Evaluation Methods for Non- Experimental Data

This paper presents a review of non-experimental methods for the evaluation of social programmes. We consider matching and selection methods and analyse each for cross-section, repeated crosssection and longitudinal data. The methods are assessed drawing on evidence from labour market programmes in the UK and in the US. JEL classification: J38, H3, C2. I. AN OVERVIEW OF THE EVALUATION PROBLEM The evaluation problem of concern here is the measurement of the impact of a policy reform or intervention — for example, a childcare subsidy or a targeted training programme — on a set of well-defined outcome variables. For the former example intervention, the outcome variables might include the child’s exam results or the mother’s labour market participation, while, for the latter, they could include individual employment durations, earnings and/or unemployment durations. Usually, individuals are identified by some observable type — for example, gender, age, education, location or marital status. The evaluation problem, therefore, is to measure the impact of the programme on *University College London and Institute for Fiscal Studies. The authors thank Costas Meghir, Barbara Sianesi, John Van Reenen, the editor and a referee for helpful comments. This review was originally prepared for the Department for Education and Employment and is also part of the programme of research of the Economic and Social Research Council (ESRC) Centre for the Microeconomic Analysis of Fiscal Policy at the Institute for Fiscal Studies. Co-funding from the Leverhulme Trust is gratefully acknowledged. The second author acknowledges financial support from Sub-Programa Ciência e Tecnologia do Segundo Quadro Comunitário de Apoio, grant number PRAXIS XXI/BD/11413/97. The usual disclaimer applies.