Electronic copy available at: http://ssrn.com/abstract=1711418

Biodiversity Conservation and Child Malaria: Microeconomic Evidence from Flores,

Indonesia

November 18, 2010

ERID Working Paper Number 85

This paper can be downloaded without charge from

The Social Science Research Network Electronic Paper Collection: http://ssrn.com/abstract=1711418

Subhrendu K. Pattanayak

Catherine G. Corey

Yewah F. Lau

Randall A. Kramer

Duke University

New York City Department of Health and Mental Hygiene

United States Forest Service

Duke University

Electronic copy available at: http://ssrn.com/abstract=1711418

1

BIODIVERSITY CONSERVATION AND CHILD MALARIA:

MICROECONOMIC EVIDENCE FROM FLORES, INDONESIA

Subhrendu K. Pattanayak†

Associate Professor, Sanford School of Public Policy & Nicholas School of the Environment,

Duke University

Catherine G. Corey

Bureau of Epidemiology Services, New York City Department of Health and Mental Hygiene

Yewah F. Lau

Forest Planner, Coconino National Forest, United States Forest Service

Randall A. Kramer

Professor, Nicholas School of the Environment, Duke University

ABSTRACT: In remote areas of developing countries, people‟s health and livelihoods are closely

intertwined with the condition of the natural environment. Unfortunately, claims regarding the role of

ecosystem degradation on disease outcomes rest on a short list of rigorous empirical studies that consider

social, cultural and economic factors that underpin both ecosystem disruptions and behaviors related to

exposure, prevention and treatment of diseases such as malaria. As the human ecological tradition suggests,

omitting behaviors can lead to erroneous interpretations regarding the nature of the relationship between

ecological changes and disease. We specify and test the relationship between child malaria prevalence and

forest conditions in a quasi-experimental setting of buffer zone villages around a protected area, which was

established to conserve biodiversity on Flores, Indonesia. Multivariate probit regressions are used to

examine this conservation and health hypothesis, controlling for several individual, family and community

variables that could confound this hypothesized link. We find that the extent of primary (protected) forest is

negatively associated with child malaria, while the extent of secondary (disturbed) forest cover is positively

correlated with child malaria, all else equal. This finding emphasizes the natural insurance value of

conservation because children are both especially vulnerable to changes in environmental risks and key

players in the future growth and prosperity of a society.

KEYWORDS: Indonesian national parks, tropical watershed conservation, ecosystem change,

environmental health, micro-econometrics, human ecology.

FULL WORD COUNT: 5699

† Address correspondence to subhrendu.pattanayak@duke.edu; Associate Professor of Public Policy and

Environmental Economics, Duke University. Yewah Lau is a Forest Planner for the Coconino National Forest

(ylau@fs.fed.us or 928 527-3411). The authors would like to thank Frans Dabukke, Sastrawan Manullang, Mariyanti

Hendro, and Nining N.P. for collection and interpretation of the data and David Butry for help with the spatial

matching of environmental statistics and survey village locations. We also acknowledge the Howard Gilman

Foundation Grant, the Josiah Charles Trent Memorial Foundation, Duke University‟s Center for Environmental

Solutions and Conservation International‟s Center for Applied Biodiversity Sciences (CABS) for partial funding for

this research. Keith Alger, Erin Sills, and seminar participants at NC State University, Camp Resources XI and

Environmental Institutions Seminars provided useful comments on earlier drafts of this paper.

Electronic copy available at: http://ssrn.com/abstract=1711418

2

The natural environment is vitally important for people living in tropical forested areas.

However, little is known about how ecosystem change impacts public health, particularly in rural

areas that lack hospitals, doctors and other public services. Widespread and rapid degradation of

tropical forests represents perhaps the most significant environmental transformation of rural

landscapes. Globally, the rate of deforestation averaged 16 million hectares from 1990 to 2000, a

three percent increase in the rate of the previous decade (FAO, 2003). Co-occurring with this

profound transformation of the landscape has been the reemergence of a number of infectious

diseases such as malaria and pockets of acute and persistence poverty. Globally, malaria ranks as

the number one vector-borne disease, causing about 2 million deaths annually and responsible for

13% of global disability and mortality, particularly among children (WHO, 2005). While simple

theories of causality cannot explain this co-occurrence, the literature makes it abundantly clear that

this juxtaposition of deprivation, deforestation, and disease is not pure coincidence (Mayer, 2000;

Wolman, 1995). The design and implementation of policies to improve human lives and conserve

forests depend critically on improving our understanding of interrelationships between forest

quality, malaria incidence and human behaviors. Through a case study from Flores, Indonesia, we

show in this paper that forest conservation is correlated with malaria prevention, as revealed by a

multivariate econometric model of household survey data combined with regional environmental

statistics.

