Air Pollution India

1

The Effect of Social and Economic Development on Air Pollution in Indian Cities

Matthew J. Holian*

August 2, 2013

Abstract

This paper presents new estimates of air pollution production functions using data from Indian cities. The resulting estimates enable tests of various hypotheses concerning the effect of income, literacy, population and other variables on four measures of air pollution: sulfur dioxide, nitrogen dioxide, and two measures of particulate matter. Controlling for multiple factors, we find the relationship between social and economic development and pollution varies across pollution type and development indicator; we find a negative relationship between income and particulate matter, no relationship between income and sulfur dioxide emissions, and a positive relationship between income and nitrogen dioxide emissions. We also test for nonlinear relationships, but do not find strong non-monotonic relationships between income and pollution, though there is some evidence for non-monotonic relationships between literacy and pollution. We present new population elasticity estimates, and document large variation in air pollution levels across regions and industries.

Keywords: urban air pollution, India, environmental Kuznets curve, economic and social development

* Associate Professor, Department of Economics, San Jose State University, One Washington Square, San Jose,

California, USA, 95192. Phone: 408-924-1371. Fax: 408-924-5406. Email: [email protected]. I thank Sahil Gandhi, Matthew Kahn, Kala Seetharam Sridhar and audiences at IIHS Bangalore and CSU East Bay for helpful discussions related to this project. Anand Saxena and Haziq Siddiqi provided valuable research assistance. This research was supported by a RSCA grant from the California State University and by a grant from the College of Social Sciences at SJSU. I alone am responsible for any errors.

2

Introduction

In 2011, 31 per cent of Indians were officially classified as residing in urban areas. In

contrast, in more developed countries this figure is around 78 per cent, suggesting that as it

develops, India will become substantially less rural than it is today. However the speed at which

India urbanises depends in part on the quality of life in Indian cities. Dangerously high levels of

urban air pollution discourage potential new residents from settling in many cities in the

developing world. This has the effect of slowing urbanisation, encouraging emigration among

the highly skilled, and lowering economic growth.1

The goal of this study is to understand the determinants of pollution levels among

medium-sized and large Indian cities. Our primary approach is to estimate pollution production

functions for four measures of ambient air pollution: sulphur dioxide, nitrogen dioxide, and two

measures of particulate matter. We combine several high-quality data sources, which enable

tests of various hypotheses concerning the effect of income, literacy, population and other

variables on pollution levels. Given the disagreement among researchers on appropriate

proxies for average income at the local level, we study the effect of two different income

measures, as well as literacy, a proxy for social development. Understanding how key variables

relate to pollution—and how to measure these key variables—contributes to the work of

scholars in a variety of disciplines, and of policy makers seeking to improve urban quality of life.

Economic growth is fuelled by urbanisation and can lead to environmental degradation.

Emissions from factories are a clear example. However wealth generated through this process

also leads to environmental improvements. Paving formally dirt roads lowers particulate

matter levels, as does converting home heating and cooking fuels from biomass to greener

3

sources. The tension between growth and the environment has been the focus of many

scholars, and our results contribute to this literature.

Controlling for multiple factors, we find the relationship between social and economic

development and pollution varies across pollution type; we find a negative relationship

between development and particulate matter, no relationship between development and

sulphur dioxide emissions, and a positive relationship between development and nitrogen

dioxide emissions. We also test for non-monotonic relationships, and find some evidence that

pollution levels at first increase and eventually decrease with rising literacy. We also provide

elasticity estimates of the magnitude by which air pollution levels increase with city size and

compare them to findings from previous studies. Other noteworthy findings include a

significant ‘South’ effect, where some pollution measures are found to be markedly lower in

this region of the country. Finally, we briefly consider the effect of industry type on the four

pollution measures, suggesting which industry types are most correlated with emissions; output

in most industries is positively correlated with sulphur dioxide and nitrogen dioxide emissions,

but is negatively correlated with particulate matter levels.

The remainder of this paper is organised as follows. The next section reviews related

literature. This is followed by a discussion of our data and methodology, which in turn is

followed by a presentation of the results. After a brief section on pollution and industry type,

the conclusion summarises the findings and draws out implications for policy and future work.

4

Literature Review

This study relates to a variety of literatures; we describe three closely related economics

literatures here. The first attempts to rank cities according to their level of pollution. The

second tests the hypothesis that pollution first rises and then falls with development. We

describe a third literature that relates to the first two, but that is specifically related to pollution

in Indian cities. Finally, we briefly discuss work by non-economists to which the present study

closely relates.

The ‘green cities’ literature describes work that attempts to rank cities according to their

level of pollution, as well as to explain the determinants of varying pollution and its

consequences. Work on this area has focused on cities in the United States (Glaeser and Kahn,

2010) and more recently China (Zheng et al., 2011), and to an increasing extent on cities around

the world.2 For a recent study focusing on Asia, see Asian Development Bank (2012).

Kahn (2006) summarises economic theories related to growth and pollution in cities.

One of the most important of such theories is the Environmental Kuznets Curve (EKC), which

hypothesises that that initially, economic development increases pollution as the scale of

output increases. However eventually further development leads to offsetting effects; changes

in the composition of consumption and production, technological improvement, and

government regulation all serve to reduce pollution, even as the scale of output continues to

increase. Following Grossman and Krueger (1995), many studies have tested this hypothesis

empirically, with mixed results. Brock and Taylor (2005) summarise this large literature. Many

studies in this literature estimate reduced form production functions like those we present

5

below. However we are not aware of any of such estimates that use city-level data from India,

thus the present study fills this important gap in the literature.

