An experiment conducted in St Louis provided 1404 girls aged between 14 and 19 with free contraceptive advice and free long-acting contraceptive devices ... The comparison group was that of similar girls in the rest of the United States who were not included in the experiment.
The researchers found that, by giving both free advice and free contraceptives, the rates of pregnancy, live birth and induced abortion declined precipitously by comparison with national averages. ... On the face of it, the experiment was a triumphant success.
Until you look a little closer at how the experimental group was "selected":
The young girls were drawn from those who resided in St Louis or sought contraceptive services in selected community clinics, and furthermore had no desire for pregnancy for at least 12 months. These girls, then, were a self-selected group...
BOTTOM LINE: the observed different between the groups is the sum of the "treatment" effect (the effect of the contraception) and the "selection" effect (that girls less likely to get pregnant were in the experimental group). This is why randomized experiments are so much easier to interpret.