5-Effect of Reducing The Preharvest Burning of Sugar Cane On Respiratory Health in Brazil

Environment and Development Economics 20: 127–140 © Cambridge University Press 2014
doi:10.1017/S1355770X14000096
The effect of reducing the pre-harvest burning of

sugar cane on respiratory health in Brazil
ALEXANDRE C. NICOLELLA
Department of Economics, University of São Paulo at Ribeirão Preto, Av.
Bandeirantes 3900, FEARP, Ribeirão Preto, São Paulo 14040-900, Brazil.
Email: anicolella@fearp.usp.br
WALTER BELLUZZO
Department of Economics, University of São Paulo at Ribeirão Preto, Brazil.
Email: belluzzo@usp.br
Submitted 18 October 2012; revised 30 August 2013; accepted 31 January 2014; first published
online 1 April 2014
ABSTRACT. This paper analyzes the effect of reducing pre-harvest burning of sugar cane
on the population’s respiratory health in Brazil. We collected data for the municipalities
in the state of São Paulo for two different periods: 2000, before the state law requiring the
gradual elimination of sugar cane area utilizing pre-burning, and 2007. We used panel
models for inpatient visits due to respiratory diseases, outpatient visits for inhalation
procedures and length of stay for inpatient visits due to respiratory diseases, controlling
for the endogeneity between health and pollution. The results show that increasing the
area of raw sugar cane harvested reduces the number of inpatient visits and does not
affect the number of inhalation procedures or length of stay.
1. Introduction
Concerns regarding global warming and the prospect of a future scarcity of
fossil fuels brought the use of biofuel into the energy policy agenda. As a
result, biofuel is gaining importance in the world energy matrix, and there
are projections that it will account for 5 per cent of liquid fuels by 2025 (EIA,
2006). In this context, ethanol arises as one of the best alternatives among
biofuels.
The author thanks the German Federal Ministry for Economic Cooperation and
Development – BMZ, German Development Institute – DIE, In Went and Keynes
College at the University of Kent.
Downloaded from https://www.cambridge.org/core. Universidade Federal de Lavras, on 27 Mar 2018 at 19:18:12, subject to the Cambridge
Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1355770X14000096
128 Alexandre C. Nicolella and Walter Belluzzo
The use of sugar cane ethanol as a substitute for gasoline is widespread

