Breakthroughs in Decision Science and Risk Analysis

Chapter 6Causal Analysis and Modeling for Decision and Risk Analysis

Louis Anthony (Tony) Cox, Jr.

Cox Associates, NextHealth Technologies, University of Colorado-Denver, Denver, CO, USA

Introduction: The Challenge of Causal Inference in Risk Analysis

In decision and risk analysis, the goal of making decisions is to change the probabilities of outcomes to make preferred outcomes more likely. The decision-maker’s choices act upon the world, causing changes in outcome probabilities. Yet, this crucial concept of causal efficacy is seldom developed in detail in decision analysis, and the fact that formal probability theory applies only to events (subsets of a sample space) rather than to actions and their consequences is seldom emphasized (Pearl, 2010). Yet, if one is careless about causation—for example, by routinely interpreting statistical associations or regression coefficients as if they were known to be causal, as is unfortunately common practice in modern epidemiology and public health applications—then the decisions that one takes based on assumptions about causality may turn out to cause quite different shifts in outcome probabilities than those that were expected and intended. Fortunately, modern methods of causal analysis can do much to prevent such unpleasant surprises. This chapter surveys some of the most useful methods for causal analysis, contrasts them with widely used but untrustworthy methods based on judgments about statistical associations, and develops their relations ...