Analysis of Bayesian quasi-experimental designs in geographical contexts to evaluate policy e ectiveness

Thumbnail Image
Issue Date
Journal Title
Journal ISSN
Volume Title
Investigating if a governmental policy implemented in a certain region has the desired e ect may be a complex task. Creating randomized controlled trials to evaluate the policy in an experimental setting is not possible. For this matter, Regression Discontinuity (RD) design has been around for a while as an implementation of quasi-experimental designs. This research provides an evaluation of the Bayesian Non-parametric Quasi-experimental Design (BNQD) framework proposed by Hinne, Van Gerven & Ambrogioni (2019) [1], in geographical settings speci cally. BNQD is a Bayesian implementation of RD that uses Gaussian Processes to perform quasi-experimental design. The design will be compared to other methods used by researchers in the eld of Geographical Regression Discontinuity designs using simulations of data. In order to be able to evaluate the e ectiveness of a policy change in one region, the model should be able to capture both spatial data and di erent time points simultaneously. Combining the two sources of data in a novel framework may rise di culties, conditions or assumptions that have to be validated in order to achieve a new valid design. The details of the framework are outlined as well as the relation with other quasi-experimental designs. Finally, the framework is applied to a real-life data set of ammonia emissions in the Netherlands to evaluate the far stricter policy applied in Noord-Brabant and Limburg aimed at restricting ammonia emissions from livestock farming. This application shows both the ease with which the framework can be applied and that the emission reduction policy did not have the expected e ect.
Faculteit der Sociale Wetenschappen