The Comparison And Optimization Of Commonly Used Ability Estimation Method In Cat Environment: Research On Synthetical Estimation Of Ability

In recent years, with the development of measurement theory and computertechnology, computerized adaptive testing (Computerized Adaptive Testing, CAT)received more and more attention. Ability estimation techniques has been playing animportant role in CAT, the accuracy of the estimated not only impact of adaptiveselection strategy, but also continuingly affect the ability estimate accuracy which ismost concerned in CAT.CAT ability estimation methods still follows several main methods used in IRT,commonly include MLE, MAP, EAP, WLE and so on. This paper carried out twostudies in the aspects of compares the development of CAT ability estimation method:Study one performed a systematic comparison for four methods using Monte Carlosimulation computer program, from the bias, RMSE,exposure rate and administrativeefficiency respectively. Study two are based on study one, depending on thecharacteristics and merits of the estimation methods, developed a new CAT abilityestimation methods-the synthetical estimation of ability, that emphasizes thedifferent stages of the ability estimate of CAT, the integrated use of the mostappropriate the ability to synthesize CAT estimation method, in order to learn fromeach other, take advantage of existing ability estimation methods, to improveadministrative efficiency while accuracy of ability estimation in CAT. The resultsshowed that:1) MLE has low bias, high RMSE, better exposure rate compared to othermethods, but administrative efficiency of the worst, and the special answer modecan’t be given valid estimates.2) WLE has smallest bias, RMSE better than MLE mostly, best exposure ratewhen using a-block selection strategy with b following uniform distribution and thehighest administrative efficiency when maximum Fisher information slection strategywere used.3) MAP has worst bias, but smallest RMSE, exposure rates in most conditionssame as WLE, EAP, and selection strategy for maximum efficiency under a-blockselection strategy. 4) EAP’s bias rank only second to MAP, but RMSE is minimized, administrativeefficiency slightly lower than MAP.5) Early and mid proposed in this study with EAP, the latter with WLE’ssynthetical estimation of ability can effectively improve bias of EAP and keep RMSEof EAP stable generally.6) Synthesis method can effectively improve bias of EAP with EAP’s RMSEunder control. EAP-improving effect of the bias may be30%to40%, while theRMSE of EAP only less than5%.7) Synthesis method in different lengths test can effectively improve the bias ofEAP,and the effect are better in short tests.