Simulating or Imputing Non-participant Intervention Durations Using a Flexible Semi-Parametric Model

Abstract

In the evaluation of labour market training programs using matching, evaluators must decide when to start comparing participant outcomes against non-participant outcomes. Measurement relative to an intervention period permits the separation of training opportunity costs from possible benefits, but an equivalent period for the comparison group must be determined. One method imputes the timing of the intervention for comparisons from that of the participant match. However, with Propensity Score Matching, this may produce biased outcome estimates. Instead, the authors develop and apply semi-parametric duration models using Human Resources and Social Development Canada data to simulate the positioning and duration of the intervention for non-participants.