freedom, growth, and the environment
TRANSCRIPT
Freedom, growth, and the environment
SCOTT BARRETTPaul H. Nitze School of Advanced International Studies, Johns HopkinsUniversity, 1619 Massachusetts Avenue, NW, Washington, DC20036–2213, USA. Tel: (202) 663–5761. Fax: (202) 663–5769. email: [email protected]
KATHRYN GRADDYExeter College, Oxford University
ABSTRACT. A number of recent papers have found that certain measures of pollutionworsen and later improve as income per head increases. It is widely believed that thedownhill portion of this inverted-U curve reflects an induced policy response; that, asincomes rise, citizens demand improvements in environmental quality, and that thesedemands are delivered by the political system. In this paper we find that, for a numberof pollution variables, an increase in civil and political freedoms significantly improvesenvironmental quality. For other pollution variables, however, we find that freedomshave no effect. The former finding suggests that political reforms may be as important aseconomic reforms in improving environmental quality worldwide. The latter findinghints that the observation that pollution levels fall with income, once income becomeshigh enough, may not always reflect an induced policy response.
IntroductionIs economic growth good or bad for the environment? Recent empiricalevidence points to a pattern reminiscent of the Kuznets curve: for anumber of indicators of environmental quality (but not all), economicgrowth seems to be accompanied by a deterioration in environmentalquality at low income levels and an improvement in environmental qualityat high income levels.1 This is a striking result; but, as we demonstrate inthis paper, it is also a partial result.
Environment and Development Economics 5 (2000): 433–456Copyright © 2000 Cambridge University Press
We are grateful to Gene Grossman and Alan Krueger for supplying us with theircomputer programs and data. We also thank Freedom House for making their dataavailable to us. Our research was partly funded by a MacArthur Foundation grantto the Global Environment and Trade Study. We have learned much about the topicof this paper from Partha Dasgupta, Paul Ehrlich, and Karl-Göran Mäler. JagdishBhagwati, Robert Deacon, Charles Perrings, and Todd Sandler provided helpfulcomments on an earlier draft, as did participants at the June 1998 World Congressof Environmental and Resource Economists, the NBER’s 1998 Summer Institute,and the GETS/MacArthur Foundation Workshop on Trade and Environment.
1 See the special issue of this journal, Volume 2, Part 4, October 1997. See alsoCropper and Griffiths (1994), Grossman and Krueger (1993, 1995), Hilton andLevinson (1998), Holtz-Eakin and Selden (1995), Selden and Song (1994),Schmalensee, Stoker, and Judson (1998), Shafik (1994), and World Bank (1992).
Though the estimated models are in a reduced form, and so cannot tellus why the inverted-U might exist, there is a kind of consensus on the prin-cipal mechanism at work. This is that there is an induced policy response:as nations become richer, their citizens demand that the non-materialaspects of their standard of living be improved. But if this reasoning iscorrect, then the observed levels of environmental quality will depend onmore than a nation’s prosperity. They will depend also on citizens beingable to acquire information about the quality of their environment, toassemble and organize, and to give voice to their preferences for environ-mental quality; and on governments having an incentive to satisfy thesepreferences by changing policy, perhaps the most powerful incentivebeing the desire to get elected or re-elected. In short, they will depend oncivil and political freedoms. These vary widely, however, both acrosscountries and over time. So the omission of these variables could beimportant. In this paper we re-estimate the relations estimated byGrossman and Krueger (1995; hereafter, G–K), including as explanatoryvariables certain measures of civil and political freedoms. In essence, weask a different question from the one that introduced this paper. We ask:Is more freedom good or bad for the environment?
For a number of measures of environmental quality, we find that ourfreedom variables are jointly highly significant.2 Inclusion of the freedomvariables does not affect the qualitative nature of the relationship betweenpollution and per capita income or the associated turning points; we findno evidence that the results reported by G–K are biased. But our results doshow that, for a number of different measures, environmental quality isincreasing in the extent of civil and political freedoms. In this sense, ourresults suggest that more freedom is good for the environment. Moreover,the effect is quantitatively and not just statistically significant. In the caseof sulfur dioxide the effect is especially strong: we find that a low freedomcountry, with an income level near the peak of the inverted-U, can reduceits pollution at least as much by increasing its freedoms as it can byincreasing its income per head. This is policy relevant if, as we argue, free-doms can be controlled independently of incomes.
Levels of some measures of environmental quality, however, seem not todepend on freedoms, even when the underlying relationship exhibits astrong inverted-U, and this hints that the downhill portion of the inverted-U
434 Scott Barrett and Kathryn Graddy
2 After writing this paper, we learned that Torras and Boyce (1998) also used theFreedom House data to estimate similar relationships. They use a different trans-formation of these data and a somewhat different model to determine the effect ofthe ‘distribution of power’ on pollution levels but obtain broadly similar results.Congleton (1992), Murdoch and Sandler (1997), Murdoch, Sandler, and Sargent(1997), and Fredriksson and Gaston (1998) also test for links between democracyand environmental policy. However, these papers are concerned with trans-boundary environmental damage—a very different situation than examined inthis paper. Bohn and Deacon (1997) examine the connections between investmentin natural resource industries and measures of both political stability and the typeof government (distinguishing, for example, between parliamentary and non-par-liamentary democracies). This again is a different kind of analysis using differentdata.
may not always result from an induced policy response. True, even dictator-ships have to please some constituency to remain in power (see Olson, 1993).So, even if our freedom variables were not significant, a kind of inducedpolicy response could still underly the inverted-U. But it does strike us aspeculiar that freedoms would show up as being very significant in somecases and not at all significant in others. We comment on possible reasons forthese results, and their implications for future research, later in the paper.
Before proceeding with the substance of the paper, we wish to empha-size that our results are not immune to the criticisms that have beenlevelled against the environmental Kuznets curve literature.3 Like pre-vious studies, we consider a selected number of pollutants only; ourresults imply nothing about optimality; and the models from which theyare derived ignore system-wide linkages. What we add in this paper is aperspective that has not been expressed previously. We believe that ourresults are important, but they are important only in this context.
Section 1 discusses the data used in our study, while sections 2 and 3present our econometric results for air and water pollution, respectively.Section 4 extends these results, by controlling for country fixed effects. Thefinal section of the paper discusses the implications of our work, both forthe literature on growth and environment and the literature on trade andenvironment.
1. DataWith the exception of our freedom variables, we rely entirely on the dataused by G–K. The pollution data were gathered by the Global Environ-mental Monitoring System and have two great virtues: they arecomparable across different countries; and they are direct measures ofenvironmental quality (as opposed to, say, pollution emission levels), andso are measures that citizens will have preferences over—preferences thatcan find expression at the ballot box or otherwise.4 Air quality is measuredin selected urban areas; water quality in selected river basins. The incomedata are from Summers and Heston (1991), and adjust for differences inpurchasing power. Note that these data are country averages, whereas thepollution readings are local.5 Time and physical features of the pollutionsites (like mean annual water temperature) are also included in our regres-sions as independent variables. For a more detailed description of thesedata, see Grossman and Krueger (1995).
Our measures of civil and political freedoms are constructed fromindices developed by Freedom House, which have been used widely in the
Environment and Development Economics 435
3 See in particular the policy forum in this journal, volume 1, part 1, February 1996.4 Shafik (1994) and the World Bank (1992) also examine a number of direct indi-
cators of environmental quality.5 Grossman and Krueger (1995, p. 361) note that the use of country-level GDP is
appropriate since ‘environmental standards are often set at a national level’. Thisstatement is correct, but there are also variations. For example, in the UnitedStates, some regions have not attained the standards that were set for them.Furthermore, environmental standards are not always uniform, even if they areset at the national level. In the United States, for example, regions with superiorair quality have been protected from any significant deterioration in air quality.
literature.6 The civil freedoms index reflects constraints on the freedom ofthe press, and on the rights of individuals to debate, to assemble, todemonstrate, and to form organizations, including political parties andpressure groups. The political freedoms index reflects whether a govern-ment came to power by election or by the gun, whether elections, if any,are free and fair, and whether an opposition exists and has the opportunityto take power at the consent of the electorate.
