Application Of Generalized Item Response Tree Model In Response Style

Personality and attitude measurement often use likert scale to assess the subjects to a set of declarative statements related to the measurement subject agree or disagree with the extent of each person in all questions by calculation on the attitude of the total score to get him or her in a scale or different states of the questionnaire.Response bias is an important factor affecting the reliability and validity of the questionnaire.Response bias refers to the behavior of the subjects in the response to Likert scale,which is not based on the content of the subject,including two types of response set and response styles.Response set related to the topic content,refers to the subjects always for a particular purpose in a systematic way instead of according to oneself real answer mental state,there are mainly social desirability effect,etc.response style has nothing to do with the topic content,that is,subjects always ignored subject content towards a particular way of response,there are mainly extreme response style,acquiescence response style,midpoint responding,response style is considered to represent a stable individual characteristics,system cultural differences,response styles exist in the test results cannot reflect measurement subject,reduce test validity.Based on this,some researchers have proposed and developed a series of studies on response style,such as frequency statistics,structural equation model,the latent class analysis,mixed IRT model,etc.However these methods still has some limitations,unable to effectively separate response style and the test content.This study applies Paul De Boeck(2015)put forward the generalized item response tree model to control response style,the model assumed that the participants in the answer likert items subject's response was produced by different cognitive states,and what is the cognitive status through which lead to observe the reaction is what we are trying to calculate ? parameters in IRT model,also is the cognitive process of numerical value measurement.Cognitive process model to be converted into a form processing tree model,and through the relatively simple technology of maximum likelihood parameter estimation,the researchers can use this method to detect cognitive status of the participants has nothing to do with the topic content,such as response style.to verify the general project tree model in dealing with reaction style of performance,this study,through a simulation study and an empirical research on the argument.In the simulation study,we demonstrate that the generalized item response tree model in the correction of extreme response style,try to estimate the accuracy of the ability parameters.The commonly used method of the recent treatment response style is the mixed partial credit model as a reference,to compare the parameter recovery of two models on the extreme response style,researcher considering the test length,thesample size of the generalized item response tree model estimation accuracy,results show that the parameter recovery of the generalized item response tree model is better than mixed partial credit model,estimation accuracy is affected by the length of test,but is not affected by sample size,the more the longer the test,the more items,the model to estimate the more accurate.The empirical research applying generalized item response tree model testing data in the actual data of the temperature of response style,further verify the model in the actual effect in the study of response style.Research of the monitoring data of Zhejiang province taizhou elementary school science education monitoring attitude questionnaire,extraction with learning difficulties,learning internal motivation and external motivation,learning self-efficacy,the parent-child relationship,peer relationship six dimensions,a total of 4189 subjects,inspecting whether they are in the process of answer performance response style.It is found that,first of all,generalized item response tree model on psychological process multiple assumption is reasonable;Seconde,Response styles exist in the test data,some responsdents have been shown extreme reaction style and middle response style,according to the generalized item response tree model to estimate the cognitive state parameters of each part as a result,the performance of extreme response style and the middle response style of test parameters and obviously on the high side,the corresponding original reaction are two extreme options and neutral options the most frequency,and on the detection index unidimensional indicators of response style,set all binary projects and multidimensional model structure is more fitting,set of binary project that not only measure is a response to a centralized style,at the same time,measuring the latent trait of questionnaire.