F.Q. Zhang, B.W. Ang r Energy Economics 23 2001 179᎐190 183
For other effects, the formulae are simply given by exchanging S with the
i j
Ž . Ž
. respective variables in Eq. 5 . Ang and Choi 1997 pointed out two problems
associated with the ADM: there is a residual after decomposition and the method Ž
. cannot accommodate zero values in the data set e.g. S s 0 . However, the
i 1
residual given by this method is generally smaller than that of the LM. 2.4. Logarithmic mean weight Di
¨
isia method LDM Ž .
Ž .
Instead of the arithmetic mean weight given in Eq. 5 , Ang et al. 1998 proposed an additive decomposition scheme using the logarithmic mean weight.
The formula for ⌬C in given by
fsh
C y C S
i 1
i 2
i 1
Ž . ⌬C
s ⌺ ln
6
fsh i
Ž .
ln C rC S
i 1
i 2
i 2
The main advantages of the LDM over the AMD are that perfect decomposition is obtained and the method can accommodate zero values in the data set. It can be
seen that the formulae of this perfect decomposition approach are much simpler Ž .
than those of the RLM. The number of terms on the right hand side of Eq. 4 Ž .
depends on the number of factors considered while Eq. 6 takes the same form irrespective of the number of factors.
3. Data and decomposition results
Our study is based on the data given in Table 1 which are taken from Ž
. International Energy Agency 1997 . The three world regions are OECD, all
countries of the former Soviet Union and all other central and eastern European Ž
. countries with economies in transition FSUrCEE , and the rest of the World
Ž .
ROW which is basically the developing world. The 1993 data set, which was also Ž
. used in Ang and Zhang 1999 , has been chosen for its completeness. Carbon
Ž .
dioxide emissions are measured in billion tonnes CO BTCO
and energy
2 2
Ž .
consumption in billion tonnes oil equivalent BTOE . For simplicity, we have estimated CO
emissions from primary fuel consumption using the following
2
emission coefficients in tonnes of CO per tonne of oil equivalent in energy
2
Ž .
consumption: coal 3.99, oil 3.07 and natural gas 2.35 see OECD, 1997 . From Table 1, OECD, FSUrCEE and ROW, respectively, accounted for 50,
17 and 33 of the world’s total energy-related CO emissions in 1993. It can be
2
seen from Table 1 that there are substantial variations in the explanatory factors Ž .
defined in Eq. 2 . The data show that GDP-related factors, i.e. GDP per capita Ž .
Ž . G
and energy intensity I , are highly dependent on GDP measure. The purchas- ing power GDP is slightly smaller than the exchange-rate-converted GDP for
OECD, but the converse is true for the other two regions. As compared with the
F.Q. Zhang, B.W. Ang r Energy Economics 23 2001 179᎐190 184
Table 1
a
Data for energy consumption, CO emissions, population and GDP for three world regions, 1993
2
OECD FSUrCEE
ROW Ž
. Primary energy consumption BTOE
4.259 1.360
2.342 Solids
1.012 0.382
0.897 Oil
1.769 0.363
1.014 Gas
0.879 0.522
0.312 Nuclear
0.471 0.068
0.031 Hydro
0.105 0.025
0.074 Geothermalrothers
0.023 0.014
Ž .
Carbon dioxide BTCO 11.537
3.866 7.427
2
Ž .
Population billion 0.872
0.416 4.226
Ž .
GDP billion 1987 US 14 522
663 3543
Ž .
GDP using PPP billion 1987 US 12 210
1741 8396
Ž .
Ž .
Ž .
Ž .
GDP per capita US 16 654 14002
1594 4183 838 1987
Ž .
Ž .
Ž .
Ž .
Energy intensity TOEr1000 0.29 0.35
2.05 0.78 0.66 0.28
Ž .
Ž .
Ž .
Ž .
CO intensity TCO r1000 0.79 0.95
5.83 2.22 2.10 0.88
2 2
Ž .
CO per capita TCO 13.23
9.29 1.76
2 2
a
Figures in parentheses are calculated using purchasing power GDP.
exchange-rate-converted values, the GDP for ROW and FSUrCEE are nearly three times as large after taking purchasing power parity into consideration. The
Table 2 Ž
. Decomposition results of the difference in CO emissions BTCO
between world regions in 1993:
2 2
OECD-FSUrCEE GDP
Method ⌬C
⌬C ⌬C
⌬C ⌬C
⌬C
tot fsh
int ypc
pop rsd
Purchasing LM
7.67 y
0.18 y
2.14 9.07
4.24 y
3.31 power GDP
y 2
y 28
118 55
y 43
RLM 7.67
y 0.42
y 7.43
9.47 6.05
y 5
y 97
123 79
ADM 7.67
y 0.04
y 6.21
9.30 5.70
y 1.08
y 1
y 81
121 74
y 14
LDM 7.67
y 0.22
y 5.58
8.35 5.12
y 3
y 73
109 67
Exchange rate LM
7.67 y
0.18 y
3.31 36.53
4.24 y
29.60 converted
y 2
y 43
476 55
y 386
GDP RLM
7.67 y
0.74 y
31.19 28.87
10.74 y
10 y
407 376
140 ADM
7.67 y
0.04 y
14.98 18.07
5.70 y
1.08 y
1 y
195 236
74 y
14 LDM
7.67 y
0.22 y
13.45 16.23
5.12 y
3 y
175 212
67
F.Q. Zhang, B.W. Ang r Energy Economics 23 2001 179᎐190 185
Table 3 Ž
. Decomposition results of the difference in CO emissions BTCO
between world regions in 1993:
2 2
OECD-ROW GDP
Method ⌬C
⌬C ⌬C
⌬C ⌬C
⌬C
tot fsh
int ypc
pop rsd
Purchasing LM
4.11 y
1.08 1.86
44.92 y
5.90 y
35.69 power GDP
y 26
45 1093
y 143
y 869
RLM 4.11
y 2.48
3.49 27.98
y 24.88
y 60
85 681
y 605
ADM 4.11
y 1.38
2.12 18.52
y 14.97
y 0.18
y 34
52 451
y 364
y 4
LDM 4.11
y 1.43
2.07 18.08
y 14.61
y 35
50 440
y 356
Exchange rate LM
4.11 y
1.08 y
4.13 140.12
y 5.90
y 124.90
converted y
26 y
101 3410
y 143
y 3039
GDP RLM
4.11 y
3.79 y
19.17 62.88
y 35.81
y 92
y 466
1530 y
871 ADM
4.11 y
1.38 y
7.71 28.34
y 14.97
y 0.18
y 34
y 188
690 y
3641 y
4 LDM
4.11 y
1.43 y
7.52 27.67
y 14.61
y 35
y 183
673 y
356
energy intensity of ROW is higher than that of OECD based on exchange-rate- converted GDP but the converse is true when it is based on purchasing power
GDP. The decomposition results given by the four decomposition methods are shown
in Tables 2᎐4. The estimated effects are also expressed as percentages of the actual total difference in emissions between regions for ease of comparison. With
the same set of emission coefficients used for all regions, ⌬C is always zero and
emc
is therefore not shown. Residuals for the RLM and LDM are also zero because Ž
these are perfect decomposition methods. In absolute terms, income effect GDP .
per capita and population effect are generally the dominant forces leading to different emission levels among the three world regions, while fuel share effect is
the smallest.
4. Impacts of variations in explanatory factors