Some issues arising in the presentation of statistical testimony

Abstract

This article describes some special problems facing expert witnesses presenting statistical evidence when they testify in court. Since the legal system, especially in civil cases, serves as a dispute resolution process that needs to reach a decision in a timely manner, its objectives and procedures differ from those in pure science. Unlike the classroom or summarizing their research in a talk where the statistician determines the presentation, in the courtroom a witness is only allowed to answer questions posed by the others, mainly the lawyers for both parties. One case where the author could not persuade the lawyer to ask a set of questions designed to show that the other ‘expert’ was using a statistical that had no power and consequently would never indicate that the data were consistent with discrimination is described. In contrast, a case where the judge understood the need for the data analysis to correspond to the time frame of the case and accepted the Mantel–Haenszel test as an appropriate way to analyse stratified data arising in a discrimination case is summarized. Several suggestions designed to improve the effectiveness of statistical testimony and related issues arising in Daubert hearings that screen proposed expert analyses for scientific reliability are given. Recently, a distinguished scientist wrote an editorial in Science that supported epidemiologists who had obtained data solely for use in a case but later submitted an article to a journal, apparently in violation of the court’s order. The last section questions this view as it diminishes the value of commitments researchers give the public when they obtain information about individuals and recommends that statisticians obtain permission to write an article based on their research at the time they are engaged.