A1154
Title: Integration of textual and quantitative responses in multidimensional item response theory
Authors: Thomas Chan - Hong Kong University of Science and Technology (Hong Kong)
Amanda Chu - The Education University of Hong Kong (China)
Mike So - The Hong Kong University of Science and Technology (Hong Kong) [presenting]
Abstract: The integration of rich unstructured textual data from open-ended responses with structured quantitative questionnaire data presents opportunities to uncover nuanced latent characteristics, but their disparate formats hinder unified analysis. A novel multidimensional item response model is proposed that seamlessly combines topic modeling for textual data and item response theory for quantitative data. Drawing on embedded space modeling, topics are projected into the latent trait space via a multivariate t density function, enabling proximity-based topic preferences that align with respondents intrinsic traits. For quantitative items, item group factors are introduced, represented by non-decreasing spline smoothing, to capture group similarities and non-linear relationships with latent traits. The model accommodates ordinal responses, handles missing data by modeling response probabilities, and assumes conditional independence. Parameter estimation employs variational Bayes. Simulation studies assess the impact of topic selection methods. Application to real-world psychosocial health data yields more profound insights into respondents patterns. This framework advances joint analysis of mixed data types, facilitating a more profound understanding across a wide range of research.
