EFFICIENCY OF TWO ITEM SELECTION METHODS IN COMPUTERIZED ADAPTIVE TESTING FOR THE TESTLET RESPONSE MODEL: A COMPARISON BETWEEN THE MONTE CARLO CAT METHOD AND THE CONSTRAINT-WEIGHTED A-STRATIFICATION METHOD
Abstract:
The purposes of this research were 1) to compare the efficiency of measurement precision between the Monte Carlo CAT method (MCC) and the constraint-weighted a-stratification method (CWA), 2) to compare the efficiency of pool utilization between MCC and CWA. This research compared the efficiency between the two methods within the testlet response model. The data were generated to explore the efficiency under two independent variables: maximum testlet exposure rate (10%, 15%, 20%, 25%) and item pool size (600 and 800). The efficiency in parameter estimation was compared in term of bias, MSE, correlation between true ability and estimated ability as well as test length. The pool utilization efficiency was examined through five criteria: observed testlets exposure rate, test overlap rate, testlets exposure rate distribution (chi-square), number of never exposed testlets and number of overexposed testlets. The experiments were repeated 10 times for each condition. The findings of the research were as follows. 1. The MCC showed greater measurement precision than the CWA for all experimental conditions. 2. The MCC provided greater efficiency in pool utilization than the CWA when maximum testlet exposure rates of less than 20% were investigated, but when maximum testlet exposure rate increased to 25% the CWA was more efficient than the MCC which was evidenced by lower number of never exposed testlets, lower observed testlets exposure rate, and lower testlets exposure rate distribution (chi-square); however, the MCC was more efficient suggested by lower test overlap rate.