Quantile Regression for Correlated Observations

摘要: We consider the problem of regression analysis for data which consist a large number independent small groups or clusters correlated observations. Instead using standard mean regression, we regress various percentiles each marginal response variable over its covariates to obtain more accurate assessment covariate effect. Our inference procedures are derived generalized estimating equations approach. The new proposal is robust and can be easily implemented. Graphical numerical methods checking adequacy fitted quantile model also proposed. illustrated with an animal study in toxicology.