Fig. 1. Estimated relationship between GNP and CO
2
, ENPC level.
coefficients are jointly significant. Fig. 1 shows how CO
2
emissions and ENPC levels change with increasing per capita income. At
income levels below 5000, environmental stress is seen to increase with energy use increasing much
more than carbon dioxide emissions. These emis- sion levels peak at around the 6000 mark, after
which energy use plateaus while CO
2
emissions decrease until income levels reach 18 000. After
this critical income point, rapid rates of increase in energy use and CO
2
emission levels are observed. These results contrast with the standard EKC curve
in that we do not find an inverted-U curve.
10
Rather, we find that the curve is a flattened inverse-S shaped curve where the slope is mostly
positive everywhere, except for the inflection point where the slope is zero. The inverse-S shape is
observed for both CO
2
and ENPC variables, with ENPC levels showing larger rates of change over
the income scale, that is, the CO
2
curve is flatter and less variable.
The results imply that we can partition the environmental stress experience of countries into
distinct phases. During the first phase when per capita incomes are low, environmental stress is
shown to increase but at a diminishing rate. During the second phase when per capita incomes are
higher, environmental stress levels appear con- trolled and no increases are observed. The third
phase occurs at extremely high incomes when emissions increase again and escalate rapidly. This
implies that the impact of income on the environ- ment is more significant at the extreme ends of the
income scale. In particular, the results show that very low and very high-income countries tend to
experience increasing stress levels in their environ- mental conditions, while there is relatively little
change in the environmental stress levels for the middle-income countries.
The results further show that population density level and levels of urbanisation are both positively
related to environmental stress while the level of income inequality is inversely related to environ-
mental quality. Hence, as a country gets more surprising, as we would expect that a rise in energy
consumption would be accompanied by a rise in pollutant emissions. Suri and Chapman 1998
explain this seemingly inconsistent result by sug- gesting that it is possible for energy consumption
to keep rising but for emission levels of local pollutants to fall, as would be the case when
end-of-pipe technology like scrubbers are used to reduce local pollutants. As the existing policies to
abate local pollution often concentrate on end-of- pipe methods and not on reducing energy consump-
tion or emission levels, it should not surprising that energy use and CO
2
emissions are not being re- duced along with reductions in the levels of local
pollutants.
5. Results
Table 3 presents results from the estimation of the environmental equation Eq. 2. It is seen that
per capita income, population density, country’s level of urbanisation, inequality in the distribution
of income as well as level of education exert significant influences on a country’s level of envi-
ronmental stress. The results are particularly strong when CO
2
emissions and ENPC are used as the dependent variables. In the case of CO
2
, we find that when all other influences are taken to be
constant, a 1000 increase in per capita GNP increases the per capita CO
2
emission level by 1 metric ton. For ENPC, a 255 kg in oil equivalent
increase in energy use results from a 1000 increase in per capita GNP.
9
The F-test shows that the
10
We do not find an inverted-U shaped relationship be- tween income and any of the pollutants. In the literature, the
EKC relationship is supported for pollutants like SO
2
and NO
x
, but not always for CO
2
, which is usually seen to increase over income.
9
The coefficients do not come out to be statistically signifi- cant for ENPC, however, the sign of the income coefficients
are very similar to the signs of the income coefficients for CO
2
emissions.
L .
Gangadharan
, M
.R .
