Abstract

Depending on the discipline, “impact evaluation” can refer to quite different problems with accordingly very different approaches. On the downside we have consequently a confusion about methods, definitions, and notations, even about the meaning of ‘evidence’, ‘assessment’ or ‘method’. On the upside we have therefore a huge bunch of methods and approaches which partly complement, partly overlap, and sometimes compete with each other. Therefore, as long as the objective of the evaluation is essentially the same, we could take advantage of this diversity by combining methods. Our goal is to offer a mostly data-driven procedure that guides the practitioner from the proper choice of indicators and causality model up to the final significant tests of particular treatment effects. We demonstrate our idea along a specific evaluation example ‘from start to finish’, and subsequently discuss general strategies for cases where data are not based on pre-designed experiments. The intention is to minimize the impact of subjective judgements of the evaluators on the final result. Graph theory and nonparametric statistics can play important roles here.