Random Assignment

Conceptual Overview and Discussion

Random assignment to treatment has three important and related inferential implications. First, everyone in the study has an equal probability of receiving the treatment. In the medical example, this means that every subject in the painkiller study has the same odds of receiving the placebo or the treatment as every other subject. Second, because everyone has an equal probability of receiving the treatment, after assignment of treatment there should be no statistically significant differences between the two (or more groups). In other words, those who receive the minor pain treatment should differ from those who receive placebos only in that they were randomly selected to receive the treatment. Although individuals will obviously differ, the important point is that the groups do not differ significantly. Third, as a result of this equivalency between groups any differences observed between the two groups can plausibly be attributed to the treatment.

Random assignment is effective because it overcomes selection effects and unobserved heterogeneity. As an example, assume a government department wanted to test the effects of two different income assistance programs on how long individuals remain on social assistance. If the depart ment wanted to make strong causal claims about the program, it would randomly assign some recipients to receive Program A while the others received Program B. Any observed differences at the end of the study would probably be the result of the differences in the programs (provided the experiment was properly administered and designed). By contrast, if the government department had allowed individuals to self-select into one program or another, differences in outcome could be a result of systematic differences between the two groups. For example, the factors that lead an individual to choose one program and not the other may also be what explain his or her difference in reliance on social assistance. Such unobserved differences and/or selection effects make program evaluation significantly more difficult.

Random assignment can take several different forms. The three most common are (1) simple, (2) blocked, and (3) matched. In simple random assignment, subjects (which could be individuals, schools, cities, etc.) are individually assigned to treatment or control. In a blocked randomization, subjects are first grouped together according to some characteristic of interest; for example, a researcher who was interested in the effects of a school voucher program on subjects from different ethnic backgrounds may first group or block subjects according to their observed ethnicity. Randomization would then occur within the group. Blocking is typically used when a researcher wants to determine whether a treatment has different effects on different groups (rather than a constant causal effect). Blocked treatments can lead to more precise or efficient causal estimates. Matched randomization is a special case of blocked randomization in which individual subjects are matched (most often in pairs) according to a matching algorithm that attempts to match subjects who are as similar as possible. Randomization then occurs between matched subjects. As with blocking, this results in more efficient or precise estimates of causal effect. The drawback of these methods is that they are often more difficult to implement.

...