The Freedom House indices of freedoms are on a scale from 1 to 7, where7 is the lowest level of freedom. Following Barro (1996), we convert theseindices to a scale of from 0 to 1, where 0 now corresponds to the lowestlevel of freedom. Since the ranking of freedoms by Freedom House isordinal, it might seem natural to represent each of the seven indexnumbers by dummies. However, there are two reasons for not doing so:there are too few observations in some freedom categories, and the inclu-sion of many dummies would result in a loss in degrees of freedom. Thissecond reason is especially important if one believes, as we do, that currentpollution levels should depend not just on current freedoms but on therecent history of freedoms. We therefore employ two different representa-tions of the freedom data. The first enters the converted indices directly,and therefore on a linear scale. The second uses grouped dummies. In cre-ating these dummies, we again follow Barro by defining a low freedomcountry as having an index number between 0 and 0.33, and medium andhigh freedom countries as having index numbers of from 0.33 to 0.67 andfrom 0.67 to 1, respectively. The low political (civil) freedom dummy takeson a value of 1 if the country is a low political (civil) freedom country andzero otherwise. The other dummies are calculated in a similar fashion. Allfreedom variables enter our regressions as a moving average of the currentyear and the previous three years to reflect the expected lag betweenchanges in freedoms and changes in environmental quality.
For the primary regressions, we use the same basic regression model asG–K.7 Of course, one might explore alternative specifications, such asincluding income–freedom interactive terms. However, G–K include manyincome variables and we use different freedom variables. It is not obviouswhich variables should be interacted or whether different variables shouldbe constructed for this purpose. Nor is it obvious that richer specificationswould be appropriate, given the data limitations. Perhaps more import-antly, our ambition for this paper is modest. We only want to determinewhether politics can be shown to intermediate between income and pol-lution, as is so often supposed without any supporting evidence. If the linkreally does exist, it should show up in our model; and if our model canshow this while making only the smallest deviation from G–K’s, which isnow a standard in this literature, so much the better. There is, however,
436 Scott Barrett and Kathryn Graddy
6 See Barro (1996), Dasgupta (1993), Helliwell (1994), Murdoch, Sandler, andSargent (1997), and Perotti (1996).
7 As in G–K, we estimate the model using generalized least squares and employ arandom effects estimator that takes into account the unbalanced nature of thepanel (see G–K for details). Note that the peak of the inverted-U can be sensitiveto the functional form used in estimation (see Hilton and Levinson, 1998).
one specification change that does seem warranted, if only as a test forrobustness, and this is to control for fixed effects. These results arereported separately, toward the end of the paper.
2. Results for air pollutantsTable 1 presents the summary statistics for three different air pollutants. Anumber of observations follow from these. First, incomes and freedoms arepositively correlated. In particular, the high freedom countries are rich; thelow and medium freedom countries are poorer. Second, pollution levelsare negatively correlated with freedom with one small exception: sulfurdioxide concentrations are about the same for countries with low andmedium political freedoms. Finally, the correlations between freedomsand income and freedoms and pollution are very similar for both politicaland civil freedoms. This is because civil and political freedoms are veryhighly correlated. The correlation coefficients between the linear trans-formations of the civil and political freedom indices are about 0.95 for thedata corresponding to each of the three pollutants.
Table 2 presents the regressions for the three measures of air pollution
Environment and Development Economics 437
Table 1. Summary statistics by freedom indices for urban air pollutants (means and standard deviations)
Political freedoms Civil freedoms
Low Medium High Low Medium High
Sulfur dioxideMedian daily 48.81 51.53 25.95 53.66 34.08 26.07observed level (41.34) (42.93) (25.72) (41.91) (36.18) (25.63)
Income 2.60 3.82 9.69 2.65 2.85 10.23(1.24) (1.17) (4.36) (1.27) (1.61) (3.95)
No. of obs. 348. 81. 889. 305. 182. 831.
No. of countries 15. 7. 25. 11. 14. 25.
SmokeMedian daily 73.86 57.67 30.32 88.94 49.40 30.01observed level (53.39) (26.79) (36.00) (55.68) (33.11) (34.99)
Income 3.43 3.86 8.12 3.70 3.43 8.32(0.77) (1.29) (2.57) (0.50) (1.18) (2.41)
No. of obs. 94. 49. 323. 65. 90. 311.
No. of countries 5. 4. 13. 3. 6. 13.
Heavy particlesMedian daily 252.62 146.23 94.81 267.39 169.11 83.36observed level (134.32) (85.34) (90.45) (133.79) (108.20) (73.50)
Income 2.28. 3.62 11.51 2.41. 2.42 12.26(1.32) (1.20) (5.16) (1.37) (1.55) (4.49)
No. of obs. 310. 78. 621. 281. 149. 579.
No. of countries 10 6 16 10 10 16
438 Scott Barrett and Kathryn Graddy
Tab
le 2
.The
det
erm
inan
ts o
f urb
an a
ir p
ollu
tion
(ra
ndom
effe
cts
esti
mat
ion)
. Dep
ende
nt v
aria
ble
is a
nnua
l med
ian
conc
entr
atio
n
Var
iabl
eSu
lfur
diox
ide
Smok
eH
eavy
par
ticl
es
12
34
56
78
9
Inco
me
(100
0s)
�3.
53�
9.76
�7.
8737
.52
33.9
633
.35
23.1
829
.17
9.83
(9.3
1)(9
.37)
(9.5
2)(2
1.55
)(2
1.40
)(2
1.27
)(2
2.33
)(2
2.75
)(2
2.33
)In
com
e sq
uare
d0.
791.
391.
27�
10.5
9�
10.0
7�
9.99
�1.
87�
2.37
�0.
21(1
.14)
(1.1
4)(1
.16)
(3.7
7)(3
.73)
(3.7
3)(2
.84)
(2.8
9)(2
.84)
Inco
me
cube
d�
0.03
�0.
05�
0.04
0.60
0.58
0.57
0.05
0.06
�0.
01(0
.04)
(0.0
4)(0
.04)
(0.1
8)(0
.18)
(0.1
8)(0
.10)
(0.1
0)(0
.10)
Lag
ged
inco
me
19.9
125
.03
25.4
0�
16.9
3�
14.3
7�
24.4
6�
66.3
2�
74.4
0�
45.5
4(9
.89)
(10.
02)
(10.
20)
(24.
55)
(24.
64)
(24.
45)
(24.
01)
(24.
60)
(24.
27)
Lag
ged
inco
me
�3.
36�
3.65
�3.
669.
539.
4911
.87
5.26
6.21
3.45
squa
red
(1.2
8)(1
.29)
(1.3
1)(4
.52)
(4.5
2)(4
.56)
(3.2
5)(3
.30)
(3.2
5)L
agge
d in
com
e0.
130.
130.
13�
0.63
�0.
64�
0.77
�0.
15�
0.18
�0.
10cu
bed
(0.0
5)(0
.05)
(0.0
5)(0
.23)
(0.2
3)(0
.23)
(0.1
2)(0
.12)
(0.1
2)L
ow p
olit
ical
�2.
0210
.82
38.4
5fr
eed
om d
umm
y(5
.94)
(9.9
9)(1
8.04
)L
ow c
ivil
26.9
924
.89
19.2
9fr
eed
om d
umm
y(5
.66)
(9.1
0)(1
4.97
)M
ediu
m p
olit
ical
8.24
1.82
67.8
2fr
eed
om d
umm
y(5
.61)
(8.0
1)(1
8.32
)M
ediu
m c
ivil
1.45
6.03
�19
.24
free
dom
dum
my
(3.7
0)(6
.56)
(9.4
1)Po
litic
al f
reed
oms
�17
.68
�19
.71
�10
4.96
(lin
ear
scal
e)(1
1.65
)(1
8.87
)(3
0.81
)
Environment and Development Economics 439
Civ
il fr
eed
oms
�15
.24
�47
.27
�8.
09(l
inea
r sc
ale)
(13.
87(2
6.73
)(3
6.79
)C
oast
�12
.44
�7.
62�
10.1
5�
35.2
3�
36.3
4�
37.7
7�
21.3
1�
29.0
7�
23.7
2(3
.82)
(3.6
3)(3
.85)
(8.2
7)(8
.18)
(8.3
3)(1
2.19
)(1
1.31
)(1
1.98
)D
eser
t0.
85�
4.13
�4.
8916
2.82
135.
8911
4.65
(11.
25)
(11.
08)
(11.
38)
(26.
35)
(25.
89)
(27.
54)
Cen
tral
cit
y3.
054.
723.
504.
782.
75�
0.51
25.6
024
.54
20.1
9(4
.39)
(4.1
1)(4
.33)
(8.8
2)(8
.80)
(8.8
5)(1
4.67
)(1
3.22
)(1
4.40
)In
dus
tria
l�
0.32
�1.