In the past decade, widely cited papers have drawn connections between ecosystem change

and diseases, many of which are synthesized in the 2005 Millennium Ecosystem Assessment

(Corvalan et al, 2005a, 2005b; Campbell-Lendrum et al, 2005; Patz et al, 2005; Colfer et al.,

2006). The variety and magnitude of environmental influences on this vector-borne disease is

enormous; climatic factors such as precipitation and temperature affect the abundance of mosquito

vectors and the development of parasites within the vectors, while anthropogenic influences

operating through deforestation, agriculture, and housing construction may influence vectorial

capacity (Wilson, 2001). Environmental factors, such as “practices regarding land use,

deforestation, water resource management, settlement siting and modified house design”, are

estimated to contribute to 42% of malaria cases worldwide as well as in southeast Asia (Pruss and

Corvalan, 2006:10).

3

A recently published synthesis of the literature on forests and malaria proposes numerous

pathways through which forest degradation (including disturbance, fragmentation and

deforestation) can affect malaria infection and disease (Pattanayak et al., 2006). First,

deforestation changes the ecology of a disease vector and its options for hosts. For example,

cleared lands are generally more sunlit and prone to the formation of puddles with more neutral

pH which can favor specific anopheline larvae development (Patz et al., 2000). Second,

deforestation can negatively impact biodiversity that favor proliferation of malaria-related species

(e.g., anopheles) by eliminating species such as dragonflies that prey on anopheles larvae. Third,

deforestation can change local climate and thereby affect the spread of disease by raising ground

temperatures which can increase the rate at which mosquitoes develop into adults, the frequency

of their blood feeding, the rate at which parasites are acquired, and the incubation of the parasite

within mosquitoes (Walsh et al., 1993). Fourth, forest degradation is often the beginning of a

variety of land use changes that may not only result in mosquito populations that have higher rates

of malaria transmission, but may also lead to increased human contact and transmission (Petney,

2001). Finally, deforestation is accompanied by migration that aids transmission. Not only do

migrants have little previous exposure and lower natural immunity, it is difficult to administer

health services to transient populations.

Unfortunately, this literature on forests and malaria has taken a predominantly biophysical

approach and overlooks the social, cultural and economic factors drivers that are crucial to

understanding anthropogenic ecosystem disruptions and their human health impacts. Only a

handful of studies consider behaviors related to exposure, prevention and treatment of malaria

(Vosti, 1990; Castilla & Sawyer, 1993; Perz, 1997; Lansang et al., 1997; Castro & Singer, 2001;

Barbieri et al. 2005). This literature shows that multiple socio-economic and community factors,

including but not limited to wealth, knowledge and awareness of the malaria problem, age, gender,

and education to be associated with malaria transmission. Omitting behavioral responses from any

analysis of malaria and ecosystem change would result in a classic case of confounding because

human behavior can mimic the risk factor and mask the ecological relationship we are attempting

to discover. As Pattanayak and Yasuoka (2008) show with three case studies, omitting behavior

leads to erroneous interpretations regarding the nature of ecological changes and disease: the size,

sign and statistical significance of regression coefficient of deforestation can be wrong.

4

To better understand the link between ecological change and human health, we draw on the

human ecology tradition that presents humans as active players (not passive hosts) in the

environment-host-agent triad (Wessen, 1972; McCormack, 1984). This approach is important

because it posits that humans (a) modify their natural environment, sometimes increasing disease

risks, and (b) ultimately adapt to the new disease risk environment. Within this human ecology

umbrella, we apply and extend a household health production model to view deforestation as a

direct input into production of children‟s heath through its role in vector ecology, particularly in

forested rural areas with limited public health infrastructure (Pattanayak and Wendland, 2007).

Forest condition may also have indirect impacts through behavioral responses to changes in the

natural resource base. This juxtaposition of proximal behavioral risks and distal ecological risks

allows us to specify econometric models that can test for correlations between forest condition and

child malaria rates, conditional on socio-economic, public infrastructure, and individual

demographic factors.

We focus on child malaria in this paper for three reasons. First, children are most

vulnerable among all sub-populations to health hazards, for example, bearing the burden of

between 75 and 90 percent of the mortality and morbidity attributed to malaria. Second, relative to

an adult, a child‟s health depends primarily on parental decisions and external factors, rather than

the child‟s personal choices and behaviors. This translates into a relatively simpler modeling task

because the estimation model does not include endogenous variables that can result in biased

estimates. An example of an endogenous behavioral variable in a model of adult (not child)

malaria includes household income or SES (which may depend on an adult‟s health status, but

unlikely to depend on a child‟s health). Third, many analyses of malaria cannot adequately

account for acquired immunity to the disease because of inadequate data. By focusing on children,

we mitigate this potential source of bias because they are unlikely to have developed immunity to

malaria in their short lifespan.

