When measuring the effect of a program or policy, the program or policy is referred to as a treatment and the causal impact as the treatment effect. The primary challenge to measuring impacts is selection bias, meaning when units (e.g., people, places, firms, and schools) select into certain levels of treatment, we can never be sure that they do not differ systematically from other units who do not.

The ability of a method to avoid this kind of bias is often called its internal validity. Experimental methods typically have the greatest internal validity because, when treatment is randomly assigned, a simple descriptive method would not suffer from selection bias. Methods that attempt to measure the impact of one treatment, corresponding to some hypothetical experimental design, are called quasi-experimental methods.