13�
1.32
�13
.61
�12
.77
�14
.53
22.5
212
.74
13.0
2(5
.38)
(5.0
2)(5
.29)
(10.
84)
(10.
56)
(10.
70)
(17.
80)
(16.
03)
(17.
52)
Res
iden
tial
�11
.16
�10
.85
�11
.14
�16
.47
�14
.33
�14
.77
6.87
�3.
95�
1.79
(4.9
4)(4
.63)
(4.8
6)(9
.36)
(9.1
4)(9
.34)
(16.
67)
(15.
24)
(16.
41)
Popl
uati
on d
ensi
ty1.
121.
141.
184.
374.
023.
03�
1.57
�.1
79�
1.03
(pop
/sq
mile
)(1
.32)
(1.3
2)(1
.33)
(1.3
4)(1
.33)
(1.3
9)(1
.97)
(1.7
9)(1
.95)
Yea
r�
1.55
�1.
66�
1.51
�0.
86�
0.70
�0.
270.
68�
0.03
0.75
(0.2
2)(0
.22)
(0.2
3)(0
.37)
(0.3
9)(0
.40)
(0.6
4)(0
.67)
(0.6
7)N
o. o
f ob
serv
atio
ns13
18.
1318
.13
18.
466.
466.
466.
1009
.10
09.
1009
.P-
valu
e (c
ombi
ned
free
dom
var
iabl
es)
�0.
0001
0.00
04�
0.00
010.
0006
�0.
0001
�0.
0001
P-va
lue
(com
bine
din
com
e va
riab
les)
�0.
0001
�0.
0001
�0.
0001
�0.
0001
0.00
1�
0.00
01�
0.00
01�
0.00
01�
0.00
01P-
valu
e (c
ombi
ned
free
dom
and
inco
me
vari
able
s)�
0.00
01�
0.00
01�
0.00
01�
0.00
01�
0.00
01�
0.00
01�
0.00
01
Not
e: R
egre
ssio
ns a
lso
incl
ude
a co
nsta
nt, a
dum
my
vari
able
to in
dic
ate
that
the
type
of
area
is u
nkno
wn,
and
a d
umm
y to
ind
icat
eth
at th
e m
easu
rem
ent d
evic
e is
a g
as b
ubbl
er.
(here and elsewhere standard errors are shown in parentheses).Regressions 1, 4, and 7 replicate the G–K regressions with the exceptionthat freedoms data for Hong Kong are unavailable and so this territory isexcluded. The omission of Hong Kong does not affect the qualitativenature of the relationship between pollution and income but it does affectthe turning point in the case of smoke. This rises from $6,151 in the G–Kstudy to $7,286 (the turning point for sulfur dioxide rises from $4,053 in theG–K study to just $4,202; both regressions reveal that heavy particlesdecrease monotonically with income).
Regressions 2, 5, and 8 add the freedom dummies. In all three regres-sions, the freedom dummies are jointly significant at the 1 per cent level,as is income. There are, however, some surprising results. These includethe low coefficient on the low political dummy in the sulfur dioxideregression and the high and low coefficients on medium political and civilfreedoms in the heavy particles regression. However, as noted earlier, pol-itical and civil freedoms are highly correlated. Countries with high civilfreedoms also have high political freedoms, and countries with low civilfreedoms have low political freedoms. Thus, while the different definitionsof freedom may have different effects, the data indicate that they are not(and, possibly, that they cannot be) controlled separately. We thereforecombine these freedom variables for the purposes of making predictions.The combined effect of political and civil freedoms—something we calldemocratic freedoms—is striking: pollution levels are monotonicallydecreasing in the extent of democratic freedoms.
The combined effect of freedoms is shown more clearly in figures 1a, 2a,and 3a. Following G–K, these graphs were constructed by assuming thatcurrent and lagged income take on the same values.8 All other variables,excluding freedoms, take on their mean values. The low democratic free-doms curve was obtained by setting the low political and low civilfreedom dummies equal to 1 and by setting all the other freedom dummiesequal to 0. The medium and high democratic freedoms curves were calcu-lated in a similar manner. The number of the shapes in these graphsindicate the number of observations for each income level in the panel dataset.9 The vertical ranges of these graphs represent four standard deviationsin the dependent variables.
Figure 1a shows that the relationship between sulfur dioxide concen-trations and freedoms is nearly linear. This is not true for the otherpollutants. For smoke, as figure 2a shows, a move from low to mediumdemocratic freedoms has the biggest effect. For heavy particles, however,figure 3a shows that the move from medium to high democratic freedomshas the more substantial effect.
For comparison, regressions 3, 6, and 9 measure civil and political free-doms on the linear scale. Of course, the results reported above suggest
440 Scott Barrett and Kathryn Graddy
8 G–K construct the variable by multiplying GDP, GDP squared, and GDP cubed bythe sum of the estimated coefficients for current and lagged GDP.
9 Note that the rising portion of the relation for sulfur dioxide at the higher incomelevels only includes data for Canada, Switzerland, and the United States. It is forthis reason that G–K discount this portion of the relation.
that, at least for smoke and heavy particles, the linear specification is a veryrough approximation. However, it must be remembered that the dummyspecification aggregates over different raw index numbers. So both repre-sentations of freedoms—the linear form and the dummies—imposerestrictions, and neither is obviously the more superior. We note, however,that the coefficients on both the civil and political freedom variables arenegative in all three regressions.
Figures 1b, 2b, and 3b depict the relationship between income and pol-lution with the freedom variables excluded (as in G–K) and included,where in the latter case the freedom dummies are set equal to their mean
Environment and Development Economics 441
Figure 1. GEMS/AIR: Sulfur dioxide in cities
values. In all three cases, the former regression predicts a higher concen-tration of pollution than the latter at low incomes (if only very slightly) anda lower concentration of pollution than the latter at high incomes. This isto be expected. Recall that freedoms and incomes are positively correlated(the correlation coefficients between the linear freedom variables andincome range between 0.72 and 0.84 for the three data sets). Low incomecountries tend to have fewer freedoms than the average, and our regres-sions indicate that they should therefore have higher pollution levels. Highincome countries tend to have more freedoms than the average and soshould have lower pollution levels.
442 Scott Barrett and Kathryn Graddy
Figure 2. GEMS/AIR: Smoke in cities
Note,finally, that thecircles infigures1b,2band3bshowthemeanresidualfor the fitted regressions including the freedom variables for each $2,000income interval, with the size scaled to reflect the number of observations ineach interval. A comparison with G–K reveals a very close correspondence.There is thus no indication that the G–K estimates are biased.
3. Results for water pollutantsTable 3 presents the summary statistics for 11 different measures of waterpollution (but note that dissolved oxygen is a measure of water quality, notpollution). These confirm a general positive association between incomeand freedom. However, in contrast with the data for air pollutants, the
Environment and Development Economics 443
Figure 3. GEMS/AIR: Heavy particles in cities
444 Scott Barrett and Kathryn Graddy
Table 3. Summary statistics by freedom indices for water pollutants (means andstandard deviations)
Political freedoms Civil freedoms
Low Medium High Low Medium High
Dissolved oxygenMean daily 7.72 6.90 8.60 7.75 6.92 8.88observed level (2.82) (5.13) (2.46) (2.93) (4.22) (2.43)
Income 2.51 4.07 8.60 2.51 2.97 9.92(1.49) (1.57) (5.20) (1.42) (2.03) (4.42)
No. of obs. 296. 296. 1007. 273. 466. 860.
No. of countries 22. 20. 31. 21. 21. 26.
Biological oxygen demandMean daily 3.32 7.99 7.34 3.58 7.19 7.55observed level (3.89) (19.02) (27.50) (4.43) (19.01) (29.11)
Income 2.51 4.06 6.62 2.51 2.97 8.11(1.49) (1.56) (4.69) (1.41) (2.02) (4.10)
No. of obs. 269. 293. 722. 251. 459. 574.
No. of countries 21. 18. 27. 20. 18. 23.
Chemical oxygen demandMean daily 18.36 59.32 51.14 18.92 48.56 57.86observed level (16.17) (102.72) (136.60) (16.08) (84.48) (158.14)
Income 1.97 4.54 6.00 1.95 2.99 8.00(1.21) (1.34) (4.91) (1.10) (2.14) (4.36)
No. of obs. 120. 200. 530. 123. 346. 381.
