Diagnostic Tree Model Based Treatment Of Rapid Guessing And Item Skip

Diagnostic classification models(DCMs)provide students with fine-grained diagnostic information to promote learning and teaching.However,when rapid guessing(RG)and item skipping(IS)are related to ability,filtering out these behaviors can lead to biased parameter estimates and decreased estimation accuracy.To address this issue,this study proposes a series of models that flexibly handle RG and IS behaviors,taking into account their relationship with ability.The current models can provide unbiased parameter estimates even when ability is highly correlated with RG and IS behaviors.They can also provide information about the degree of correlation between ability and RG/IS behaviors,allowing for the evaluation of the appropriateness of handling these behaviors.Three studies were conducted to investigate how to detect and handle RG and IS behaviors.Each study includes an empirical part and a simulation part.The empirical component demonstrates the application of the current models and compares them with existing research,while the simulation component examines the psychometric properties of the models.Study 1 focuses on detecting and handling RG behaviors and proposes a hierarchical joint tree model(DINA-RGTM)that decomposes response behavior into RG and non-RG behaviors and constructs corresponding indicators to detect RG behaviors.The hierarchical joint modeling considers the relationship between RG behaviors and ability.The results show that DINA-RGTM can accurately detect RG behaviors and provide unbiased parameter estimates even when ability and RG behaviors are highly correlated.Moreover,the parameter estimation accuracy is higher than when RG behaviors are directly filtered out.Additionally,the method used to handle RG behaviors mainly affects the diagnostic classification accuracy of the RG behavior group.Study 2 focuses on handling IS behaviors and proposes another hierarchical joint tree model(DINA-SM)that considers the relationship between IS behaviors and ability through the hierarchical joint modeling.The results show that DINA-SM can provide unbiased parameter estimates even when ability and IS behaviors are highly correlated,and the parameter estimation accuracy is higher than when IS behaviors are directly filtered out.Additionally,the method used to handle IS behaviors mainly affects the diagnostic classification accuracy of the IS behavior group.Study 3 focuses on situations where both RG and IS behaviors occur simultaneously and constructs the DINA-S-RGTM model by combining DINA-RGTM and DINA-SM.The study finds that ignoring either RG or IS behaviors can lead to decreased parameter estimation accuracy,while DINA-S-RGTM can handle the test situations with both RG and IS behaviors and provide satisfactory fidelity.