U.S. manufacturing aggregate energy intensity decomposition: The application of multivariate regression analysis

Abstract

This paper introduces the use of a multivariate regression analysis to explain factors that impact aggregate energy intensity. This kind of study is useful to evaluate the past and predicts the future trends for energy-policy evaluation. Historical aggregate fuel and electricity intensities of the entire U.S. manufacturing sector (Standard Industrial Classification, SIC, codes of 20–39) over the 1977–1998 period are used to develop the proposed multivariate regression model. The proposed model allows identifying the structural effect of aggregate energy intensity changes without relying on detailed disaggregated energy data. Its results are validated by comparison with those from conventional decomposition techniques based on economic index numbers. For illustration, the historical aggregate fuel intensity of the U.S. primary metal industry (SIC 33) is used as an example of a situation for which economic index numbers fail to decompose the historical aggregate energy intensity since the disaggregated energy data are unavailable, while the multivariate regression analysis can still be applied. Empirical results show that a structural shift contributes to decreases of about 28, 41 and 19% of total declines of U.S. manufacturing aggregate fuel, U.S. manufacturing aggregate electricity, and U.S. primary metal industry aggregate fuel intensities, respectively, for the 1977–1998 period. The method based on multivariate regression models estimates the time series structural effects within deviation averages of 8.5 and 7.0% of the time series structural effect estimates based on the economic index numbers for the U.S. manufacturing aggregate fuel and electricity intensities, respectively, even though the multivariate regression model does not require disaggregated energy data. Copyright © 2007 John Wiley & Sons, Ltd.