Using internal replication to establish a treatment effect

In many cases assignment to a treatment may affect concomitant variables. I show how a concomitant variable can be used to corroborate evidence from an observational study. In the observational study two types of training programs are compared. One program is part of regular Swedish labor market training while the other program was run by Swedish industry during 1998-2000. A large and positive effect on employment is found from this latter program. In this program it was much easier to get employer contact than in the regular program. From a survey I have information about employer contacts in the two programs. I find the same positive effect on employment from employer contacts in either program and no effects from the new program when conditioning on employer contacts. I interpret this as a causal effect on employment from employer contacts. In addition, this effect is found to be more pronounced for individuals with a weak position in the labor market.

Keyword: Evaluation; Active labor market training; Information technology; Employment rate; Propensity score matching; Internal replication
JEL: C14, C52, J68