The present study most closely relates to two recent papers. Managi and Jena (2008)

regress an environmental efficiency index on Gross State Product and other variables, and find

it has declined more in high-income states in India. Although their approach is an improvement

over the standard reduced form modelling in the EKC literature, the focus on states rather than

cities limits the applicability of their results for some important purposes. Greenstone and

Hanna (2013) also move beyond reduced form modelling, and measure the causal impact of

pollution regulations on infant mortality rates in India. While the present study estimates

reduced form air pollution production functions, it uses different, arguably superior income

measures. In sum, by analysing geographically more refined units, with better proxies for

average income, the present analysis is complementary to these two closely related studies.

Finally, several studies explore pollution trends in India, and make cross country

comparisons. Gupta and Kumar (2005) document trends in particulate matter in Delhi,

Mumbai, Kolkata and Chennai and find that particulate matter levels have generally fallen over

the period 1991 to 2003; data presented in Sridhar and Kumar (2013) show particulate matter

levels falling by 17 per cent and sulphur dioxide levels falling by 37.6 per cent from 1990 to

2007, but no change in nitrogen dioxide over this period. Gurjar et al. (2007) present pollution

data from eighteen world cities with population over 10 million, including three from India;

while for some measures the Indian cities have below average pollution levels, particulate

matter levels are among the highest.

6

Data and Methodology

Our methodological approach to estimating air pollution production functions is to use

the data described below to estimate the following model using ordinary least squares:

AVGPOLLUTIONi = β0 + β1lnPOPi + β2lnRAINi + β3PCTMANUFi + β4SOUTHi

+ β5DEVi + β6DEVi2 + ui (1)

where, AVGPOLLUTIONi is the dependent variable and refers to one of the four measures of

pollution (described below) for city i, the independent variables are as described below and are

also at the city level except for income which is measured at the district (county) level, the β’s

are coefficients to be estimated, and ui is an error term with the usual properties. This model is

quite similar to one estimated by Zheng et al. (2010) using data from Chinese cities; for

examples from the United States, see Kahn (1997, 2006).

We estimate several versions of this model. We have four measures of pollution, and

three measures of socioeconomic development, which yields twelve versions of equation (1).

In addition we estimate a restricted version of (1) where β6 is constrained to be zero. As should

be clear, this is to distinguish between linear and nonlinear effects of socioeconomic

development on pollution.

We obtained our data from multiple sources. The four air pollution measures were

taken from the Environmental Data Bank of India's Central Pollution Control Board (CPCB).3

These include two measures of particulate matter, Suspended Particular Matter (SPM), and

Respirable Particulate Matter (RSPM), the difference between them being particle size;

respirable particles are smaller (less than 10 cubic microgrammes.)4 The other two air pollution

7

measures are sulphur dioxide (SO2) and nitrogen dioxide (NO2). This data bank allows the user

to download daily pollution data at the level of the monitoring station for over 100 cities and

mainly covers the years 2006 to present. Our study utilizes data from 2006-2011; later in this

section, and in more detail in the Appendix, we describe how we used these data to construct

average pollution estimates for cities.

The main independent variable of interest in equation (1) is DEVi, which represents

socioeconomic development. We utilise data from three sources to proxy for socioeconomic

development. The most obvious economic development indicator is income, however average

income data is notoriously difficult to obtain for Indian cities. We therefore collected two

sources of district-level income proxies and assign these to cities based on the district in which

they are located. The first income proxy is monthly per capita consumption expenditure

(MPCE), collected by the National Sample Survey Organisation (NSSO). Chaudhuri and Gupta

(2009) present district-level estimates of MPCE for urban and rural households using data from

a recent NSSO survey; Greenstone and Hanna (2013) used district-level estimates from earlier

NSSO surveys as their city-level income proxies, but the more recent NSSO surveys were

designed to overcome certain deficiencies.5 For example, we are able to use MPCE for urban

households, while earlier NSSO surveys were not designed to distinguish between urban and

rural households.

The second income proxy is also a district-level measure and is distributed by the

Planning Commission of the Government of India. These consist of District Domestic Product

(DDP) estimates for most but not all districts in India. We have obtained data from the 2004-

2005 study. We use per capita DDP as our second income proxy. We also take advantage of

8

the fact that the DDP data lists output for eighteen industry classifications. In equation (1), the

variable PCTMANUFi was calculated using these data; it is total output in manufacturing divided

by total output in all categories in the district in which city i is located.

The third measure used to proxy for DEVi, is a measure of social development. We use

the literacy rate which is a city-level measure and is collected by the Census. These three

development indicators are moderately correlated with one another. In our sample, the simple

correlation between DDP and LIT_RATE is 0.34; between DDP and MPCE it is 0.51, and between

MPCE and LIT_RATE it is 0.40.6

The variable SOUTHi is a dummy variable indicating whether or not city i is in the south.7

This region has higher levels of income and literacy generally, and can therefore itself be

thought of as a development indicator; however we include it mainly to enable estimating the

within-region relationship between social and economic development and ambient pollution.

