Identifying Risk Factors for Severe Childhood Malnutrition by Boosting Additive Quantile Regression

Ordinary linear and generalized linear regression models relate the mean of a response variable to a linear combination of covariate effects and, as a consequence, focus on average properties of the response. Analyzing childhood malnutrition in developing or transition countries based on such a regression model implies that the estimated effects describe the average nutritional status. However, it is of even larger interest to analyze quantiles of the response distribution such as the 5% or 10% quantile that relate to the risk of children for extreme malnutrition. In this paper, we analyze data on childhood malnutrition collected in the 2005/2006 India Demographic and Health Survey based on a semiparametric extension of quantile regression models where nonlinear effects are included in the model equation, leading to additive quantile regression. The variable selection and model choice problems associated with estimating an additive quantile regression model are addressed by a novel boosting approach. Based on this rather general class of statistical learning procedures for empirical risk minimization, we develop, evaluate and apply a boosting algorithm for quantile regression. Our proposal allows for data-driven determination of the amount of smoothness required for the nonlinear effects and combines model selection with an automatic variable selection property. The results of our empirical evaluation suggest that boosting is an appropriate tool for estimation in linear and additive quantile regression models and helps to identify yet unknown risk factors for childhood malnutrition.