No. of countries 10. 9. 19. 9. 12. 18.
NitratesMean daily 0.88 1.21 1.79 1.04 0.84 2.07observed level (1.25) (1.47) (4.63) (1.35) (1.31) (5.12)
Income 2.67 3.70 7.97 2.65 2.31 9.79(1.61) (1.66) (5.54) (1.55) (1.89) (4.72)
No. of obs. 210. 131. 676. 187. 292. 538.
No. of countries 18. 13. 21. 15. 18. 18.
Fecal coliformsMean daily 218370. 67967. 80584. 237319. 107919. 59701.observed level (788312) (210883) (609272) (819458) (450290) (585941)
Income 2.65 4.09 8.78 2.51 3.02 10.20(1.49) (1.60) (5.36) (1.41) (2.04) (4.53)
No. of obs. 224. 227. 810. 206. 368. 687.
No. of countries 17. 14. 27. 15. 18. 23.
Total coliformsMean daily 25050.70 106294.40 203578.10 60051.24 187972.80 182737.00observed level (99607.51) (263936.40) (1059427.00) (206218.10) (901880.50) (1031001.00)
Income 2.70 4.24 4.54 2.66 2.05 6.73(1.53) (1.32) (4.68) (1.34) (1.91) (4.65)
No. of obs. 25. 82. 387. 27. 223. 244.
No. of countries 5 9 14. 7. 11. 11.
Environment and Development Economics 445
relationship between water pollution levels and freedom is not alwaysnegative. Levels of biological and chemical oxygen demand, nitrates, totalcoliforms, and cadmium generally increase with the extent of freedom.10
Table 3. Continued
Political freedoms Civil freedoms
Low Medium High Low Medium High
LeadMean daily 0.071 0.031 0.013 0.073 0.026 0.013observed level (0.099) (0.067) (0.028) (0.098) (0.064) (0.029)
Income 2.46 4.21 11.44 2.26 4.30 11.46(1.51) (0.80) (2.92) (1.17) (0.87) (2.88)
No. of obs. 43.68 499. 44. 68. 498.
No. of countries 4. 8. 14. 5. 7. 12.
CadmiumMean daily 0.021 0.003 0.050 0.023 0.003 0.050observed level (0.063) (0.007) (0.179) (0.066) (0.007) (0.179)
Income 2.47 4.10 11.33 2.18 4.12 11.31(1.51) (1.07) (2.83) (1.20) (1.17) (2.84)
No. of obs. 43. 62. 544. 39. 64. 546.
No. of countries 5. 8. 19. 4. 9. 17.
ArsenicMean daily 0.007 0.010 0.005 0.005 0.010 0.005observed level (0.008) (0.011) (0.006) (0.006) (0.011) (0.006)
Income 2.14 4.44 11.63 1.96 4.45 11.61(0.76) (0.70) (2.94) (0.38) (0.69) (2.96)
No. of obs. 17. 49. 302. 16. 49. 303.
No. of countries 3. 4. 12. 3. 5. 11.
MercuryMean daily 0.345 0.176 0.297 0.335 0.153 0.301observed level (0.238) (0.344) (0.832) (0.261) (0.306) (0.832)
Income 2.72 4.29 11.29 3.25 4.31 11.28(1.77) (0.82) (2.81) (1.55) (0.95) (2.80)
No. of obs. 17. 62. 558. 14. 64. 559.
No. of countries 5. 8. 21. 5. 8. 19.
NickelMean daily 0.010 0.009 0.010 0.009observed level (0.010) (0.011) (0.011) (0.011)
Income 4.22 12.05 4.26 12.01(0.67) (2.58) (0.72) (2.64)
No. of obs. 14. 336. 12. 338.
No. of countries 3. 8. 3. 8.
10 For nickel, there is only one observation for low political freedoms (the Philip-pines in 1982) and no observations for low civil freedoms. Hence, we cannotinclude these dummy variables in our regression. Instead, we include the single observation for low political freedoms in the medium political freedomsdummy.
446 Scott Barrett and Kathryn GraddyT
able
4.T
he d
eter
min
ants
of w
ater
pol
luti
on (
rand
om e
ffect
s es
tim
atio
n). S
peci
ficat
ion
wit
h du
mm
y po
litic
al a
nd fr
eedo
m v
aria
bles
Var
iabl
eD
OB
OD
CO
DN
ITFE
CC
OL
LEA
DC
AD
AR
SM
ER
CN
ICK
EL
Inco
me
(100
0s)
1.33
475
1.81
421
�21
.931
43�
0.84
600
�0.
8794
0�
2.41
657
�0.
0074
6�
0.01
626
0.00
599
0.11
420
�0.
0279
1(0
.702
84)
(4.5
8019
)(3
7.32
588)
(1.3
5881
)(0
.624
40)
(2.1
7754
)(0
.035
39)
(0.1
2283
)(0
.003
66)
(0.5
7727
)(0
.021
01)
Inco
me
squa
red
�0.
1251
40.
3027
10.
6478
80.
1455
60.
1373
20.
2783
2�
0.00
032
0.01
020
�0.
0004
7�
0.02
906
0.00
242
(0.0
8136
)(0
.620
89)
(4.7
8137
)(0
.173
16)
(0.0
7169
)(0
.333
71)
(0.0
0315
)(0
.010
82)
(0.0
0032
)(0
.051
60)
(0.0
0170
)
Inco
me
cube
d0.
0037
6�
0.02
101
0.01
167
�0.
0052
9�
0.00
497
�0.
0081
80.
0000
4�
0.00
046
0.00
001
0.00
129
�0.
0000
7(0
.002
77)
(0.0
2469
)(0
.185
42)
(0.0
0602
)(0
.002
40)
(0.0
1435
)(0
.000
09)
(0.0
0032
)(0
.000
01)
(0.0
0154
)(0
.000
05)
Lag
ged
inco
me
�1.
3790
2�
0.63
921
35.4
4237
0.49
990
0.82
265
3.73
928
0.01
076
�0.
0082
8�
0.00
107
0.12
265
0.02
705
(0.7
449)
(4.8
4620
)(4
1.23
138)
(1.4
1613
)(0
.663
78)
(2.4
4551
)(0
.036
13)
(0.1
3492
)(0
.003
02)
(0.5
6977
)(0
.022
28)
Lag
ged
inco
me
0.13
380
�0.
3587
7�
1.71
738
�0.
0159
9�
0.08
117
�0.
5908
2�
0.00
079
�0.
0059
8�
0.00
007
�0.
0080
7�
0.00
258
squa
red
(0.0
9232
)(0
.697
23)
(5.6
1768
)(0
.181
75)
(0.0
8157
)(0
.407
26)
(0.0
0344
)(0
.012
58)
(0.0
0029
)(0
0.05
440)
(0.0
0196
)
Lag
ged
inco
me
�0.
0034
80.
0200
5�
0.00
117
�0.
0021
0�
0.00
112
0.02
659
0.00
001
0.00
027
0.00
000
0.00
019
0.00
008
cube
d(0
.003
38)
(0.0
2936
)(0
.229
85)
(0.0
0704
)(0
.002
96)
(0.0
1011
)(0
.000
11)
(0.0
0039
)(0
.000
01)
(0.0
0173
)(0
.000
06)
Low
pol
itic
al�
0.07
768
�0.
0341
20.
8145
9�
0.43
365
0.68
480
�0.
9339
4�
0.01
829
�0.
0100
10.
0067
2�
0.05
999
�0.
0065
7fr
eed
om d
umm
y(0
.416
42)
(2.2
7928
)(1
8.12
813)
(0.6
8836
)(0
.416
18)
(0.9
1966
)(0
.023
84)
(0.0
7552
)(0
.002
91)
(0.3
8424
)(0
.006
83)
Low
civ
il0.
2401
50.
5344
4�
9.26
354
�0.
5439
60.
9732
3�
1.65
061
0.02
834
0.00
847
�0.
0022
7�
0.11
15�
0.00
145
free
dom
dum
my
(0.3
9912
)(2
.099
85)
(15.
3047
4)(0
.598
90)
(0;4
0379
)(0
.670
11)
(0.0
2357
)(0
.045
23)
(0.0
0416
)(0
.392
08)
(0.0
0647
)
Med
ium
pol
itic
al�
0.04
876
1.29
264
�6.
7094
0�
0.01
457
0.48
174
0.83
057
�0.
0029
20.
0310
30.