Valenzuela
Ecological Economics
36 2001
513 –
531
523 Table 3
Impact of GNP and other explanatory variables on the environment
a
SO
2
NO
x
EMW DEFRTE
CO
2
Variable ENPC
TSP 844416.1
1.034 1.513
13.760 Constant
− 0.895
0.485, 0.651 −
233.047 1.011
7.941
b
897.500 697726.6
3.401 1.519
0.001 −
0.002 GNP
− 1.93E-04
0.255 −
1.85E-04 −
129.00 −
4.05E-05 201.261
0.001
c
2.33E-04 0.208
2.61E-04 0.001
2.40E-04 −
2.11E-05 6.91E-09
− 1.12E-07
GNP
2
1.07E-08 1.21E-07
0.002 −
9.46E-10, 0.016
2.01E-08 1.52E-08
1.88E-05 1.70E-08
5.62E-08
c
1.11E-07 −
5.38E-14 −
2.41E-12 −
1.65E-13 6.53E-10
1.13E-07 3.02E-12
GNP
3
6.71E-14, 3.12E-13
3.68E-07 2.38E-12
4.65E-13 4.81E-10
1.47E-12
c
3.60E-13 0.199
9.84E-05 0.001
POPDEN −
7.21E-05 0.001
− 152.317
− 5.45E-05,
143.116 1.82E-04
1.20E-04 0.213
1.43E-04 0.001
b
0.001 −
9.916 1989.433
5.034 −
1.421 UPOP
− 1.016
− 1010368
0.201, 1.288 1.956
795.405
b
1152394 1.736
2.869
b
7.185 GINI
0.011 0.017
11058.34 0.018, 0.015
− 36.024
0.020 −
0.090 0.021
0.096 14.988
b
0.024 0.059
14480.17 6086.675
0.010 0.002
0.012 −
0.008, 0.008 18.411
EDU 0.034
0.016 0.011
0.046 7135.055
0.034 7.757
b
4.44 0.53
1.08 2.45
6.96 0.71
F-test 0.83
a
Figures in parenthesis indicate robust S.E.
b
Significant at 10 level.
c
Significant at 5 level.
crowded more people on a fixed area of land, the higher will be their CO
2
emissions and per capita energy use. This can be due to the fact that as
population density increases, there is increasing pressure to use the existing land more intensively.
The creation of multi-storey residential and com- mercial buildings in high population density coun-
tries is a good example of this problem. Lifestyle adjustments for residents in these countries imply
more energy consumption and this leads to abnor- mally high levels of CO
2
emissions. Singapore is a case in point; its population density in 1996 was
4990 persons per km
2
and the commercial energy use was 7162 kg of oil equivalent per capita. In
contrast, the corresponding average levels for our sample of 51 countries are 206 and 2850, respec-
tively. Clearly, the high population density in Singapore exerts a major influence on its extremely
high level of energy use. Further, the percentage of population living in urban areas UPOP impacts
positively on the levels of CO
2
and ENPC, with emissions rising as urban population increases.
We observe a positive coefficient for the educa- tion variable EDU, which runs counter to expec-
tations. The results show that higher levels of education aggravate, rather than improve, environ-
mental conditions. On the other hand, any im- provement in the inequalities between the rich and
the poor is found to be detrimental to the environ- ment. While counter intuitive in the first instance,
this makes empirical sense because a move towards more equal standards of living implies more people
are able to afford the use of electricity, cars and other luxuries — which leads to increased energy
use.
11
For such a cross-section of countries, the explanatory power of these two models is fairly
high adjusted R
2
= 67 for CO
2
and 79 for ENPC.
Eq. 2 was also estimated using data on other specific pollutants such as TSP, SO
2
, NO
x
, EMW and DEFRTE. The magnitude and signs of the
estimated coefficients are very sensitive to the pollutant used, and are very unstable. Further, the
explanatory power of the models are greatly re- duced with F-test results simultaneously indicating
inappropriate models. We note that many environ- mental studies used CO
2
and ENPC precisely because the data on these variables are well devel-
oped. Also, we note here that trend results are similar for CO
2
and ENPC because CO
2
is a major component of ENPC. As seen in Table 2, these two
environmental stress variables have a high and positive correlation between them.
5
.
1
. Impact on health Results of the two-stage least squares 2SLS
estimation of Eq. 3 are presented in Tables 4 and 5. In these estimations, we use alternative indicators
of a population’s health status — namely, life expectancy LEXP, healthy life expectancy HLE,
infant mortality rate IMR and child mortality rate CMR — and treat the environmental stress
variable as endogenous. We find that if we ignore the potential endogeniety of the environmental
variables, the results obtained are inconsistent. The Davidson and MacKinnon 1993 augmented re-
gression test shows that the null hypothesis of an exogenous environmental stress variable is strongly
rejected for all the alternative types of pollutants.
