Targeting data collection in games based assessment

Abstract

Educational game performance data has the potential to allow new types of complex, procedural skills to be assessed. However, prior work has shown that gameplay data do not readily align to existing assessment validation paradigms, and game performance scores are difficult to use for proficiency testing. A new assessment paradigm that can cope with the nature of gameplay data has not emerged. In this paper, we uncovered a range of structural issues in data collection caused by, and potentially solved by, the engineered environments in games. Choice and the iterative nature of games were found to allow curriculum specialisation. We found evidence that early attempts at new games are less reliable and perhaps best discarded, and we propose a solution to weight scores to reflect novelty in repeated tasks. We found capturing the effect of competitor or collaborator ability on performance challenging but propose the potential for bots to resolve this. Finally, we also investigated the use of response time as a proxy for ability. The physical measure of time proved difficult and potentially unfair to use, but we propose a possible stochastic treatment of speed that could allow scoring some skills in some games using response time.