in Brazil.1 The success of the Brazilian program is due to several factors.
On the supply side, there were significant cost reductions over the last
several decades due to heavy investments in technology that increased
sugar cane productivity from 34 t.ha −1 in 1960 to 69 t.ha −1 in 2006 (IBGE,
2010a). On the demand side, there were three important factors: the imple-
mentation of flex-fuel technology, which allows automobiles to run on any
proportion of ethanol and gasoline2 , the mandatory blend of ethanol and
gasoline,3 and, perhaps more importantly, the increase in oil prices.
The effect of these shifts on supply and demand in the recent decades
caused ethanol production and the sugar cane area to soar. In 2009, Brazil
was responsible for about one–third of the world production of ethanol,
while the area grew from 4.9 million ha in 2000 to 8.2 million ha in 2008,
a growth rate of 6.7 per cent per year. Such a rapid expansion certainly
has had a social and environmental impact, which has been addressed
in the literature. There are studies on soil degradation (Giampietro et al.,
1997; Moreira and Goldemberg, 1999; Oliveira et al., 2005), water pollution
(Moreira and Goldemberg, 1999; Gunkel et al., 2007) and greenhouse gases
emissions (Moreira and Goldemberg, 1999; Moreira, 2000; Oliveira et al.,
2005; Crutzen et al., 2008; Börjesson, 2009), and air pollution (Allen et al.,
2004; Lara et al., 2005; Andrade et al., 2010).
The connection between sugar cane ethanol production and air pollu-
tion is the harvesting of sugar cane. Sugar cane is a semi-perennial culture
and has two different harvest systems: the mechanical harvest of raw sugar
cane or the manual harvest of previously burned sugar cane. Because large
areas are burned at once, this sort of harvesting has a significant impact on
air pollution. In fact, there are studies presenting evidence that pre-harvest
burning is responsible for the increase of fine particulate matter, coarse
particulate matter and black carbon concentrations, especially in the time
period the burning occurs (Lara et al., 2005), which increases concentrations
of substances such as nitrites, sulfites, carbon oxide and other substances
(Allen et al., 2004). As indicated in the literature, short- and long-term expo-
sure to these pollutants can negatively affect human health capital (Sicard
et al., 2010), especially for the young and elderly (Braga et al., 1999; Farhat
et al., 2005; Gonçalves et al., 2005; Roseiro and Angela, 2006).
Although the pollution from sugar cane pre-harvest burning may be as
harmful as traffic and industrial pollution (Mazzoli-Rocha et al., 2008), there
are few studies addressing the effect of the pre-harvest burning of sugar
cane on health. Our literature review revealed only a handful of studies
from recent years: Uriarte et al. (2009), Ribeiro (2008), Arbex et al. (2007),
1 In 2009, Brazilian consumption of ethanol reached 11.3 million tonnes of oil equiv-
alent (toe), compared to 14.7 million toe of gasoline. Additionally, sugar cane
accounts for 18.1 pre cent of the Brazilian energy matrix (EPE, 2010).
2 The percentage of flex-fuel small-sized cars sold in Brazil in 2005 was 39 per cent
rising to 87 per cent in 2009 (ANFAVEA, 2010).
3 Brazil produces an anhydrous ethanol (99.6 GL) as an octane enhancer in gaso-
line with blending rates that range from 20 per cent to 26 per cent, and hydrated
ethanol (95.5 GL) for neat-ethanol engines and flex fuel engines
Environment and Development Economics 129
Cançado et al. (2006), and Arbex et al. (2000). Most of these studies are
restricted to the effect in specific municipalities and only Uriarte et al. (2009)
considers a larger geographic area. Moreover, these studies do not present
a well-defined identification strategy.
The identification problem of accessing the effect of the pre-harvest burn-
ing of sugar cane on respiratory health is the bias arising from unobserved
heterogeneity and endogeneity of the total sugar cane area. Because total
sugar cane area is likely to be endogenous in the equation for respiratory
diseases, and the parameter of interest cannot be estimated consistently, it
is therefore not identified.
The main contribution of this article is to present an estimation strat-
egy that addresses unobserved heterogeneity and endogeneity, considering
data from a larger geographic area and a long time span. In particular, we
address unobserved heterogeneity using panel data methods and avoid
endogeneity by using the change in the area harvested mechanically, with-
out fire, instead of the total area harvested. We argue that exogeneity of
the area harvested mechanically follows from the enactment of a state law
establishing a schedule for the gradual elimination of pre-harvest burning.
Additionally, this paper contrasts with the existing literature by con-
trolling for other sources of pollution, such as the automobile fleet and
industrial production, socioeconomic variables, and the extent of the pri-
vate sector health supply system. The introduction of additional control
variables and the consideration of inpatient and outpatient visits separately
provides a clearer picture of the effect of sugar cane harvesting on health.
To achieve these objectives, we collected data for the municipalities in
the state of São Paulo, Brazil, for two different periods, 2000 and 2007,
and estimate a fixed effects model considering three different proxies for
health: hospital admissions due to respiratory conditions; the number of
inhalation procedures; and the average length of hospitalization due to a
respiratory condition.
2. Methodology
2.1. Estimation strategy
The main estimation problem to access the effect of sugar cane pre-harvest
burning on respiratory health is the bias arising from the unobserved het-
erogeneity and endogeneity of the total sugar cane area. Total sugar cane
area is led by economic growth, among other factors, which is likely to be
related to other sources of air pollution and/or the quality of the health
system. Thus, the change in the total sugar cane area is endogenous in the
equation for respiratory diseases, and the coefficient of interest cannot be
estimated consistently.
Our strategy to tackle the endogeneity problem is to estimate a fixed
effects panel data model using the change in the area harvested mechani-
cally instead of the total area. The changes in the area harvested mechani-
cally are induced by the enactment of São Paulo State Law 11,241 in 2002.
This state law requires farms with less than a 12 pre cent slope to com-
pletely eliminate burning by 2021, starting with a 20 per cent reduction
in the burned area immediately. Farms with a greater than 12 per cent
slope must completely eliminate buring by 2031, starting with a 10 per cent
reduction in 2002.4
Producers started adapting rapidly to State Law 11,241, and in 2007
approximately 47 per cent of the sugar cane area in the state of São Paulo
was already harvested without burning (SMA, 2010). Apparently, such
rapid adaptation facilitated accelerating the deadlines for the abolishment
of pre-harvest burning to 2014 and 2017, in a protocol signed in 2007 by the
São Paulo State Secretary of the Environment and the Brazilian Sugar Cane
Industry Association (União da Indústria da Cana-de-Açúcar – UNICA).
We argue that the change in the area harvested mechanically induced
by State Law 11,241 is exogenous in the equation for respiratory health
because it is not related to factors such as economic growth. The reasoning
is that the changes in area across municipalities depend basically on geo-
graphic characteristics, such as terrain slope, which are independent of the
change in variables measuring respiratory diseases.
Because all municipalities are subject to the law, but it imposes differ-
ent requirements for areas with slopes greater than 12 per cent, assuming
exogeneity of the changes in the area harvested mechanically implies
assuming that slopes are not related to respiratory health through unob-
served factors. Thus we assume that respiratory health is conditionally
not correlated to the area harvested mechanically or the terrain slope.
Any unobserved factors that may be correlated are assumed to be time
invariant, such that the coefficient of interest can be estimated consistently
through a fixed effects model.
2.2. Data
As discussed before, our estimation relies on the variation in the area har-
vested of sugar cane without burning. This variation was induced by the
São Paulo state law, enacted in 2002. For this reason, we limit our analy-
sis to the state of Sao Paulo. It is worth noting that the state is the main
producer of sugar cane, corresponding to 60 per cent of the total Brazil-
ian production and 55 per cent of the total sugar cane area in 2008 (IBGE,
2010b).
We collected data from the 643 municipalities for 2000, before the state
law, and 2007, five years later. We assume that in 2000 only pre-harvest
burning was used.
In order to provide a more nuanced analysis of the effect of pre-harvest
burning on health, three proxy variables for health were used. The first
proxy is the number of inpatient visits or hospital admissions due to
a respiratory condition. The second dependent variable is the number
of outpatient visits due to the necessity of inhalation procedures. The
third variable is the average length of hospitalization due to a respiratory
condition.
Data on these proxy variables were collected from the database of the
Brazilian Ministry of Health (DATASUS), according to the International
4 According to (Aguiar et al., 2009), about 25 per cent of the total area harvested has
slopes greater than 12 per cent.
Classification of Diseases 10th codes j00 to j99 (DATASUS, 2010). Because