The final source of data is the 2001 Town Directory, produced by the Census of India, from

which we obtained population and average precipitation measures, the latter a control variable

included in previous studies.8 We convert both of these variables into natural logarithms

following common practice in the literature.

In calculating average pollution data for cities, we only used estimates from cities for

which the CPCB contained data from at least 365 days. The CPCB describes monitoring stations

as residential, industrial or sensitive; we used data from all of these station types because we

are interested in average city pollution.9 Table 1 summarises the discussion above by listing the

variables used below, their definitions and the source of the data.

9

Table 1: Variable definitions and sources Variable Definition Source

AVGSO2 Average sulphur dioxide emissions CPCB

AVGNO2 Average nitrogen dioxide emissions CPCB

AVGRSPM Average respirable particulate matter CPCB

AVGSPM Average suspended particulate matter CPCB

POP City population, 2001 Town Directory (Census)

AVG_RAIN Average precipitation Town Directory (Census)

PCTMANUF Per cent of DDP in Manufacturing Planning Commission, India

SOUTH Regional indicator for cities in South Author’s calculation

MPCE Monthly per capita consumption expenditure National Sample Survey Org DDP Per capita District Domestic Product Planning Commission, India

LIT_RATE Literate fraction of population Census Sources: (1) Central Pollution Control Board, Environmental Data Bank (accessed 4/15/2012); (2) Town Directory, produced by the Indian Census Bureau (2001); (3) National Sample Survey Organization's Consumer Expenditure Survey (2004-2005), as reported by Chaudhuri and Gupta (2009); (4) Planning Commission, Government of India website (accessed 5/15/2013); (5) 2001 Census, Table 3, Population, population in the age group 0-6 and literates by sex - Cities/Towns.

Table 2 provides summary statistics for the variables discussed above.

Table 2: Summary Statistics Variable Obs Mean Std. Dev. Min Max

AVGSO2 64 10.74 6.77 3.60 37.18

AVGNO2 67 25.33 11.43 7.54 59.16 AVGRSPM 68 107.76 54.36 33.26 254.78 AVGSPM 66 224.92 103.10 48.50 463.78 POP 71 1,076,771 1,633,140 103,099 11,978,450 SOUTH 71 0.42 0.50 0 1 AVG_RAIN 71 1183.46 657.55 102.6 3852 PCTMANUF 71 0.16 0.09 0.02 0.55 MPCE 71 1.13 0.37 0.55 1.87 DDP 71 31.15 12.77 12.60 69.68 LIT_RATE 71 67.54 9.43 34.83 86.44

Sources: (1) Central Pollution Control Board, Environmental Data Bank (accessed 4/15/2012); (2) Town Directory, produced by the Indian Census Bureau (2001); (3) National Sample Survey Organization's Consumer Expenditure Survey (2004-2005), as reported by Chaudhuri and Gupta (2009); (4) Planning Commission, Government of India website (accessed 5/15/2013); (5) 2001 Census, Table 3, Population, population in the age group 0-6 and literates by sex - Cities/Towns.

10

Results

In Table 3 below we begin by presenting the estimates of the restricted version of

equation (1), for each of the four measures of pollution, using each of the three development

proxies. From this table one can see that most of the coefficient estimates on the economic

and social development variables, MPCE, DDP and LIT_RATE, are statistically insignificant.

There are three exceptions. In two of the models that use AVGSPM as the dependent variable,

both MPCE and LIT_RATE are associated with lower levels of pollution. A one standard

deviation increase in MPCE is associated with a fall in average SPM of 22.61 cubic

microgrammes, which is about a one-fifth standard deviation. A one standard deviation

increase in LIT_RATE is associated with a slightly smaller fall in AVGSPM of 21.15 cubic

microgrammes.

The third statistically significant coefficient on a development proxy in Table 3 is on DDP

in the model that uses NO2 as a dependent variable. Unlike the models of particulate matter,

this coefficient estimate is positive. A one standard deviation in DDP is associated with a rise in

average NO2 of 3.2 cubic microgrammes, which is just over a fourth of a standard deviation.

Before discussing the other estimates, we will explore whether the linear (restricted) or

nonlinear (unrestricted) models better fit the data. Comparing the adjusted R2 in Table 3 and

Table 4, we find the linear model is a slightly better fit in seven out of twelve cases. In terms of

statistical significance, none of the coefficients on the development indicators in Table 4 are

significant at the five per cent level, except in one case. In the model where AVGSPM is the

dependent variable and DDP is the development proxy, the negative coefficient on DDP and the

11

positive coefficient on DDP2 suggest there is a U-shaped relationship between DDP and

AVGSPM, with pollution falling until DDP reaches 43.51 and rising after. However, from Table 2

we see that 43.51 is one standard deviation above the mean of DDP. Also, of the 66 cities used

in estimating that model, only eight have a value of DDP above 43.51. Therefore, for most

cities, rising DDP is associated with falling AVGSPM.

12

.

