Document Type
Dissertation - Open Access
Award Date
2021
Degree Name
Doctor of Philosophy (PhD)
Department / School
Mathematics and Statistics
First Advisor
Christopher Saunders
Abstract
Many scholars have proposed the use of a Bayes factor to quantify the weight of forensic evidence. However, due to the complex and high-dimensional nature of pattern evidence, likelihood functions are intractable and thus, Bayes factors cannot be assigned using traditional methods. Approximate Bayesian Computation (ABC) model selection algorithms provide likelihood-free methods to assign Bayes factors. ABC Bayes factors leverage the use of the scoring functions commonly used in recent years in forensic statistics in a rigorous statistical manner. However, traditional methods for assigning ABC Bayes factors are subject of several criticisms. In this dissertation, one of the main criticisms of traditional ABC Bayes factors is alleviated by deriving a relationship between ABC Bayes factors and ROC curves. Additionally, the use of the ROC curve allows for an intuitive communication of the ABC Bayes factor. A simple example is outlined to illustrate the implementation of a ROC-ABC algorithm. Asymptotic properties of the ROC-ABC Bayes factor are explored. The ROC-ABC algorithm is implemented to quantify the weight of fingerprint evidence.
Library of Congress Subject Headings
Bayesian statistical decision theory.
Forensic sciences -- Statistical methods.
Forensic statistics.
Evidence, Criminal.
Receiver operating characteristic curves.
Number of Pages
180
Publisher
South Dakota State University
Recommended Citation
Hendricks, Jessie, "Development and Properties of the ROC-ABC Bayes Factor for the Quantification of the Weight of Forensic Evidence" (2021). Electronic Theses and Dissertations. 228.
https://openprairie.sdstate.edu/etd2/228
Included in
Rights Statement
