A New Parameter Estimating Method For IRT 3PLM-Genetic Algorithm

Basing on IRT, in this paper several prevalent parameter estimation methods are compared. To 3PLM of IRT, the traditional parameter estimate method is often unsatisfactory, so we use an improved Genetic Algorithm to resolve parameter estimates, this new approach doesn't need the information of derivative and prior distributionof the unknown paramaters, and conquers shortcoming in iteration in the tradition parameter estimating that is highly sensitive to initial values. The results of Monte Carlo simulation show that this new algorithm is more effective than traditional parameter estimation techniques on contributions to raise the capability of parameters recovery, especially to raise the precision of item difficulty and asymptote parameter estimates. Encircling the techno-foundation and characteristic-research of Genetic Algorithm and application in 3PLM estimation algorithm, we mainly made the following researching:1.In this paper, we collect, regulate and summarize internal and external approaches of parameter estimation on IRT 3PLM in recent years. We systematically study and discuss the basal principle, constitutes, character and application of algorithm in IRT.2.The structure and characteristic of Genetic Algorithm is introduced.3.Based on the above work, in this paper Genetic Algorithm is closely checked and improved.4.In order to validate the unique advantage of parameter estimation based on Genetic Algorithm, this paper apply the improved Genetic Algorithm to the estimation of item parameters based on the three parameters logistic model for several fit-functions The creative points of this paper are:1. We synthesize several improved methods of existing genetic algorithm, present a new parameter estimating method "The Improved Real Coded Genetic Algorithm", then apply it to 3PLM parameter estimating of IRT, and receive favorable result.2. Apply this parameter estimating method to 3PLM,which was used in two parameters estimating and required known ability Chi-square test in the past. And this new method broadens adaptive scope. (Not require known ability)3. Rewrite several test statistics, and apply them to parameter estimating for IRT3PLM.4. Based on the new parameter estimating method, we present several fit-functions, and make simulating experiments for the performance contrast.