Date of Graduation

7-2020

Document Type

Thesis

Degree Name

Master of Science in Statistics and Analytics (MS)

Degree Level

Graduate

Department

Statistics and Analytics

Advisor/Mentor

Robinson, Samantha E.

Committee Member

Datta, Jyotishka

Second Committee Member

Chakraborty, Avishek A.

Third Committee Member

Petris, Giovanni G.

Keywords

Differential Item Functioning; Item Response Theory; Model-based Recursive Partitioning; Perceived Stress Scale

Abstract

When an item on a test functions differently for subgroups of respondents with respect to an exogenous variable (or covariate) after conditioning on the latent variable of interest, the item is said to exhibit Differential Item Functioning (DIF). The 10-item Perceived Stress Scale (PSS10) is administered to respondents via MTurk to quantify “perceived stress” and identify if items on the scale function differently for specific subgroups defined by age, sex, race, marital status, number of children, employment status and social media usage.

The purpose of this study was to compare traditional DIF detection approaches (Mantel-Haenszel, logistic regression, likelihood ratio test and Raju’s area measure) with Rasch trees (a model-based recursive partitioning approach to DIF detection). The former are methods that can only compare two subgroups and so require continuous covariates to be arbitrarily dichotomized so the subgroups can be pre-specified; with the latter approach, subgroups do not need to be pre-specified and as a result, DIF can be detected in combinations of covariates.

Current results corroborate that while traditional methods are able to detect DIF in items based upon sex, race, employment status and social media usage, the Rasch tree approach is able to detect DIF resulting from a combination of employment and social media covariates.

Download

Share

COinS