Web Survey Bibliography

Title Complications When Using Nonrandomized Job Training Data to Draw Causal Inferences
Source 55th ISI Session 2005, Invited Paper Session on "Inferential Potentials of Non-Probability Samples"
Year 2005
Access date 24.11.2005
Full text doc (47k)

Immense reform efforts is going on in . In particular, the so called Hartz reforms aim to improve the efficiency of the Federal Employment Agency (BA), and its local departments. Amongst other things this requires measures for the efficiency of the large variety of labor market policy programs financed by the Agency. In the Hartz reforms many of the evaluation projects are based on the IEB data source. The IEB (Integrierte Erwerbsbiografien) merges data from different data sources. The current version includes process data from the following administrative records:
1.      BeH: Employment spells from social security data records,
2.      LeH: Unemployment spells for registered unemployed persons receiving unemployment insurance payments (Arbeitslosengeld), stately financed unemployment assistance (Arbeitslosenhilfe),
3.      MTG: Spells of participation in labor market programs, and
4.      ASU: Search spells for people registered as unemployed and searching for a job.
Since the assignment to any of the possible treatments typically is not based on a random process, complications and corrections have to be considered before drawing causal inference based on these data. The particular question we are confronted with is: Which kind of treatment is the most effective and efficient one for each person registered as unemployed? This requires knowledge about potential outcomes – how would each individual’s employment history had evolved under different programs?
Generally, approaches to predict the outcomes of several labor market programs are known as targeting systems. In this talk, we will first describe the magnitude of the problem and then previous attempts to address it. This includes two systems that have been suggested for a use as Targeting Systems throughout during the last years. According to complications with the existing approaches a new treatment effect and prediction system, called TrEffeR, will be proposed. TrEffeR analyzes, which program should be assigned to a certain unemployed person and what causal effect a program will have for his future employment history.

Access/Direct link Conference homepage
Year of publication2005
Bibliographic typeConferences, workshops, tutorials, presentations