Randomized experiments are ubiquitous in field research and social evaluation design. In addition to the numerous advantages they present for researchers (see previous post), randomized trials in the social sciences can also create special ethical problems. To my knowledge though, social science researchers have not openly discussed ethical issues.
To me, the most important is the problem of over-subscription. Encouraging programs to have over-subscription is often seen as an ethical solution for situations where randomization is not immediately feasible. However, this is not necessarily an ethical solution if, as is likely, many individuals are more in need of the program than others.
Even when a researcher is faced with equipoise (unsureness about the outcome of a program), encouraging more people to apply may then not only produce unrealistic expectations, but also reduce the average neediness of beneficiaries, especially if it is easier to reach better-off groups.
Given the possiblity of such a situation, it is important to remember that patients who participate in medicine trials expect that they are not to be sacrificed for the sake of the trial. The World Medical Association in 1964 declared that “concern for the interests of the subject must always prevail over the interests of science and society.”
Thus, it is the Kantian requirement of do no harm that is paramount, not the utilitarian argument of most good to the most people. While some may argue that social programs are not as harmful as medical trials, there is still no excuse to sacrifice individuals for the sake of a study.
Of course, even given all of their advantages, randomized controlled trials are not the only way to do science. As Royall (1991) notes, “science desires randomized clinical trials, it does not demand them”.
Since the invention of the randomized trial there has been debate among doctors and statisticians on the preferencing of randomized trials over other methods, as well as debate regarding the ability of randomized trials to really solve all of the problems they were designed to solve (Urbach (1985) makes a Bayesian argument against randomiuzed trials).
From a social optimization perspective, there may be other solutions than a randomized trial. For instance, when we know unequivocally that the effect is positive, we can avoid random study for other trials. This is equally important for social programs where resources are very constrained, and so need to do their best to reach the very needy.
Even in situations where we are not 100% sure, randomized trials may not be necessary.
I am worried that the current trend to preference randomization could lead to a mis(over)use of the method. It is the researcher’s responsibility to prove that a randomized experiment is necessary; we should not assume this is so.