the observational units are municipalities, visits were measured per 1,000
inhabitants, per year. Additionally, because individuals living in smaller
cities might seek treatment in neighboring cities, we accounted for the
patient’s city of residence to build the variables.
A shortcoming of using the Ministry of Health data is that only the free
public health system visits are accounted for. This is important, because
the private health system is relatively large in the State of São Paulo,5
and the distribution between public and private systems varies consid-
erably among municipalities. If patients using the private system were
independent and identically distributed (iid), the share would be constant,
facilitating the correction required to make inferences about the whole
population. As discussed later, handling this complication requires an
additional hypothesis to define the proper correction.
Considering that the characterization of air pollution is usually linked to
the concentration of pollutants in the atmosphere, we use three indicators
to account for the volume of pollutants per m3 of air, each controlling a
different source of pollution: (i) area in thousands of hectares of raw sugar
cane that was harvested without pre-burning (IBGE, 2010b; SMA, 2010);
(ii) total fleet in 2002 as a proxy 2000 fleet, in thousands (SEADE, 2010);
and (iii) the 2001 total consumption of industrial energy, as a proxy for
2000’s consumption, in 1,000 MWh (SEADE, 2010). Implicit is the hypoth-
esis that energy consumption is related to production, which is related to
the amount of pollution generated.
Because the number of visits and length of hospitalization are likely to
be correlated with environmental factors facilitating the spread of viruses
and with the age of the population, we control for the municipalities’ pop-
ulation densities and the percentage of the population above 60 years old
to control for age (SEADE, 2010).
For the investment in health, two variables were used: the number of
health professionals and nurses per 1,000 inhabitants registered in the
regional council for each municipality (SEADE, 2010) and the percent-
age of the population covered by private health insurance, coded in three
categories: all ages, 14 years old or younger and 60 years old or older
(DATASUS, 2010).
Due to the fact that wages and education have a high correlation, we use
the average of real salaries for formal workers to account for both factors
that can affect the capacity to invest in health and the efficiency of such
investment (SEADE, 2010).
Although the weather is an important variable for explaining respiratory
conditions, the yearly average temperature, precipitation and humidity do
not vary substantially for each city from 2000 to 2007 (CIIAGRO, 2010).
For this reason, we assume that the annual average climate variables are
constant variables over time, so that heterogeneity is captured by the panel
data model.
5 In 2007, approximately 38 per cent of the state’s population was covered by

private insurance.
2.3. Econometric specification

To circumvent the problem of the unobserved heterogeneity and endogene-
ity of the total sugar cane area presented above, we proposed a panel data
model to estimate the effect of increasing the area of raw sugar cane harvest
as follows:
h it = β1 itc + β2 oit + β3 Sit + β4 Mit + β5 Wit + εit

(1)
εit = u it + θi , and u it ∼ N (0, 1)
where i represents the 643 municipalities and t the two periods, 2000 and
2007. The variable h it∗ is the number of inpatient visits per 1,000 inhabitants,
the number of inhalation procedures per 1,000 inhabitants, and the average
length of stay. itc is the harvest area of raw sugar-cane, oit is the total
fleet and total industrial energy consumption, Sit is the population density
and percentage of the population above 60, Mit is the number of health
professionals and nurses per thousand inhabitants and the percentage of
the population covered by health insurance, and Wit is the average salary.
The error term u it is homoscedastic and non-autocorrelated and θi is the
time invariant individual and municipality specific effects, such as weather
and individual behavior in the presence of pollution.6
3. Results
Descriptive statistics for the variables used in the models are presented in
table 1, and the correlations among independent variables are presented in
table 2. We note that these correlations are relatively small, even though
there are some correlation coefficients close to |0.60|. Nonetheless, stability
of the estimated coefficients and the results obtained for the t- and F-tests
suggest that multicollinearity is not a concern.
To analyze model sensitivity, we successively added blocks of inde-
pendent variables and evaluated the changes in magnitude, sign and
significance of the parameter estimates. Table 3 presents the fixed effect
regressions with cluster robust standard errors for inpatient visits due to
respiratory diseases per 1,000 inhabitants.
The first equation uses just the environmental factors, the second adds
the pollution factors, the third adds health goods, and the fourth is the
complete equation with salary. We observe that the magnitudes, signs and
significance of the parameter estimates are reasonably stable. For instance,
the effect of area of raw sugar-cane harvested on inpatient visits was −0.172
without controlling for health goods, education or salary and a decrease
in module to −0.129 for the complete model. So, by not controlling for
6 An alternative econometric specification for the error term is to consider the

spatial dependence among neighboring municipalities. In this particular case,
because there is no spatially lagged dependent variable in the model, ignoring
spatial dependence does not produce any bias, affecting only the variance of
the estimated coefficients. We chose to use clustered robust standard errors as
a variance correction.
Table 1. Summary statistics for the variables used in the models
Variable Mean Std. Dev. N
Inpatient visits/1,000 inhabitants 11.143 7.472 1288