Table 3: Regression Results, Restricted Model VARIABLES AVGSO2 AVGNO2 AVGRSPM AVGSPM AVGSO2 AVGNO2 AVGRSPM AVGSPM AVGSO2 AVGNO2 AVGRSPM AVGSPM

logPOP 0.987 3.465** 12.23** 38.84*** 0.832 2.821** 12.86** 38.15*** 0.703 3.508** 12.36** 32.35***

(0.609) (1.528) (5.778) (8.715) (0.632) (1.366) (6.114) (9.140) (0.672) (1.623) (5.699) (8.878)

logRAIN 1.946 2.209 -16.09 -23.7 1.259 0.154 -13.56 -29.47* 1.522 2.092 -13.78 -29.08*

(1.232) (2.247) (9.925) (16.600) (1.518) (2.284) (9.247) (16.670) (1.140) (2.116) (8.469) (16.900)

PCTMANUF 27.59*** 38.16*** 110.4** 186.8 26.43** 26.37* 120.6** 218.6* 25.56** 38.51*** 113.9** 160.2

(9.984) (14.240) (49.820) (115.800) (11.580) (15.750) (51.990) (126.600) (10.400) (13.770) (51.010) (111.800)

SOUTH 0.586 -4.141 -64.05*** -122.7*** 0.654 -5.511** -63.52*** -118.1*** 0.996 -4.268 -63.28*** -114.1***

(1.807) (2.672) (9.623) (17.920) (1.757) (2.689) (9.737) (18.800) (2.048) (2.783) (9.709) (18.820)

MPCE -3.049 0.0279 7.282 -61.11**

(1.843) (3.335) (19.260) (25.040)

DDP

-0.00248 0.251** -0.119 -1.158

(0.072) (0.103) (0.475) (0.774)

LIT_RATE

-0.0805 0.0311 -0.175 -2.243**

(0.094) (0.118) (0.456) (0.893)

Constant -17.24* -40.79 57.74 -38.63 -13.66 -23.23 41.82 -28.12 -8.4 -42.64 59.06 169.6

(10.100) (25.460) (104.400) (160.400) (12.470) (23.740) (109.100) (176.300) (13.760) (28.890) (97.150) (163.200)

Observations 64 67 68 66 64 67 68 66 64 67 68 66 R-squared 0.187 0.222 0.483 0.581 0.164 0.279 0.482 0.562 0.174 0.223 0.482 0.586

adj R-squared 0.117 0.159 0.441 0.546 0.0919 0.22 0.44 0.525 0.103 0.159 0.44 0.551 Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1

13

Table 4: Regression Results, Unrestricted Model VARIABLES AVGSO2 AVGNO2 AVGRSPM AVGSPM AVGSO2 AVGNO2 AVGRSPM AVGSPM AVGSO2 AVGNO2 AVGRSPM AVGSPM

logPOP 0.881 3.231** 14.59*** 38.91*** 0.795 3.086** 12.90* 33.97*** 0.711 3.546** 12.23** 32.34***

(0.669) (1.577) (5.303) (8.912) (0.630) (1.464) (6.520) (9.487) (0.643) (1.611) (5.767) (8.960)

logRAIN 2.003 2.203 -16.71* -23.85 1.231 0.368 -13.54 -31.35* 2.001* 2.656 -10.74 -28.27

(1.262) (2.312) (9.461) (16.960) (1.553) (2.320) (9.241) (17.160) (1.147) (2.181) (8.692) (18.220)

PCTMANUF 27.19*** 37.17** 115.6** 187.4 26.19** 28.63* 120.8** 189.6 22.49** 34.79** 95.10* 155.5

(9.911) (14.020) (46.260) (117.400) (12.220) (15.420) (51.840) (123.400) (10.470) (14.690) (50.850) (115.700)

SOUTH 0.623 -4.054 -67.02*** -122.9*** 0.662 -5.526** -63.53*** -115.9*** 1.563 -3.615 -60.32*** -113.2***

(1.823) (2.690) (10.250) (18.810) (1.756) (2.687) (9.845) (18.660) (2.169) (2.811) (9.540) (19.640)

MPCE 5.845 19.59 -209.9* -70.6

(13.680) (18.710) (119.300) (176.000)

MPCE2 -3.654 -8.08 89.23* 4.046

(5.490) (7.659) (47.700) (71.630)

DDP

-0.0448 0.595 -0.0698 -6.188**

(0.241) (0.386) (1.698) (2.460)

DDP2

0.000604 -0.0049 -0.00069 0.0711**

(0.003) (0.005) (0.021) (0.032)

LIT_RATE

0.977 1.318 5.762* -0.684

(0.586) (0.991) (2.923) (6.083)

LIT_RATE2

-0.00840* -0.0102 -0.0472* -0.0124

(0.005) (0.008) (0.024) (0.049)

Constant -21.07* -48.23* 151.1 -33.51 -12.31 -33.73 40.35 120.2 -43.92** -86.31* -139.9 116.8

(12.150) (28.200) (118.700) (188.000) (14.460) (28.750) (121.600) (192.300) (18.960) (45.430) (133.800) (279.000)

Observations 64 67 68 66 64 67 68 66 64 67 68 66 R-squared 0.192 0.23 0.521 0.581 0.164 0.286 0.482 0.581 0.206 0.239 0.497 0.586 adj R-squared 0.107 0.153 0.474 0.539 0.0763 0.215 0.431 0.539 0.122 0.163 0.448 0.544 Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1

14

Moreover, this U-shaped finding is inconsistent with the significant, linear relationships

we found between MPCE and AVGSPM, and between LIT_RATE and AVGSPM in Table 3. Taken

together, these results suggest to us a negative relationship between income and particulate

matter. However, regardless of whether one prefers to characterise the relationship between

AVGSPM and DDP as linear or U-shaped, neither characterisation is consistent with the Kuznets

curve hypothesis, which predicts an inverted U-shaped relationship.

