William L. Meadow & Cass R. Sunstein, Causation in Tort: General Populations vs. Individual Cases (U. Chi. L. & Econ., Olin Working Paper No. 130, U. Chi. Pub. L. Working Paper No. 179, Sept. 17, 2007).
Abstract: To establish causation, a tort plaintiff must show that it is "more probable than not" that the harm would not have occurred if the defendant had followed the relevant standard of care. Statistical evidence, based on aggregate data, is sometimes introduced to show that the defendant's conduct created a statistically significant increase in the likelihood that the harm would occur. But there is a serious problem with the use of such evidence: It does not establish that in the particular case, the injury was more likely than not to have occurred because the defendant behaved negligently. Under existing doctrine, a plaintiff should not be able to establish liability on the basis of a showing of a statistically significant increase in risk. This point has general implications for the use of statistical evidence in tort cases. It also raises complex issues about the relationship between individual cases and general deterrence: Optimal deterrence might be obtained by imposing liability on defendants who engage in certain behavior, even though a failure to engage in such behavior cannot be connected with the plaintiff's harm by reference to the ordinary standards of causation.