Title Information
Type of Resource (primo): dissertations
Name: Personal
Name: Personal
Name: Personal
Name: Corporate
Origin Information
Physical Description
Note: thesis: Thesis (Sc. M.)--Brown University, 2025
Genre (aat): theses
Abstract: Fractional outcomes with excess zeros and ones are a type of data with increasing attention across various disciplines. The dataset used in this study was obtained from a longitudinal assessment of the financial burden on colon or rectal cancer patients undergoing curative-intent treatment. Due to the nature of our data, the evaluation of Work Productivity and Activity Impairment (WPAI) among patients exhibits an unusually high frequency of observations at zero and one. This study evaluates three statistical frameworks for modeling zero-one inflated data, namely logistic regression, beta regression, and zero-one inflated beta regression. Beta regression is applied using two different approaches, (1) transformation of the outcome variable and (2) exclusion of excess zeros and ones. We conduct 1,000 simulations to assess model performance by examining bias in estimated coefficients, mean squared error (MSE), and rejection rates. Additionally, hyperparameters such as the percentage of excess boundary data and sample size are tuned to determine the most effective model for different scenarios. Performance measures indicate that logistic regression produces highly biased coefficient estimates and the highest MSE across all models. Zero-one inflated beta regression separately models the continuous and discrete parts of the data; consequently, the coefficients estimated under this model are equivalent to those of beta regression with excess zeros and ones excluded. Among the models, beta regression with transformed outcome data demonstrates the least bias in coefficient estimation and the lowest MSE. For outcome variables with 40% or less inflation at zero and one, beta regression with transformed data yields the best result for analyzing this unique data characteristic. However, this method has limitations when applied to highly inflated data with large sample sizes > 2000. Logistic regression consistently yields the worst results among the three models across all simulation scenarios. Zero-one inflated beta regression shows no significant advantage over beta regression, as its estimates for the continuous portion align with those of standard beta regression.
Subject
Subject
Subject (fast) (authorityURI="http://id.worldcat.org/fast", valueURI="http://id.worldcat.org/fast/01432090")
Subject
Subject
Language
Record Information