Empirical Priors on Theta to Improve Precision in Computerized Adaptive Tests with Item Exposure Control
This study utilizes an idea set forth by Matteucci and Veldkamp (2012) in which empirical prior distributions were used to improve parameter recovery in computerized adaptive testing. This study applies this procedure to a series of computerized adaptive tests under different constraints in the presence of Sympson-Hetter (1985) item exposure control. Latent trait estimation precision and item exposure control are often thought to be opposite goals and this study seeks to marry the two in a simple set of procedures. The use of the empirical prior resulted in greater efficiency and accuracy when the item bank was large and the correlation between the prior and true values was high. The procedure resulted in less accurate and less efficient tests than using a standard prior when the empirical prior correlated weakly with the true values.
Duffy, Liam, "Empirical Priors on Theta to Improve Precision in Computerized Adaptive Tests with Item Exposure Control" (2017). ETD Collection for Fordham University. AAI10253818.