Inpatient visits with private insurance/1,000 inhabitants 12.939 9.374 1288
Inpatient visits <15 years/1,000 inhabitants 18.347 12.669 1288
Inpatient visits >60 years/1,000 inhabitants 30.105 20.064 1288
Inhalation procedure/1,000 inhabitants 442.449 382.447 1288
Average length of stay (days) 5.031 1.465 1288
Population density 284.069 1141.331 1288
Population <15 years old 0.249 0.038 1288
Population >60 years old 0.112 0.028 1288
Harvest w/burning (1,000 ha) 1.411 3.617 1288
Total fleet (1,000 unit) 22.118 194.219 1288
Industry energy (1,000 MWh) 71.850 320.318 1286
Health professionals/1,000 inhabitants 4.799 2.782 1288
Nurses/1,000 inhabitants 4.18 2.373 1288
Private insurance/population 0.151 0.153 1288
Private insurance/population <15 years 0.146 0.167 1288
Private insurance/population >60 years 0.155 0.141 1288
Salary (2007 R$)a 1052.553 320.991 1288
Note: a This series is in real price of 2007 deflated by INPC (National Price
Consumer Index) for income (IPEADATA, 2010).
Table 2. Linear correlation among variables used in the models
Pop. Pop. Raw Total Ind. Health Private

Variables Density >60 sugar-cane fleet energy profes. benef. Salary
Population density 1.000

Population >60 −0.265 1.000
Harvest w/burning −0.068 0.075 1.000
Total fleet 0.303 −0.045 −0.000 1.000
Industrial energy 0.310 −0.183 −0.007 0.582 1.000
Health professionals 0.191 0.149 0.188 0.188 0.200 1.000
Private insurance 0.314 −0.214 0.160 0.197 0.415 0.332 1.000
Salary (2007 R$) 0.367 −0.355 0.113 0.226 0.517 0.295 0.587 1.000
those factors, one might overestimate the effect of increasing the area of
raw sugar cane harvest.
Considering the possibility of influential observations, we identified the
city of São Paulo as the only obvious candidate. With a population of
around 11 million and being the largest city in South America, São Paulo
holds a considerable share of the Brazilian industry and of Brazil’s vehi-
cle fleet, which is certainly unlike any other Brazilian city. For this reason
we also estimated the complete model, dropping the city of São Paulo, as
shown in the last column of table 3. Comparing the parameter estimates
Table 3. Sensitivity analysis of the panel data model for inpatient visits due to
respiratory condition
Dependent variable – inpatient visits per 1,000

inhabitants due to respiratory diseases
Independent St St , St , , Mt St , , Mt , E t , Wt St , , Mt , E t , Wt
variables FE model FE model FE model FE model FE model–SP
Population 0.00121 0.000615 0.00145 0.00101 0.00118

density (0.00122) (0.00124) (0.00120) (0.00123) (0.00118)
Population >60 −147.6∗∗∗ −125.4∗∗∗ −75.54∗∗ −66.84∗∗ −66.36∗∗
(21.37) (23.69) (31.93) (31.72) (31.91)
Harvest −0.172∗∗∗ −0.140∗∗∗ −0.129∗∗∗ −0.127∗∗∗
w/burning (0.0379) (0.0355) (0.0355) (0.0355)
Total fleet 0.00269∗∗∗ 0.00342∗∗∗ 0.00267∗∗∗ −0.00246
(0.000949) (0.000893) (0.000955) (0.0154)
Industrial 0.00128∗∗∗ 0.00159∗∗∗ 0.00161∗∗∗ 0.00168∗∗∗
energy (0.000421) (0.000457) (0.000423) (0.000428)
Health −0.492∗∗ −0.480∗∗ −0.476∗∗
professionals (0.194) (0.191) (0.192)
Private −3.860∗∗ −3.486∗ −3.520∗
insurance (1.928) (1.904) (1.918)
Salary (2007 R$) −0.00297∗∗ −0.00301∗∗
(0.00134) (0.00136)
Constant 27.27∗∗∗ 25.07∗∗∗ 22.12∗∗∗ 24.29∗∗∗ 24.32∗∗∗
(2.300) (2.514) (2.864) (2.933) (2.933)
Observations 1,288 1,286 1,286 1,286 1,284
R-squared 0.113 0.130 0.145 0.152 0.152
Number of 644 643 643 643 642
municipalities
Notes: Clustered robust standard errors in parentheses.

