10/31/2014

Post date: Nov 3, 2014 6:51:27 PM

"Estimating conditional quantiles in the presence of missing covariates,"

Speaker: Benjamin Sherwood, Department of Biostatistics, JHU

Abstract: Quantile regression is used to estimate conditional quantiles without making any parametric assumption about the error term. This estimate can be biased if missing values in the data are ignored. A weighted objective function is used to handle the potential bias caused by missing values. The main idea is to provide weight to completely observed subjects that have similar observations to subjects with missing data. The weighted method is introduced for both linear and partial linear quantile regression. A modified BIC and penalized objective functions are used for variable selection in the presence of missing data. Simulations show that using weighted methods will reduce the bias caused by missing data. The weighted method is applied to patient cost data with missing