B0862
Title: Quantile regression with a mixture of function-valued and a scalar-valued covariate prone to classical measurement error
Authors: Carmen Tekwe - Indiana University - Bloomington (United States) [presenting]
Abstract: Current recommendations for dietary intake (DI) and physical activity (PA) to minimize risks for chronic health conditions are based on statistical analyses of data prone to measurement error, including those collected from self-reported questionnaires and wearable devices. Self-reported measures based on food frequency questionnaires are often used in DI assessments; however, they are prone to recall bias. Wearable devices enable the continuous monitoring of PA but generate complex functional data with poorly characterized systematic errors. We propose the quantile regression model with function- and scalar- valued covariates prone to measurement errors. We develop semiparametric and parametric approaches to correct measurement errors associated with the mixture of functional and scalar covariates prone to errors in quantile regression settings. Simulations are performed to assess the finite sample properties. The developed methods are applied to investigate the influence of wearable-device-based PA and self-reported measures of total caloric intake on the quantile function of body mass index (BMI). The device-based measures of PA are assumed to be prone to functional covariates prone to complex arbitrary heteroscedastic errors. In contrast, DI is assumed to be a scalar-valued covariate prone to error. The developed methods are applied to assess the relationship between PA and DI with quantile functions of BMI among community-dwelling adults living in the US.
