DescriptionIn educational assessment, several questions must be answered when constructing a test, such as “How many items are necessary for adequate knowledge measurement precision?”, “How many field-test students are needed to adequately calibrate model parameters?”, or “Which computerized adaptive testing (CAT) algorithm performs best?” For complex non-linear models, these questions are typically approached by simulation: Model parameters are calibrated (as if unknown) from simulated student item responses, or the emergent properties of particular CAT algorithms are investigated with a large number of simulated test takers. Since the design space grows quickly, many simulations are necessary to understand general trends. Match-for-OSG: Simulations throughout the test design space can be run independently, requiring little coordination between cores. Computations generally do not have high memory requirements or unusual library/code dependencies, and computations can be recovered from checkpoints easily. The large number of simulations suggests parallel computing, but the independence allows an asynchronous, distributed environment, such as OSG.
OrganizationUniversity of Illinois, Urbana-Champaign
Sponsor Campus GridOSG Connect
Principal Investigator
Michael J. Culbertson
Field Of ScienceEducational Psychology