A1337
Title: A model-aware approach to quantile regression for cross-sectional data with zero-inflated counts
Authors: Derek Young - University of Kentucky (United States) [presenting]
Xuan Shi - University of Kentucky (United States)
Carlos Lamarche - University of Kentucky (United States)
Abstract: Quantile regression for cross-sectional data with (potentially zero-inflated) count responses is addressed using a novel model-aware framework. The model-aware framework transfers the assumed conditional (zero-inflated) discrete distribution to its continuous analogue, thus allowing one to leverage existing quantile regression methods to estimate the conditional quantiles of interest. The proposed approach is followed by fitting a nonlinear function to the estimated conditional quantiles, which yields the estimated quantile effects. One major benefit to this approach is that it mitigates the issue of quantiles crossing, which can occur with existing jittering-based quantile regression methods for count data. Identification, large sample results, and bootstrap routines for the estimated quantile effects are discussed. The small sample performance of the procedure is compared with existing approaches. An analysis of the famous Oregon Health Insurance Experiment data is presented.