Assignment vs. choice: lessons from training vouchers

For instrumental-variable estimation using binary instruments, we offer simple methods to obtain policy-relevant insights beyond local average treatment effects. We demonstrate the methods by evaluating an element of choice introduced into active labour market policy in Germany by a reform in 2003. Instead of being assigned to training courses by caseworkers, unemployed job seekers receive vouchers allowing them to choose among approved courses. Our data record the receipt of a voucher and the participation in a course. We use exogenous variation in treatment probabilities across locally integrated labour markets as an instrument. Applying our theoretical methods, we can make statements on the distribution of treatment effects and thereby infer the priorities of caseworkers. We further examine the consequences of choice for policy effectiveness and finally explore the distributions of gains or losses from choice.