INTRODUCTION

The approach to causal inference outlined in the first chapter has important antecedents in the literature. In this chapter we review some of these antecedents to put the potential outcomes approach in perspective. The two most important early developments, in quick succession in the 1920s, are the introduction of potential outcomes in randomized experiments by Neyman (Neyman, 1923, translated and reprinted in Neyman, 1990), and the introduction of randomization as the “reasoned basis” for inference by Fisher (Fisher 1935, p. 14).

Once introduced, the basic idea that causal effects are the comparisons of potential outcomes may seem so obvious that one might expect it to be a long-established tenet of scientific thought. Yet, although the seeds of the idea can be traced back at least to the eighteenth century, the formal notation for potential outcomes was not introduced until 1923 by Neyman. Even then, however, the concept of potential outcomes was used exclusively in the context of randomized experiments, not in observational studies. The same statisticians, analyzing both experimental and observational data with the goal of inferring causal effects, would regularly use the notation of potential outcomes in experimental studies but switch to a notation purely in terms of realized and observed outcomes for observational studies. It is only more recently, starting in the early seventies with the work of Donald Rubin (1974), that the language and reasoning of potential outcomes was put front and center in observational study settings, and it took another quarter century before it found widespread acceptance as a natural way to define and assess causal effects, irrespective of the setting.

Moreover, before the twentieth century there appears to have been only limited awareness of the concept of the assignment mechanism. Although by the 1930s randomized experiments were firmly established in some areas of scientific investigation, notably in agricultural experiments, there was no formal statement for a general assignment mechanism and, moreover, not even formal arguments in favor of randomization until Fisher (1925).