REID (Research and Evaluation in Education) (Dec 2019)
Parallel tests viewed from the arrangement of item numbers and alternative answers
Abstract
This research aims to prove that a parallel test can be constructed by randomizing the test item numbers and or alternative answers' order. This study used the experimental method with a post-test only non-equivalent control group design, involving junior high schools students in Yogyakarta City with a sample of 320 students of State Junior High School (SMPN) 5 Yogyakarta and 320 students of SMPN 8 Yogyakarta established using the stratified proportional random sampling technique. The instrument used is a mathematics test in the form of an objective test consisting of a five-question package and each package contains 40 items with four alternatives. The test package is randomized in the item numbers' order from the smallest to the largest and vice versa. The options in each item are also randomized from A to D and vice versa. Each item is analyzed using the Classical Test Theory and Item Response Theory approaches, while data analysis is done using the discrimination index with Kruskal-Wallis test technique to see the differences among the five-question packages. The study reveals that the result of item analysis using the Classical Test Theory and Item Response Theory approaches shows no significant difference in the difficulty index among Package 1 until Package 5. Nevertheless, according to the Classical Test Theory, there is a category shift of the difficulty index of Package 2 until Package 5 when compared to Package 1 – the original package – which is, in general, not a good package, because it contains too easy items.
Keywords