∗∗∗p < 0.01; ∗∗p < 0.05; ∗p < 0.1.
with the previous estimates we can see that, in general, they agree in both
sign and significance and magnitude, except for ‘total fleet’.
3.1. Inpatient visits

We specified four different models to account for the effect of the area of
raw sugar cane harvest on inpatient visits due to respiratory diseases. The
estimated models are presented in table 4. According to Hausman tests,
presented at the bottom of table 4, we reject at 1 per cent the null hypothesis
that the differences are in the coefficients of both models, suggesting that a
fixed effects model is more appropriate.
The first model is for the number of inpatient visits, controlling for
the percentage of individuals covered by private health insurance. In the
second model, we assume that all inpatient visits correspond to uninsured
individuals, covered only by the free public health system. We assume
further that visits are equally distributed for both public and private sys-
tems. Therefore, the total number of visits were approximated by dividing
Table 4. Estimation results for distinct measures of inpatient visits due to

respiratory conditions
Dependent variable – inpatient visits

due to respiratory diseases
Inpatient with Inpatient Inpatient
Inpatient private sector <15 visits >60
Variable Fixed effect model
Population density 0.00101 −0.00133 0.00282 −0.00229

(0.00123) (0.00215) (0.00214) (0.00360)
Population >60 −66.84∗∗ −66.94∗∗
(31.72) (33.88)
Harvest w/burning −0.129∗∗∗ −0.148∗∗ −0.192∗∗ −0.378∗∗∗
(0.0355) (0.0678) (0.0772) (0.103)
Total fleet 0.00267∗∗∗ 0.00203 0.00271 0.00609∗∗∗
(0.000955) (0.00349) (0.00283) (0.00233)
Industrial energy 0.00161∗∗∗ 0.0168∗∗∗ 0.00203∗∗ 0.00336∗∗∗
(0.000423) (0.00203) (0.000864) (0.000860)
Health professionals −0.480∗∗ −0.628∗∗∗ −0.871∗∗∗ −1.737∗∗∗
(0.191) (0.221) (0.241) (0.429)
Private insurance −3.486∗
(1.904)
Private insurance <15 −2.275
(2.560)
Private insurance >60 −4.507
(8.136)
Salary (2007 R$) −0.00297∗∗ −0.00295∗ −0.00294 −0.00863∗∗
(0.00134) (0.00159) (0.00203) (0.00419)
Constant 24.29∗∗∗ 25.87∗∗∗ 25.24∗∗∗ 49.06∗∗∗
(2.933) (3.236) (2.341) (4.641)
F 19.24∗∗∗ 20.14∗∗∗ 6.27∗∗∗ 13.25∗∗∗
Observations 1286 1286 1286 1286
Hausman test 60.40∗∗∗ 53.15∗∗∗ 32.20∗∗∗ 34.69∗∗∗

∗∗∗p < 0.01; ∗∗p < 0.05; ∗p < 0.1.
the observed number of visits in the public system by the proportion of the
population not covered by private insurance.
The third model considers the number of inpatient visits by individuals
younger than 15 years old, and the last model considers the number of vis-
its for the population older than 60 years old. Note that in these last two
models, there is no correction to take into account visits in the private sys-
tem, but they do include controls for the private coverage share, computed
within each age bracket.
The variable of interest, the raw sugar cane harvest area ( c ), has a
negative and significant effect in all estimated models. The increase of
1,000 hectares harvested without burning decreases the number of inpa-
tient visits by 0.129 per 1,000 inhabitants. Correcting the number of visits
for private coverage, the estimated effect increases, in absolute value, to