12
Table 6 presents the coefficients of the environmen- tal stress variables for the different health indica-
tors and compares the OLS and the 2SLS estimates. The coefficients obtained from the 2SLS estimation
have signs in the expected direction and the magni- tudes are larger compared with the coefficients
from the OLS estimation. This implies that the impact of the environmental stress variable on
11
This issue has been analysed in greater detail in Torras and Boyce 1998, Scruggs 1998 with mixed results. While Torras
and Boyce 1998 find that more equitable distributions of income and power tend to result in better environmental quality,
Scruggs 1998 shows that equality does not necessarily lead to lower environmental degradation. Magnani 2000 finds that
higher levels of income would increase environmental quality provided the negative effect of production of goods and services
on pollution levels is more than counterbalanced by the positive effect of growth on the demand for pollution abatement policy.
The demand for environmental quality will be affected by inequality levels in the country. As the level of per capita income
increases above a critical level, income equality becomes posi- tively correlated with environmental protection; however, be-
yond a certain threshold level of income, the correlation between income and environmental protection turns negative.
12
The augmented regression is formed by including the predicted value of the endogenous right-hand side variable as
a function of the all exogenous variables, in a regression of the original model.
L .
Gangadharan
, M
.R .
Valenzuela
Ecological Economics
36 2001
513 –
531
525 Table 4
Impact of GNP and environment on health: with LEXP and HLE as dependent variables
a
SO
2
NO
x
EMW DEFRTE
Variable ENPC
CO
2
TSP A Dependent 6ariable: life expectancy
62.728 6.948
b
65.456 12.590
b
63.381 5.336
b
58.002 3.962
b
Constant 57.228 3.597
b
53.078 3.250
b
62.143 5.048
b
− 4.152 5.439
− 4.220E-06 0.470E-06
c
− 1.803 3.430
c
− 5.144 5.156
− 0.623 0.440
Environmental −
0.002 0.001
b
− 0.380 0.156
b
stress variable 3.3E-04 1.2E-04
b
4.6E-04 1.4E-04
b
5.01E-04 1.1E-04
b
4.0E-04 8.7E-05
b
0.001 1.6E-04
c
GNP 4.1E-04 7.4E-05
b
0.001 9.3E-05
b
0.063 0.069 0.087 0.081
0.056 0.104 0.110 0.082
0.095 0.062 IMM
0.064 0.038
c
0.091 0.033
b
0.012 0.009 0.009 0.004
b
0.007 0.004
c
− 0.001 0.017
0.013 0.004
b
0.006 0.005 0.005 0.007
DOC −
0.004 0.040 0.012 0.022
0.006 0.025 0.002 0.041
0.030 0.023 0.008 0.022
0.016 0.021 EDU
− 3.737 4.363
3.664 4.506 −
4.054 6.700 −
5.541 5.614 UPOP
− 5.383 4.526
4.932 3.301 3.846 2.813
9.40 9.43
12.59 21.61
17.27 F-test
16.16 23.60
B Dependent 6ariable: healthy life expectancy 47.126 9.874
b
41.164 6.873
b
37.139 5.767 44.883 8.480
b
47.426 10.820
b
52.610 20.266
b
40.236 6.759
b
Constant −
0.326 0.237 −
7.256 7.622 −
6.626 8.987 −
5.890E-06
c
3.690E-06 0.320 4.680
− 0.002
b
00.001 Environmental
− 0.603 0.627
stress variable 3.053E-04 1.807E-04
c
5.029E-04 2.108E-04
b
5.604E-04 1.665E-04
b
4.652E-04 1.564E-04
b
7.393E-04 1.678E-04
b
GNP 4.352E-04 1.115E-04
b
0.001 1.357E-04
b
0.146 0.123 0.112 0.099
0.120 0.172 0.196 0.138
IMM 0.157 0.097
c
0.146 0.061
b
0.126 0.070
c
0.028 0.016
c
0.022 0.008
b
0.020 0.009
c
0.021 0.026 0.025 0.006
b
0.019 0.009
b
0.017 0.011 DOC
− 0.004 0.048
0.014 0.039 −
0.020 0.070 0.012 0.042
0.004 0.038 EDU
0.008 0.039 0.023 0.037
− 4.771 10.689
6.504 3.639
c
− 3.463 6.259
4.662 4.797 7.176 3.348
b
− 2.234 7.330
− 5.738 8.015
UPOP 7.72
7.14 9.43
17.97 19.50
F-test 14.11
18.16
a
Figures in parenthesis indicate robust S.E.
b
Significant at 5 level.
c
Significant at 10 level.
L .
