Assessment of Dimensionality in Social Science Subtest.
Ozbek Bastug, Ozlem Yesim · Educational Sciences Theory & Practice · 2012
Abstract Most of the literature on dimensionsionality focused on either comparison of parametric and nonparametric dimensionality detection procedures or showing the effectiveness of one type of procedure. There is no known study to shown how to do combined parametric and nonparametric dimensionality analysis on real data. The current study is aimed to fill this missing part in the literature by illustrating how to do combined parametric and nonparametric dimensionality analysis. The purpose of this study is to describe dimensionality structure of social science subtest of the Secondary School Institutions Student Selection and Placement Test using combined parametric and nonparametric dimensionality analysis. The data from the social science subtests of the Secondary School Institutions Student Selection and Placement Test of 1999, 2000, and 2001 were used for this study. The study indicated multidimensionality for the social science subtest. Because the results indicated multidimensionality does exist in social science subtest, it would be helpful to describe multidimensionality structure and, finally, score separately by these unidimensional grouping. Key Words Dimensionality, Unidimensionality, Validity, Parametric Methods, Nonparametric Methods, Social Science Subtest. Claiming unidimensionality does not itself ensure the validity of the test, and any assumption of unidimensionality should be checked. Unidimensionality defined as the existence of one latent trait or construct underlying a set of measures (Hattie, 1985; McDonald, 1981). Procedures used to assess the dimensionality are profound and based on various techniques. However, most of the literature focused on either comparison of parametric and nonparametric dimensionality detection procedures (e.g., Finch & Habing, 2003; Mroch & Bolt, 2006) or showing the effectiveness of one type of procedure (e.g., Roussos, Stout, & Marden, 1998; Stout, Froelich, & Gao, 2001). Only a few studies have illustrated how to do dimensionality analysis either using parametric or nonparametric methods on real data (e.g., Douglas, Kim, Roussos, Stout, & Zhang, 1999; Jang & Roussos, 2007). Also, no known study has shown to how to do combined parametric and nonparametric dimensionality analysis on real data. Therefore, current study is aimed to fill this missing part in the literature by illustrating how to do combined parametric and nonparametric dimensionality analysis. The second purpose of this study is to describe dimensionality structure of social science subtest of the Secondary School Institutions Student Selection and Placement Test (SSISSPT). Describing the dimensional structure (e.g., verification of unidimensionality or multidimensionality) is important to confirm the construct equivalence of social science subtest across gender and forms. Furthermore, verification of unidimensionality is important because many IRT techniques (e.g., BILOG) presume unidimensionality of the data. Use of these IRT procedures can be justified by a statistical analysis to confirm approximate unidimensionality or by statistical argument to claim that the departure from unidimensionality is not serious enough to jeopardize use of specific tools (Stout, 1987). Method Data The data from the social science subtests of the Secondary School Institutions Student Selection and Placement Test of 1999, 2000, and 2001 in Turkey were used for this study. Each year's data contains responses from approximately 350,000 examinees. Two random samples of 4000 examinees were drawn from each data set with equal number of female and male examinees to perform dimensionality analyses. The social science subtest was constructed to measure students' general social science knowledge (e.g., remembering the particular knowledge on history, geography), social science conception and notion knowledge (e.g., being able to interpret graphs and maps, providing examples, transforming particular knowledge), application skills (e. …