0.148. The effect of the area of raw sugar cane harvest is more important for
the population younger than 15 years old and older than 60 years old. The
effect of an increase of 1,000 hectares of sugar cane harvested without burn-
ing decrease the inpatient visits by 0.192 and 0.378, respectively. Finally,
it is worth noting that those coefficients are lower than the ones shown
in table 3 without controlling for health goods, indicating that omitting
control variables may result in overestimating effects.
These results relating the pre-harvest burning to health are in line with
the literature. Uriarte et al. (2009), Arbex et al. (2007), Cançado et al. (2006)
and Arbex et al. (2000), for example, present evidence of a positive relation-
ship between pre-harvest burning and respiratory diseases. The literature
on the relationship between urban air pollution and health is much more
extensive, with a considerable number of studies presenting evidence in
line with our results, such as Farhat et al. (2005); Gonçalves et al. (2005);
Roseiro and Angela (2006); Sicard et al. (2010) and Braga et al. (1999).
Even though the controlling variables are included in the model mainly
to isolate the effect of the variable of interest, the corresponding parameter
estimates may shed some light on the internal validity of the models. An
inspection of table 4 reveals that most parameter estimates for the control-
ling variables have the correct sign in most cases. An obvious exception is
that the percentage of the population over 60 years old has a significant
negative effect in the models regardless of correcting for private coverage.
In other words, an increase in the population over 60 decreases the number
of inpatient visits per 1,000 inhabitants.
For the variable controlling for other sources of pollution ( o ), we
observe that the coefficient for the total fleet is positive and significant
for the model with the sample restricted to those older than 60 years old.
Industrial energy consumption has a positive and significant effect in all
the estimated models, with a slightly greater effect for the sample restricted
to older individuals.
The number of health professionals has a negative and significant effect
on the number of inpatient visits. This may be expected because better
supply conditions facilitates access to health care, which in turn leads to
less acute problems requiring hospitalization. On the other hand, more
health goods imply better health conditions. This is especially important
for individuals younger than 15 and over 60 years old. Thus, according to
the inpatient visits model, an increase of one health professional per 1,000
inhabitants decreases the inpatient visits by 0.48 for the whole population,
0.628 for inpatient care in the private health sector model, 0.871 for the
population younger than 15 years old, and 1.737 for the population over 60
years of age.
The percentage of the population in the municipalities covered by pri-
vate health insurance has a negative and significant effect (at the 10 per cent
level) in the model considering the whole population, and is not signif-
icant for the models with their samples restricted to younger and older
individuals.
Finally, the average salary (W ) in a municipality has a negative effect on
inpatient visits. The argument here is that better salaries are correlated with
Table 5. Estimation results for outpatient visits and length of stay due to respiratory
conditions
Inhalation Inhalation Length of stay
Variable FE Model RE Model FE Model
Population density −0.0315 −0.00584 0.000421

(0.0523) (0.00956) (0.000385)
Population <15 2,043∗∗∗ 1,633∗∗∗
(627.5) (298.7)
Population >60 −10.59∗
(5.470)
Harvest w/burning 0.701 −1.770 0.0106
(2.707) (2.327) (0.00996)
Total fleet 0.0725 −0.0167 0.000690
(0.0882) (0.0418) (0.000657)
Industrial energy −0.122∗∗ −0.0434 0.000350
(0.0483) (0.0411) (0.000264)
Nurses −6.053 −11.17∗∗∗
(9.986) (4.106)
Health professionals 0.0458
(0.0402)
Private insurance −12.10 220.6∗∗ 0.456
(153.0) (109.0) (0.518)
Salary (2007 R$) 0.0316 −0.0794∗ −0.000397
(0.0781) (0.0471) (0.000274)
Constant −61.17 136.6 6.166∗∗∗
(222.4) (86.04) (0.554)
F 7.78∗∗∗ 1535.74∗∗∗ 1.55
Observations 1286 1286 1286
Hausman test 9.52 9.52 12.19∗∗

∗∗∗p < 0.01; ∗∗p < 0.05; ∗p < 0.1.
better education, implying a greater and more efficient capacity to invest in

health and so fewer inpatient visits. The effect of salary was higher for the
population over 60 years old and it was not significant for the population
under 15 years of age.
3.2. Outpatient visits and length of stay

