Causality, Impartiality and Evidence-Based Policy - Mechanism and Causality in Biology and Economics - page 211

It is an empirical question to be solved on a case-by-case basis if disguising
randomisation as a lottery influences participation. Banerjee and Duflo certainly
acknowledge that even fair lotteries can provoke a self-selection depending on the
way they are presented: if the participants in the control group are told that the
experimental treatment will be available to them in the future (once the resources
are gathered), this may affect their willingness to participate or their compliance.
In addition, organising a lottery to distribute aid seems to be politically contro-
versial for governments that are expected to serve an entire population (Duflo et al.
2007 , p. 21).
Hiding randomisation altogether from participants seems a more effective strat-
egy. As Banerjee and Duflo observe, 'ethics committees typically grant an exemp-
tion from full disclosure until the endline survey is completed, at least when the fact
of being studied in the control group does not present any risk to the subject' ( 2009 ,
p. 20). Participants in the experimental group will not know how they got involved,
and those in the control group may never know they have been excluded. If the
latter live in different villages, as it often happens in trials run in developing
countries, they may not get to know about the experimental treatment. In this
way a totally different scenario arises: in order to avoid a self-selection bias, we
deceive the participants about the comparative structure of the experiment. The
experimenters are assuming here that participants only care taking part in a lottery,
but, as a matter of fact, they may also have preferences about the treatments tested.
They may want to get one rather than the other. Or, if they understand the nature of
the experiment, they may even have a favourite treatment that they want to see
succeed - e.g. we may well imagine parents preferring direct allocations of cash to
send their kids to school rather than paid meals. If these preferences exist, disguis-
ing randomisation will only succeed to the extent that the disguise is successful: the
participants have been 'blinded' to the comparison, but shall we just assume that
such blinding is successful?
There is some evidence that deception in medical trials can fail. Patients have
preferences about treatments, and they usually neither understand nor like
randomisation (Featherstone and Donovan 2002 ): their compliance is usually
explained by their lack of alternatives to get access to experimental treatments,
they would not get the medication outside the trial. And they play by the research
protocol only to a point: they try to find out which treatment they are receiving (and if
they succeed, this has an effect on the experiment). However, in most medical trials,
the researchers have means to make patients comply with the research protocol,
e.g. they may mask the treatments well enough for an ordinary participant not to be
able to distinguish them. They would need a laboratory. Whether they have access to
a laboratory often depends on the social organisation of the patients. The testing of
early anti-AIDS treatments in the USA, documented by Epstein ( 1996 ), illustrates
this point: the participants wanted to have experimental treatments and not placebos,
so they resorted to all sort of strategies to make sure they would receive the
treatment, drawing on their connections in the gay activism networks. Many
abstained from taking part in trials if they didn't think the drug was promising
enough (in order to remain 'clean' and thus eligible for other tests); those who