Measuring Racial Discrimination (review)

Discrimination has long been a fascinating but frustrating subject for social science. It is fascinating because it represents a powerful mechanism underlying many historical and contemporary patterns of inequality; frustrating because it is elusive and difficult to measure. The impressive new volume from the National Research Council, written by a team of interdisciplinary experts (chaired by Rebecca Blank), takes stock of the existing approaches to measuring racial discrimination, guiding readers through the possibilities and potential pitfalls of a wide range of available alternatives. Although focused on racial discrimination (primarily against African Americans), the insights of this book can be readily applied to the study of discrimination more broadly (e.g., by gender, national origin, age, etc.). Indeed, this volume is likely to become required reading for all those interested in the subject of discrimination, broadly defined.

The book is divided into three sections. Part one focuses on concepts, with definitions of race, discrimination, and theories of discrimination. A majority of research on discrimination (and, correspondingly, a majority of the approaches discussed in this book) focuses on discrimination in one context at one time. The authors encourage researchers to begin thinking more dynamically. Here, and in a final chapter, they urge the development of models of discrimination capable of going beyond the individual level, and/or at single points in time, to instead consider the organizational/structural processes that constitute forms of racial discrimination as well as the cumulative features of discrimination across multiple domains and over time.

Part two represents the heart of the volume. After a brief overview of the problems of causality and counterfactuals, the following three chapters explore the major approaches to measuring racial discrimination — including experimental methods, large-scale statistical analyses, and attitudinal and behavioral indicators. For each method, the authors present a thorough overview of the key design features, highlighting the various strengths and weaknesses, primarily with respect to the problems of causal inference and generalizability. These chapters are written at a high level of sophistication, but central points are explained clearly and will be readily accessible to a wide readership of researchers and students.

The first of the three methods chapters focuses on experimental approaches, which offer a clear view of causal mechanisms by randomly assigning subjects to treatment conditions and holding constant all other influences. These studies are considered the "gold standard" for establishing causal effects; their key limitation, however, is the problem of limited (unknown) generalizability from lab settings and undergraduate research subjects to the real settings in which discrimination emerges. Field experiments, used most often to study discrimination in housing and labor markets, relax certain controls over environmental influences in order to better simulate real-world interactions. While retaining the key experimental features of matching and random assignment important for assertions of causality, this approach relies on real contexts (e.g., actual real estate markets or employment searches) for its staged measurement techniques. Field experiments, however, are often costly and difficult to implement, and can only be used for selective decision-points (i.e., hiring decisions but not promotions).

The next chapter in this section discusses statistical analyses of secondary datasets. Such analyses have advantages of large numbers, allowing researchers to take into account of wide range of influences and interactions. The primary limitations of this approach, however, are threats to causality: omitted variable bias (failing to control for a variable related to both race and the outcome variable) and sample selection (when certain segments of the relevant population are over- (under-) represented in the sample). This chapter is the most technical of the volume, including discussions of statistical techniques for strengthening causal inference such as instrumental variables, propensity scores and matching, panel data methods, and natural experiments.

The last of the methods chapters deals with attitudinal indicators from surveys and interviews. Questions measuring prejudice and other attitudinal indicators of race relations, as well as individuals' own experiences with discrimination, can provide a valuable source of information. This chapter presents an overview of the major sources of survey data, discusses the various measurement problems...