Skip to main content Skip to main navigation


Faster Confidence Intervals for Item Response Theory via an Approximate Likelihood

Benjamin Paaßen; Christina Göpfert; Niels Pinkwart
In: Alexandra I. Cristea; Chris Brown; Tanja Mitrovic; Nigel Bosch (Hrsg.). Proceedings of the 15th International Conference on Educational Data Mining (EDM 2022). International Conference on Educational Data Mining (EDM-2022), July 24-27, Durham, United Kingdom, International Educational Data Mining Society, 2022.


Item response theory models the probability of correct student responses based on two interacting parameters: student ability and item difficulty. Whenever we estimate student ability, students have a legitimate interest in knowing how certain the estimate is. Confidence intervals are a natural measure of uncertainty. Unfortunately, computing confidence intervals can be computationally demanding. In this paper, we show that confidence intervals can be expressed as the solution to a feature relevance optimization problem. We use this insight to develop a novel solver for confidence intervals and thus achieve speedups by 4-50x while retaining near-indistinguishable results to the state-of-the-art approach.


Weitere Links