Distributed adaptive e-assessment in a higher education environment

The rapid growth of Information Communication Technology (ICT) has promoted the development of paperless assessment. Most of the e-Assessment systems available nowadays, whether as an independent system or as a built-in module of a Virtual Learning Environment (VLE), are fixed-form e-Assessment systems based on the Classical Test Theory (CTT). In the meantime, the development of psychometrics has also proven the potential for e-Assessment systems to benefit from adaptive assessment theories. This research focuses on the applicability of adaptive e-Assessment in daily teaching and attempts to create an extensible web-based framework to accommodate different adaptive assessment strategies for future research. Real-data simulation and Monte Carlo simulation were adopted in the study to examine the performance of adaptive e-Assessment in a real environment and an ideal environment respectively. The proposed framework employs a management service as the core module which manages the connection from distributed test services to coordinate the assessment. The results of this study indicate that adaptive e-Assessment can reduce test length compared to fixed-form e-Assessment, while maintaining the consistency of the psychometric properties of the test. However, for a precise ability measurement, even a simple adaptive assessment model would demand a sizable question bank with ideally over 200 questions on a single latent trait. The requirements of the two categories of stakeholders (pedagogical researchers and educational application developers), as well as the variety and complexity of adaptive models, call for a framework with good accessibility for users, considerable extensibility and flexibility for implementing different assessment models, and the ability to deliver excessive computational power in extreme cases. The designed framework employs a distributed architecture with cross-language support based on the Apache Thrift framework to allow flexible collaboration of users with different programming language skills. The framework also allows different functional components to be deployed distributedly and to collaborate over a network.