The data for this analysis are drawn from a household survey in the Manggarai district of

Flores island, Indonesia in 1996 around a protected area (Ruteng Park) that was established to

conserve biodiversity. This data is supplemented with environmental statistics (e.g., amount of

primary and secondary forest cover) and public infrastructure data (e.g., road network and sub-

regional health care facilities). Multivariate probit regression models suggest that primary

5

(protected) forest cover is negatively associated with lower prevalence of child malaria, while

secondary (disturbed) forest cover is positively correlated with malaria prevalence. These results

are robust to the inclusion of child, family, community and environmental factors in the model and

a variety of specification checks. Thus, we find that environmental conservation can complement

public health policies and provide a form of natural insurance, at least in tropical forested

watersheds such as Ruteng.

RUTENG PARK IN FLORES, INDONESIA – A CASE STUDY

The setting for our empirical analyses is the buffer zone area of Ruteng Park on Flores

Island in Indonesia where children have experienced different rates of malaria. Ruteng Park is an

Integrated Conservation and Development Project (ICDP) established in 1993 on approximately

32,000 hectares of rugged terrain encompassing seven volcanic ridges that are between 900 and

2400 meters. The park protects the best submontane and montane forests of the fragmented

forests of Flores, forms a critical watershed for the population of Ruteng town and its surrounding

farms, and conserves biodiversity (cave bats and Komodo rats are examples of two known

endemic species). Villages in the buffer zone of the Park watersheds continue to face varying

degrees of forest degradation because of unequal forest protection. This quasi-experimental

setting thus allows us to investigate whether there is any systematic spatial correlation between

child malaria and forest condition, controlling for key individual, household and community

factors.

The data for this paper primarily come from three sources. Individual and household data

are from a survey conducted in Ruteng in 1996 as part of a larger economic analysis project on

protected areas (Kramer et al., 1997). Five hundred households were selected from 13,700

farming households living in 48 village clusters (desas) in the buffer zone of Ruteng Park using

stratified random sampling and weighted based on population density of the desas. The household

surveys gathered detailed information on socio-economic characteristics including annual income,

assets (e.g., consumer durables), and housing conditions. Additionally, the surveys collected

information on age, gender, occupation, education levels and diseases for each member of the

household.

6

Data on public infrastructure such as hospitals, health care clinics, and schools were

collected from secondary and administrative sources. District-level (kecamatan) health

infrastructure information included data on the number of health care workers such as doctors,

nurses, midwives, and paramedics in the district, as well as the number of medical facilities such

as hospitals, health centers, clinics and pharmacies. These estimates were scaled by the proportion

of the district population living in a particular village to develop village level approximations of

health infrastructure.

Finally, the household and administrative data was integrated with ecological data on

forest cover within a geographical information system (GIS). Priyanto (1996) provides estimates

of secondary (or regenerating) and primary (undisturbed) forest cover by micro-watershed of

rivers and streams that emerge from Ruteng Park. By overlaying environmental (e.g., watershed

boundaries) and administrative (e.g., desa boundaries) data in a GIS, we obtain estimates of forest

cover by desa (see Figures 2 and 3).

Data on malaria was based on survey questions answered by the primary care giver on the

individual diseases in the 12 months prior to the survey. This generated a binary variable

measuring whether or not an individual had malaria. Table 1 presents the prevalence rates of major

diseases among children in the study area. The occurrence of malaria generally increases with age.

Approximately 27 percent in those less than 5 years were reported to have experienced malaria,

while nearly 42 percent of children over 10 and under 16 years were reported to have experienced

malaria. Males in the youngest cohort were reported to have experienced approximately 33 percent

higher rates of malaria than females, though in the older cohorts females had slightly higher rates

of malaria than males. Figure 1 shows how the average incidence of malaria in children varies

across the sample villages. Darker shading of the villages indicates higher disease rates in

children, with a range of 0 to 88%.

Laboratory analysis, which could have improved the accuracy of the malaria records, was

beyond the scope and resources of our study. Instead we rely on three features that justify the use

of our data for the coarse level hypothesis testing pursued in this paper. First, maternal recall is a

common method used to assess child health status in large household surveys (e.g., Demographic

Health Surveys that are conducted all around the world). Second, unlike Sub-Saharan Africa

7

(where sensitivity estimates of maternal recall is high due to different endemicities), self-diagnosis

is a smaller problem in an endemic area such as the eastern Indonesian islands where malaria

transmission is stable, follows a predictable cycle and is familiar to the local populations. Third, if

significant measurement error persists, it is likely to be of a classical variety that is absorbed into

the model error in our regression analysis. There is no persuasive argument suggesting any

systematic correlation between the reporting error and the forest variables, which could confound

our inference regarding the forest variables.

