SMABS 2004 Jena University
SMABS 2004 Home Organization About Jena Sponsors Links Imprint / Contact SMABS Home

European Association of Methodology

Department of methodology and evaluation research

Jena University

Contributions: Abstract

Standard statistical analyses of data from quasi-experiments will generally yield biased results since they do not take into account the non-equivalence of the treatment and control groups. In the past several methods have been proposed to correct for this non-equivalence and to reduce the risk of wrong causal conclusions.
A first class of methods is based on Rubin's causal model and uses the propensity score to control for pre-existing differences between control and treatment groups. Theoretical and applied research has shown that these methods, which equate control and treatment groups on the propensity score by either matching, stratification or covariance adjustment, perform quite well provided treatment assignment is ignorable given the covariates that are used to estimate the propensity score.
A broad second class of models encompasses different approaches that try to relax the assumption of ignorable treatment assignment. Most of the models within this class are adaptations of models for analysing nonignorable nonresponse in surveys and experiments and fall in two subcategories: selection models and pattern mixture models. Models of this type make very strong assumptions about the treatment assignment mechanism and may not be very robust against violations of these assumptions.
In this presentation both class of models will be compared in terms of their underlying assumptions and will be evaluated with respect to their applicability in social and behavioural science research. The results of relevant simulation studies will be presented.