Application of Generalized Item Response Tree Models in Social Work Research

Constructs in social work research and practice such as depression and anxiety are often measured with Likert-type scales. The response categories in Likert-type scales are typically coded in ascending order, and researchers proposed various ordinal models, such as ordinal item response theory models, rating scale models, and graded response models, to analyze Likert-type scale item response data. It is important to acknowledge that these models assume that the response categories are ordinal.

However, this assumption may not be true, if the response categories show qualitative differences rather than ranking or intensity differences. For example, a middle response category such as "Undecided" may not be part of a true order with the other response categories, such as agree or disagree. Such response categories always involve internal cognitive or psychological decision process.

A tree structure will allow for a partial ordering of response categories and thus account for the decision process. With a tree structure, the generalized item response tree model (GIRT) could describe a postulated internal decision process, which is composed of sub-trees and their corresponding internal nodes and branches. The tree continues to diverge through branches until it reaches leaves which are the terminal nodes that represent the observed categorical item responses.

The GIRT is especially relevant to social work research, since it can capture the nuanced differences in latent traits in psychological and educational assessment more accurately by accounting for the decision process. Given the lack of both methodological and empirical studies based on the GIRT in social work research, this workshop will discuss the steps in building a GIRT, strategies for model appraisal, and reporting guidelines based on real data.

Objectives

Upon completion of the workshop, participants will be able to:

1. recognize how the GIRT is promising in latent variable modeling,

2. demonstrate understanding of the steps of model building for GIRT, and

3. use the flirt package in R to fit a GIRT.

Content

Based on the pedagogical techniques of learning-science-by-doing-science and problem-based learning, this workshop aims to address the following content:

1. summary of the characteristics and advantages of GIRT,

2. a concise and systematic set of procedures to establish a GIRT,

3. application of GIRT in social work research using a problem example based on real data,

4. a step-by-step demonstration of model construction, estimation, visualization, and diagnostics using the flirt package, and

5. how to draw inferences and conclusions.

Implications

Overall, this focus on the generalized item response tree model can be invaluable to social work researchers, faculty, students, and professionals who wish to leverage state-of-the-art item response models to derive more rigorous inferences and conclusions. Featuring real data analyzed with a cutting-edge statistical package, this workshop simplifies esoteric item response models making them more applicable to social work research. By becoming familiar with the best practices in applying the generalized item response tree model, attendees can gain the expertise necessary to use a powerful statistical tool that can help enhance the examination of latent traits and advance research in social work.