Dissertation
Bayesian hierarchical growth curve methods with applications in linguistics and audiological sciences
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Spring 2024
DOI: 10.25820/etd.007410
Abstract
This dissertation focuses on the prospects of Bayesian hierarchical modeling as an alternative to traditional statistical modeling techniques in the field of hearing science. In this field, there is a continued motivation to link how hearing, and by extension hearing loss, is associated with the development of cognitive, communicative, and academic abilities.
In one aspect, Bayesian hierarchical modeling is explored in the paradigm of psychophysics and psychometrics as applied to hearing science. In these fields, an integral method to the disciplines involves charting how a person’s response pattern changes according to a continuum of stimuli. For instance, in hearing science, Visual Analogue Scaling (VAS) tasks are experiments in which listeners hear sounds across a speech continuum and give a numeric rating between 0 and 100 conveying whether the sound they heard was more like word “a” or more like word “b” (i.e., each participant is giving a continuous categorization response (CCR)). By taking all the CCRs across the speech continuum, a parametric curve model can be fit to the data and used to analyze any individual’s response pattern by speech continuum. Standard statistical modeling techniques are not able to accommodate all of the specific requirements needed to analyze these data. Thus, Bayesian hierarchical modeling techniques are proposed in order to accommodate group-level non-linear curves, individual-specific non-linear curves, continuum-level random effects, and a subject-specific variance that is predicted by other model parameters. These techniques are employed both in a cross-sectional setting analyzing a VAS study consisting of mono-lingual and bi-lingual participants and in a longitudinal setting exploring the Growing Words project, a 4-year longitudinal study tracking reading and language development in children in the Iowa City / Cedar Rapids area. Overall, the model results generally highlighted the negative association between individual slope and variance, suggesting those persons who had a more swift transition in response across the speech continuum were more consistent in their responses across trials.
In another aspect, Bayesian modeling techniques are explored in the context of hearing status affecting academic outcomes. This is motivated by a study of adolescents across primary school with various degrees of hearing loss whom were observed in order to map the potential linkage between hearing difficulty and academic outcomes. As part of the study, participants were evaluated across multiple testing items meant to assess reading comprehension and word reading ability. While this study lends itself well to an item response theory (IRT) modeling approach, standard implementations of IRT modeling are unable to handle all the statistical considerations of the study, which includes, correlated latent abilities, participants being measured across time, testing items being applied at different rates across time, and a non-linear latent growth trend over time. Thus, a Bayesian hierarchical model using the two-parameter logistic function as a framework is proposed which can adequately accomplish all the goals of the study. Overall, negative associations between reading, both general comprehension and word-level ability, and better ear pure tone average were witnessed.
Details
- Title: Subtitle
- Bayesian hierarchical growth curve methods with applications in linguistics and audiological sciences
- Creators
- Eldon Sorensen
- Contributors
- Jacob Oleson (Advisor)Grant Brown (Committee Member)Elizabeth Walker (Committee Member)Daniel Sewell (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Biostatistics
- Date degree season
- Spring 2024
- Publisher
- University of Iowa
- DOI
- 10.25820/etd.007410
- Number of pages
- xv, 140 pages
- Copyright
- Copyright 2024 Eldon Sorensen
- Language
- English
- Date submitted
- 04/10/2024
- Description illustrations
- illustrations, tables, graphs
- Description bibliographic
- Includes bibliographical references (pages 137-140).
- Public Abstract (ETD)
- In hearing science, there is a continued interest in how hearing is tied to cognitive, communicative, and academic ability. However, experiments looking at this relationship are particularly involved. Therefore, we aim to demonstrate how Bayesian modeling techniques can be used as an alternative to traditional methods that can accomplish the research goals of present and future studies. In one application, we look at how people respond to different levels of a speech stimuli. Specifically, by using numeric values given by individuals who are rating how much they think the stimuli sounds more like word ”a” vs. word ”b”, our modeling approach fits appropriate 4-parameter logistic growth curve functions. By using a function defined by a small, definite number of parameters, our methodology can identify how people uniquely respond to each type of stimuli. In addition, our research highlights that the approach is generalizable to different types of studies by being able to use different types of curve functions. In another scenario, our research focuses on adolescents with various degrees of hearing loss who, while going through primary school, were tested on reading comprehension and word reading skills. Due to participants being measured across time and having to go through multiple different testing items, traditional modeling techniques were unable to handle all aspects of the study. However, through a model built upon the two-parameter logistic function, we were able to effectively account for all the data complications while identifying relationships between hearing loss and reading skills. Overall, as hearing loss increased, reading ability was found to decrease. Our research also looked at associations between reading ability and other explanatory variables of interest, including hearing aid access and pure tone average.
- Academic Unit
- Biostatistics
- Record Identifier
- 9984647257302771
Metrics
4 File views/ downloads
26 Record Views