Date of Award


Degree Type


Degree Name

Master of Science in Statistics


Computer Science and Statistics

First Advisor

Gavino Puggioni


Drug overdose deaths in the United States have increased at an alarming rate over the last two decades prompting a declaration of a public health emergency in 2017. Drug and opioid addiction have deeply negative social and economic effects on families and communities causing a significant strain on federal and local resources which are needed to fight the epidemic. The United States now has the unwanted designation as a country with the highest drug overdose death rates among the high-income nations. For these many reasons, more research should be focused on this health problem in order to advise policy-making for the most effective treatments and mitigation measures. This thesis explores data for Connecticut and Rhode Island collected by the Medical Examiner’s Office on drug overdose fatalities using the Bayesian spatiotemporal statistical model in order to assess the socio-economic factors which may impact overdose death risk, and to estimate overdose death risk at township level.

Bayesian hierarchical models, and more specifically conditional autoregressive models are widely accepted as the most appropriate modelling options for areal data. Analysis of areal data over a specific time period raises the underlying problem of residual spatio-temporal autocorrelation which can remain after accounting for covariate effects. For these reasons, we use the spatially autocorrelated autoregressive time series model which allows for residual spatio-temporal autocorrelation and provides effect estimates of different socio-economic factors on the drug overdose death risk at the township level.

Our analysis shows that higher overdose death risk is associated with areas where greater proportion of the population has only a high school diploma or equivalent and where greater proportion of the population is living below the poverty line. Lower overdose death risk is associated with townships where median home values are higher. Estimates of the relative risk provided by our autoregressive CAR model point to consistently elevated risk in townships of Hartford, Newington, New Britain, New Haven, Norwich and Waterbury in Connecticut, and Providence and Woonsocket in Rhode Island. These townships share a history of having enjoyed economic prosperity built on manufacturing in the first half of the twentieth century followed by decline. This common trait should be explored further to understand fully why drug dependence disorders are more prevalent in such communities. Lastly, the findings presented in this thesis work do not indicate causal inference since they are made at aggregate town-level data. In order to validate the relationships presented between drug overdose death risk and socio-economic factors, individual-level data should be analyzed.

Available for download on Wednesday, May 04, 2022