CAJ | 학술논문

认知诊断目前在国内外越来越受到重视,而要真正实现认知诊断,需借助一定的计量学模型即认知诊断模型。国际上,目前已开发近60种认知诊断模型,不同模型又各具特点。本研究重点分析国外研究较成熟的模型——HO-DINA/DINA模型,具体探讨该模型参数估计的实现及模型性能,采用Monte Carlo模拟方法进行,研究发现：（1）本研究对项目参数、属性参数和被试参数估计返真性较好、稳健性较强、HO-DINA模型具有较高的判断率,MCMC算法可行;（2）诊断的属性个数越多,诊断的模式正确率越低,建议实际使用该模型时,诊断的属性个数不宜超过7个;（3）用于诊断的项目数越多,诊断的模式正确率越高,在实际工作中,若要保证有80%以上的模式判准率,则4个属性的至少需20题,5个、6个和7个属性的至少需要40题,8个属性至少需60题。实际运用者应根据实际情况考虑适当的项目数及属性数。
인지진단목전재국내외월래월수도중시,이요진정실현인지진단,수차조일정적계량학모형즉인지진단모형。국제상,목전이개발근60충인지진단모형,불동모형우각구특점。본연구중점분석국외연구교성숙적모형——HO-DINA/DINA모형,구체탐토해모형삼수고계적실현급모형성능,채용Monte Carlo모의방법진행,연구발현：（1）본연구대항목삼수、속성삼수화피시삼수고계반진성교호、은건성교강、HO-DINA모형구유교고적판단솔,MCMC산법가행;（2）진단적속성개수월다,진단적모식정학솔월저,건의실제사용해모형시,진단적속성개수불의초과7개;（3）용우진단적항목수월다,진단적모식정학솔월고,재실제공작중,약요보증유80%이상적모식판준솔,칙4개속성적지소수20제,5개、6개화7개속성적지소수요40제,8개속성지소수60제。실제운용자응근거실제정황고필괄당적항목수급속성수。
Recently Cognitive Diagnosis（CD） has attracted increasing importance internationally.But the realization of CD must depend on a special psychometrical model called the Cognitive Diagnosis Model（CDM）.The present paper focused on the HO-DINA model to investigate its parameter estimation and its properties that can be the inference of users. To explore the feasibility of MCMC algorithm and the estimated precision,and to probe the properties of the HO-DINA model,the Monte Carlo method was used.Two experiments were made： Experiment 1：with the fixed number of cognitive attributes（5）,of test items（60） and of examinees（1000）,this experiment was to explore the feasibility of MCMC algorithm and the estimated precision.Experiment 2：this experiment was to study the properties of the HO-DINA model,particularly the correctness rate of diagnosis.Two factors were considered：the number of test items varied with possible values of 20,40,60,80,100,120;and the number of cognitive attributes varied with possible values of4,5,6,7,8.The number of examinees was fixed as 1000. We reached the conclusions as follows：（1） The estimation method of MCMC algorithm holds fairly robustness,and its precision of items and ability parameters are preferably great,which indicates the MCMC algorithm method is feasible.（2） With the number of attributes fixed,the increasing of the number of text items can enhance the pattern correctness rate effectively.When the number of test items varies from 20 to 40,the extent of increasing is the greatest.While the number varies from 40 to 60,the degree of increasing is less.In actual work,if we want to obtain a pattern correctness rate higher than 80%,then the number of test items needs to be 20 with the number of attributes as 4,40 with the number of attributes as 5,6,7 and possibly,60 with the number of attribute as 8.（3） With the number of test items fixed,the increasing of the number of attributes will decrease the pattern correctness rate.The extent of decreasing is the greatest with the number of test items as 20,while the degree of decreasing is the least with the number of test items as 120. When the number of attribute varies from 5 to 7,the degree of deceasing is less.Thus the number of seven attributes is the turning point.So in actual work,it is suggested the number of attributes should not exceed seven.