Nonparametric Kernel Density Estimation with Automatic Data-driven Bandwidth Selection Method: An Application to National Panel Survey Data Wave 4 in Tanzania

Abstract

Bandwidth selection is of great importance when statisticians want to estimate the functional forms of a data set by using nonparametric method. Bandwidth can be selected by several methods, but least squares cross-validation method may provide
optimal bandwidth. This study used nonparametric kernel density estimation approach to estimate the function form of consumption of the household and computing an optimal bandwidth parameter, a bias-variance trade –off by using the least squares cross-validation data –driven automatic selection method. The study used National Panel Survey data wave 4 (2014/2015) in Tanzania with 3,352 households based on stratified, multi-stage cluster sample design to estimate the function form of the
households’ food shares.