0065
6�
0.03
621
free
dom
dum
my
(0.3
3225
)(1
.789
04)
(14.
8241
7)(0
.613
82)
(0.3
2173
)(0
.476
20)
(0.0
1240
)(0
.083
04)
(0.0
0157
)(0
.204
05)
Med
ium
civ
il0.
0990
10.
1093
8�
2.28
055
�0.
5496
30.
0834
0�
0.90
901
0.02
585
0.00
423
�0.
0003
7�
0.11
713
free
dom
dum
my
(0.2
5443
)(1
.329
26)
(7.6
5207
)(0
.393
87)
(0.2
4156
)(0
.279
45)
(0.0
1517
)(0
.048
29)
(0.0
0235
)(0
.248
30)
No.
of
obse
rvat
ions
1599
.12
84.
850.
1017
.12
61.
494.
610.
649.
368.
637.
350.
P-va
lue
(com
bine
dfr
eed
om v
aria
bles
)0.
9795
0.88
990.
9488
0.36
29�
0.00
010.
0004
0.22
920.
9915
0.00
010.
9702
0.18
73P-
valu
e (c
ombi
ned
inco
me
vari
able
s)0.
0007
0.07
710.
2269
0.00
23�
0.00
010.
1796
0.38
910.
0176
0.00
120.
6305
0.10
66P-
valu
e (c
ombi
ned
free
dom
and
in
com
e va
riab
les)
0.00
480.
2352
0.54
100.
0009
�0.
0001
�0.
0001
0.00
770.
1065
�0.
0001
0.91
890.
2213
Not
e: T
he d
epen
den
t var
iabl
e in
all
regr
essi
on e
quat
ions
exc
ept f
or f
ecal
and
tota
l col
ifor
ms
is a
nnua
l mea
n co
ncen
trat
ion.
For
fec
al a
nd to
tal c
olif
orm
s th
e d
epen
den
tva
riab
le is
1 �
log
annu
al m
ean
conc
entr
atio
n. A
ll eq
uati
ons
incl
ude
mea
n te
mpe
ratu
re, y
ear
and
an
inte
rcep
t as
cova
riat
es.
Environment and Development Economics 447
Freedom data were available for all the countries and years for whichG–K had water pollution data, and so we were able to replicate exactly theG–K results for these measures of water pollution. For this reason, we onlyreport here the regression results with the freedoms variables included.These are shown in tables 4 and 5.
The regressions show that freedoms do not significantly affect theoxygen regime of rivers, as measured by dissolved oxygen (DO), biologicaloxygen demand (BOD), or chemical oxygen demand (COD). The concen-tration of nitrates (NIT) is significantly affected by freedoms, but onlywhen freedoms are entered in the linear form. Even then, the coefficientson political and civil freedoms cancel each other out. An increase in demo-cratic freedoms thus has no discernible quantitative effect on nitrate levels.The income variables are jointly significant in the dissolved oxygen andnitrates regressions, however, and the estimated relationships closelyresemble those estimated by G–K: dissolved oxygen follows the U-shapedrelationship and nitrates the inverted-U. That freedoms do not affect thesepollution levels suggests that the negative relationship between pollutionand income at higher income levels may not reflect an induced policyresponse. This may seem surprising, but oxygen loss does not threatenhuman health directly and nor are nitrates especially dangerous.11
Fecal coliform is different. It is a direct health hazard. Fecal contami-nated water carries infectious diseases like cholera and typhoid and isimplicated in many cases of diarrheal disease, roundworm infection, andschistosomiasis. We should expect levels of this pollution to depend onfreedoms, and this is confirmed by our regressions. An increase infreedom, however measured (civil or political, entered either in linear formor as dummies), reduces fecal contamination (denoted FEC in tables 4 and5). Figure 4a illustrates our results, and one can confirm that, as in the caseof the various air pollution measures, the underlying relationship is nearlyidentical to the G–K result. For the same income level, countries withhigher freedoms have lower pollution readings, but increases in income doseem to be needed to bring these readings down to very low levels.
Our results for total coliform are also highly significant, but the signs onthe freedom variables are the wrong way round (see also figure 5a). G–Kalso obtain unexpected results for total coliform—results which theydescribe as ‘baffling’.12 But unlike fecal coliform, total coliforms are notnecessarily disease producing. So at one level our results should not causemuch concern. However, it is also possible that our results reflect a samplebias: the regressions for total coliform rely on a much smaller number ofobservations. To see whether sample bias may be a problem, we re-esti-
11 Nitrates may cause ‘blue baby syndrome’ in newborns, but the condition is veryrare.
12 Specifically, G–K find that total coliforms increase sharply with incomes at highincome levels. Note, however, that Shafik (1994) obtains a very similar relation-ship for measures of fecal coliform. Shafik also finds that dissolved oxygendecreases monitonically with income. These differences can probably be traced tothe use of different data, but they are nonetheless worrying. Neither Shafik norG–K comment on these different results.
448 Scott Barrett and Kathryn Graddy
Tab
le 5
.The
det
erm
inan
ts o
f wat
er p
ollu
tion
(ra
ndom
effe
cts
esti
mat
ion)
. Spe
cific
atio
n w
ith
linea
r po
litic
al a
nd fr
eedo
m v
aria
bles
Var
iabl
eD
OB
OD
CO
DN
ITFE
CC
OL
LEA
DC
AD
AR
SM
ER
CN
ICK
EL
Inco
me
(100
0s)
1.31
259
1.39
026
�26
.737
34�
1.22
271
�0.
9162
4�
3.56
248
�0.
0191
5�
0.02
764
0.01
277
�0.
0497
6�
0.02
501
(0.7
0041
)(4
.555
44)
(36.
7916
4)(1
.365
06)
(0.6
2583
)(2
.194
39)
(0.0
3531
)(0
.116
85)
(0.0
0364
)(0
.559
81)
(0.0
2125
)
Inco
me
squa
red
�0.
1265
20.
3434
00.
8506
40.
2191
00.
1330
60.
4035
50.
0002
60.
0111
5�
0.00
100
�0.
0159
20.
0022
2(0
.081
14)
(0.6
1937
)(4
.710
49)
(0.1
7436
)(0
.072
15)
(0.3
3825
)(0
.003
13)
(0.0
1048
)(0
.000
31)
(0.0
5045
)(0
.001
73)
Inco
me
cube
d0.
0038
5�
0.02
233
0.01
151
�0.
0078
6�
0.00
471
�0.
0121
70.
0000
3�
0.00
048
0.00
003
0.00
095
�0.
0000
6(0
.002
76)
(0.0
2466
)(0
.183
07)
(0.0
0607
)(0
.002
42)
(0.0
1459
)(0
.000
09)
(.000
32)
(0.0
0001
)(0
.001
52)
(0.0
0005
)
Lag
ged
inco
me
�1.
3691
70.
1669
839
.287
060.
8965
40.
7908
26.
5864
20.
0269
60.
0073
7�
0.00
419
0.17
147
0.01
673
(0.7
3310
)(4
.772
77)
(39.
7071
7)(1
.405
16)
(0.6
5636
)(2
.403
55)
(0.0
3505
)(0
.127
01)
(0.0
0320
)(0
.542
39)
(0.0
2303
)
Lag
ged
inco
me
0.13
469
�0.
4631
2�
1.84
324
�0.
0843
9�
0.06
795
�1.
0649
5�
0.00
198
�0.
0075
80.
0002
2�
0.01
377
�0.
0018
0sq
uare
d(0
.091
10)
(0.6
8916
)(5
.398
02)
(0.1
8998
)(0
.081
01)
(0.3
9921
)(0
.003
39)
(0.0
1202
)(0
.000
30)
(0.0
5284
)(0
.002
03)
Lag
ged
inco
me
�0.
0035
40.
0240
0�
0.00
557
0.00
028
0.00
059
0.04
897
0.00
004
0.00
032
0.00
000
0.00
037
0.00
006
cube
d(0
.003
34)
(0.0
2911
)(0
.221
80)
(0.0
0696
)(0
.002
95)
(0.0
1860
)(0
.000
11)
(0.0
0038
)(0
.000
01)
(0.0
0170
)(0
.000
06)
Polit
ical
fre
edom
s�
0.61
956
0.36
574
�36
.556
456.
4448
9�
1.14
736
1.19
741
�0.
1038
80.
1352
60.
0012
60.
3715
50.
0167
2(0
.925
50(5
.119
23)
(31.
