The Bank wants to determine whether certain activities cause an alleviation of poverty. The key word here is “cause,” to verify which requires constructing an unbiased counterfactual. To this end, one can measure a group before and after an intervention, which is rarely useful because it assumes that there is no statistical regression, that no seasonal or maturational changes would otherwise have occurred, that no historical events taking place during the same period could have caused the same outcome. However, the basic approach is sometimes useful if it is extended to create an interrupted time series—that is, many measures made over time on the same population both before and after the intervention, at all times using the same measure as the effect of interest. Then it is possible to estimate the preintervention trend and test whether there are any subsequent, reliable deviations from it. However, this design also requires that regression and cyclical maturational patterns do not masquerade as treatment effects, that the intervention has a sudden onset, and that the delay period within which an effect can be expected is known. These are difficult conditions to meet. Moreover, interrupted time series clearly benefit enormously from studying a control population during the same period that does not get the intervention but that does experience the same measures, with the control group being selected to maximize its multivariate similarity to the intervention group.