Racial prejudice refers to how some people feel and think about blacks, racial discrimination to how they treat them. Plainly, the public opinion survey is well-suited for the study of prejudice; but consensually, it is not suited for the study of discrimination, since by definition an opinion survey only records opinions, not behaviors.

We want to urge a less self-denying ordinance, a slight relaxation of the definition of discrimination. Our interest is politics, so by racial discrimination we shall mean a person honoring a claim for government assistance for a white but refusing to honor exactly the same claim, made on exactly the same grounds, for a black.

Chapter 13 introduces a technique for the assessment of racial discrimination, so defined. As with Chapters 11 and 12, the technique capitalizes on computer-assisted interviewing. The key experiment – the “laid-off worker” experiment – makes particularly plain how our introduction of planned variations has burst the constraints of the traditional paper-and-pencil split-ballot technique. In the laidoff worker experiment five attributes of an unemployed worker are randomly varied: race, gender, age, marital–parental status, and work history. Because each is varied independently of the other, there are ninety-six different combinations in all, far more than could possibly be accommodated in the traditional technique.

One word of caution. Chapter 13 reports initial results. We have a high degree of confidence in them, but they do not represent our complete analysis of discrimination. In particular, it would be unwarranted – indeed, flatly wrong – to infer from the results of the laid-off worker experiment that racial discrimination no longer occurs. […]