The Trouble with Statistical Evidence

Abstract

Statistical evidence-frequently known as mere statistical evidence-has suffered at the hands of its critics for some time now. The preceding piece by Professor Judith Jarvis Thomson2 continues to refine the attack on such evidence-at least where it is the only evidence to support a court's judgment. In candor, I must admit that I am far from sure that Professor Thomson's attack is wrong, though I think it probably is. Perhaps it would be better to say that I find her primary argument unconvincing. Professor Thomson mounts her argument largely in three hypothetical cases to which she applies the usual rules of evidence (or, in the last case, epistemology) to reach the conclusions that a civil defendant should be found liable, that a criminal defendant should be found guilty, and that a fact may be said to be known. Professor Thomson regards these conclusions as highly unsatisfactory, however, and isolates the problem in the statistical nature of the evidence supporting each result. She offers the implication that statistical at least by itself, provides insufficient grounds for a verdict in favor of the proponent of that primarily because it lacks a causal link to that which is to be proved. What is needed, she argues, is individualized evidence, which, by her definition of the term, provides the necessary causal linkage. A very brief recap of the three central hypotheticals may be appropriate. Of the three, the first seems clearly the most important. It involves a hit and run automobile accident caused by a recklessly driven taxi. Identification of the taxi proves to be impossible, but it is known that the Red Cab Company owned and operated sixty percent of the taxis in circulation on the night in question. Since by assumption no other relevant facts are known, it is asserted that the probability that the accident was caused by Red Cab Company is sixty percent.3 Because this evidence meets the preponderance