Gangadharan
, M
.R .
Valenzuela
Ecological Economics
36 2001
513 –
531
Table 5 Impact of GNP and environment on health: with IMR and CMR as dependent variables
a
EMW Variable
DEFRTE CO
2
ENPC TSP
SO
2
NO
x
A Dependent 6ariable
:
infant mortality rate 59.308
74.065 56.816
75.416 82.829
Constant 45.716
60.419 15.342
b
12.086
b
46.084
b
13.419 11.594
b
12.095
b
20.929
b
2.113 16.079
16.439 1.540E-05
12.889 0.173
Environmental stress 0.004
variable 14.902
10.253 18.806
1.178
c
0.003 0.619
8.420E-06
c
− 1.270E-03
− 0.001
− 0.001
− 0.001
− 0.001
GNP −
0.001 −
0.002 3.391E-04
b
3.586E-04
c
4.523E-04
b
2.181E-04
b
3.580E-04
b
0.001
b
3.285E-04
b
IMM −
0.388 −
0.214 −
0.216 −
0.283 −
0.275 −
0.348 −
0.442 0.209
c
0.349 0.159
b
0.141
b
0.256
c
0.129
c
0.135
b
− 0.051
− 0.066
− 0.056
− 0.047
DOC −
0.073 −
0.048 0.003
0.016
b
0.031
b
0.053 0.020
b
0.013
b
0.014
b
0.011
b
− 0.038
− 0.091
− 0.064
− 0.066
EDU 0.011
− 0.023
0.034 0.074
0.071 0.133
0.066 0.132
0.064 0.066
− 17.604
11.772 −
13.266 UPOP
11.144 12.711
10.838 −
24.073 15.122
22.007 16.440
10.654
c
16.183 9.619
12.768 22.62
7.64 16.53
13.39 17.98
17.64 F-test
10.85 B
Dependent 6ariable
:
child mortality rate 113.820
90.029 71.226
89.890 127.268
82.166 112.428
Constant 36.188
b
23.089
b
27.228
b
71.380 22.437
c
21.052
b
21.322
b
0.416 Environmental stress
3.433 0.007
24.473 23.110
17.854 2.190E-05
15.862 28.130
22.826 1.250E-05
c
1.948
c
0.851 0.005
c
variable −
0.001 GNP
− 0.001
− 0.002
− 0.001
− 0.001
− 0.002
− 0.001
0.001
b
0.001 4.825E-04
c
3.176E-04
b
0.001
b
0.001
b
4.639E-04
b
DPT −
0.467 −
0.468 −
0.723 −
0.779 −
0.571 −
0.512 0.683
0.419
b
0.363
b
0.547 0.299
b
0.266
c
0.227
b
0.223
b
− 0.095
− 0.059
0.011 −
0.057 DOC
− 0.062
− 0.088
− 0.071
0.021
b
0.080 0.030
c
0.017
b
0.046
b
0.020
b
0.021
b
− 0.111
0.068 −
0.063 0.043
− 0.010
SEC −
0.022 0.071
0.095 0.195
0.094 0.107
0.208 0.103
0.095 13.224
20.191 15.370
− 22.663
− 36.789
UPOP 12.990
− 29.605
19.265 24.261
33.370 15.172
c
13.364
c
21.196 24.349
10.61 10.82
17.97 5.36
11.99 6.44
13.91 F-test
a
Figures in parenthesis indicate robust S.E.
b
Significant at 5 level.
c
Significant at 10 level.
L .
Gangadharan
, M
.R .
