Der Arbeitsbereich Methoden und Evaluation/Qualitätssicherung von Prof. Dr. Steffi Pohl lädt herzlich ein zum

Gastvortrag von Prof. Dr. Peter Steiner (University of Wisconsin - Madison, USA)
Multilevel Matching Strategies for Observational Data

am Donnerstag, den 2. Juni 2016  von 16:00 - 18:00 Uhr
in der Freie Universität - Habelschwerdter Allee 45 - Raum KL 24/122d

Zum Vortrag:
Propensity score (PS) techniques for matching treated and control units are frequently used for estimating causal treatment effects. With observational multilevel data, where selection into treatment takes place at the lowest level (for instance, students select into treatment and control conditions within schools), different matching strategies are possible. We discuss the strengths and weaknesses of three strategies for matching level-one units (students): (i) within-cluster matching where matches are only allowed within clusters (schools), (ii) unrestricted across-cluster matching where treatment and control units are matched across all clusters, and (iii) across-cluster matching within homogeneous groups of clusters. We define group homogeneity with regard to the similarity of clusters’ selection processes or outcome models. Latent group memberships are estimated via multilevel mixture modeling. In addition to the matching strategies, we also address covariate selection issues for removing confounding bias.

Alle Interessierten sind herzlich zum Vortrag eingeladen. Die Teilnahme ist kostenlos und eine Voranmeldung nicht erforderlich.