Causality, Impartiality and Evidence-Based Policy - Mechanism and Causality in Biology and Economics - page 210

Pr D ¼

Pr D ð ¼ 1 j c Þ ¼ ð 1 j c ; p Þ:

In other words, we need to assume either that:
1. Randomisation does not influence participation, or
2. If it does influence participation, the effect is the same for all the potential
participants, or
3. If the effect is different, it does not influence their decision to take part in the
programme.

Heckman's main objection is that randomisation tends to eliminate risk-averse
persons. This is only acceptable if risk aversion is an irrelevant trait for the outcome
under investigation - i.e. it does not feature in C . However, even if irrelevant, it
compels experimenters to deal with bigger pools of potential participants in order to
meet the desired sample size, so the exclusion of risk-averse subjects does not
disrupt recruitment. But bigger pools may affect in turn the quality of the experi-
ment, if it implies higher costs. One way or another, argues Heckman, rando-
misation is not neutral regarding the results of the experiment.
Heckman's analysis is causal: randomisation can create a self-selection bias
distorting the sample of participants on which any inference should rest. We are
going to argue that it is impossible to correct this self-selection bias without putting
in question the impartiality of the trial. The threat of partiality does not come in this
case from the researchers but from the participants themselves. In RFEs,
participants may have their own preferences about the compared treatments, and
the risk aversion elicited by randomisation is just one of them: people may prefer to
make choices about treatments. In order to preserve randomisation and to correct
self-selection biases, Duflo and her coauthors try to control the participants'
preferences by blinding, i.e. by disguising or hiding the randomised nature of the
experiment. We argue that these attempts assume the indifference of the
participants regarding the experimental outcome. If the participants have strong
preferences about the outcome, masking randomisation will not be enough to
enforce the experimental protocol.
According to Banerjee and Duflo ( 2009 ), we can avoid the self-selection bias if
we either disguise or hide randomisation. Both solutions are feasible in many
programmes, at least if we conduct the experiment in a developing country. As to
the former, randomisation can be disguised as a lottery by which the scarce
resources of the programme are allocated. If the potential participants perceive
this lottery as fair, it may not dissuade them from taking part in it. The fairness of
lotteries as allocating procedures can be certainly defended on theoretical grounds
(Stone 2007 ), and we know that there is empirical evidence about the acceptability
of unequal outcomes when they come from a lottery perceived as fair (Bolton et al.
2005 ). However, not everybody likes lotteries, even fair ones: for instance, surveys
show that people oppose the use of lotteries by colleges and universities in order to
choose which students are admitted (Carnevale et al. 2003 ).