J. Heckman et al., Substitution and dropout bias in social experiments: A study of an influential social experiment, Q J ECON, 115(2), 2000, pp. 651-694
This paper considers the interpretation of evidence from social experiments
when persons randomized out of a program being evaluated have good substit
utes for it, and when persons randomized into a program drop out to pursue
better alternatives. Using data from an experimental evaluation of a classr
oom training program, we document the empirical importance of control group
substitution and treatment group dropping out. Evidence that one program i
s ineffective relative to close substitutes is not evidence that the type o
f service provided by all of the programs is ineffective, although that is
the way experimental evidence is often interpreted.