Leo Breiman’s Two Cultures: Introspection, Debate, and Discussion 20 Years Later

• Activity Number: 406 - Leo Breiman’s Two Cultures: Introspection, Debate, and Discussion 20 Years Later
• Type: Invited
• Date/Time: Wednesday, August 10, 2022 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Mike Baiocchi* and Andrew Gelman* and Debashis Ghosh* and Arman Oganisian* and Ani Eloyan* and Elizabeth Ogburn* and Anna Neufeld*
• Companies: Stanford University and Columbia University and Colorado School of Public Health and Brown University and Brown University and Johns Hopkins University and University of Washington
• Keywords: modeling; algorithms; causal inference; machine learning; fairness; paradigms

Abstract

The journal Observational Studies recently published a special issue in which 28 commentators celebrate, examine, and discuss Leo Breiman’s influential paper Statistical Modeling: The Two Cultures (Statistical Science, 2001) on its 20th anniversary. In his original paper, Breiman contrasts the modeling and algorithmic cultures within the field of statistics. In this panel discussion, we have chosen seven of those 28 commentators who will present a diversity of opinions and foster debate and discussion with the audience on aspects of Breiman’s work that still hold today and ways that our field has evolved over the past 20 years. The panelists will present their current views on (1) the blending and cross-fertilization of modeling cultures (not just two distinct ones) under historical, foundational, and flexible paradigms; (2) the importance of interpretable and fair algorithms, understanding the data, outcome reasoning, and social responsibility; (3) causal modeling; (4) Bayesian inference; and (5) distributed, targeted, and supervised learning. A robust discussion among the audience and panelists will be encouraged.