Establishing Measurement Invariance of the Woodcock-Johnson Subtests for Deaf and Hard-of-Hearing Children

• In Event: Language and Literacy Descriptions and Interventions From the Center on Literacy and Deafness

Abstract

Purpose
This study investigate the factorial validity and measurement invariance of the Woodcock-Johnson Tests of Achievement-III, Letter-Word Identification (WJ-LWID), Passage Comprehension (WJ-PC), and Picture Vocabulary (WJ-PV) for deaf and hard-of-hearing (DHH) children using a large sample of young DHH children.
Theoretical Framework
Literacy development in DHH children may vary as a function of their ability to access spoken language and communication modality, thus measurement of literacy for DHH children should differ depending on whether they have functional hearing (via cochlear implant or hearing aids) or whether they use sign language, spoken language, or both. Furthermore, the WJs are developed for hearing children and each of the subtests are developed under unidimensionality assumption. Thus, it is imperative to investigate whether the WJs’ items scores obtained from DHH children are valid indicators of the construct being measured for DHH children and whether the items have equivalent functionality across the subpopulations of the DHH children to support its utility for DHH population.
Method/Techniques
We conducted confirmatory factor analyses (CFA for binary items or item response theory, i.e., IRT) and multiple-group IRT via Mplus with least square estimation with mean- and variance-adjusted chi-square statistics.
Data Source
Responses from 576 DHH children in prekindergarten through second grade were used in the initial study with WJ-LWID (Mage = 72.2 months, SDage = 16.4 months; 51% boys). We conducted a 2-parameter IRT model on the first 30 items on the WJ-LWID because more than 80% of the children reached ceiling after item 30, resulting in too little item variance for multiple-group modeling. The CFA were followed by multiple-group IRT tests for differences among the three DHH groups, signing only (n = 150), spoken only (n = 222), and bimodal group (n = 204).
Results
The model fit for the unidmensional model was excellent across the groups; χ2 ranging from 478.6 to 613.9, RMSEA ranging from .01 to .05, and CFI ranging from .97 to 1.00. All factor loadings were significant with standardized loadings ranging from .40 to .99, indicating that the items on the WJ-LWID are valid indicators of the construct being measure across the groups. Test of measurement invariance revealed that the WJ-LWID items have equal functionality across the subgroups of DHH children (see Table 1).
Scientific Significance of the Study
Our initial analyses shows that items on the WJ-LWID reflect one underlying ability as stated by the test developers and that the items function similarly across the three subgroups of DHH children. Establishing measurement invariance across the groups is a prerequisite when comparing group performances and examining patterns of relations with scores on other theoretically related measures. The current study will conduct the same modeling approach to the WJ-PC and WJ-PV items. This study provides important psychometric information of these three WJ subtests for DHH children and its utility across the DHH subgroups. Thereby, the study will address important substantive research questions that support for construct validity of the scores on these measures.

Authors

• Mi-Young Lee Webb, Georgia State University

• Lee Branum-Martin, Georgia State University

• Amy R. Lederberg, Georgia State University