Table 5 shows the estimation results for the other two proxies for health,
namely outpatient visits and length of hospitalization due to a respiratory
condition. The model for outpatient visits includes the number of inhala-
tion procedures per 1,000 inhabitants as the dependent variable, and the
corresponding estimates are shown in the first column of table 5. The
second covers inpatient length of stay due to respiratory diseases.
Unlike the model for inpatient visits, the null hypothesis of the Haus-
man test is not rejected for the outpatient visits model. Rejection of the null
hypothesis suggests that the random effects model may be more appropri-
ate in this case. Nonetheless, because the random effects model is difficult
to justify in this setting, we present both fixed and random effects models.
For the length of stay model, the null hypothesis is rejected, and thus only
the fixed effects model is presented. See the test results at the bottom of
table 5.
Starting with the variable of interest, the area of raw sugar cane harvest,
it is clear from the results presented that there is no statistically significant
effect on either outpatient visits or length of hospitalization. The contrast
of this finding with the results obtained earlier for inpatient visits sug-
gests that pre-harvest burning imposes a harsher toll on acute respiratory
conditions, which require brief hospitalization and reduced variance in
the length of stay. It also indicates that focusing on outpatient visits for
inhalation may underestimate the health effect of air pollution.
Also contrasting with the model for inpatient visits, few of the control
variables have statistically significant coefficients. The model for length of
stay did not reveal any interesting health effect associated with the control
variables, nor did the model for the number of outpatient visits. In this case,
the proportion of the population younger than 15 years old seems to hold
the main effect, with a positive and significant effect on outpatient visits, in
both the random effects and fixed effects models.
4. Conclusion
This paper presented an analysis of the respiratory health effect of the
air pollution caused by the pre-harvest burning of sugar-cane. As sugar
cane has increased in importance in the energy agenda in Brazil and in the
world, it is important to shed some light on the impact of its production on
the environment and the health of the population. This article contributes
to this body of literature by presenting evidence that there may be a sig-
nificant health effect, using a novel identification strategy and a unique
data set.
The proposed identification strategy relies on the enactment of a state
law reducing the pre-harvest-burned area over time. After collecting data
before and after the law took effect, we estimate a series of panel data mod-
els that control for endogeneity and make it possible to evaluate the health
effect of reducing the area burned.
The results obtained suggest that reducing the area where sugar cane is
harvested after burning reduced the number of inpatient visits in the state
of São Paulo. Interestingly, we found that the effect of pre-harvest burning
is relatively large, as compared to the estimated effects of the total vehicle
fleet and industrial pollution, included as control variables in the models.
On the other hand, the models estimated for the number of outpatient
visits for inhalation procedures and for the length of stay did not reveal
any significant relationship between pre-harvest burning and respiratory
conditions.
These results suggest that the effect may be restricted to acute respiratory
conditions, which require brief hospitalization and reduced variance in the
length of stay. Moreover, based on these results, we may argue that the state
law contributed to the improvement of the population’s respiratory health
in the state of São Paulo.
References
Aguiar, D.A., B.F.T. Rudorff, M. Adami, and Y.E. Shimabukuro (2009), ‘Imagens
de sensoriamento remoto no monitoramento da colheita da cana-de-açúcar’,
Engenharia Agrı́cola 29(3): 440–451.
Allen, A., A. Cardoso, and G. da Rocha (2004), ‘Influence of sugar cane burning on
aerosol soluble ion composition in Southeastern Brazil’, Atmospheric Environment
38(30): 5025–5038.
Andrade, S.J., J. Cristale, F.S. Silva, G.J. Zocolo, and M.R. Marchi (2010), ‘Contribu-
tion of sugar-cane harvesting season to atmospheric contamination by polycyclic
aromatic hydrocarbons (PAHs) in Araraquara city, Southeast Brazil’, Atmospheric
Environment 44, 2913–2919.
ANFAVEA (2010), ‘Brazilian automotive production’, Historical data, Associação
Nacional dos Fabricantes de Veı́culos Automotores São Paulo, Brazil.
Arbex, M., G. Böhm, P. Saldiva, III, A.P.G. Conceição and A. Braga (2000), ‘Assess-
ment of the effects of sugar cane plantation burning on daily counts of inhalation
therapy’, Journal of the Air and Waste Management Association 50: 1745–1749.
Arbex, M.A., L.C. Martins, R.C. de Oliveira, L.A.A. Pereira, F.F. Arbex, J.E.D.
Cançado, P.H.N. Saldiva, and A.L.F. Braga (2007), ‘Air pollution from biomass
burning and asthma hospital admissions in a sugar cane plantation area in Brazil’,
Journal of Epidemiology and Community Health 61(5): 395–400.
Börjesson, P. (2009), ‘Good or bad bioethanol from a greenhouse gas perspective –
what determines this?’, Applied Energy 86(5): 589–594.
Braga, A.L., G.M. Conceição, L.A. Pereira, H.S. Kishi, J.C. Pereira, M.F. Andrade,
F.L. Gonçalves, P.H. Saldiva, and M.R. Latorre (1999), ‘Air pollution and pedi-
atric respiratory hospital admissions in São Paulo, Brazil’, Journal of Environmental
Medicine 1(2): 95–102.
Cançado, J.E.D., P.H.N. Saldiva, L.A.A. Pereira, L.B.L.S. Lara, P. Artaxo, L.A. Mar-
tinelli, M.A. Arbex, A. Zanobetti, and A.L.F. Braga (2006), ‘The impact of sugar
cane burning emissions on the respiratory system of children and the elderly.’,
Environmental Health Perspectives 114(5): 725–729.
CIIAGRO (2010), Centro Integrado de Informações Agrometeorológicas (Center
of Agrometeorological Information), Technical Report, Instituto Agronômico de
Campinas, São Paulo, Brazil.
Crutzen, P.J., A.R. Mosier, K.A. Smith, and W. Winiwarter (2008), ‘N2 O release from
agro-biofuel production negates global warming reduction by replacing fossil
fuels’, Atmospheric Chemistry and Physics 8(2): 389–395.
DATASUS (2010), Informações de Saúde, Technical report, Ministério da Saúde.
EIA (2006), International Energy Outlook, Report, Energy Information Administra-
tion, D.F. Brasilia, Brazil.
EPE (2010), Balanço Energético Nacional 2010, Technical Report, Empresa de
Pesquisa Energética, Ministério de Minas e Energia, D.F. Brası́lia, Brazil.
Farhat, S., R. Paulo, T. Shimoda, G. Conceição, C. Lin, A. Braga, M. Warth, and
P. Saldiva (2005), ‘Effect of air pollution on pediatric respiratory emergency room
visits and hospital admissions’, Brazilian Journal of Medical and Biological Research
38(2): 227–235.
Giampietro, M., S. Ulgiati, and D. Pimentel (1997), ‘Feasibility of large-scale bio-
fuel production – does an enlargement of scale change the picture?’, Bioscience
47(9): 587–600.
Gonçalves, F., L. Carvalho, F. Conde, M. Latorre, P. Saldiva, and A. Braga (2005), ‘The
effects of air pollution and meteorological parameters on respiratory morbidity
during the summer in São Paulo city’, Environment International 31(3): 343–349.
Gunkel, G., J. Kosmol, M. Sobral, H. Rohn, S. Montenegro, and J. Aureliano (2007),
‘Sugar cane industry as a source of water pollution – case study on the situation in
Ipojuca river, Pernambuco, Brazil’, Water Air and Soil Pollution 180(1–4): 261–269.
IBGE (2010a), Censo Agropecuário, Historical data, Instituto Brasileiro de Geografia
e Estatı́stica, Rio de Janeiro, Brazil.
IBGE (2010b), Pesquisa Agropecuário Municipal, Historical data, Instituto Brasileiro
de Geografia e Estatı́stica, Rio de Janeiro, Brazil.
IPEADATA (2010), ‘Índice Nacional de Preços ao Consumidor’, Technical Report,
Instituto de Pesquisa Econômica Aplicada.
Lara, L., P. Artaxo, L. Martinelli, P. Camargo, R. Victoria, and E. Ferraz (2005),
‘Properties of aerosols from sugar-cane burning emissions in Southeastern Brazil’,
Atmospheric Environment 39(26): 4627–4637.
Mazzoli-Rocha, F., C.B. Magalhães, O. Malm, P.H.N. Saldiva, W.A. Zin, and D.S.
Faffe (2008), ‘Comparative respiratory toxicity of particles produced by traffic and
sugar cane burning’, Environmental Research 108(1): 35–41.
Moreira, J.R. (2000), ‘Sugarcane for energy – recent results and progress in Brazil’,
Energy for Sustainable Development 4(3): 43–54.
Moreira, J.R. and J. Goldemberg (1999), ‘The alcohol program’, Energy Policy
27(4): 229–245.
Oliveira, M.E. D.d., B.E. Vaughan, and E.J. Rykiel Jr. (2005), ‘Ethanol as fuel: energy,
carbon dioxide balances, and ecological footprint.’, Bioscience 55(7): 593–602.
Ribeiro, H. (2008), ‘Queimadas de cana-de-açúcar no Brasil: efeitos à saúde respi-
ratória’, Revista de Saúde Pública 42: 370–376.
Roseiro, M.N.V. and M.M.T. Angela (2006), ‘Morbidade por problemas respiratórios
em Ribeirão Preto (SP) de 1995 a 2001, segundo indicadores ambientais, sociais e
econômicos’, Revista Paulista de Pediatria 24(2): 163–170.
SEADE (2010), Informações dos municı́pios paulistas – imp, Technical report,
Fundação Sistema Estadual de Análise de Dados, São Paulo, Brazil.
Sicard, P., A. Mangin, P. Hebel, and P. Malléa (2010), ‘Detection and estimation
trends linked to air quality and mortality on French Riviera over the 1990–2005
period’, Science of the Total Environment 408(8): 1943–1950.
SMA (2010), Colheita mecanizada – projeto CANASAT, Technical Report, Secretaria
do Meio Ambiente do Estado de São Paulo, São Paulo, Brazil.
Uriarte, M., C.B. Yackulic, T. Cooper, D. Flynn, M. Cortes, T. Crk, G. Cullman,
M. McGinty, and J. Sircely (2009), ‘Expansion of sugarcane production in São
Paulo, Brazil: implications for fire occurrence and respiratory health’, Agriculture,
Ecosystems & Environment 132(1–2): 48–56.

