A1063
Title: Bayesian monotone single-index quantile regression model with bounded response and misaligned functional covariates
Authors: Debajyoti Sinha - Florida State University (United States) [presenting]
Abstract: The main goal is to understand how existing scalar variables, as well as multiple functional covariates measuring neural response to rewards, are associated with future adolescent depression. Unlike previous studies using simple linear regression to index all covariates, a novel Bayesian quantile regression model is proposed using a single-index summary of all scalar and functional covariates along with an associated monotone link function to accommodate unknown functional forms as well as interactions among the covariates. Compared to existing methods, the new method also addresses the following practical challenges: an index with a clinical interpretation similar to a linear model, a fitted value of the pre-specified quantile within the same bounds as the response, and accommodation of the uncertainty in the registration/alignment of the observed functional covariates within the data analysis. In the simulation, the new method outperforms existing unrestricted single-index-based models in the presence of both scalar and even pre-registered functional covariates. The practical advantages and implications of the method are illustrated by analyzing a large existing adolescent depression study and, in the process, developing a new statistically principled summary of the functional covariates measuring neural response to rewards.
