Does the threshold distribution in polytomous items matter in computer adaptive testing?

Activity: Talk or presentationScience to science


Computer adaptive testing (CAT) contributes to efficiency in measurement by administering the smallest possible number of items while obtaining a predefined level of the standard error of measurement. Since polytomous items provide more information than dichotomous items, they are also suitable for CAT. Information depends on the distribution of thresholds raising the question whether more broadly or more narrowly distributed thresholds are preferable. A series of simulations using CAT software revealed that narrowly distributed thresholds contribute more to measurement precision even at an early stage of CAT provided the items cover a broad range of the continuum and proper targeting is ensured. While broadly distributed thresholds may compensate for small variation in the items’ overall locations, we argue that in the interest of validity and
interpretation more variation in the items is desirable and fine graded, but distinguishable, categories appear to be the best option for CAT and fixed scales.
Period10 Apr 201812 Apr 2018
Event titleIOMW (International Objective Measurement Workshop/Conference)
Event typeUnkonwn
Degree of RecognitionInternational

Austrian Classification of Fields of Science and Technology (ÖFOS)

  • 502019 Marketing
  • 509
  • 303012 Health sciences
  • 502020 Market research
  • 502052 Business administration