Validating a Pictorial Vocabulary Size Test via the 3PL-IRT Model
Wen-Ta Tseng
National Taiwan Normal University
doi: http://dx.doi.org/10.7820/vli.v02.1.tseng
Download this article (pdf)

Abstract
The paper presented a newly conceived vocabulary size test based on
pictorial cues: Pictorial Vocabulary Size Test (PVST). A model-based
(1-2-3 parameter logistic item response theory model comparisons)
approach was taken to check which model could absorb the most
information from the data. Junior high school and primary school
students participated in the study (N = 1,354). Subjects’ ability estimates
and item parameter estimates were computed based on expected a
posteriori (EAP) method, one type of Bayesian method. BILOG-MG 3
was adopted to execute parameter estimates and model comparisons. The
results showed that the 3PL-IRT model best fit the empirical data. It was
then argued that test takers’ English vocabulary size could be best
captured under the 3PL-IRT model, as not only the discrimination
parameter, but also the guessing parameter has a fundamental role to
play in consideration of the test format adopted in the PVST. The article
concluded that the PVST could have positive washback effects on test
development and English vocabulary instruction.

Citation
Tseng, W.-T. (2013). Validating a pictorial vocabulary size test via the 3PL-IRT model. Vocabulary Learning and Instruction, 2 (1), 64-73. doi: 10.7820/vli.v02.1.tseng