It is one of the dimensions of parliamentary oversight. In this dimension, the parliament is engaged in a preventative evaluation of governmental conduct that precedes the formal adoption by the government of a decision, action or public policy. Parliaments scrutinise governmental proposals or draft documents to address future policy choices and exercise an influence over the sphere of action of the executive.

For further details, see: Parliamentary oversight

To estimate the effect of a policy, the individuals treated and those not treated should have the same initial characteristics, so that any changes in the result variable can be attributed solely to the public policy. When the initial characteristic are different, however, a selection bias occurs. We speak of bias because it prevents the use, to estimate the effect, of a comparison between the average result variable for those treated and the average for those not treated. The bias is caused by the selection because it depends on differences in the characteristics of treated and the non-treated individuals which existed before the public policy. These influence the allocation of individuals to the treated group or the non-treated group. These influence whether individuals are allocated to the treated group or the non-treated group.

Statistical matching is a method used to define a control group ex post byselecting, from the individuals excluded from the treatment (non-treated), those most similar (controls) to the individuals admitted to the treatment (treated), on the basis of observable characteristics. If the matching is carried out correctly, the effect of the treatment can be calculated simply as the difference between the average of the result variable for the treated individuals and the matched non-treated individuals. This represents an estimate of the counterfactual.

Statistical matching is said to use observations that have a common support, since it is conducted with individuals from the two most comparable groups on the basis of observable characteristics.

If there are a high number of characteristics to be considered, it can be particularly difficult to find the most similar individuals. To resolve this problem, a propensity score is generally used, i.e. the probability of an individual being treated, given his or her observable characteristics. It has been demonstrated that the essential condition for the elimination of selection bias on the basis of observable characteristics is met even if these characteristics are encapsulated by the propensity score alone.