CMStatistics 2016: Start Registration
View Submission - CMStatistics
B1066
Title: Proxy variables in a matching analysis of data from the Swedish social insurance agency Authors:  Philip Fowler - Department of Statistics, USBE, Umea University (Sweden) [presenting]
Xavier de Luna - Umea University (Sweden)
Per Johansson - Uppsala University (Sweden)
Petra Ornstein - The Swedish Social Insurance Agency (Sweden)
Sofia Bill - The Swedish Social Insurance Agency (Sweden)
Peje Bengtsson - The Swedish Social Insurance Agency (Sweden)
Abstract: An enhanced cooperation between the Public Employment Service (PES) and the Social Insurance Agency (SIA) in Sweden was implemented in 2012. The target group of the joint efforts were individuals identified in need of support in order to regain work ability. Such individuals partook in a joint assessment, i.e. meetings with both aforementioned agencies, with the aim to assess the individuals' work abilities. The idea being that such an assessment could lead to a better rehabilitation plan and thus quicker reintroduction to the labour market. To evaluate this, we perform a matching analysis on data from PES, SIA and Statistics Sweden in order to reduce covariate differences between the treatment and control groups, that otherwise could bias the results of the study. Furthermore, a prediction of the individuals' duration in sick leave without intervention was made by case workers at the Social Insurance Agency. This prediction is used as a proxy variable for potential unmeasured confounders in our analysis. Data on the outcome for the treated individuals was purposely not available to the researchers in the matching process and did thus not influence the choice of matching.