There are two broad ways for thinking about the determinants of child health, including

forest conditions: the integrated socio-biological framework from demography (Mosley and Chen,

1984) and the household health production approach used by economists (Berman et al. 1994).

Irrespective of the approach, two broad conclusions emerge. First, health outcomes have both

proximate causes (e.g., nutrition and household prevention) and distal causes (e.g., socio-

economic status, education, and cultural factors [diet, hygiene and what economists call

preferences]). Some factors will mediate between underlying and immediate causes (e.g., mother‟s

education will moderate the impact of bed net supply on child malaria). Second, determinants of

health outcomes can be classified as individual-specific, household and community characteristics.

Alternatively, the determinants are classified as biological (often individual-specific), SES (often

household-specific) and environmental and cultural. We can represent a model of health at various

scales: the individual, the household, and the village. However, we analyze health within a

household primarily because many determinants of health, particularly those risk factors affecting

an individual, operate at the household level and aggregation to the village level results in a loss of

information and increase in measurement bias.

Next we describe the empirical surrogates for each of the child, household, community and

environmental factors that influence the relationship between malaria and forest condition.

Starting with our key ecological risk variables, we measure forest quality for each household in

terms of the amount of primary (undisturbed) and secondary (regenerating) forests associated with

the buffer zone desa that they live in. On average, a typical desa has approximately 280 hectares

of primary forests and approximately 116 hectares of secondary forests.

8

Agricultural fields in this area were originally carved out of the forest through slash and

burn agriculture, but households are no longer expanding their fields, as reported in the survey

(93% of households said they did not intend to expand fields in next year), probably largely due to

the park. Even though the current park management came into existence in 1993, the core area has

been protected for almost 100 years (Indonesian Ministry of Forestry, 1995). Secondary forests are

largely the result of logging by family or commercial firms from outside our study area

(Blomkvist, 1995). Therefore, it is possible that the secondary forests are impacted by households

in a desa (and neighboring desas) to some degree (Moelino, 1995). Overall, we maintain that both

forest variables are exogenous to a particular household‟s health. Two further technical reasons

ensure that the forest variables are pre-determined and exogenous in our econometric model of

child malaria. First, we are using lagged values of the ecological data, i.e., from years preceding

the household survey, to capture overall environmental conditions: lagged values are typically pre-

determined in any estimation. Second, forest cover is measured at the sub-watershed level, which

is an order of magnitudes larger than any individual farm or family. Actions and health of

individual farm households cannot impact overall watershed conditions.

Child specific demographic factors – age and gender – constitute our second set of factors.

Age, for example, may be correlated with exposure because as the child gets older, he or she

participates in a greater number of chores, such as fuelwood and water collection, that involve

going into forests. Alternatively, as a child grows, he or she may develop resistance to malaria

through previous malaria incidences. The mean age of children is 8 years. We include three age

cohorts in our analysis: 0–5 years of age, 5–10 years of age, and 10–15 years of age. Preferential

treatment of boys or different behaviors between boys and girls could lead to differences in

prevalence (Kondrashin & Orlov, 1989).

Household characteristics constitute our third set of malaria determinants. First, the survey

provides several measures of SES (annual household expenditure, annual farm value-added, land

and livestock ownership, and assets) that are highly correlated. We employ an asset index that is a

simple count of household possessions: radio, a television, electricity, wall clock, wrist watches,

kerosene stove, and motorcycle. The typical household owns less than one consumer durable.

Previous analysis and publications with this variable confirm that it serves as a good proxy for

SES and is correlated with the other SES measures.

9

Second, we might expect a child‟s health and care giving to depend on the human capital

of the mother (Jalan & Ravallion, 2001; Perz, 1997). Thus, we might expect to see the mother‟s

occupation, health stock, education, and age impacting child health outcomes. There is practically

no variation in the mother‟s occupation in our data set (all are farmers) and so we exclude this

variable. We proxy the mother‟s health stock by a count of the number of activities in which she

participates on a day to day basis. The typical mother is nearly 35 years old. Nearly 70 percent of

the mothers have some formal education. Caregivers typically experienced one illness during the

12 months prior to the survey. The final specification includes health stock, age and education.

Third, housing quality can enhance malaria transmission (Barbieri et al. 2005). Based on

the condition of the walls and floor, we find that approximately 10% of our sample of houses fell

into this category. Specifically, a house was deemed as poor if the walls were constructed from

bamboo and the floor of the dwelling was erected on stilts.

