In "A Practical Solution to the Reference Class Problem," Edward K. Cheng usefully surveys the ways in which the problem arises in legal contexts. Cheng argues that a practical solution to the problem lies in modern "model selection" methods which decide on the appropriate complexity of a model. This Essay argues that a simpler area of recent statistics, the theory of feature selection methods, is more relevant. Part I of this Essay argues that the correct reference class with which to compare a case is the set of cases which share with it all relevant features. Part II discusses model complexity and argues that Cheng's approach is workable, but that the statistical literature provides equally credible alternative approaches, based on smoothness instead of simplicity.
