Defining and Identifying the Effect of Treatment on the Treated

One rôle of labour economics is to evaluate the impact of government initiatives, such as employment and education schemes, on economic indicators such as income, the purpose being to inform the introduction of future schemes and policy changes: Should more adult training programmes be funded? Should education be made compulsory until the age of 18? Evaluating the effect of such policies is far from straightforward, most especially on account of the fundamental problem of self-selection (Heckman, 1979). Because of self-selection it is typically unclear whether, and to what extent, changes in economic indicators can be attributed to government programmes where participation is voluntary, since individuals that take part in such programmes (i.e., the self-selected) tend to be more motivated and receive higher incomes, irrespective of participation. A similar problem emerges in epidemiologic contexts. A recent ruling in the US (Okie, 2006) gives terminal cancer patients the right to be treated with experimental (Phase I) drugs. This means that patients can self-select themselves into the treatment group, without randomisation. Data on response from this group of patients will not yield reliable estimates of the average causal effect (ACE) as estimates will be confounded by the patients’ attitude and health. We will use the following two examples, one from labour economics and another from epidemiology, where effect evaluation is hampered by unknown selection criteria, to illustrate aspects of the methodology we develop in this paper.