Preface

In the Analects of Confucius, it is written that "He who sets to work upon a different strand destroys the whole fabric."* In most civil rights cases, the evidence brought to bear on disputed facts consists of the testimony of witnesses who have observed the events in question or who have knowledge of the surrounding circumstances. Such testimony-which is the fabric of the typical case that comes to court-can be understood and analyzed by jurors, judges, and attorneys who lack specialized training or esoteric knowledge.

In recent years, however, many of those participating in the litigation and resolution of cases alleging illegal discrimination have set to work on a different strand. The new strand being woven into the traditional fabric is statistical inference. When expert witnesses testify as to what quantitative data show about the relationships among variables, they draw on a rich tradition of statistical thought unknown to most other participants in litigation. Confronted with such evidence, the courts have remarked that "statistics are not irrefutable," International Brotherhood of Teamsters v. United States, 431 U.S. 324 (1977), that they "come in infinite variety," id., and that to be relied upon, "[s] tatistical evidence ... must be meaningful," Lewis v. NLRB, 750 F.2d 1266, (5th Cir. 1985), quoting Pouncy v. Prudential Life Insurance Co., 669 F.2d 795, 802 (5th Cir. 1982). Given the limitations on the enthusiasm of attorneys and courts for quantitative proof, it seems doubtful that the new strand of proof in discrimination cases threatens to destroy the whole fabric.

Nevertheless, statisticians and other experts appearing in discrimination litigation have a professional responsibility to use methods that are well adapted to the problems to which they are applied, and to choose from the "infinite variety" of statistics those that are most "meaningful." This volume is intended to aid in this process. Too often, we fear, statistical methods developed in other contexts have been applied, without sufficient deliberation or refinement, to legal problems. Thus, this book has three major purposes: to describe the more or less standard methods being brought into court, to identify some of the difficulties that can arise with these methods in this context, and to suggest at least some directions for enhancing the usefulness of statistical analysis in these cases.

Chapters 1-4 sketch some of the legal doctrines that underlie discrimination litigation. Chapter 1 outlines important features of the Equal Protection Clause of the Fourteenth Amendment to the United States Constitution, and it identifies the points at which statistical evidence may help detect unconstitutional discrimination at work in such areas as the selection of jurors and the meting out of capital sentences to persons convicted of murder.

Chapter 2 looks in more detail at the use of statistical analysis to show racial or other prohibited discrimination in jury selection. It is in this area that the United States Supreme Court first recognized that statistical proof of discrimination could be compelling, and it is an area in which some courts perform their own hypothesis tests. Because of the unusual degree of acceptance of formal statistical methods in jury discrimination cases, this chapter describes the process by which jurors are selected to serve on grand and petit juries, and it considers the distinct statistics that may be employed to measure the degree of discrimination, the tests that may be performed on these statistics, and the role of these tests in deciding whether discrimination exists.

Chapter 3 explains the provisions of Title VII of the Civil Rights Act of 1964. These provisions construing the most important prohibition against discrimination in the work force, and claims brought under this act probably account for the largest number of instances in which statistical evidence of discrimination is adduced. In constituting this act, the courts have erected intricate structures, with shifting burdens of proof, for establishing the existence of discrimination, and federal administrators have promulgated regulations on what procedures may be followed in selecting employees. Understanding the elements of Title VII and their relation to the administrative guidelines is important to an appreciation of the role that the statistical proof can play in Title VII litigation.

Chapter 4 focuses on a fundamental issue in Title VII cases alleging discrimination in hiring. Any statistical analysis must begin with a conception of the population from which employees are drawn. Often, the most conceptually satisfying definition cannot be used in practice, and an operationally viable alternative must be found. In such cases, the analyst will be guided by the rules that the courts establish for defining this population. This chapter therefore articulates the legal principles involved in defining the relevant market from which an employer hires.

Chapters 5-9 describe and probe frequently seen statistical methods. The technique most commonly used to understand how one variable is related to several others probably is multiple regression analysis. Some judicial opinions seem to say that regression is a simple, ideal device for examining such relationships, while other courts suggest that regression models are too esoteric and capture too little of the litigants' world to be of much value. Chapter 5, therefore, provides a broad introduction to this technique and its pertinence to discrimination cases, and mentions some of the cautions that should attend its use.

Chapter 6 deals with one aspect of the form that a regression model should take. It clarifies two competing definitions of fairness and argues that in most applications "reverse regression" will be unreasonable.

Chapter 7 identifies a related issue in regression studies of employment discrimination. It demonstrates that in cases where less than fully reliable proxy measures of productivity are used, linear regression analysis inevitably produces biased and misleading assessments of discriminatory effects. This alarming conclusion points to the need for more sophisticated data analysis techniques in discrimination litigation.

Chapter 8 looks at the procedures for validating job skills tests and at how regression can shed light on the fairness of such employee selection and promotion procedures. This chapter addresses the competing psychometric definitions of fairness and the companion issue of establishing reasonably valid and reliable measures of qualities or achievements that an employer should consider in selecting and rewarding employees.

Chapter 9 is a wide-ranging essay inspired by much of the preceding material. It argues that the historical development of statistical methods has not been oriented toward resolving legal problems and that further development of the emerging discipline of legal statistics is warranted. Seeking to put the methods and views of earlier chapters in perspective and to indicate some of their limitations, it considers a variety of specific techniques in moderate detail.

Most chapters presuppose that the reader is conversant with statistics but not fully initiated into the mysteries of the legal profession. We hope that statisticians and other experts will find this book helpful in at least two respects: in orienting themselves to the lawyer's world of discrimination litigation, and in applying statistical methods that are truly appropriate for assisting the trier of fact who must decide whether a claim of discrimination is justified.

At the same time, we hope that this work also will give attorneys and courts some guidance in what they can and should expect from experts who undertake statistical studies of discrimination. The essays here reveal that although it surely is possible and desirable to enumerate many procedures that must be followed for a particular method to give good results, a great many methodological questions remain open. Thus, we hope that this volume not only will aid the courts in recognizing such matters as when a regression analysis is incomplete or superficial, but that it also will lead them to remain open to intelligent variations or improvements on traditional methods. In our view, the courts must become sensitive to the elements of good statistical work, but they should be wary of dictating a "common law" of statistical practice.

Author: Kaye