Join us for a Quantitative Psychology Colloquium with Dr. Yang Liu (University of Maryland, College Park)!
Those unavailable to attend the talk in person may join via this Zoom link.
Title: Two Tales About a Curious Psychometric Model
Abstract: Magnus & Liu (2022) formulated a multidimensional hurdle item response model for psychopathology symptom measures. The model features two latent variables: susceptibility, which relates to the presence of symptoms, and severity, which relates to the frequency of symptoms. We illustrated empirically that the resulting factor scores for both dimensions are reliable and that they uniquely and differentially predict health outcomes. In this talk, I present two recent studies extending the methodological depth and rigor of the 2022 paper. First, Liu, Pek, & Maydeu-Olivares (2025) proposed a unified theoretical framework for understanding measurement precision. Specifically, we distinguished measurement from prediction as two different regressions: the former concerns regressing an observed score on latent variables (i.e., constructs), whereas the latter pertains to regressing a latent score on manifest variables (i.e., responses). These two types of regression yield qualitatively different precision quantifications: reliability and proportional reduction in mean squared error (PRMSE), respectively. We further developed a Monte Carlo procedure to compute reliability and PRMSE for complex measurement models where analytical calculation is challenging. The second study introduced a general bias-correction strategy for second-stage regression analysis using factor scores (i.e., factor score regression). Unlike extant work, our method achieves root-n-consistency for a broad range of measurement models and factor score predictors. Moreover, the bias-corrected estimator of structural parameters can be computed efficiently via plug-and-play stochastic approximation, circumventing involved analytical derivations. We evaluated the estimator in a sequence of Monte Carlo experiments, demonstrating its accuracy and efficiency in recovering the true structural parameters. I conclude the presentation by discussing the implications of these studies and outlining future avenues of research.
About Yang Liu: Yang Liu is currently an Associate Professor in the Quantitative Methodology: Measurement and Statistics (QMMS) program of the Department of Human Development and Quantitative Methodology (HDQM) at University of Maryland, College Park. His research focuses on the development of statistical methods for analyzing item response data, as well as applications of measurement models to psychological, educational, and health-related research. Yang Liu received his M.S. in Statistics in 2014 and Ph.D. in Quantitative Psychology in 2015 from the University of North Carolina at Chapel Hill. He worked as an Assistant Professor in the School of Social Sciences, Humanities, and Arts at University of California, Merced between 2015 and 2017. He then joined University of Maryland, College Park in 2017 and was promoted to an Associate Professor in 2022. Yang Liu served as an Associate Editor for Psychometrika between 2020 and 2023. He is currently an Associate Editor for the Journal of Educational Measurement and a Consulting Editor for the British Journal of Mathematical and Statistical Psychology. He was elected into the Society of Multivariate Experimental Psychology (SMEP) in 2022.