Estimating Total Energy Demand from Incomplete Data Using Non-parametric Analysis

Abstract

The validity and usefulness of empirical data requires that the data analyst ascertains the cleanliness of the collected data before any statistical analysis commence. In this study, petroleum demand data for a period of 24 hours was collected from 1515 households in 10 clusters. The primary sampling units were stratified into three economic classes of which 50% were drawn from low class, 28% from medium class and 22% from high class. 63.6% of the questionnaires were completed whereas incomplete data was computed using multivariate imputation by chained equation with the aid of auxiliary information from past survey. The proportion of missing data and its pattern was ascertained. The study assumed that missing data was at random. Nonparametric methods namely Nadaraya Watson, Local Polynomial and a design estimator Horvitz Thompson were fitted to aid in the estimation of the total demand for petroleum which has no close substitute. The performance of the three estimators were compared and the study found that the Local Polynomial approach appeared to be more efficient and competitive with low bias. Local polynomial estimator took care of the boundary bias better as compared to Nadaraya Watson and Horvitz Thompson estimators. The results were used to estimate the long time gaps in petroleum demand in Nairobi county, Kenya.