The Influence of Parceling on the Implied Factor Structure of Multidimensional Item Response Data

Public Deposited

Analytics

Last Modified

Creator

Magnus, Brooke Erin
- Affiliation: College of Arts and Sciences, Department of Psychology and Neuroscience

Abstract

Parceling is a method researchers often use to circumvent issues that arise in handling item-level data; however, the degree to which the true factor structure is preserved after parceling remains ambiguous in the literature. The goal of this thesis was to examine the effects of parceling on the implied factor structure of multidimensional data using both simulation and analytic techniques: does the estimated factor change after parceling? This question was addressed across three studies. Item covariance matrices were computed from bifactor models comprising continuous or dichotomous item responses. The item covariance matrices were then parceled and a one-factor confirmatory factor analysis was fit to the parcel covariance matrices. Additionally, a simulation was carried out in which factor scores from the CFA were compared with the latent variable values from the generating model. Results of both studies suggest that parceling does change the estimated factor. Furthermore, fit statistics overwhelmingly indicate good fit despite a misspecified model. Finally, to illustrate how parceling is used in practice, an application using empirical data is shown. Practical implications are discussed.

Date of publication