7274
8)(1
.481
18)
(0.8
8336
)(1
.795
87)
(0.0
4607
)(0
.128
97)
(0.0
0456
)(0
.731
80)
(0.0
2178
)
Civ
il fr
eed
oms
0.72
506
�0.
7388
765
.415
66�
6.88
875
�0.
6620
63.
3945
90.
0869
8�
0.15
639
�0.
0241
00.
4680
30.
0265
5(1
.140
17)
(6.1
6151
)(4
2.76
095)
(1.9
4052
)(1
.077
70)
(1.9
1052
)(0
.045
58)
(0.1
2254
)(0
.005
94)
(0.7
3128
)(0
.022
06)
No.
of
obse
rvat
ions
1599
1284
850
1017
1261
494
610
649
368
637
350
P-va
lue
(com
bine
dfr
eed
om v
aria
bles
)0.
7909
0.99
180.
3108
0.00
010.
0066
0.00
150.
0633
0.44
090.
0001
0.39
120.
0449
P-va
lue
(com
bine
din
com
e va
riab
les)
0.00
360.
0790
0.14
310.
0003
�0.
0001
0.00
080.
2831
0.01
67�
0.00
010.
4797
0.04
26P-
valu
e (c
ombi
ned
free
dom
and
inco
me
vari
able
s)0.
0014
0.16
540.
2242
0.00
00�
0.00
01�
0.00
010.
0012
0.02
76�
0.00
010.
6569
0.09
35
Not
e: T
he d
epen
den
t var
iabl
e in
all
regr
essi
on e
quat
ions
for
fec
al a
nd to
tal c
olif
orm
s is
ann
ual m
ean
conc
entr
atio
n. F
or f
ecal
and
tota
l col
ifor
ms
the
dep
end
ent v
aria
ble
is
1 �
log
annu
al m
ean
conc
entr
atio
n. A
ll eq
uati
ons
incl
ude
mea
n te
mpe
ratu
re, y
ear
and
an
inte
rcep
t as
cova
riat
es.
mated the relationships for both measures of coliform for a consistent setof data. The results are given in table 6 and illustrated in the figures 4b and5b. The fecal and total coliform regressions are now fairly consistent witheach other. Moreover, both the combined freedom variables and the com-bined income variables are no longer significant in the fecal coliformregressions. We conclude from this exercise that, as a result of the muchlarger data set, the fecal coliform relationships estimated in tables 4 and 5are the more reliable.
We turn finally to the results for heavy metals. Freedoms are jointly sig-
Environment and Development Economics 449
Figure 4. GEMS/WATER: Fecal coliforms in rivers
450 Scott Barrett and Kathryn Graddy
Table 6. Selected sample regressions for fecal and total coliforms (random effectsestimation). Dependent variable is 1 � log annual mean concentration
Variable Fecal coliforms Total coliforms
Income (1000s) 0.316 �0.972 �1.027 �2.064(2.519) (2.543) (2.250) (2.268)
Income squared �0.054 0.100 0.105 0.223(0.384) (0.390) (0.343) (0.347)
Income cubed 0.004 �0.002 �0.002 �0.006(0.016) (0.017) (0.015) (0.015)
Lagged income �0.316 2.009 1.495 3.690(2.872) (2.844) (2.565) (2.535)
Lagged income 0.035 �0.325 �0.256 �0.613squared (0.477) (0.470) (0.425) (0.419)
Lagged income �0.001 0.016 0.013 0.029cubed (0.022) (0.022) (0.020) (0.019)
Low political 0.010 �0.416freedom dummy (1.209) (1.078)
Low civil �0.333 �0.795freedom dummy (1.053) (0.940)
Medium political 1.267 0.829freedom dummy (0.579) (0.517)
Medium civil �0.610 �0.707freedom dummy (0.339) (0.303)
Political freedoms �0.367 �0.058(linear scale) (2.322) (2.067)
Civil freedoms 0.737 2.453(linear scale) (2.359) (2.100)
No. of observations 431. 431. 431. 431.
P-value (combined 0.181 0.945 0.106 0.048freedom variables)P-value (combined 0.730 0.450 0.135 0.203income variables)P-value (combined 0.181 0.479 0.003 0.005freedom and incomevariables)
Note: All equations include mean temperature, year and an intercept ascovariates.
nificant only in the case of arsenic (denoted ARS in tables 4 and 5). Whenfreedoms are entered in linear form, an increase in democratic freedomsreduces arsenic concentrations. When entered as dummies, arsenic levelsincrease slightly and then decrease with democratic freedoms. Theseresults are broadly consistent with the hypothesis of an induced policyresponse, but we note that these regressions rely on a small sample andmay not be reliable. Interestingly, we find that the relationship between
lead levels (LEAD) and income is not significant when freedoms areincluded as independent variables (the relationship between lead levelsand income is significant in G–K’s regression). However, the combinedeffect of income and freedom is very significant. This suggests that the gen-erally negative relationship between incomes and lead levels detected byG–K may indeed result from an induced policy response. As for the otherheavy metals, G–K found a statistically significant relationship betweenconcentrations of cadmium (CAD) and income, but the relationships fornickel (NICKEL) and mercury (MERC) were not significant.
Environment and Development Economics 451
Figure 5. GEMS/WATER: Total coliforms in rivers
Tab
le 7
.The
det
erm
inan
ts o
f pol
luti
on (
coun
try
fixed
effe
cts)
Spe
cific
atio
n w
ith
dum
my
polit
ical
and
free
dom
var
iabl
es
Var
iabl
eSO
2Sm
oke
H.P
art.
DO
BO
DC
OD
NIT
FEC
CO
LLE
AD
CA
DA
RS
ME
RC
NIC
KE
L
Inco
me
(100
0s)
�18
.395
6739
.847
8171
.952
180.
1413
13.
7738
8�
42.4
9774
�0.
0924
9�
2.64
295
�5.
7150
90.
0762
20.
0019
50.
0145
20.
0805
3�
0.03
611
(9.2
4979
)(2
1.71
658)
(21.
1561
8)(0
.820
16)
(5.2
0027
)(4
3.21
95)
(1.5
6464
)(0
.690
59)
(2.5
6909
)(0
.042
60)
(0.2
0186
)(0
.005
15)
(0.7
7917
)(0
.036
29)
Inco
me
squa
red
1.75
833
�9.
8813
0�
4.42
766
�0.
0139
70.
1069
22.
1904
9�
0.01
994
0.28
179
0.71
891
�0.
0071
30.
0073
2�
0.00
106
�0.
0197
10.
0032
4(1
.116
52)
(3.7
7838
)(2
.659
69)
(0.0
9092
)(0
.679
66)
(5.2
7734
)(0
.186
62)
(0.0
7679
)(0
.385
80)
(0.0
0368
)(0
.017
13)
(0.0
0043
)(0
.067
25)
(0.0
0290
)
Inco
me
cube
d�
0.05
411
0.54
370
0.09
658
0.00
062
�0.
0153
1�
0.02
530
0.00
128
�0.
0087
9�
0.02
579
0.00
021
�0.
0003
30.
0000
20.
0010
0�
0.00
009
(0.0
3825
)(0
.180
59)
(0.0
8940
)(0
.003
01)
(0.0
2636
)(0
.199
41)
(0.0
0634
)(0
.002
53)
(0.0
1655
)(0
.000
11)
(0.0
0048
)(0
.000
01)
(0.0
0192
)(0
.000
08)
Lag
ged
inco
me
28.9
9138
�16
.143
3311
.099
94�
2.30
027
1.02
451
10.3
6041
�0.
4580
9�
0.45
352
�2.
2590
4�
0.03
655
�0.
0444
3�
0.00
100
0.36
043
0.01
583
(10.
2968
3)(2
6.55
506)
(23.
3408
1)(0
.893
02)
(6.7
8312
)(7
6.97
612)
(1.8
1907
)(0
.757
08)
(4.2
9734
)(0
.035
38)
(0.1
7003
)(0
.003
47)
(0.6
3971
)(0
.030
52)
Lag
ged
inco
me
�5.
0551
510
.978
390.
4392
20.
1832
9�
0.56
164
0.49
379
0.09
522
�0.
0041
30.
1233
00.
0027
4�
0.00
359
�0.
0000
6�
0.01
559
�0.
0015
2sq
uare
d(1
.293
54)
(4.8
5435
7)(3
.121
59)
(0.1
1082
)(1
.063
27)
(11.