The findings related to the third development indicator in Table 4 do, however, provide

some evidence that is consistent with the Kuznets curve hypothesis. While none of these

coefficient estimates are significant at the five per cent level, in all models using LIT_RATE as

the development indicator except for the one where AVGSPM is the dependent variable, the

positive coefficient on LIT_RATE and negative coefficient on LIT_RATE2 suggest a Kuznets curve

(an inverted U-shaped) relationship between social development and pollution. For AVGSO2,

the top of the inverted U occurs when LIT_RATE equals 61.1; for AVGNO2 this value is 64.6, and

for AVGRSPM this value is 61.0. Thus, these findings indicate that air pollution increases as

literacy increases, but after literacy reaches about sixty per cent, further increases in literacy

are associated with decreasing pollution.10

Before concluding this section, we briefly describe the effect of the other independent

variables on the four pollution measures. Across all models, logPOP is associated with higher

levels of air pollution. The estimates we report above are useful for comparing the scale effects

of city size on pollution in India to that from other countries, but in order to make our results

more comparable with those from Zheng et al. (2010), in non-reported results, we estimated

models identical to those in Tables 3 and 4 but that used the natural logarithm of the four

15

average pollution measures as a dependent variable. 11 This log-log specification has a desirable

property in that the coefficient estimates can be directly interpreted as elasticites. In these

results, the average of the coefficients on logPOP for the models with logAVGSO2 as a

dependent variable was 0.12, meaning a 10 per cent increase in population leads to a 1.2 per

cent increase in AVGSO2. For the models of logAVGNO2, the average elasticity was 0.15 and

for the two particulate matter measures the average elasticity was 0.17. It is interesting to note

that the average population elasticity reported by Zheng et al. (2010) for particulate matter for

Chinese cities was identical (to two decimal places) to what we found for Indian cities. On the

other hand, in Zheng et al. (2010) the average elasticity for SO2 was 0.36, which is exactly three

times higher than what we found in our unreported results.

We do not find statistically significant evidence that cities with higher levels of

precipitation enjoy less pollution, however we do find strong evidence that cities with a higher

concentration of manufacturing have higher levels of all four measures of pollution. Finally,

southern India enjoys much lower levels of particulate matter pollution, and to some degree,

lower levels of NO2. This is also apparent from Table 6 in the Appendix, which presents a list of

cities analysed in this study, ranked from lowest to highest in terms of AVGRSPM. Most of the

cities with the lowest particulate matter levels are located in the south.12

16

Pollution and Industry Type

It is important to emphasise that these results reported above control for the fraction of

output in manufacturing and a variety of other factors. The large, positive coefficient on

PCTMANUF suggests that some types of development are worse than others with regard to

pollution. In order to shed light on which industries are most strongly associated with air

pollution, below we present a table containing the simple correlation between the level of

economic activity in the various industry categories identified by the Planning Commission and

our four measures of pollution.13

Table 5: Correlation between DDP by Category and Pollution

AVGSO2 AVGNO2 AVGRSPM AVGSPM

Agriculture 0.06 0.16 0.10 -0.03 Forestry and Logging 0.13 0.12 -0.23 -0.10 Fishing -0.06 0.22 -0.13 -0.11 Mining and Quarrying 0.19 0.22 -0.04 0.05 Manufacturing 0.29 0.34 -0.02 0.00 Registered Manufacturing 0.30 0.34 -0.03 0.00

Unregistered Manufacturing 0.21 0.26 0.02 -0.01 Electricity, Gas and Water 0.20 0.31 0.01 0.01 Construction 0.10 0.25 -0.14 -0.16 Trade, Hotels and Restaurants 0.20 0.35 -0.07 -0.03 Railways 0.20 0.37 0.16 0.23

Other Transport 0.16 0.31 -0.07 -0.07 Storage 0.17 0.43 0.21 0.14 Communication 0.20 0.31 -0.06 -0.06 Banking and Insurance 0.23 0.28 -0.07 -0.02 Real Estate and Legal Services 0.21 0.33 -0.07 -0.03 Public Administration 0.10 0.41 0.03 0.06

Other Services 0.14 0.41 -0.05 -0.05 Total DDP 0.24 0.38 -0.05 -0.03 DDP per capita 0.21 0.30 -0.09 -0.20

17

From this table, one can see that most categories are positively correlated with SO2

emissions, while only the category fishing is negatively correlated with SO2. The results are

slightly different for NO2; the highest positive correlations here are with storage, public

administration, and other services, and no category is negatively correlated with NO2.

The results are much different for the two particular matter measures (RSPM and SPM);

no industry category was highly positively correlated with either measure (the highest was

storage, and railways), and output in most categories was negatively correlated with particulate

matter levels.

The correlations identified in Table 5 correspond well with the findings identified above.

