Research Of Comparing Compensatory MIRT Model To Noncompensatory MIRT Model And Its Application

In this paper, by using BMIRT software, set up different experimental conditions:the sample size (1000and3000), number of items (25and50) and latent abilitycorrelation(0.3and0.7).Under these simulate conditions, generating compensatoryMIRT dates and noncompensatory MIRT dates. Then using the compensatory MIRTmodel and noncompensatory MIRT model estimate those dates’ parameters. Bycomparing the RMSE of the item parameters and ability parameters, to compare theparameter recovery of compensatory MIRT model and noncompensatory MIRTmodel under the different experimental conditions. It was found that both estimatescompensatory MIRT dates and noncompensatory MIRT dates’ parameters, theparameter recovery of compensatory MIRT model estimates parameters better thancompensatory MIRT model’s. Especially, estimating compensatory MIRT dates, thevalues of RMSE form compensatory MIRT model’s is almost half of thenoncompensatory MIRT model’s, significantly better than noncompensatory MIRTmodel’s.This article applying compensatory MIRT model and noncompensatory MIRTmodel to the analysis of Raven’s Advanced Progressive Matrices test. Found that thenoncompensatory MIRT model is not fitting as well as the compensatory MIRTmodel. Using the compensatory MIRT model to elaborate analyses the Raven’sAdvanced Progressive Matrices test. Exploration the quality, difficulty and the maincognitive component of items of the Raven’s Advanced Progressive Matrices test.The results showed the entirety discrimination of the Raven’s Advanced ProgressiveMatrices test is well. And the difficulty size of items is increasing with the orderitems’ increasing. In five dimensions of the Raven’s Advanced Progressive Matricestest, the difficulty of the cognitive component of the items is increasing by this order:A/S, CR, PP, D2and D3. And the ability of participants in A/S dimension is thehighest, the second was followed by CR, PP, D2and D3. Finally, at the base on theability parameters estimates of the Raven’s Advanced Progressive Matrices test bycompensatory MIRT model and noncompensatory MIRT model, exploring analyzedthe interactions of the abilities that the subjects solving the items. Found that theability of CR, PP and D3have the compensation relationship between eachdimensions. And the ability of A/S and D2exist the compensation relationship, too. Finally, at the end of this paper discussed the deficiency of this study, and expectthe future studies of multidimensional item response theory.