8598
5)(0
.235
73)
(0.0
9224
)(0
.699
90)
(0.0
0332
)(0
.015
55)
(0.0
0032
)(0
.059
75)
(0.0
0263
)
Lag
ged
inco
me
0.18
719
�0.
7302
1�
0.02
834
�0.
0034
00.
0266
9�
0.12
861
�0.
0030
0�
0.00
029
�0.
0006
6�
0.00
009
�0.
0002
60.
0000
00.
0004
30.
0000
5cu
bed
(0.0
4690
)(0
.241
65)
(0.1
1177
)(0
.004
25)
(0.0
4811
)(0
.536
93)
(0.0
0870
)(0
.003
45)
(0.0
3208
)(0
.000
11)
(0.0
0048
)(0
.000
01)
(0.0
0189
)(0
.000
07)
Low
pol
itic
al0.
1601
60.
4891
0�
50.4
8287
�0.
3303
12.
1323
111
.944
08�
0.09
322
0.86
889
1.84
166
�0.
0421
7�
0.00
636
0.02
722
�0.
1394
5�
0.00
887
free
dom
dum
my
(9.7
0882
)(1
2.50
615)
(50.
9419
6)(0
.536
56)
(2.7
9515
)(2
8.17
388)
(1.0
0371
)(0
.490
79)
(1.5
7049
)(0
.026
09)
(0.1
1450
)(0
.013
61)
(0.4
7033
)(0
.008
84)
Low
civ
il18
.313
7620
.918
9742
.235
790.
4810
5�
0.41
577
�11
.866
18�
0.54
050
1.15
279
�1.
5803
10.
0191
00.
0463
9�
0.00
463
�0.
9871
00.
0038
4fr
eed
om d
umm
y(5
.979
14)
(9.2
0029
)(1
4.52
888)
(0.4
2961
)(2
.266
40)
(16.
2700
7)(0
.647
70)
(0.4
1679
)(0
.702
58)
(0.0
2427
)(0
.122
79)
(0.0
0525
)(0
.470
82)
(0.0
0783
)
Med
ium
Pol
itic
al6.
3862
4�
5.89
065
�9.
8515
5�
0.27
199
2.82
724
�1.
4547
30.
3414
30.
4776
90.
9319
90.
0052
40.
0111
70.
0087
5�
0.03
845
free
dom
dum
my
(7.1
1051
)(8
.548
18)
(48.
9768
2)(0
.408
36)
(2.1
0275
)(2
1.85
618)
(0.8
5361
)(0
.378
75)
(0.6
1643
)(0
.013
31)
(0.0
6394
)(0
.001
70)
(0.2
4553
)
Med
ium
civ
il4.
1657
56.
5192
2�
10.6
7042
0.27
307
�0.
6771
1�
6.31
023
�0.
4967
20.
2366
8�
0.85
480
0.01
423
0.03
290
�0.
0011
2�
0.07
295
free
dom
dum
my
(3.5
6992
)(6
.563
73)
(8.8
7939
)(0
.269
01)
(1.4
0411
)(8
.242
74)
(0.4
1496
)(0
.250
99)
(0.3
1949
)(0
.016
62)
(0.0
7198
)(0
.003
49)
(0.3
0727
)
Cou
ntry
dum
mie
s40
1627
5045
3039
4121
1926
1427
9P-
valu
e<
0.00
01<
0.00
01<
0.00
01<
0.00
010.
1890
0.08
78<
0.00
01<
0.00
010.
0004
<0.
0001
<0.
0001
0.00
05<
0.00
010.
0071
P-va
lue
(com
bine
dfr
eed
om v
aria
bles
)0.
0076
0.12
24<
0.00
010.
8244
0.74
370.
8863
0.56
730.
0001
0.06
840
0.24
610
0.97
180
0.00
000
0.99
330
0.09
420
P-va
lue
(com
bine
din
com
e va
riab
les)
<0.
0001
<0.
0001
<0.
0001
0.00
080.
0990
0.19
020.
9256
<0.
0001
0.02
230.
2621
0.33
590.
0198
0.34
390.
4107
P-va
lue
(com
bine
dfr
eed
om a
nd in
com
eva
riab
les)
<0.
0001
<0.
0001
<0.
0001
0.00
620.
3257
0.51
500.
8488
<0.
0001
<0.
0001
0.31
30.
7186
0.00
00.
735
0.36
1
Not
e: T
he d
epen
den
t var
iabl
e in
all
regr
essi
on e
quat
ions
exc
ept f
or f
ecal
and
tota
l col
ifor
ms
is a
nnua
l mea
n co
ncen
trat
ion.
For
fec
al a
nd to
tal c
olif
orm
s th
e d
epen
den
t var
iabl
e is
1 �
log
annu
alm
ean
conc
entr
atio
n. U
nrep
orte
d, i
nclu
ded
cov
aria
tes
and
num
bers
of
obse
rvat
ions
are
as
in p
revi
ous
tabl
es.
4. Fixed effectsBecause political and civil freedoms vary little within countries over time,we have tested our results for robustness by including separate countrydummy variables. Essentially, our fixed effects regressions test whetherour results might reflect omitted country characteristics like geographythat just happen to be correlated with freedoms.
As shown in table 7, we find that our main results hold up surprisinglywell. The relationship between freedom and pollution is still highly sig-nificant for SO2, heavy particles, and fecal coliforms. While the direction ofthe effect for heavy particles appears somewhat ambiguous in theseregressions, if we include only a combined low civil and political freedomsdummy, the essence of our earlier results is still supported (the coefficienton the low freedoms dummy is 33.272 with a standard error of 12.680).Similarly, while the combined political and civil freedom variables are nolonger significant at the 5 per cent level in the smoke regression, the com-bined civil freedom variables on their own are significant.
We wish to stress that the fixed effects model provides an overly strongtest of the effect of freedoms on pollution levels. A country is what it ispartly because of the political and civil institutions it embraces. Countrieswith more repressive regimes may also subsidize energy consumption orconcentrate industrial production, so that in our earlier regressions theeffects of the subsidies and planning decisions may be reflected in thefreedom variables. But of course it is partly because freedoms are low inthese countries that such subsidies and central planning decisions can besustained. For this reason, it is best to consider the random and fixedeffects models as alternatives, each with its own advantages and disad-vantages.
5. ImplicationsEffective environmental protection requires both economic reforms, asArrow et al. (1995) and the World Bank (1992) have emphasized, and—aswe have shown—political reforms. Which type of reform ought to comefirst? We cannot say, but our results do clash with a related literaturewhich is less equivocal on this question. Barro (1996) argues that politicalfreedoms in poor countries should not be promoted directly but that econ-omic freedoms should be. One reason for this is that Barro finds that anincrease in political freedoms has a small adverse effect on growth, forcountries that have already attained a medium level of freedom. We find,however, that some measures of environmental quality are monotonicallyincreasing in the extent of freedoms. If GDP per head were synonomouswith welfare, then one might still agree with Barro that political freedomsought not be promoted (ignoring, of course, the direct contribution thatfreedom makes to well-being). But we know that GDP per head is an inad-equate measure of welfare, not least because it excludes the benefits ofenvironment improvements (Dasgupta and Mäler, 1995). To the extent thatfreedoms are instrumental in correcting for market failures, they mayincrease welfare, even if at the expense of growth.
Barro’s recommendation that economic reforms should precede politicalreforms is also supported by his finding that political freedoms tend to
Environment and Development Economics 453
erode over time if they get out of line with a country’s standard of living.The implication is that freedoms cannot be controlled independently ofincomes. But Barro’s estimates only reveal that democratic freedoms tendto increase with incomes; they do not indicate that a country cannotincrease freedom by more than the average for its income level, and onecan point to a number of countries that have increased freedoms rapidly(Barro might say prematurely) without ever sliding back: in our samplealone, these include Argentina, Spain, Portugal, Greece, and Brazil, not tomention a number of the former Communist countries of Eastern Europe.As Dasgupta (1993: 116) has put it, ‘the claim that the circumstances thatmake for poverty are also those that make it necessary for governments todeny their citizens political and civil liberties is simply false’. This meansthat our results do have real implications for policy. If the environmentalKuznets curve suggests that growth need not harm the environment, ourresults show that an expansion of freedoms may lead to an improvementin environmental quality.
