Development Of A Generalized Forced Choice Diagnostic Classification Model And Its Application To The Assessment Of Personality Disorders

As the most commonly used self-report instrument in the field of personality assessment,Likert-type scales fail to prevent their participants from producing fairly common response biases,e.g.,socially desirable responses or responses influenced by response styles,which are even more frequent in high-stake tests.To eliminate response biases in non-cognitive assessments,multidimensional forced-choice questionnaires(FCQs)can be considered as an alternative to Likert-type scales.FCQs usually consist of blocks including at least two statements,all of which have matchable social desirability.There are three formats of FCQs,i.e.,RANK,MOLE and PICK,and subjects are asked to respond according to different instructions of these three formats.However,FCQs have been criticized for their traditional scoring,which is actually ipsative data and unable to make comparison between individuals.In the last decade or so,the development of item response theory(IRT)has made it possible to obtain normative information from FCQs,and previous studies have proposed a number of forced-choice IRT models,cleverly circumventing the problem of ipsative data as before.However,IRT provides no methods to capture the mental processes behind the tests,in this way a new generation of psychometric theory,cognitive diagnosis theory(CDT),has emerged.The application of cognitive diagnostic assessment to personality assessments provides timely feedback to participants on diagnostic latent categories,meeting the need to assess personality or psychological disorders in clinical diagnostic contexts where the FCQs are also applicable.Thus,the combination of cognitive diagnosis and FCQs can certainly complement each other.Few cognitive classification models(DCMs)had been developed in the past and all of them are limited to the format of applicable FCQs or the number of statements contained in one block.Such limitations make the existing forced-choice DCM incapable in some possible clinical situations.Therefore,this study aims to develop a ranking forced choice diagnostic classification model.By designing and implementing two simulation studies and an empirical study the performance of the newly proposed model,RANK-GDINA,and its effectiveness in practice can be validated.The main findings of these studies are as follows.In the simulation studies,RANK-GDINA model performed best in the RANK form of FCQs,slightly outperforming the MOLE form,and the results of MOLE form outperformed that of the PICK form under the majority of experimental conditions,whether in between-statement multidimensionality or within-statement multidimensionality.Other experimental factors can also affect the estimation accuracy of the newly proposed model.To be specific,the classification accuracy of the model increases when there is an increase in the sample size,the number of blocks or the number of statements contained in each block.The classification accuracy decreases as the number of attributes or the correlation between attributes increases.In the empirical study,the RANK-GDINA model yielded highly estimated consistency rates in the RANK and MOLE forms,while each of the two forms had a much lower consistency rate with the PICK form.The empirical studies also explore issues such as attribute correlation in the application of the model and provide a visual interpretation of the diagnostic classification results.Overall,results of the above studies collectively suggest that the ranking forced choice diagnostic classification model has broad applicability and good estimation accuracy.The RANK format of FCQs generally outperforms the other two forms in terms of estimation accuracy.Also,there are other factors that can affect the estimation results of the proposed model.