Community and environmental variables constitute our final set of malaria risk factors.

First, the extent of public health facilities (e.g., clinics and health posts) could impact malaria

outcomes by improving access to health care and serving as a forum for malaria prevention and

control. In this remote part of rural Indonesia, health care facilities tended to be located at sub-

district headquarters, at least in 1996 at the time of this study. The population in every village in

that sub-district had to share the clinic facilities and the associated health care. So we created a

per-capita public health facility variable to account for a typical households‟ access to health care

and malaria prevention.

Second, large populations and crowded living conditions can encourage the spread of

communicable diseases. So we include the village population density to approximate these

influences. On average, villages consist of 2,500 inhabitants and cover approximately as many

hectares. Third, we also include village elevation to represent weather conditions and approximate

temperature influences. Descriptive statistics for this and all other determinants of child malaria

discussed in this section are reported in Table 1.

10

EMPIRICAL RESULTS – PROBIT REGRESSIONS

Child‟s malaria prevalence is recorded in the survey response as a binary variable that

takes values 0 or 1 (Mi). Thus, we implement a discrete choice model by formulating apriori that

Prob[Mi =1] = F(‟x), where F is any function of the index ‟x that satisfies the axioms of

probability, and x is a vector of individual, household, community and environmental factors

related to child malaria. This implies that

)x|1M(F*1)x|0M(F*0)x'(F)x|M(E AFiii

where F(‟x) is the conditional mean function that can be estimated as a probit regression model

using maximum likelihood estimation.

All the variables listed in Table 1 and discussed in the previous section are included in

estimating the probit model. We take the natural log of the forest cover and elevation variables to

reduce scale differences (which can cause convergence problems in maximum likelihood

estimation), improve linearity, and pull in outliers (Hamilton, 1992). All the standard errors (and

resulting probability value estimates) are corrected to account for the fact that many of the

observations belong to sub-groups or „clusters‟ with shared factors (i.e., children within the same

household) and may therefore violate the assumption of identical and independent regression

errors across observations. This approach adjusts the coefficients‟ standard errors and level of

significance, but does not affect the size and sign of the estimated coefficients.

Table 2 presents regression results for three child cohorts and the full sample of all those

under 16 years of age. Results are reported in terms of marginal effects and not regression

coefficients (which do not measure the impact of a unit change in the independent variables in a

non-linear model such as probit regressions). The overall models are statistically significant, i.e.,

we can clearly reject the joint hypothesis that all variables in the regression are uncorrelated with

malaria prevalence.

Our key variables of interest are primary and secondary forests. We find that the amount of

primary forests in a village is negatively associated with malaria incidence among children in the

two youngest cohorts, all else equal. We also find that the extent of secondary forest in a village is

11

positively associated with occurrence of malaria only in the youngest cohort. The lack of statistical

significance in the model of malaria in older children suggests possibly confounding effects of

various behavioral patterns that are difficult to fully characterize – older children may have

developed resistance, take precautionary actions, work in the fields and forests, and are otherwise

exposing and protecting themselves in many more complicated ways.

Specific child demographic factors appear to have weak correlations with malaria rates. In

the youngest cohort, males are more likely to experience malaria, relative to females. However, in

the two older cohorts, gender is not associated with disease at a high level of statistical

significance. Children born in malaria endemic areas may be less susceptible because they had

developed immunity; unfortunately we do not have data on where the migrating children came

from. A variable recording whether the child was born locally or not (86% are born locally) was

found to be statistically irrelevant.

Household wealth generally is negatively correlated with malaria. Poor housing quality is

positively correlated with malaria in the youngest cohort. As suggested by Barbieri et al. (2005) in

their study of forest communities in Brazil, housing materials such as bamboo and thatch could be

poor screens against mosquitoes.

We find that the caregiver‟s age is positively associated with malaria occurrence in the

youngest cohort. Older caregivers may be able to dedicate less time and effort to childcare.

Caregiver‟s level of education was also included in the probit models but was found to be

statistically insignificant in the full specification. Simple models including only the caregiver‟s

education showed a statistically weak negative relationship, possibly because formal education is

not correlated with health literacy concerning malaria prevention and or treatment. Healthier

mothers (proxied by greater levels of activity) have children with lower malaria rates in the 0-5

cohort.

Our public health indicator is negatively correlated with the malaria in all cohorts. This

could be because of greater levels of malaria prevention, rapid diagnosis and or better treatment of

malaria. Population density is not statistically significant. Finally, villages at higher elevation have

lower malaria, presumably because of climatic factors such as lower temperatures.

