Identification, estimation, and q-matrix validation of hierarchically structured attributes in cognitive diagnosis
Citation & Export
Hide
Simple citation
Akbay, Lokman.
Identification, estimation, and q-matrix validation of hierarchically structured attributes in cognitive diagnosis. Retrieved from
https://doi.org/doi:10.7282/T3RR21JVExport
Description
TitleIdentification, estimation, and q-matrix validation of hierarchically structured attributes in cognitive diagnosis
Date Created2016
Other Date2016-10 (degree)
Extent1 online resource (xi, 115 p. : ill.)
DescriptionMany cognitive diagnosis model (CDM) examples assume independent cognitive skills; however, cognitive skills need not be investigated in isolation (Kuhn, 2011; Tatsuoka, 1995). Kuhn (2001) argues that some preliminary knowledge can be the foundation for more sophisticated knowledge or skills. When this type of hierarchical relationships among the attributes are not taken into account, estimation results of the conventional CDMs may be biased or less accurate. Hence, this dissertation investigates the change in the degree of accuracy and precision in the item parameter estimates and correct attribute classification rates of different estimation approaches based on modi cation of either the Q-matrix or prior distribution. Modi fication of the prior distribution and the Q-matrix depend on the assumed hierarchical structure, as such, identifying the correct hierarchical structure is of the essence. To address the subjectivity in the conventional methods for attribute structure identification (i.e., expert opinions via content analysis and verbal data analyses such as interviews and think-aloud protocols), this dissertation proposes a likelihood-ratio test based exhaustive empirical search for identifying hierarchical structures. It further suggests a likelihood-approach for selection of the most accurate hierarchical structure when multiple candidates are present. Furthermore, implementation of the CDMs requires construction of a Q-matrix to indicate the associations between test items and attributes required for successful completion of the items (de la Torre, 2008; Chiu, 2013). Q-matrix construction heavily depends on content expert opinions, as such this subjective process may result in misspecifications in the Q-matrix. Up to date, several parametric and nonparametric Q-matrix validation methods have been proposed to address the misspeci fications that may emerge due to fallible judgments of experts in Q-matrix construction (Chiu, 2013). Yet, although they have been examined under various conditions, none of these methods was tested under hierarchical attribute structures. Therefore, this dissertation further investigates the reciprocal impact of misspeci fied Q-matrix and hierarchical structure on hierarchy identification and Q-matrix validation. The results showed that structured prior distribution led to the most accurate and precise item parameter estimation, and highest correct examinee classification. When an unstructured prior was employed, impact of structured Q-matrix was different for compensatory and noncompensatory CDMs. Furthermore, study results showed that likelihood-based exhaustive search was promising in identification/validation of hierarchical attribute structure. Lastly, results indicated that performance of Q-matrix validation methods might not be as high when they are used as is under hierarchical attribute structures.
NotePh.D.
NoteIncludes bibliographical references
Noteby Lokman Akbay
Genretheses, ETD doctoral
Languageeng
CollectionGraduate School - New Brunswick Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.
Statistical Profile