ADAM estimation for the hierarchical joint model of response accuracy and response time

• In Event: AI, Machine Learning, and Data Science in Modeling and Prediction: New Strategies and Applications

Abstract

Jointly modeling response accuracy and response time has become an essential approach for understanding respondents’ problem-solving processes. The most widely used framework is the hierarchical model proposed by \cite{van2007hierarchical}, which offers flexibility and generalizability that broaden research possibilities. However, it is currently estimable only within a Bayesian framework, which suffers from slow mixing, long run times for large-scale data, and requires careful tuning of sampling schemes. To address these challenges, we introduce an ADAM-based estimation pipeline tailored to the hierarchical joint model. Simulation studies demonstrate that the ADAM estimator achieves over ten times greater efficiency than the traditional Bayesian approach while maintaining comparable accuracy in parameter recovery.

Authors

• Yingshi Huang, University of California - Los Angeles

• Minjeong Jeon, University of California - Los Angeles