Though these simple correlations do not allow for nuanced interpretation as do our multiple

regression results, they do suggest that, in general, economic development is positively

correlated with SO2 and NO2 and is negatively correlated with particulate matter. The high

positive correlation between NO2 and services and public administration may be surprising,

until one considers that NO2 emissions are largely attributable to vehicular transportation, and

there are likely many residents of these cities who can afford to commute in motorised

vehicles.14

Conclusion

We have estimated air pollution production functions that have controlled for multiple

factors. Ordinary least squares results indicate that, controlling for a variety of factors, there is

a negative relationship between income and particulate matter levels, a positive relationship

between income and nitrogen dioxide, and no relationship between income and sulphur

dioxide. Though our results do not indicate a Kuznets curve relationship between income and

18

air pollution at the sub-national level for India, we are not able to reject the Kuznets curve-type

relationship between literacy and air pollution with a high level of confidence. We also provide

an estimate of the magnitude by which air pollution levels increase with city size; the

population elasticity of pollution for Indian cities is identical to that found in previous research

for Chinese cities for particulate matter, but is much smaller with respect to sulphur dioxide.

Finally, we briefly considered the effect of output in various industries on the four pollution

measures. Output in most industries is positively correlated with sulphur dioxide and nitrogen

dioxide levels, but is negatively correlated with particulate matter levels.

The policy implications of these findings suggest to us that, while cost-benefit analysis

should be used to set all pollution reduction goals, special emphasis on developing new

technological options for reducing SO2 and NO2 may be needed, as it appears that existing

technology to mitigate particulates can be adopted with rising development.

From a scholarly perspective, by demonstrating how high quality, publicly available data

sources can be combined to generate new knowledge concerning the relationship between

socioeconomic development and pollution, we believe a major contribution of this study is to

facilitate future empirical research. The estimates we present can also be directly used in a

variety of studies. Finally, while we have made progress in answering the question of which

proxies for average income measures at the city level are most appropriate for use in research,

we have not resolved this issue and future research here is warranted.

19

References

Asian Development Bank. 2012. Green Urbanization in Asia: Key Indicators for Asia and the Pacific. Special chapter in Indicators for Asia and the Pacific 2012. Manila: Asian Development Bank. Brock, W., and S. Taylor. 2005. Economic Growth and the Environment: a Review of Theory and Empirics. In S. Durlauf and P. Aghion (eds.) The Handbook of Economic Growth. Amsterdam: North Holland. Chaudhuri and Gupta (2009); Levels of Living and Poverty Patterns: A District-Wise Analysis for India. Economic and Political Weekly. Vol. 44 No. 9, pp. 94-110. Dasgupta, Susmita, Kirk Hamilton and Kiran D. Pandey (2006). Environment during growth: Accounting for governance and vulnerability. World Development. 34(9), p. 1597-1611. Economist Intelligence Unit, Asian Green City Index: Assessing the environmental performance of Asia's major cities. A research project sponsored by Siemens. http://www.siemens.com/entry/cc/en/greencityindex.htm (Accessed March 15, 2013) Glaeser, E.L., Kahn, M.E., 2010. The greenness of cities: Carbon dioxide emissions and urban development. Journal of Urban Economics 67, 404–418. Greenstone, Michael and Rema Hanna, 2011. Environmental Regulations, Air and Water Pollution, and Infant Mortality in India. CEEPR WP 2011-014 Grossman, G., and A. Krueger. 1995. Economic Growth and the Environment. The Quarterly Journal of Economics, 110(2): 353–77. Guilmoto, Christophe Z. and S. Irudaya Rajan. District Level Estimates of Fertility from India's 2001 Census. Economic and Political Weekly, Vol. 37, No. 7 (Feb. 16-22, 2002), pp. 665-672 Gupta, Indrani and Rakesh Kumar, 2005. Trends of particulate matter in four cities in India. Atmospheric Enviornment, 40(14), pp. 2552-2566. Gurjar, B.R., T.M. Butler, M.G. Lawrence, J. Lelieveld, 2007. Evaluation of emissions and air quality in megacities. Atmospheric Environment 42(7), pp. 1593-1606. Kathuria, Vinish, 2002. Vehicular pollution control in Delhi. Transportation Research Part D: Transport and Environment. 7(5), 373-387. Kahn, Matthew E. 1997. Particulate pollution trends in the United States. Regional Science and Urban Economics 27, 87-107.

20

Kahn, M.E. 2006. Green Cities: Urban Growth and the Environment. Brookings Institution Press, Washington, D.C. Managi, Shunsuke and Pradyot Ranjan Jena, 2008. Environmental productivity and Kuznets curve in India. Ecological Economics. Volume 65, Issue 2, pp. 432–440. Planning Commission, Government of India, http://www.planningcommission.nic.in/plans/stateplan/index.php?state=ssphdbody.htm (Accessed May 15, 2013) Sankhe, S., Vittal, I., Dobbs, R., Mohan, A., Gulati, A., Ablett, J., Gupta, S., Kim, A., Paul, S., Sanghvi, A., & Sethy, G. (2010). India’s urban awakening: Building inclusive cities and sustaining economic growth (April). McKinsey Global Institute. McKinsey & Company. Sridhar, Kala Seetharam and Surender Kumar. 2013. India’s Urban Environment: Air/Water Pollution and Pollution Abatement. Economic and Political Weekly. Vol 48, No. 6, pp. 22-25. Suzuki, Hiroaki, and Arish Dastur, Sebastian Moffatt, Nanae Yabuki, Hinako Maruyama, 2010. Eco2 Cities: Ecological Cities as Economic Cities. World Bank Publications. Uchida, Hirotsugu and Andrew Nelson. 2009. Agglomeration Index: Towards a New Measure of Urban Concentration. Washington, DC: World Bank. United Nations, Department of Economic and Social Affairs, Population Division (2012). World Urbanization Prospects: The 2011 Revision, CD-ROM Edition. Zheng, Siqi, Matthew E. Kahn and Hongyu Liu. 2010. Towards a system of open cities in China: Home prices, FDI flows and air quality in 35 major cities. Regional Science and Urban Economics 40, 1–10. Zheng, Siqi, Rui Wang, Edward L. Glaeser, and Matthew E. Kahn. 2011. The greenness of China: household carbon dioxide emissions and urban development J Econ Geogr, 11(5): 761-792.

