Abstract
Rasch measurement has been widely applied in the human sciences, including education, psychology, health sciences, sports, management, sociology and political sciences. The major beauty of Rasch measurement is that it diagnoses noise in test or survey data and converts ordinal item response or test raw score into a linear measure such that subsequent parametric statistical analysis (e.g., t-test, ANOVA, correlation and regression) becomes feasible, and intra-person growth and inter-person difference can be quantified. Recent decades have witnessed the blooming of Rasch measurement. In this paper, I highlight several important developments where Rasch models have been extended to deal with complicated testing situations: (a) polytomous items (e.g., constructed-response items) (b) multiple facets (e.g., rater effect) (c) multilevels (e.g., gender difference in math, school effect) (d) mixture models (i.e., latent class plus latent trait) (e) testlet items (i.e., a set of items are connected by a common stimulus of passage or figure) (f) multiple dimensions (e.g., tests with subtests) (g) hierarchical latent traits (e.g., Quality of life includes physical, psychological, social and environmental domains, and each domain may include subdomains.) (h) structural equation modeling with categorical data (i) differential item functioning (i.e., items function differently for different groups of test-takers, an issue of test fairness) (j) computerized adaptive testing / computerized classification testing (k) person-item interaction in rating scale items or Likert items (e.g., my strongly agree is equal to your agree, which is equal to his neutral). Copyright © 2010 The Hong Kong Institute of Education.
Original language | English |
---|---|
Place of Publication | Hong Kong |
Publisher | The Hong Kong Institute of Education |
Publication status | Published - 28 Jan 2010 |