Data for Brazil provide an anecdotal illustration of this contention.Between 1977 and 1987, per capita income increased, decreased, and thenincreased again. It was less than 10 per cent greater at the end of the periodthan at the beginning and averaged around $4,000 over this period, a levelnear the peak of the inverted-U for the sulfur dioxide curve (figure 1a).Such small changes in income, especially near the peak of the inverted-U,should have translated into small changes in pollution levels. However,sulfur dioxide concentrations in Brazil fell by half over this period. At thesame time, freedoms increased. The raw score of political and civil free-doms fell from 4 and 5, respectively in 1977 to 2 and 2 in 1987, enough tolift Brazil from being a medium to a high freedom country. Of course, thisis only an illustration, but it supports our interpretation of the statisticalevidence that freedoms can change independently of incomes, and have ademonstrable effect on environmental quality.
This conclusion has important implications for the trade and environ-ment debate. Environmentalists and protectionists claim that weakerenvironmental standards in developing countries should be brought intoalignment with the higher standards found in industrialized countries. AsBhagwati has repeatedly emphasized, however, diversity is usually to becheered.13 Certainly, in a model in which environmental standards arechosen by governments to maximize the welfare of their citizenry, diver-sity is beneficial. The hole in this argument is that only textbookgovernments choose standards in this way. Actual standards are chosen bya political process. Of course, we know that no form of government can besure of sustaining first-best outcomes every time. However, there is reasonto believe that countries with greater civil and political freedoms will do abetter job at supplying public goods (see Olson, 1993). The point is that, tothe extent that the interests of the citizens of some countries are poorly rep-resented, the appropriate remedy is not to impose an outcome (aharmonized standard) but rather to promote a move toward greater free-doms. As Bhagwati and Srinivasan (1996: 170) put it:
454 Scott Barrett and Kathryn Graddy
13 See, for examples, Bhagwati (1993, 1996).
even if one argues that the decisions made undemocratically by a dic-tatorship or an oligarchy are vitiated, there is no reason to believe thatthe higher standards being pursued by a foreign country representingthe competitive interests of a foreign industry or labor union in anindustry are what a more democratic process would yield. The correctapproach should rather to be encourage a shift to more democraticprocedures in arriving at social and economic legislation, includingenvironmental policy. Process, not outcomes (especially outcomessought by self-serving groups elsewhere), is what we should aim at incountries that lack democratic ways.
This paper shows that the promotion of freedoms will in many casesactually lead to improvements in environmental quality. That is, pro-motion of a more democratic process is not the only ‘right’ thing to do butit will also lead to the changes in outcomes desired by the environmentaland protectionist lobbies.
The mystery in our results is why freedoms should affect some measuresof environmental quality and not others. With the exception of heavymetals and total coliforms, which may suffer from small sample problems,freedoms significantly affect environmental quality measures that relatedirectly to human health: all three of our air pollution measures in additionto fecal contamination. But they do not affect DO, BOD, COD, or nitrates.Our results for the oxygen-related measures of water quality are especiallysurprising, because sewage treatment intended to reduce fecal coliformwould tend to increase DO and reduce BOD and COD in the bargain. Theresults seem to hint that something other than an induced policy responsemay lie behind the inverted-U relationships for these measures of waterquality. The distinction is important, for if the relationship is determinedby, say, fixed factors, it may not be stable; growth in the middle incomecountries may not be accompanied by reductions in pollution, assuggested by the inverted-U.
The latter explanation is at least plausible in the case of nitrates. Toreduce nitrates in rivers would require reductions in the use of fertilisers,or more specifically in fertiliser run-off, and changes in cropping tech-niques. The richer countries have done little to regulate these activities(occasionally, nitrates are removed from drinking water). Moreover, therelease of both unused fertiliser and organic nitrogen into rivers varieswith the type of crop that is planted as well as climatic conditions. Sosomething other than an induced policy response may explain theinverted-U, at least in this case. Research that links the inverted-U to actualpolicies would seem to be badly needed.
ReferencesArrow, K. et al. (1995), ‘Economic growth, carrying capacity, and the environment’,
Science, 268: 520–521.Barro, R.J. (1996), ‘Democracy and growth’, Journal of Economic Growth, 1: 1–27.Bhagwati, J. (1993), ‘Trade and the environment: the false conflict?’, in D. Zaelke, P.
Orbuch, and R.F. Housman (eds.), Trade and the Environment: Law, Economics, andPolicy, Washington, DC: Island Press, pp. 159–190.
Bhagwati, J. (1996), ‘The demands to reduce domestic diversity among trading
Environment and Development Economics 455
nations’, in J. Bhagwati and R.E. Hudec (eds.), Fair Trade and Harmonization:Prerequisites for Free Trade? Volume 1: Economic Analysis, Cambridge, MA: MITPress, pp. 9–40.
Bhagwati, J. and T.N. Srinivasan (1996), ‘Trade and the environment: does environ-mental diversity detract from the case for free trade?’, in J. Bhagwati and R.E.Hudec (eds.), Fair Trade and Harmonization: Prerequisites for Free Trade? Volume 1:Economic Analysis, Cambridge, MA: MIT Press, pp. 159–223.
Bohn, H. and R.T. Deacon (1997), ‘Ownership risk, investment, and the use ofnatural resources’, mimeo, Department of Economics, University of California,Santa Barbara.
Congleton, R.D. (1992), ‘Political institutions and pollution control’, Review ofEconomics and Statistics, 74: 412–421.
Cropper, M. and C. Griffiths (1994), ‘The interaction of population growth andenvironmental quality’, American Economic Review, 84: 250–254.
Dasgupta, P. (1993), An Inquiry into Well-Being and Destitution, Oxford: ClarendonPress.
Dasgupta, P. and K.G. Mäler (1995), ‘Poverty, institutions, and the environmentalresource-base’, in J. Behrman and T.N. Srinivasan (eds.), Handbook of DevelopmentEconomics, Amsterdam: Elsevier.
Fredriksson, P.G. and N. Gaston (1998), ‘Ratification of the 1992 climate change con-vention: what determines legislative delay?’, mimeo, World Bank.
Grossman, G.M. and A.B. Krueger (1993), ‘Environmental impacts of a NorthAmerican Free Trade Agreement’, in P. Garber (ed.), The US–Mexico Free TradeAgreement, Cambridge, MA: MIT Press.
Grossman, G.M. and A.B. Krueger (1995), ‘Economic growth and the environment’,Quarterly Journal of Economics, 110: 353–377.
Helliwell, J.F. (1994), ‘Empirical linkages between democracy and economicgrowth’, British Journal of Political Science, 24: 225–248.
Hilton, F.G.H. and A. Levinson (1998), ‘Factoring the environmental Kuznets curve:evidence from automotive lead emissions’, Journal of Environmental Economic andManagement, 35: 126–141.
Holtz-Eakin, D. and T.M. Selden (1995), ‘Stoking the fires? CO2 emissions and econ-omic growth’, Journal of Public Economics, 57: 85–101.
Murdoch, J.C. and T. Sandler (1997), ‘Voluntary cutbacks and pretreaty behavior:the Helsinki Protocol and sulfur emissions’, Public Finance Review, 25: 139–162.
Murdoch, J.C., T. Sandler, and K. Sargent (1997), ‘A tale of two collectives: sulphurversus nitrogen oxides emission reduction in Europe’, Economica, 64: 281–301.
Olson, M. (1993), ‘Dictatorship, democracy, and development’, American PoliticalScience Review, 87(3): 567–576.
Perotti, R. (1996), ‘Growth, income distribution, and democracy: what the data say’,Journal of Economic Growth, 1: 149–187.
Selden, T.M. and D. Song (1994), ‘Environmental quality and development: is therea Kuznets curve for air pollution emissions?’, Journal of Environmental Economicsand Management, 27: 147–162.
Schmalensee, R., T.M. Stoker, and R.A. Judson (1998), ‘World carbon dioxide emis-sions: 1950–2050’, Review of Economics and Statistics, 80: 15–27.
Shafik, N. (1994), ‘Economic development and environmental quality: an econo-metric analysis’, Oxford Economic Papers, 46: 757–773.
Summers, R. and A. Heston (1991), ‘The Penn World Table (Mark 5): an expandedset of international comparisons, 1950–1988’, Quarterly Journal of Economics, 106:327–369.
Torras, M. and J. Boyce (1998), ‘Income, inequality, and pollution: a reassessment ofthe environmental Kuznets curve’, Ecological Economics, 25: 147–160.
World Bank (1992), World Development Report 1992, Oxford: Oxford University Press.
456 Scott Barrett and Kathryn Graddy