Consensus formation from observation of complex systems with limited intervention: Why statistics needs to be absorbed into epistemology
Professor Sander Greenland
Department of Epidemiology and Department of Statistics, University of California, Los Angeles, California
Experienced epidemiologists recognize that current statistical formalisms are inadequate for inference in risk assessment and health policy; astute social scientists recognize analogously that those formalism are inadequate for inference in economic and social policy. The inference problems in these fields are examples of consensus formation from observation of complex systems with limited intervention (CFOCLI). CFOCLI is a conceptually hard problem being addressed only slowly the statistics profession, in part because addressing it requires an epistemology both far more broad and more detailed than found in conventional statistical theory. There are alternative inference systems under development in computer science which claim to address at least some of the more severe limits of current statistical formalisms; their ability to address CFOCLI is intriguing but as yet far from demonstrated. Some crucial elements for CFOCLI absent from most formalisms include coalescing inferences from multiple agents that each have severe cognitive biases. These biases may be viewed as priors that down-weight large regions of the parameter space to the point of no influence. Shared biases develop and can lead to disastrous loss when all agents severely down-weight a region that contains reality in its interior. Examples include the history of diet and health recommendations by academically-based societies, as well as the collapse of hedge funds run by econometric theories. Such examples underscore the need to incorporate concepts from antifragility as well as information and cognitive sciences into the core of CFOCLI.
Hosted by: Nicola Orsini, Institute of Environmental Medicine, Karolinska Institutet.
Supported by: Strategic Research Program in EpidemiologyContact person: Nicola Orsini