Workshop B: Propensity score matching

Tuesday afternoon, June 14, 13.30-17.00 CET

This workshop will introduce how propensity score matching methods can be used to address cause and effect questions in empirical research.

Causal research questions aim to isolate effects of a treatment, independent of other effects and differences between the treated and untreated (i.e., selection bias). The most straightforward way to answer causal research questions is to conduct randomized trials like in pharmaceutical studies, for instance. In many fields including education, randomized trials are however very difficult to conduct, due to practical, ethical, or financial reasons, among others. Therefore, educational research often has to rely on observational data, that is, information that stems from pure observations of educational processes and outcomes without an interference of the researchers. This workshop will address how propensity score matching methods can be utilized for causal inference. The basic aim of these methods is to make treatment and control groups as comparable as possible to remove selection bias. Ideally, this allows to interpret outcome differences between control and treatment groups as causal effects.

The workshop will address propensity score matching methods from both a theoretical and an applied perspective. This workshop will

  • revisit Rubin’s potential outcome framework, a model that formalizes cause and effect questions, and the issue of selection bias,
  • introduce different propensity score matching (e.g., nearest-neighbour matching, optimal matching) and balance check methods (e.g., love plots),
  • address the estimation of effects, and
  • discuss the central advantages of propensity score matching over other methods.

Example studies will help to understand the assumptions and prerequisites behind propensity score matching and to evaluate its scope critically. Demonstrations in R will help to showcase how the methods can be applied in statistical software. The workshop materials and recommended readings will be provided after the course.

Target group and previous knowledge requirements

The workshop is an introduction to the topic of propensity score matching and is primarily targeted at researchers who have not yet worked with the method themselves. For more experienced participants, further literature will be provided. Basic knowledge of causal inference is welcome but not a prerequisite. The R demonstrations are for illustrative purposes and will be designed so that participants without R experience can follow as well. Laptops and software are not necessary.

Workshop leader

The workshop will be led by Dr. Isa Steinmann, Centre for Educational Measurement at the University of Oslo (CEMO), Norway

Photo of Isa Steinmann