Logo image
Back
Propensity scores for spatial, temporal, and spatiotemporal data
Dissertation   Open access

Propensity scores for spatial, temporal, and spatiotemporal data

Sarukkalige Shanika Achini De Silva
Doctor of Philosophy (Ph.D.), Drexel University
Dec 2022
DOI:
https://doi.org/10.17918/00001427
pdf
De-Silva_Sarukkalige_20224.55 MBDownloadView
PDF Open Access Open Access (License Unspecified)

Abstract

Dynamics--Mathematical models Propensity scores Spatiotemporal data
Propensity score analyses allow researchers to estimate treatment, intervention, or exposure effects by mimicking the characteristics of a randomized controlled trial. However, traditional methods rely on strong assumptions that often do not hold for data collected over space and/or time. When these assumptions are violated, estimates of treatment effects will be biased. Of particular interest to this study is the assumption of "no unmeasured confounders". In this dissertation, we investigate and present extensions of the traditional propensity score matching algorithm to accommodate spatial, temporal, and spatiotemporal data, respectively. We leverage the properties of spatially and temporally structured data to recover the unconfoundedness assumption within a Bayesian framework. We study the operating characteristics of the proposed models under different settings in a series of simulation studies. We observe that accounting for space or temporality when the observational data is spatial, temporal, or spatiotemporal can help recover the true treatment effect. Finally, we apply the proposed methods to examine the relationships between county-level exposure to PM2.5 and county-level preterm birth rates in the US between 2003 and 2006.

Metrics

58 File views/ downloads
68 Record Views

Details

Logo image