Valenzuela
Ecological Economics
36 2001
513 –
531
527
Table 6 Comparative test results from alternative estimation methods used
a
Health indicator Environmental
stress variable CMR
IMR LEXP
HLE t-Value
b
OLS 2SLS
OLS t-Value
b
OLS 2SLS
t-Value
b
2SLS t-Value
b
OLS 2SLS
3.51
c
− 0.22 0.20
0.17 0.62 3.78
c
− 0.08 0.39
− 0.35
c
0.20 0.42 0.85
CO
2
4.19
c
− 0.33 0.24
6.63
c
− 0.38
b
0.16 −
0.16
c
0.09 −
6.02E-04
c
3.67E-04 −
0.002
b
0.001 3.43
c
3.00E-04 8.12E-04 0.004 0.003
3.76
c
1.50E-03 1.09E-03 0.007
c
0.005 4.03
c
− 0.002
b
0.001 6.45
c
ENPC −
1.33E-03
b
5.75E-04 0.42 0.33
− 7.26 7.62
3.81
c
0.53 1.11 16.44 14.90
3.56
c
2.34E-03 1.40 24.47 22.83
4.06
c
− 5.14 5.16
SO
2
8.35
c
0.21 0.24 4.01
c
0.31 0.98 16.08 18.81
3.62
c
− 0.19 1.35
23.11 28.13 −
6.63 8.99 4.13
c
8.48
c
0.46 0.31 NO
x
0.28 0.19 −
4.15 5.44 −
0.60 0.63 0.02 0.07
3.74
c
0.33 0.37 2.11
c
1.18 3.45
c
0.34 0.53 3.43
c
1.95 3.96
c
− 0.62 0.44
8.45
c
0.10 0.10 TSP
3.53
c
EMW 1.17E-06 1.53E-06
6.86E-08 4.58E-07 1.54E-05
c
8.42E-06 3.56
c
1.20E-06 2.11E-06 2.19E-05
c
1.25E-05 4.21
c
− 4.22E-06
c
2.47E-06 6.98
c
3.45E-07 5.99E-07 −
5.89E-06
c
3.69E-06 3.43
c
0.03 2.10 12.89 10.25
3.70
c
− 0.27 3.08
17.85 15.86 0.32 4.68
4.29
c
− 0.16 0.77
DEFRTE −
0.25 0.44 −
1.80
c
3.43 7.46
c a
Figures in parenthesis indicate robust S.E. , Significant at 10; , significant at 5.
b
t-Value from Davidson–Mackinnon augmented regression test for exogeneity.
c
Indicates the null hypothesis of exogeneity is rejected at 5 significance level.
health is bigger when we take the endogeniety into account. The t-statistic from the Davidson –
Mackinnon augmented regression test for exo- geneity
shows that
the null
hypothesis of
exogeneity is rejected at the 5 significance level.
Panel A in Table 4 presents the results when life expectancy is the health indicator used. In
general, the environmental stress variables have the correct sign negative and are significant for
certain pollutants such as CO
2
, EMW, ENPC and DEFRTE.
13
GNP is always significant and increases in income levels lead to an increase in
life expectancy.
The level
of immunisation
IMM is seen to increase life expectancy when the pollutant used is CO
2
or ENPC. Education level EDU does not seem to be significant for
improving health. The availability of doctors DOC as a proportion of the population has a
significant impact on improving life expectancy, particularly for the CO
2
, ENPC and EMW pol- lutants. When the health indicator used is
healthy life expectancy panel B of Table 4, the results are similar. The availability of doctors
increases healthy life expectancy significantly; the absolute impact measured by size of the
coefficients of this variable is also greater with HLE, as the dependent variable. The impact of
urbanisation UPOP on health is large and sig- nificant particularly for HLE as the dependent
variable. This positive impact reflects the soci- ety’s benefits from improvements in the provi-
sion of better waste disposal and sanitation facilities, which would come with urbanisation.
When infant mortality is taken as the health indicator panel A of Table 5, we find that in-
creases in TSP emissions and water pollutant emissions levels lead to significantly high infant
mortality. Coefficients derived from the estima- tion of the model using the child mortality rates
are very similar to the infant mortality results panel B of Table 5. In general, the results
show that income, immunisation rates, access to doctors and urbanisation levels all make large
positive and significant impacts on both infant and child mortality rates. Only the education
variable fails to make a significant impact on mortality rates.
We also use log linear models to check for robustness of results. Per capita GNP, purchas-
ing power parity, is found to be significant in all cases in improving health. Using log TSP, log
SO
2
, log NO
x
and log EMW as the environmen- tal stress variable, it is found that the coeffi-
cients are negative and significant for log SO
2
and log NO
x
in explaining health outcomes. Hence, the environmental variable is significant
in explaining changes in health levels in a popu- lation. The estimated coefficients for log CO
2
and log ENPC, however, have positive coeffi- cients and are significant. This is contrary to
what we would expect. We find, therefore, that in some cases the results could be sensitive to
the functional form used.
14
6. Conclusion and further research