The second approach is to use either naturally occurring experiments (such as when some people are enrolled in a program and others are not) or clever econometrics techniques that manage to detect the impact of treatments even though no one deliberately designed the situation for that purpose.
