Causal Inference of Spectral Discontinuities using Gaussian Processes A Bayesian Non-parametric Method for Spectral Analysis in Quasi-Experimental Design
Keywords
Loading...
Authors
Issue Date
2020-07-10
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Quasi-experimental designs are the next best inference technique when a randomized control trial
is not available. Despite their popularity, common quasi-experimental frameworks have exclusively
considered the time-domain representation of a signal. In this paper, we propose a new method for
inferring spectral discontinuities in quasi-experimental designs. By using a Gaussian Process model
with spectral kernels, a
exible method for inferring discontinuities in the periodic features of a
signal is obtained. To measure the average treatment e ect, two di erent e ect sizes are proposed.
The consistency of the method is shown by applying it to simulated periodic data with a known
discontinuity. Lastly, the method is applied to real-world examples by determining the e ect of
Russia's 2006 alcohol policy on the monthly suicides, as well as inferring spectral discontinuities in a
heart rate signal.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen