detecting and comparing learning effect in adaptive IRT tests I have two IRT adaptive online ability tests.
Both have a relatively large item bank (n>200) relatively few items displayed in a test (n=25) and quite a lot of test results (n>10k) that I got using adaptive testing - so I have two large, but sparse response matrices.
My hypothesis is that there is a significantly larger learning effect (people get better by doing the test, though I know this violates IRT assumptions) in one of the tests. How could I test this hypothesis?
 A: This seems to me to be some kind of item position effect on the person side -- "reverse test fatigue" if you want. You might want to have a look at Davis and Ferdous (2005) and De Boeck and Wilson (2004).
Also, there is a German-language dissertation on context effects in large-scale assessment by Sebastian Weirich (http://edoc.hu-berlin.de/dissertationen/weirich-sebastian-2015-07-13/PDF/weirich.pdf). Have a look at the appendices. You'll find the reference to Weirich, Hecht and Böhme (2014, APM), as well as an anglophone yet unpublished manuscript titled "Item Position Effects are Moderated by Changes in Test-Taking Effort" which might both be of interest to you.
Davis, J. & Ferdous, A. (2005). Using item difficulty and item position to measure test fatigue. American Institutes for Research, Washington. Paris.
De Boeck, P. & Wilson, M. (Hrsg.). (2004). Explanatory item response models: A generalized linear and nonlinear approach. New York: Springer.
Weirich, S., Hecht, M., & Böhme, K. (2014). Modeling item position effects using generalized linear mixed models. Applied Psychological Measurement, 38(7), 535-548. doi: 10.1177/0146621614534955 
