Targeting labour market programmes — results from a randomized experiment
- Civil servants
- Job seekers
- Earnings and income
- Technology adoption
- Digital and mobile
In many countries, there has recently been an increased interest in targeting labor market programs to unemployed persons who would benefit the most from them. Researchers evaluated the impact of providing a statistical targeting system to case workers in government employment agencies on their choice of labor market programs for jobseekers. Results showed that caseworkers largely ignored the statistical information provided by the targeting system, and that offering this information did not have any impact on their choice of program.
Active labor market programs, which provide job-search counseling, subsidized employment, training, work practice, or further education, have often had moderate or even negative impacts when evaluated in Europe. These results suggest the importance of targeting programs to the jobseekers who are most likely to benefit from them. Providing information on which programs are most likely to help a given job seeker find work may help caseworkers match job seekers to the most effective programs. Researchers partnered with the Swiss State Secretariat for Economic Affairs to evaluate the impact of providing customized information on the best labor market programs for each client on caseworkers’ decisions when assigning job seekers to programs.
Context of the evaluation
This evaluation took place in five different regions of Switzerland: Basel, Bern, St. Gallen, Zürich, and Geneva. In the 1990s, Switzerland experienced an increase in unemployment to previously unseen levels of over five percent. In response, the government introduced mandatory active labor market programs to provide training and work experience to the unemployed. Caseworkers in government employment agencies used their own discretion to allocate job seekers to different programs, and non-compliance would lead to the suspension of a job seeker’s unemployment benefits. While some caseworkers focused on helping jobseekers gain employment quickly, others focused on helping job seekers find a good match. In an effort to improve the targeting of these programs, in 2005 the government decided to pilot an online software that provided caseworkers with concrete recommendations on which programs were more likely to help individual job seekers.
Details of the intervention
Researchers partnered with the Swiss State Secretariat for Economic Affairs to test the impact of the Statistically Assisted Programme Selection (SAPS) software on case workers’ choices of labor market programs. The study was implemented in 21 employment offices from May to December 2005. In each of these employment offices, around half of the regular caseworkers were randomly assigned to receive access to the prediction technology (18,713 caseworkers) while the other half served as the comparison group (16,677 caseworkers).
Caseworkers in the treatment group received a one-hour training session on the SAPS software and could download an individual report for each of the job seekers they were counseling. Each report predicted the number of months job seekers would spend in stable employment (3 months or more) after participating in each of the available programs, and ranked the programs into three groups: significantly best, intermediate, and worst programs. Caseworkers were encouraged to choose a program from the significantly best category. However, the caseworkers maintained full control over which program they assigned to each client.
Researchers used administrative data from Switzerland’s unemployment insurance system and the social security and pensions system to measure whether or not caseworkers chose the programs that had the best predicted outcomes according to the prediction technology.
Results and policy lessons
Providing information on the predicted effectiveness of different employment programs had no effect on caseworkers’ choices of programs for job seekers. Caseworkers who had access to additional information on the most effective programs for each job seeker were not more likely to assign job seekers to the most recommended employment program compared to caseworkers who did not receive this additional information.
Caseworkers’ compliance with the statistical recommendations was low or nonexistent, meaning that for most jobseekers the SAPS-recommended program did not coincide with the first program they were assigned by a caseworker within the subsequent 90 days. In both the treatment and the comparison group, 29 percent were assigned a program from the set of significantly best ones, while only 12 percent of jobseekers were assigned the most recommended program.
Researchers believe that caseworkers ignored these predictions either because they trusted their own judgements more or because they were unwilling to comply with predictions made by an external tool which they thought could restrict their future autonomy. For nearly two-thirds of jobseekers, their caseworker never analyzed their prediction report.
Overall, these findings suggest that leaving the allocation of a program to caseworkers’ discretion may hinder the potential of statistical targeting systems. One way through which future programs could improve compliance is by offering rewards to caseworkers who comply with the system’s recommendations.