Specification of the index decomposition method and data

L.A. Greening et al. r Energy Economics 23 2001 153᎐178 156 Ž . 17᎐41 for nine of the countries . These two terms are most comparable with the analysis presented here, and patterns of change for both terms for each of the countries are the same between the two studies. However, the magnitudes vary as a result of the specification of five terms and the indexing method used. The work presented here analyzes development of carbon emission trends from residential energy consumption in 10 OECD countries: Denmark, Finland, France, Ž . Germany West , Italy, Japan, Norway, Sweden, the UK, and the US. Decomposi- tion of aggregate carbon intensity allows attribution of changes in this measure to changes in the primary fuel mix for the generation of electricity, changes in final fuel mix for all residential end uses, and changes in energy intensity and end use Ž . activity mix structure. As with our previous analyses of sectoral emissions trends, we use a modified Adaptive Weighting Divisia index specification. As with our other sectoral studies, declines in energy intensity made a substantial contribution to declines in residential carbon intensity in the majority of countries in this analysis. In addition, declines in aggregate carbon measures also result from significant shifts towards a less carbon-intensive mix for both primary fuels used in the generation of electricity and final fuels. However, for the majority of countries, shifts in the activity mix or structure of end uses offset either partially or totally those declines in aggregate carbon intensity resulting from changes in these other measures. As a result, cumulative changes in per capita carbon emissions, our measure of aggregate carbon intensity, show a great deal of variability. Six of the countries exhibit declines ranging from almost 8 for the US to almost 72 for Sweden. The other four countries exhibit increases of less than 1 to well over 94. The remainder of this paper is organized into several sections. Section 2 provides an overview of the parametric framework for the carbon decomposition index and discusses the data used in the analysis. The complete technical development of the index decomposition method is presented in Appendix A. Section 3 of the main body of the text discusses the results of our analysis. The final section provides brief concluding remarks.

