• Going from theory to item and item to theory
  • What needs to go into an item
  • What needs to come out?
• Diagnosing levels of student understanding
• while simultaneously validating them
• Evaluating “fit” - ways we thought about fit before might not apply
• Characterizing growth over time

(article: zhidong_zhang_jingyan_lu)

Task-centered approach to performance assessment (Messick)

Starting from things we expect of doctors on the job (evidence), map cognitive processes, hierarchy of “explanatory” latent variables

Use a BIN to make probabilist inferences

Similar to evidence-centered design (Mislevey), BAS (Wilson et al), Messick: performance assessment

Seems to build from literature in learning sciences - parallell traditions

“Conventional assessment procedures, such as MC questions, are ineffective because they lack construct validity”

“Idea Unit Analysis” seems like a fragile way to establish a confirmatory structure for cognitive model. How was the fit tested?

Study design: CR item scores may be influenced by prior exposure to OMC or MTF

Partial credit model, assumes continuos latent construct - learner progression models usually use discrete levels

Article: (daisy_rutstein)

Bayesian inference network reduces to latent class model.

Tatsuoka, Leighton & Gierl, Rupp & Templin

Repeated measures design - we don't give kids the same test, there are linking designs to create new tests, assumes continuous latent variable

Forthcoming books: Learning Progressions in Science, Gotwals & Alonzo