12

Robustness checks

Space constraints limit lengthier discussions of robustness checks that confirm the central

results of this paper. Thus, we briefly summarize them in the paragraphs below (the interested

reader is welcome to contact the authors for additional results). Potential endogeneity can be a

persistent concern in cross-sectional data sets. Previously we suggested that the aggregate and pre-

determined nature of the forest variables diminishes the threat from endogeneity. To further probe

and rule out this concern regarding our main variables, we apply an instrumental variable method,

which is an increasingly popular approach for dealing with endogeneity (Pattanayak and Yasuoka,

2008). The key to this strategy lies in identifying good instruments or exogenous data that are

correlated with the exposure (deforestation) but uncorrelated with the outcome (malaria), typically

a difficult task. In this case, we use regional environmental factors – rainfall, elevation, distance to

highways, and population density to instrument for the potentially endogenous forest cover

variables. Overall, the IV models confirm the original result: the marginal effects are larger and

statistically significant.

Moreover, the connections between the ecological exposure and human health are

consistent with related analyses. Additional regression analyses shows lower rates of respiratory

infections in children in villages with higher levels of primary (undisturbed) forests; the

probability values of the negative coefficients are 0.03 and 0.003 for the oldest and youngest

cohorts (using similar specification). There are two possible explanations for these findings. First,

households in villages with protected forests use less bio-mass fuels (firewood from forests) in

cooking and therefore have lower levels of indoor air pollution. Second, households living around

protected watersheds have more water, which can improve their personal hygiene and reduce

„water-washed‟ infections. Additionally, Pattanayak and Wendland (2007) find that diarrhea and

typhoid are lower in Ruteng watersheds with greater water availability. Collectively, these results

suggest that forest conservation by Ruteng Park may be providing the unintended health benefit in

surrounding communities.

13

DISCUSSION: THE UNINTENDED IMPACTS OF FOREST CONSERVATION

Rapid economic development in recent decades has resulted in dramatic rates of global

deforestation without a full accounting of the social costs. Unfortunately, the long list of claims

regarding the role of ecosystem degradation on diseases outcomes (see 2005 Millennium

Ecosystem Assessment) rests on a very short list of studies that consider social, cultural and

economic factors drivers that underpin both sets of factors – the ecosystem disruptions and

behaviors related to exposure, prevention and treatment of malaria. Omitting socio-economic

factors can lead to erroneous interpretations regarding the nature of the relationship between

ecological changes and disease. Drawing from a human ecology perspective, we specify and test

the functional relationship between child malaria prevalence and forest conditions in a quasi-

experimental setting of buffer zone villages around a protected area in Flores, Indonesia.

Multivariate regressions are used to test the conservation and health hypothesis, controlling

for several individual, family, community and environmental variables that could confound this

test. These models confirm that primary forests are negatively correlated with malaria in the

younger children (aged 0 – 10), ceteris paribus. They also show that the secondary (disturbed)

forests are positively associated with malaria in the youngest age cohort (0-5 years), all else equal.

Contrasted to primary forests, the secondary forests in this setting are basically disturbed forests

that have micro-habitats (e.g., temperatures, humidity and rainfall) that encourage the

proliferation, density, behavior, variety, viability, and maturation of mosquito populations.

Disturbed forests also signal increased exposure of human beings who enter these forests to collect

forest products and commute between villages, in contrast to the primary (undisturbed) forests that

are closed and difficult to penetrate. Collectively, these results confirm the reasons attributed to

vector and human ecology in the literature regarding malaria transmission (Pattanayak and

Yasuoka, 2008).

Although this study is an important contribution to a thin empirical literature on forest

malaria, consider two caveats. As with any empirical work, the data used in our study utilizes

approximations of how forests, mosquitoes, households, and children interact. Spatial activity

diaries, richer characterization of human interactions, and household level forest mapping would

undoubtedly enhance the precision of this type of model building and our inferences regarding

14

ecology and economics of child malaria. Furthermore, it is difficult, if not impossible, to establish

causality with a cross-sectional observational data set that does not permit us to fully examine the

coupled dynamics of transmission and prevention. Thus, our robustness checks provide useful

verification of the main conservation-health result.

Despite these limitations, this study is one of the few that has brought microeconomic data

to bear on the role of forest degradation on malaria, while controlling for variety of socio-

economic, cultural, and environmental factors that underlie household and community behaviors.

This type of analysis is important because complex and dynamic relationships between

deforestation, malaria and poverty make it difficult to determine whether health and forest policies

complement or conflict with each other‟s goals for improving the lives of people. We have

focused on forest conservation throughout our discussion because deforestation is the beginning of

an entire chain of activities that affect malaria risks; it can trigger human behavioral changes

through accompanying increases or decreases in wealth; it can lock communities into a vicious

cycle of poverty, illness and environmental degradation; and it is an integral part of the landscape

and therefore of donor agencies‟ and policy makers‟ focus (Pattanayak et al., 2006). Our results

indicate possibilities for designing complementary „natural insurance‟ policies for environmental

conservation and health promotion.

