its value creation. Evidence from 244,034 apps launched by software developers for Apple’s iPhone platform during 2008-2013 highlight these two faces of value creation in platform-based ecosystems.