A COMPARATIVE ANALYSIS OF QUALITY ASSESSMENT MODELS OF SCIENCE INSTRUCTIONAL MANAGEMENT: AN APPLICATION OF MULTILEVEL VALUE-ADDED MODEL WITH DIFFERENTIAL ITEM FUNCTIONING AND TEST FUNCTIONING ANALYSES
Abstract:
The purposes of this research were to 1) examine the differential item functioning (DIF) and differential test functioning (DTF) of dichotomous and polytomously scored science literacy items, 2) compare the efficiency of the two value-added models between the model without the inclusion of differential item functioning and differential test functioning (Model1) and the model with the inclusion of differential item functioning and differential test functioning (Model 2). For the Model 2, DIF and DTF analyses were used to identify flawed items and subsequently items identified as DIF or differential step functioning (DSF) were deleted and not used in the value-added model. The efficiency of the models was evaluated by the correlation between value-added values provided by Model1 and Model2 and the three control variables: socioeconomic status, home educational resources, and family wealth, the consistency between rankings of value-added values from the two models and the consistency of school classifications based on the value-added values from the two models. Data used in this study were the secondary data of Thailands 2009 student literacy assessment which were from Programme for International Students Assessment (PISA). Data were consisted of 9 booklets and responses of 4,292 students in 230 schools throughout Thailand were used. The major research findings were found as follows: 1) The differential item functioning analysis indicated that there were items identified as DIF items favoring females (3 items), tutoring students beyond science classrooms (1 item), low-socioeconomic status students (2 items), students with high home educational resources (5 items), and students from high wealth family (2 items). The reliability coefficient from Model 2 decreased as the DIF items were deleted. Differential test functioning (DTF) analysis of the data for the two booklets showed that the magnitude of DTF decreased as DIF items were deleted, while the magnitudes of DTF were unchanged for other seven booklets. After some DIF items in the Model 2 were removed, confirmatory factor analysis was applied to the total test and to individual booklet, and it was found that construct validity evidences were acceptable. 2) Model 1 was more efficient than Model 2 in terms of proportion of variance explained (R2); R2 for Model 1 was .49946 which was higher than R2 for Model2 (.49507). 2.1) The correlation between value-added measures and three control variables were statistically significant at .05 and getting smaller as DIF items were deleted in the value-added estimation process compared with when DIF items were not deleted. 2.2) The Wilcoxon Signed Ranks Test applied to two school ranking data based on the ranking of school value-added measures from the Model 1 and the Model 2 was not statistically significant. However; when schools were classified into five grades based on their value-added measures from Model 1 and Model 2, there was a statistically consistent between school classifications provided by the two model (Cohens Kappa = .865, p <.05).