Online Program

Small Area Estimation using Generalized Kernel Regression
*Jean Opsomer, Colorado State University 


Keywords: categorical data, ordinal data, National Compensation Survey

We describe a nonparametric regression model for the case of a continuous dependent variable and a mixture of continuous, ordinal and nominal independent variables. The estimation method is generalized product kernel regression, which combines kernels appropriate for all three types of covariates.  This work is motivated by a small area estimation problem of the U.S. Bureau of Labor Statistics, who are tasked with estimating the mean wage by job type, job level and geographic area based on data from the National Compensation Survey, a nation-wide establishment survey.