Tuesday, June 16, 2009

The ethics of randomized trials

Since the infamous social experiments of Milgram, the ethics of human subjects has become a hot point for researchers. By hot point I mean that we all moan and groan at having to go through the tedious application process of the institutional review board (IRB) that oversees all human subjects research.

Lest we think though that unethical use of human subjects is really just a problem of the past, a recent conflict over some work Jared Diamond did in Papua New Guinea should remind us all that this is still a serious issue.

Most researchers know that they should be careful with the people they are interviewing, but I find it strange that there is little (i.e. none that I can find) debate in the social sciences about the ethics of randomized evaluations. The benefits should by now be obvious to everyone, but why has there been no discussion about the costs?

Randomized evaluations have their roots in medical science. After 100 years of use though, there is still a debate among doctors about the ethics of randomizing. I see the debate as falling into two competing camps: the utilitarian (do the most good to the greatest number) and the Kantian (do not use people as a means to an end), with the Kantians winning in medicine. 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 even declared that “concern for the interests of the subject must always prevail over the interests of science and society.”

There is also plenty of debate today still about the usefulness of randomized trials in medicine, and in statistics in general, but I am only aware of Dani Rodrick's excellent dissention for economists.

For me, oversubscription, while theoretically fine, is often used as a cruch. Sometimes programs purposefully look for oversubscription, but describe it as spontaneous in order to tell a more human story. Some researchers even suggest purposefully looking for it as a way to avoid legal troubles. Encouraging more people to apply though can reduce the average neediness of beneficiaries, especially if it is easier to reach better-off groups.

But why is oversubscription actually ethical? We would have to believe that everyone in the program deserves it equally, but that is rarely the case. Identifying the needy may not be easy, but we can do it, oftentimes at very little additional programing cost.

And when was the last time you heard of someone invoking the stopping rule for a social experiment?

There are potential solutions. Phasing in programs so that the control remains without the program for only a year or so is a way to be less heartless, but then it means asking people to wait for the program.

Of course, randomized experiments do have some positive effects. Most importantly, they can be used as a transparency tool, especially in programs that are already beset by corruption problems. Few can argue that a coin has been unfair to them, but the head of a program in a country beset by corruption is normally assumed to be biased.

I am a big fan of randomization. It is the best tool we have to answer questions that can at first seem incredibly difficult. But it is not the only way to evaluate a program or a hypothesis. I think a good debate about the ethics of randomization would serve both the public and the researchers in the end.

1 comment:

grant said...

Agreed. As a fan of RCT, I use it as the least worst method often.

However, I've got issues with randomizing targeting of the intervention. In conflict zones there are usually places and sub-populations in greater need of an intervention (often to the point of where absence of the intervention would result in grave harms). This means you can't really randomize at site or individual units of measurement. And given that you aren't testing the relative efficacy of two different types of intervention, its seems unethical to randomize and deny needier beneficiaries goods in the name of methods. I think this only applies in conflict zones and complex emergencies, but is a serious challenge to figuring out how RCT can be incorporated into humanitarian policy.