5-Effect of Reducing The Preharvest Burning of Sugar Cane On Respiratory Health in Brazil

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

5-Effect of Reducing The Preharvest Burning of Sugar Cane On Respiratory Health in Brazil

Uploaded by

Copyright:

Available Formats

Environment and Development Economics 20: 127–140 © Cambridge University Press 2014

The effect of reducing the pre-harvest burning of

The use of sugar cane ethanol as a substitute for gasoline is widespread

Classification of Diseases 10th codes j00 to j99 (DATASUS, 2010). Because

5 In 2007, approximately 38 per cent of the state’s population was covered by

2.3. Econometric specification

h it = β1 itc + β2 oit + β3 Sit + β4 Mit + β5 Wit + εit

6 An alternative econometric specification for the error term is to consider the

Table 1. Summary statistics for the variables used in the models

Variable Mean Std. Dev. N

Inpatient visits/1,000 inhabitants 11.143 7.472 1288

Table 2. Linear correlation among variables used in the models

Pop. Pop. Raw Total Ind. Health Private

Population density 1.000

Dependent variable – inpatient visits per 1,000

Population 0.00121 0.000615 0.00145 0.00101 0.00118

Notes: Clustered robust standard errors in parentheses.

3.1. Inpatient visits

Table 4. Estimation results for distinct measures of inpatient visits due to

Dependent variable – inpatient visits

Variable Fixed effect model

Population density 0.00101 −0.00133 0.00282 −0.00229

Notes: Clustered robust standard errors in parentheses.

for private coverage, the estimated effect increases, in absolute value, to

Population density −0.0315 −0.00584 0.000421

Notes: Clustered robust standard errors in parentheses.

better education, implying a greater and more efficient capacity to invest in

3.2. Outpatient visits and length of stay

You might also like