Regression discontinuity approaches
Like DD methods, regression discontinuity methods involve a pretest–posttest group comparison.
Groups, however, are assigned not by the programme or policy but rather by the analyst based on an
observed covariate that is expected to be impacted by the programme or policy, and that has values
on either side of a critical fixed threshold (e.g. the poverty line, 0.08 blood alcohol content level, or
possession of 1 ounce of marijuana). Observations in the data are assigned to groups based on the
cut-off score of the measured covariate. Observations below the cut-off are assigned to one group,
and observations above the cut-off are assigned to the other group. The effect of the programme or
policy is estimated by the disconnect in the regression lines (slopes) or functions (for higher order
polynomials) obtained from separate models estimated on each of these groups below and above
the threshold value. Unlike DD, regression discontinuity does not provide an estimate of the average
effect but rather identifies whether a difference in average behaviour exists at the critical threshold
value. Thus, it can identify if a programme or policy had an effect, but it cannot provide a good
estimate of the average effect of the programme or policy on the populations as a whole.