15

REFERENCES

Berman, P., Kendall, C., & Bhattacharyya, K. (1994.) The Household Production of Health:

Integrating Social Science Perspectives on Micro-Level Health Determinants. Social

Science and Medicine, 38(2), 205–215.

Barbieri, A., D. Sawyer, and B. Soares-Filho. 2005. Population and Land Use Effects on Malaria

Prevalence in the Southern Brazilian Amazon. Human Ecology 33(6) : 847-874.

Blomkvist, G. (1995). Forestry and Silviculture Specialist Report on Ruteng. Jakarta, Indonesia:

Directorate General of Forest Protection and Nature Conservation, Ministry of Forestry.

Campbell-Lendrum, D, Molyneux, D. et al (2005) „Ecosystems and vector-borne disease control‟,

in Ecosystems and Human Well-being: Policy Responses, Volume 3, Millennium

Ecosystem Assessment, pp353–372

Castilla, R., & Sawyer, D. (1993). Malaria rates and fate: a socio-economic study of malaria and

Brazil. Social Science and Medicine, 37(9), 1137–1145.

Castro, M.C., & Singer, B. (2001). Malaria foci and colonization processes on the Amazon

frontier: New evidence from a spatial analysis and GIS approach. Mimeo, Princeton

University, Office of Population Research.

Corvalan, C., Hales, S., McMichael, A. et al (2005a) „Ecosystems and human well-being: Human

health synthesis‟, Millennium Ecosystem Assessment

Corvalan, C., Hales, S., Woodward, A. et al (2005b) „Consequences and Options for Human

Health‟, in Ecosystems and Human Well-being: Policy Responses, Volume 3, Millennium

Ecosystem Assessment, pp467–486

16

Colfer, C.J.P., D. Sheil, D. Kaimowitz and M. Kishi. 2006. Forests and human health in the

tropics: some important connections. Unasylva 224 (57): 3-10.

Food and Agriculture Organization (FAO). (2003). Global Forest Resource Assessment 2000.

FAO Forestry Paper 140. Retrieved January 5, 2003, from

http://www.fao.org/forestry/index.jsp.

Hamilton, L. (1992). Regression with graphics: A second course in applied statistics (pp. 363).

California: Duxbury Press.

Indonesian Ministry of Forestry. (1995). Integrated conservation and management plan for

Ruteng nature recreation park: 1995-2020 (Volume I). Jakarta, Indonesia: Directorate

General of Forest Protection and Nature Conservation.

Jalan, J., & Ravallion, M. (2003). Does piped water reduce diarrhea for children in rural India?

Journal of Econometrics, 111(1), 153–173.

Kondrashin, A.V., & Orlov, V.S. (1989). Migration and Malaria. In M.W. Service (Ed.),

Demography and vector-borne diseases (pp. 353–365). Boca Raton, Florida: CRC Press,

Inc.

Kramer, R., Pattanayak, S.K., Sills, E., & Simanjuntak, S. (1997, May). The economics of Siberut

and Ruteng protected areas. Prepared for the Directorate General of Forest Protection and

Nature Conservation, Indonesia: Authors.

Lansang, M.A.D., Belizariob, V.Y., Bustosa, M.D.G., Saulc, A., & Aguirrea, A. (1997). Risk

factors for infection with malaria in a low endemic community in Bataan, the Philippines.

17

Acta Tropica, 63, 257–265.

Mayer, J. (2000). Geography, ecology, and emerging infectious diseases. Social Science and

Medicine 50 (7-8), 937–952.

MacCormack, C.P. (1984) „Human ecology and behavior in malaria control in tropical Africa‟,

Bulletin of the World Health Organization 62: 81–87 Suppl. S

Moeliono, M. (1995). Wood Use in the Manggarai. Ruteng, Flores, Indonesia: Intercooperation.

Mosley, WH and LC Chen (1984) An Analytical Framework for the Study of Child Survival in

Developing Countries. Population and Development Review, Supplement to Vol. 10: 25-

45.

Pattanayak, S.K., and K. J. Wendland. 2007. “Nature's Care: Diarrhea, Watershed Protection and

Biodiversity Conservation in Flores, Indonesia”. Biodiversity and Conservation. 16(10):

2801-2819.

