The Estimation Stability Comparison of Participants' Abilities on Scientific Literacy Test Using Rasch and One-Parameter Logistic Model

Abstract

This research aims at comparing the estimation stability of participants' abilities on scientific literacy test based on the integrated Sciences using Rasch Dichotomous Model and One-parameter Logistic Model (1-PL IRT). The research data was the responses of scientific literacy test based on the integrated Sciences involving 310 students joining the Mathematics and Natural Sciences Programs at SMA N 2 and SMA N 3 Tegal City, Indonesia. The estimation of test participants' abilities with those two models was assisted with the eRm package and ltm package contained in the R program version 4.0.2. The next stage included analyzing mean, median, and variance of the estimation scores resulted from both models. Bootstrapping was then conducted to the parameter estimation using the new samples which have the same size with the original samples starting from 500, 800, 1000, 1500, 2000 2500, 3000 to 4000. The research result showed that the estimation of participants' abilities in scientific literacy test based on the integrated Sciences using the Rasch Dichotomous Model was more stable than that using the 1- PL IRT seen from both bias and error standard. This means that Rasch modeling on high-stakes test assessments is more precise than using the One-Parameter Logistic Model.