2. Specification of the index decomposition method and data

A modified or rolling base year specification of the Adaptive Weighting Divisia Ž . AWD Index method is used to decompose and attribute changes in aggregate carbon intensity to several different factors. This method has been previously used to decompose sectoral aggregate carbon for manufacturing, freight and personal Ž . transportation Greening et al., 1996, 1998a,c . In comparison to other index decomposition methods, this indexing method is more robust, exhibiting a smaller Ž . residual term with less variation Greening et al., 1997 . As with these previous analyses, this decomposition method was applied to time series of carbon emissions from fuel consumption, energy consumption and other characteristics developed by researchers at Lawrence Berkeley National Laboratory. L.A. Greening et al. r Energy Economics 23 2001 153᎐178 157 2.1. Methods of decomposition Aggregate carbon intensity may be expressed by a multiplicative relationship, and for residential end uses, changes in aggregate carbon may be attributed to four different factors. For a four-term index decomposition, changes in the aggregate carbon intensity index may be attributed to changes in the primary fuel mix used for electricity generation, changes in the final fuel mix, changes in energy intensity and changes in the structure or mix of residential energy services consumed. This relationship can be expressed as follows: Ž . Ž . Ž . Ž . 1 q ⌬G s 1 q ⌬G 1 q ⌬G 1 q ⌬G tot emissions fuel intensity Ž . Ž . 1 q ⌬G 1 q D end-use Ž . where D is the unexplained residual or approximation error represented by the quotient of the product of the four terms on the right hand side and the actual change in aggregate carbon intensity. The AWD provides more robust estimates of changes in aggregate carbon intensity by reconciling the results from a discrete and a continuous index decom- position method. The reconciliation process is performed through the application of a weighting scheme to the difference between the current year and the base year of the various factors of attribution. The weighting scheme is derived through Ž equating the two end points of a parametric family of indices Liu et al., 1992; . Ž . Greening et al., 1997 . The end points are defined by a discrete Laspeyre’s index Ž . decomposition method, and a continuous simple Divisia index decomposition Ž . method Greening et al., 1997 . Each of these end points provides for different assumptions concerning the path of integration, and as such, the results of each method will be slightly different. The weights defining the value of the index between the two end points change from year to year as emissions, energy consumption, and various other measures change. The technical derivation of the AWD and the weighting scheme is provided in Appendix A. Table A-1 provides the variable and notational definitions for that development. 2.2. Data Data for this analysis was taken from files maintained at Lawrence Berkeley National Laboratory. Time series of final energy consumption by fuel type, primary energy consumption by fuel type used for electricity generation, population, the number of households, energy consumption by major residential end-use, the number of square meters of housing and the number of major house appliances were used in this analysis. These series were collected from a variety of official or industry sources, including energy companies, utilities, appliance manufacturers Ž . and consumer associations Schipper et al., 1992; IEA, 1997 . Due to the periodic nature of official surveys, or inconsistencies resulting from changes in reporting L.A. Greening et al. r Energy Economics 23 2001 153᎐178 158 methods and differences between data sources, 3 efforts have been made to recon- cile those data sources, and missing values have been interpolated. Allocations of end-use energy consumption are made on the basis of observed relationships between an activity and energy consumption from survey data or similar instru- ments. 4 For space conditioning, energy consumption has been normalized for the number of heating degree days. 5 As a result of these modifications to the data, particularly for the countries 6 of Finland, France and Italy, interpretation of results, especially the inference of long-term trends, must be done with care. To avoid some of the potential for misinterpretation of results, we have adopted the practice of reporting period averages as a means of smoothing out some these discrepancies. Available carbon emissions were estimated using the latest methods recom- Ž . mended by the Intergovernmental Panel on Climate Change IPCC, 1995 . The carbon emissions factors used by the IPCC are the result of recent contributions to the literature in this area and for most fuels are slightly higher than previously published results. However, our estimates of aggregate carbon do not reflect the Ž . effects of other greenhouse gases GHG , e.g. nitrogen oxides, which may have a Ž . greater radiative forcing effect on the atmosphere Shine et al., 1990 . Estimates of other GHG emissions from residential energy consumption require a number of additional assumptions on combustion efficiencies and the types of combustion technologies in use. These types of assumptions cannot be made with aggregate data. As a result this analysis is restricted to examining changes in the rates of growth of available carbon. Available carbon estimates by sector are developed using fuel specific carbon coefficients and adjusting for the heating value of each fuel across countries. To allow for incomplete combustion, only 99 of available carbon is assumed com- busted. Waste biomass fuels, which may be a substantial energy source in the generation of electricity, are assigned an emissions factor of zero based on the assumption that the carbon released from combustion of this source is equivalent to the carbon sequestered by biomass replacement. Since electricity is a major final energy source for residential consumption, and the primary energy types used in Ž electrical generation have been shifting from solid to other fuel types natural gas, . nuclear, biomass and hydro , carbon coefficients for electricity were calculated based on primary energy shares for each year of the time period under evaluation. 3 For the countries Finland, Sweden and Denmark, data must be combined from several sources Ž . IEA, 1997 . 4 Allocation of energy consumed for each end use is based primarily on survey data, and converted to Ž . Ž . unit energy consumption values UEC Schipper et al., 1985 . 5 The energy consumption for space heating has been weather normalized by multiplying it by the inverse of the percentage deviation of that year’s degree-days from the long-term average. The cut-off Ž . varies with the country, depending upon climate IEA, 1997 . 6 Of the countries included in this analysis, the national authorities of these countries do not publish Ž . energy balances by end use for the residential sector IEA, 1997 . L.A. Greening et al. r Energy Economics 23 2001 153᎐178 159 This procedure captures not only changes in fuel mix, but also changes in generation technologies and capacity utilization, and improvements in generation efficiencies. Similar assumptions were not made for the emissions coefficients of other primary fuels. The carbon emissions coefficients for these fuels were held constant over the entire period of analysis. We must acknowledge that emissions for other primary fuels have changed over time due to changes in grades of fuel, however, these changes are expected to be small and data allowing for evaluation of such changes are not available. This simplification means that the emissions index term, R , which in the broader methodological framework would capture changes in i jt emissions from other fuels, is restricted to only changes in emissions from electrical generation.

3. Discussion of results