Comparison of methods for detecting anomalous behavior on large-scale computer-based exams based on response time and responses

Abstract

This study aims to determine the anomalous index (indeks anomali or IA) that considers both response time and responses and compares it with response time effort (RTE) or rapid guessing (tebakan cepat or TC) on various thresholds. Response time and responses from 732 examinees are in natural science subjects consist of 40 multiple choice items with four answer choices. Response time and responses are analyzed to obtain descriptive statistics related to them, calculate the TC and IA index using two methods of the threshold, the first method (M1) is a visualization of identification, and the second method (M2) is based on the amount of time spent responding to each item related to the complexity of items, as proposed by Nitko. The performance of the IA and TC scores is compared related to validity and reliability. The coefficient alpha of IAM1 score 0.84, the coefficient alpha of IAM2 0.82. Both values of the alpha coefficient have fulfilled the reliability requirements of the index determination. The IA proposed in this study has a high correlation with ERP, which is commonly used to determine the solution behavior's magnitude and rapid guessing. The correlation value of IAM1 with TCM1 0.86, the correlation value of IAM2 with TCM2 0.89, and this high correlation value shows that there is a strong relationship between IA and TC. Determination of threshold time uses three categories of multiple choices item that reveal IA and TC distributions that are close to normal distribution so that it reflects natural empirical conditions.