21

Appendix A: Method of Calculating Average Pollution by City

For each state, daily air pollution data at the monitoring station level was downloaded

from the CPCB’s online Environmental Data Bank. We assembled these data into one file and

this yielded 176,292 observations. Fewer than 1,000 observations were taken prior to 2005.

From 2006 to 2011, total observations ranged from 9,529 (in 2010) and 34,387 (in 2007). We

then aggregated the data by city. This yielded a total number of cities with average pollution

data equal to 164.

Next we matched the city name from the CPCB data to the town, district and state

codes contained in the town directory file. There were 15 cities that we could not match.15

Finally, we merged the pollution and town directory data to the literacy, MPCE and DDP data

described above; variables from at least one of these sources was not available for 30 cities. In

estimating our models, we followed several data rules. First, we restricted the sample to cities

whose average particulate matter estimates were constructed with at least 365 daily pollution

observations. In calculating summary statistics in Table 2, we dropped cities without at least

one valid average pollution estimate. Finally, we dropped 14 cities with population less than

100,000. This left us with a sample that ranged from 57 to 64 depending on the pollution

measure.

Appendix B: Average Pollution in Select Indian Cities

In the table below, we present the average pollution figures used in the analysis. The

cities are ranked based on average respirable particular matter. Full data used in this study is

available upon request from the author.

22

Table 6: Average Pollution Levels by City (sorted by average RSPM) city state AVGSO2 AVGNO2 AVGRSPM AVGSPM

Belgaum Karnataka - 15.0 33.3 70.7 Tirupati Andhra Pradesh 4.1 9.1 33.9 103.4 Kozhikode Kerala - 8.9 36.3 83.5 Madurai Tamil Nadu 10.4 24.2 41.1 98.8 Kottayam Kerala 5.4 18.1 45.0 48.5 Kochi Kerala 7.4 13.1 45.5 78.8 Mysore Karnataka 15.7 25.7 45.8 88.4 Dibrugarh Assam 5.3 12.4 49.6 90.4 Sambalpur Orissa 3.9 13.2 49.9 127.6 Shimla Himachal Pradesh 4.7 11.4 53.3 105.5

Mangalore Karnataka 7.2 - 53.7 119.8 Thane Maharashtra 10.7 13.8 54.1 108.7 Salem Tamil Nadu 7.9 26.5 61.8 99.2 Hassan Karnataka 4.8 21.6 61.8 152.1 Singrauli Madhya Pradesh - - 67.2 307.8 Trivandrum Kerala 9.4 24.5 67.9 77.3 Chennai Tamil Nadu 12.0 19.2 68.9 153.7 Coimbatore Tamil Nadu 7.1 31.8 70.3 140.7 Haldia West Bengal 8.6 44.0 72.3 166.7 Gulbarga Karnataka - 13.4 73.8 200.3 Palakkad Kerala - - 74.2 133.3 Amravati Maharashtra 10.4 13.3 74.8 -

Nashik Maharashtra 32.8 29.3 78.0 152.4 Kurnool Andhra Pradesh 4.0 13.7 78.4 185.4 Kolhapur Maharashtra 9.6 20.9 78.5 194.2 Dewas Madhya Pradesh 15.6 22.4 78.8 192.0 Cuttack Orissa - 24.5 79.3 226.2

Hyderabad Andhra Pradesh 5.2 25.9 81.4 224.7 Bangalore Karnataka 14.7 40.1 85.5 223.8 Vijayawada Andhra Pradesh 5.8 24.9 87.5 208.2 Bhopal Madhya Pradesh 6.0 16.5 88.4 233.6 Visakhapatnam Andhra Pradesh 10.9 28.6 88.9 187.2 Ujjain Madhya Pradesh 12.0 12.8 90.6 189.1

Hubli-Dharwad Karnataka 4.0 10.5 90.7 198.3 Udaipur Rajasthan 7.2 34.6 90.7 281.4 Solapur Maharashtra 17.0 35.8 93.7 287.0 Nagpur Maharashtra 9.5 30.7 99.2 194.7 Korba Chhattisgarh 13.2 21.0 101.8 211.5 Guwahati Assam 7.2 15.8 108.4 184.1

Note: A missing value indicates there were fewer than 365 daily observations for this pollution measure for a given city.

