Skip to main content

Search NYU Steinhardt

Learners' Resilience in Tests and Collaborative Filtering

Modeling the interaction between resilience and ability in assessments with allowances for multiple attempts

Published in Computers and Human Behavior, Dr. Yoav Bergner and colleagues discuss how allowing multiple chances to answer on a test turns any such test, unwittingly perhaps, into a test of resilience, not just ability. They then show how resilience can be explicitly modeled and estimated from such tests using tree-based IRT models. This raises new questions about the potential understanding of test question design and also leads to a new measurement approach to resilience that is not based on self-report.

Read the paper 

Multidimensional Item Response Theory in the Style of Collaborative Filtering

Forthcoming in Psychometrika, Dr. Bergner and colleagues show how collaborative filtering—a machine learning approach commonly used in recommender systems—provides an alternative way of thinking about multidimensional item response theory methods that are well established in educational and psychological measurement. The collaborative filtering methods can be used to estimate very high (20+) dimensional models quickly. Although these models are not directly interpretable, they can be used to understand auxiliary data, such as what practice questions are helpful when learning to solve new problems.