Progressive-restricted method decreases item exposure of a Computerized Adaptive Testing without compromising precision

Abstract

The Progressive-Restricted method (PR) increases the security (in terms of item exposure) of Computerized Adaptive Testing (CAT). However, little is known about the effect of the acceleration parameter on precision and security. Therefore, we evaluated the PR with different acceleration parameters in CATs with fixed and variable-length. We combined item selection methods (Maximum Fisher Information – MFI – and PR) with stopping criteria (Fixed length, Standard error of 0.30 and Error reduction of 0.015) and simulated CATs for each Enem test. With the error reduction criterion, the precision with MFI was lower. In the other CATs, the precision was similar. Security has increased with larger acceleration parameters. At last, we compared the linear format of the Enem with the 20-item CAT. The latter had greater precision.