ABS Criterion For Test Equating And The Application Of Heuristic Algorithms

Computerized intelligent construction of test paper and Computerized Adaptive Test (CAT) require a huge item bank. Item banking necessarily needs test equating. By analyzing the subsistent methods for test equating based on Item Response Theory (IRT), two new equating methods, Common Absolute Value equating method and Square Root Absolute Value equating method, are introduced in the paper.The acknowledged criteria of test equating have good mathematical properities and we generally adopt the Newton-Raphson algorithm which cann't deal with the new criteria. Therefore, two kinds of modern heuristic algorithms, Simulated Annealing (SA) and Evolutionary Strategy (ES), are proposed to gain the coefficients.By statistic analysis, the result of Monte Carlo stimulation study shows that the new methods for test equating method are more robust than subsistent ones in common.