Bayesian density regression for count data

TL;DR

This paper introduces a Bayesian Dirichlet process mixture model for count data quantile regression, addressing limitations of jittering methods and enabling flexible, accurate estimation of conditional quantiles without crossing issues.

Contribution

It proposes a novel Bayesian DP mixture approach for count data quantile regression that avoids jittering and crossing quantiles, improving estimation accuracy.

Findings

The method effectively estimates conditional quantiles for count data.

It overcomes issues associated with jittering and crossing quantiles.

The approach is applicable to distributions with intractable likelihoods.

Abstract

Despite the increasing popularity of quantile regression models for continuous responses, models for count data have so far received little attention. The main quantile regression technique for count data involves adding uniform random noise or "jittering", thus overcoming the problem that the conditional quantile function is not a continuous function of the parameters of interest. Although jittering allows estimating the conditional quantiles, it has the drawback that, for small values of the response variable $Y,$ the added noise can have a large influence on the estimated quantiles. In addition, quantile regression can lead to "crossing" quantiles. We propose a Bayesian Dirichlet process (DP)-based approach to quantile regression for count data. The approach is based on an adaptive DP mixture (DPM) of COM-Poisson regression models and determines the quantiles by estimating the density of the data, thus eliminating all the aforementioned problems. Taking advantage of the exchange algorithm, the proposed MCMC algorithm can be applied to distributions on which the likelihood can only be computed up to a normalising constant.