23

Table 6 (continued): Average Pollution Levels by City (sorted by average RSPM)

city state AVGSO2 AVGNO2 AVGRSPM AVGSPM

Greater Mumbai Maharashtra 21.2 43.8 108.6 257.1 Navi Mumbai Maharashtra 16.4 36.3 109.1 270.4 Hisar Haryana 7.9 - 111.1 195.9 Varanasi Uttar Pradesh 16.2 18.8 112.2 354.1 Jabalpur Madhya Pradesh

22.8 112.7 247.0

Kolkata West Bengal 8.9 57.5 115.4 234.2 Bhilai Nagar Chhattisgarh 16.9 24.4 116.1 207.6 Kota Rajasthan 8.0 24.1 116.2 251.1 Patna Bihar 8.8 37.4 120.6 306.2 Jaipur Rajasthan 5.9 31.8 129.4 301.2

Asansol West Bengal 7.1 59.2 133.2 283.1 Chandrapur Maharashtra 32.8 41.8 134.9 221.7 Alwar Rajasthan 7.6 21.3 137.3 250.4 Jodhpur Rajasthan 6.4 22.3 143.0 371.8 Indore Madhya Pradesh 9.3 16.5 147.6 243.2 Noida Uttar Pradesh 13.7 44.5 148.9 443.0 Jamshedpur Jharkhand 37.2 50.1 162.7 314.9 Jalandhar Punjab 11.8 29.7 164.2 - Allahabad Uttar Pradesh 8.4 31.2 171.1 407.2 Agra Uttar Pradesh 6.7 22.0 181.0 377.9 Lucknow Uttar Pradesh 9.8 32.7 192.4 407.0 Kanpur Uttar Pradesh 6.9 23.6 193.2 425.1

Satna Madhya Pradesh 3.6 7.5 193.5 290.9 Gwalior Madhya Pradesh 8.9 16.6 198.5 305.2 Firozabad Uttar Pradesh 20.4 30.9 200.9 411.0 Amritsar Punjab 13.9 34.3 212.3 407.7 Khanna Punjab 10.0 34.3 234.2

Ludhiana Punjab 12.0 35.6 238.0 - Ghaziabad Uttar Pradesh 19.1 19.9 254.8 463.8 Bathinda Punjab 10.1 24.1 - 227.4 Sagar Madhya Pradesh 4.3 15.5 - 239.5 Meerut Uttar Pradesh - - - 640.2 Patiala Punjab 7.1 19.7 - -

Note: A missing value indicates there were fewer than 365 daily observations for this pollution measure for a given city.

24

Endnotes 1 Urbanisation can also lead to environmental improvements. As one example, women in urban areas have fewer

children and therefore place fewer demands on the environment. In 2001, the average fertility rate in districts in India that are majority urban was 2.38, while a woman in a rural district had on average 3.38 children (these calculations were made using fertility estimates presented in Guilmoto and Rajan, 2002). 2 Economist Intelligence Unit (2011).

3 Greenstone and Hanna (2013) describe the legislation that enabled the collection of this data.

4 Many researchers refer to RSPM as ‘PM10’ though here we use the CPCB terminology.

5 ‘It was only in the 61st round survey of NSS (2004-05) that the sampling design defined rural and urban parts of

districts as strata for selection of sample villages and urban blocks respectively. This has paved the way for generating unbiased estimates of important socio-economic parameters at the district-level adequately supported by the sample design.’ Chaudhuri and Gupta (2009, p. 94). 6 When using all districts in India for which DDP, MPCE and literacy data are available, the correlation between DDP

and LIT_RATE is 0.62; between DDP and MPCE it is 0.73, and between MPCE and LIT_RATE it is 0.62. 7 Our definition of the south includes the following states: Andhra Pradesh, Goa, Karnataka, Kerala, Maharashtra

and Tamil Nadu. 8 Both our literacy rate and population measures are from the year 2001. As of writing, only provisional figures

have been released from the 2011 Census. We have repeated the analysis presented below using these more recent but provisional figures and overall the results were quite similar. These results are available upon request. 9 The WHO Outdoor Pollution Database includes CPCB data on about 30 cities in India. However by only including

data from residential monitoring stations for one year, their sample size is only about half as large as what we were able to achieve by using more years of data and all station types. 10

From Table 2, the average literacy rate among cities in our sample was 67.54. 11

These results are available from the author upon request. In terms of sign and statistical significance of the coefficient estimates, there were only small differences across the level and log specifications. 12

In both Tables 3 and 4, the coefficients on the South dummy were either both negative and significant, or, as in all cases with SO2, not statistically different from zero. 13

Note the industry measures are of the total level of economic activity (or output) in an industry category, not the fraction of output in the district in that category. 14

According to Greenstone and Hanna (2013, pp. 10-11), ‘The monitored pollutants can be attributed to a variety of sources. PM is regarded by the CPCB as a general indicator of pollution…SO2 emissions, on the other hand, are predominantly a byproduct of thermal power generation; globally, 80 per cent of sulphur emissions in 1990 were attributable to fossil fuel use…NO2 is viewed by the CPCB as an indicator of vehicular pollution, though it is produced in almost all combustion reactions.’ 15

Aurangabad (MS), Balasore, Berhampur, Byrnihat, Damtal, Daranga, Dawki, Imphal, Jhansi, Lote, Naya Nangal, North Lakhimpur Town, Thoothukudi, Vasco, Wayanad.