Pattanayak, S.K., and J. Yasuoka. 2008. “Deforestation and Malaria: Revisiting the Human

Ecology Perspective”. Chapter 9 in C.J.P. Colfer (ed.), Forests, People and Health: A Global

Interdisciplinary Overview. Earthscan Publishers. 197-217.

Pattanayak, S.K., K. Dickinson, C. Corey, E.O. Sills, B.C. Murray, and R. Kramer (2006).

“Deforestation, Malaria, and Poverty: A Call for Transdisciplinary Research to Design

Cross-Sectoral Policies”. Sustainability: Science, Practice and Policy. 2(2): 1-12.

Patz, J.A., Graczyk, T.K., Geller, N., & Vittor, A.Y. (2000). Effects of environmental change on

emerging parasitic diseases. International Journal for Parasitology, 30(12-13), 1395–

1405.

18

Perz, S.G. (1997). The environment as a determinant of child mortality among migrants in frontier

areas of Para and Rondonia, Brazil, 1980. Population and Environment, 18(3), 301–324.

Petney, T.N. (2001). Environmental, cultural, and social changes and their influence on parasite

infections. International Journal for Parasitology, 31(9), 919–932.

Priyanto, A. (1996). Hydrology Specialist Report on Ruteng. Jakarta, Indonesia: Directorate

General of Forest Protection and Nature Conservation, Ministry of Forestry.

Prüss-Üstün, A. and C. Corvalan (2006). Preventing disease through healthy environments.

Towards an estimate of the environmental burden of disease. World Health Organization

(WHO): Geneva, Switzerland.

Walsh, J.F., Molyneux, D.H., & Birley, M.H. (1993). Deforestation: effects on vector-borne

disease. Parasitology, 106(Suppl.), S55–S75.

Vosti, S. (1990). Malaria among gold-miners in southern Para, Brazil: estimates of determinants

and individual costs. Social Science and Medicine, 30(10), 1097–1105.

Wessen, A.F. (1972) „Human ecology and malaria‟, American Journal of Tropical Medicine and

Hygiene 21(1): 658–662

Wilson, M.L. (2001). Ecology and Infectious Disease. In J. Aron, and J.A. Patz. (Eds.), Ecosystem

Change and Public Health (pp. 285–291). Baltimore: The Johns Hopkins University Press.

Wolman, R. (1995). Human and ecosystem health: Management despite some incompatibility.

Ecosystem Health, 1, 35–40.

WHO 2005. World health statistics 2005. Geneva, Switzerland.

19

TABLE 1. CHARACTERISTICS OF HOUSEHOLDS IN STUDY AREA

VARIABLE MEAN STD. DEV.

Household & individual data

Family size 5.6 1.7

Number of children 2.2 1.4

Male (children) 0.53 0.5

Percentage households native born 0.86 0.35

Child age 7.7 4.5

Caregiver age 34.6 10.2

Caregiver health (number of daily activities) 5.5 2.6

Caregiver education (avg. years) 4.6 3.2

Household wealth index 0.9 1.3

(number of consumer durables)

Housing quality (bad) 0.08 0.28

Village-level data

Village public health facility 0.1 0.07

Village population 2,504 1,379

Village area (hectares) 2,881 1,903

Environmental statistics

Primary forest cover (hectares) 280 263

Secondary forest cover (hectares) 116 93

Village elevation (meters) 926 264

20

TABLE 2. MARGINAL EFFECTS FROM A PROBIT REGRESSION MODEL OF CHILD MALARIA

Ch0-5 Ch5-10 Ch10-15

dy/dx P>|z| dy/dx P>|z| dy/dx P>|z|

Ln (Primary Forest) -0.06 0.057 -0.07 0.045 0.035 0.386

Ln (Secondary Forest) 0.11 0.001 0.04 0.202 0.012 0.747

Male 0.10 0.031 0.01 0.853 -0.028 0.613

Wealth index -0.01 0.651 -0.05 0.028 -0.027 0.266

Housing Quality (Bad) 0.28 0.027 0.02 0.796 0.009 0.929

Mother’s health stock -0.02 0.190 0.00 0.748 -0.021 0.083

Mother’s age 0.01 0.030 0.00 0.653 0.004 0.250

Public health facility -2.00 0.001 -1.15 0.036 -0.767 0.223

Population density 0.02 0.343 0.01 0.468 0.007 0.741

Ln (Elevation) -0.20 0.032 -0.18 0.090 -0.145 0.237

N 339 432 342

Pseudo R 0.123 0.037 0.042

Wald Chi2 (10) 36.89 0.000 15.86 0.104 13.33 0.206

21

Figure 1. Average

malaria rates in

children by village.

Figure 2. Primary

(undisturbed,

protected) forest by

village (desa)

Figure 3. Secondary

(disturbed, open)

forest by village

(desa)