Dissertations, Theses, and Capstone Projects

Date of Degree

6-2024

Document Type

Dissertation

Degree Name

Ph.D.

Program

Educational Psychology

Advisor

David Rindskopf

Committee Members

Bixi Zhang

Patricia J. Brooks

Wei Wang

Teresa M. Ober

Subject Categories

Cognitive Science | Quantitative Psychology

Keywords

Bayesian, meta-analysis, processing speed, autism, language disorder

Abstract

Meta-analysis is the systematic review and quantitative synthesis of specific areas in literature and is an important quantitative tool for researchers interested in synthesizing a particular body of research. The current research used meta-analysis to investigate processing speed in two neurodevelopmental disorders. This dissertation consisted of four meta-analytic papers. The first paper was a meta-analysis that synthesized a large body of research on processing speed, measured via reaction time (RT) measures, in groups of individuals diagnosed with autism spectrum disorder (ASD) versus age-matched neurotypical comparison groups. This research was motivated by two previous meta-analyses in the literature on processing speed in ASD arriving at discrepant results, as the authors used different statistical methodologies and systematic review criteria. The findings from the first meta-analysis demonstrated generalized slowing across processing speed domains in groups of individuals diagnosed with ASD. Generalized slowing was also investigated in a separate meta-analytic paper on RT-based tasks in groups of individuals diagnosed with Developmental Language Disorder (DLD) versus age-matched neurotypical comparison groups. This second meta-analysis in this dissertation was motivated by the need for a literature synthesis in a field of research on processing speed in DLD dating back almost four decades. Random-effects meta-analysis with Robust Variance Estimation (RE RVE) was used in both papers, as both meta-analytic datasets included a level of complexity attributed to effect size dependencies. The results from both the ASD and DLD meta-analyses have implications on the cognitive processing profile, the diagnostic screening, and the potential interventions for each disorder.

The third meta-analysis in this dissertation critiqued the frequentist RE RVE analysis approach of the two previously published meta-analyses and aimed to provide a Bayesian re-analysis of both datasets. In frequentist meta-analysis, researchers commonly report an effect size estimate and a heterogeneity estimate, along with their confidence intervals (CI), but cannot make probabilistic statements regarding the degree of belief that these parameters take on a certain value, and statistical significance testing using classical p-value comparison to a š¯›¼ level is heavily relied on to make conclusions. Rather than being able to state that the calculated CI has a 95% chance of containing the parameter of interest within that interval, frequentists must interpret the CI such that, if the meta-analysis was completed repeatedly, then 95% of the CIs will contain the true parameter value. The Bayesian approach allows researchers to express a degree of belief regarding parameters of interest, and ultimately allows for a more intuitive interpretation of meta-analytic results. The Bayesian framework also allows researchers to include prior information about parameters and generate probabilistic statements from posterior distributions of parameters (e.g., P[Ī¼ > 0]), which is nonsensical in the frequentist framework. The current re-analyses were conducted by utilizing a hierarchical RE Bayesian meta-analytic model to account for effect sizes that were nested within studies. The Bayesian re-analysis aimed to showcase the statistical methodology that should be adopted by researchers performing meta-analyses on complex datasets who desire to escape binary null hypothesis significance testing and interpret results in a more intuitive manner.

The fourth and final meta-analytic paper applied the Bayesian methodology to investigate the potential diagnostic overlap between ASD and DLD in terms of processing speed deficits. Previous literature has suggested that ASD and DLD share diagnostic features. The results of the comparative Bayesian meta-analysis suggested that groups of individuals with ASD and groups of individuals with DLD both exhibit processing speed deficits relative to their age-matched neurotypical comparisons. However, more heterogeneity was observed in the ASD meta-analytic dataset, potentially due to the classification of the disorder as a spectrum and the variety in diagnostic profiles across autistic individuals.

Download

Share

COinS