New Methods for Molding Models to Specific Cases to Enhance Policy Predictions

University Of California-San Diego, La Jolla CA

Investigators

Abstract

General Audience Summary This project aims to enhance policy effectiveness by focusing on ways to improve models that are used to estimate whether a social policy will achieve its intended outcomes when implemented in the setting at hand, in the way it would be implemented in that setting. The project will provide a systematic, theoretically grounded approach for deciding what kinds of evidence are good to collect, and for deciding how to put the evidence together to ascertain what it shows about the chances of success. Policy outcomes will always be uncertain, often greatly so. This project has promise to substantially improve predicting outcomes in individual cases by providing methods for molding models to the cases at hand. The results of this research will have potential to improve policy effectiveness and to suggest new kinds of scientific studies to support policy prediction; they will be disseminated across a number of policy domains. Technical Summary This project is an investigation of the kinds of evidence, both local and scientific, that can be used to build full enough models to make reasonable, albeit uncertain, policy predictions. The methodology is primarily analytic, and it builds on studies in the history, philosophy, and sociology of science on evidence, objectivity, and causal modeling. It also builds on the PI's recent work on evidencing singular causal claims and on causal mechanisms. The project involves in-depth study of cases in the natural and the social sciences, engineering, and the law of successful single-case prediction and post-hoc evaluation in complicated systems. It will develop categories of evidence, and it will refine and test the proposed model structure to see how well it fits successful cases. The project should contribute to the philosophical understanding of singular causation, currently a big topic in philosophy. It should advance understanding of causal modeling, in principle and in practice, especially in the social sciences.