"Nonlinear psychometrics"
⇧ ⇖ ⇘

Systems like the GRE provide "adaptive testing": if you get an easy question right, they give you a harder question, and if you get a hard question wrong, they give you an easier question. Hard questions count for more points. Can we apply a similar sort of approach to learning design, not just to testing? This is the thematic question for "adaptive tutoring". Now, what happens if we take social factors into account? (This is where the "nonlinear" part comes into play!) We focus on the concrete case of informal study of mathematics at the undergraduate-level in a knowledge-rich online wiki-like community. We will study learner actions and interactions, and look for ways to extract information that later can be fed into an adaptive tutoring system.

• Research Problem
  • Come up with analytic criteria to describe and measure learning as it happens in a peer learning context for mathematics.
  • (The main difference between this paper and the earlier paper for the Networked Learning Conference is that this one should provide an expanded discussion and more empirical evidence, whereas the NLC paper is basically just methodological, i.e. looking at the literature to think about the feasibility of this approach.)
• Interaction design
  • added problem P,
  • added solution S,
  • edited article A,
  • added hint H on problem P2,
  • viewed solution C to problem P3,
  • commented on forum post F,
  • evaluate someone else's solution S2,
  • etc.
• Analytical techniques
  • The typical thing for Microgenetic Analysis is ANOVA, but there are other approaches to consider as well. The main question is whether we can gather rich enough data in the above!
  • "If microgenetic studies are superior to longitudinal and cross-sectional designs it is because the intensive nature of data can reveal detailed information about what actually occurs during periods of rapid change."
• Findings
  • Can we give evidence that what we're doing works?
• Future/related work
  • The UCIAD project (http://uciad.info, http://github.com/uciad) and other similar projects take up the challenge of aggregating (and hopefully analysing) lots and lots of activity data. Making a solid first pass in one domain-specific context would be important for showing that this project can lead to useful things.
• References
  • Fischer, K. W and Granott, N., Beyond one-dimensional change: Parallel, concurrent, socially distributed processes in learning and development, Human Development, 38, 1995, pp. 302--314.
  • Cheshire, A. and Muldoon, K. P and Francis, B. and Lewis, C. N and Ball, L. J, Modelling change: New opportunities in the analysis of microgenetic data, Infant and Child Development, 16(1), 2007, pp. 119--134
  • Matteo Baldoni, Cristina Baroglio and Viviana Patti, Web-Based Adaptive Tutoring: An Approach Based on Logic Agents and Reasoning about Actions, Artificial Intelligence Review, Volume 22 Issue 1, September 2004
  •  
  • See also: http://www.zotero.org/groups/nonlinear_psychometrics