Investigating Methods for Calculating Model-Based Reliability When Analyzing Categorical Data

• In Event: 34.094 - Fairmont Royal York Roundtable Session Six
  In Roundtable Session: 34.094-4 - Finer Details in Modeling Latent Interactions, Moderated Mediation, and Multicategory Indicators

Abstract

A Monte Carlo simulation study is proposed to examine different methods for estimating reliability within the SEM framework. More specifically, reliability was estimated using Cronbach’s alpha, alpha for ordinal variables, McDonald’s omega with categorical data analyzed by weighted least squares mean- and variance adjusted estimator, and omega calculated with parceled data and the robust Maximum Likelihood estimator. Using a 2-factor CFA, the following conditions are examined: response categories (2, 3, and 5), model specification (correctly, misspecified), indicator distributions (symmetric, asymmetric), loading values (.50 , .80), and sample size (250, 500, 1000). The goal of the proposal is to guide applied SEM researchers interested in estimating reliability with ordinal data.

Authors

• Kelvin T. Pompey, University of South Carolina

• Christine DiStefano, University of South Carolina