Nonparametric kernel regression to investigate factors affecting heating/cooling efficiency of building

Abstract

Statistical inference is used to make generalisations about an unknown parameter from the collected data. There are two inferential approaches (parametric and nonparametric) and the choice is often a trade-off between efficiency and generality. In parametric inference, efficiency outweighs generality. Given a good model that can estimate the population parameters, this inferential approach uses methods (confidence intervals and hypothesis testing) to make efficient parameter generalisations. In nonparametric inference, generality outweighs efficiency. Given a set of minimal and weak assumptions (smothness of a density and the existence of moments of random variables), this inferential approach provides methods that are consistent for most situations. Nonparametric approaches are useful when good parametric models are not available or as goodness-of-fit tests to validate parametric models. Using the facilities of STATA, this nonparametric regression approach is used on a dataset of 768 observations to relate the heating/cooling efficiency of buildings to the relative compactness and glazing area of the buildings. The heating/cooling efficiency is expressed as a percentage, the building relative compactness is the ratio of the volume of the building to its surface area, and the building glazing area is the ratio of the aperture glass area to the wall area.