Date of Degree

6-2022

Document Type

Dissertation

Degree Name

Ph.D.

Program

Psychology

Advisor

Bertram O. Ploog

Committee Members

Daniel M. Fienup

Robert Lanson

Subject Categories

Data Science | Experimental Analysis of Behavior

Keywords

visual inspection, visual analysis, functional measurement, signal detection theory, machine learning

Abstract

In behavior analysis, data are usually analyzed using visual analysis of the graphed data. There are a wide range of methods used to visually analyze data, from a basic ‘textbook’ style approach to the use of visual aids, decision-rubrics, and computer-based approaches. In the literature, there have been some comparisons of the efficacy of different approaches. Visual analysis as a behavior can be taught using a variety of methods, independent of how the skill itself is to be performed. Teaching methods include lecture, online instruction, and equivalence-based instruction. There is not much research on the teaching of visual analysis specifically, though there are a wide range of behavioral teaching approaches to choose from. Finally, there are a variety of methods for assessing visual analysis, from interrater reliability to different measures of accuracy, to signal detection theory. Advantages and disadvantages of assessment methods, as well as additional assessment methods from other areas of psychological and behavior analytic research are discussed. The present study used an adapted form of multiple-exemplar training to train naïve participants how to visually analyze graphs. Different aspects of the training were systematically manipulated to examine their effects. The naïve participants’ performance was compared to that of machine learning algorithms which were trained using similar methods, as well as to the performance of experts. Participants’ visual analysis decisions are discussed in terms of interrater reliability, different measures of accuracy, signal detection theory, and functional measurement.

Share

COinS