The Effect of Malaria on Settlement and Land Use:
Empirical Evidence from the Brazilian Amazon
Shufang Zhang, Marcia Caldas de Castro, and David Canning
Harvard School of Public HealthMay, 2010
Malaria and Development• Malaria burden:
• Potentially large effects on income levels• Mechanisms
• Health care costs, prevention and treatment• Labor productivity • Childhood exposure and cognitive development• Educational Attainment and adult productivity• Population pressure
• Avoidance
Effect of Malaria on Land Use
• Land use • Avoidance can reduce the burden of
malaria but had costs• Settlement, choice of crop
• Evidence of Effect of Malaria on Crop Choice• Paraguay, Conly (1975)• Kenya, Wang'ombe and Mwabu (1993)• Vietnam, Laxminarayan (2004)
Machadinho Resettlement Project
• Resettlement project in Brazil in the 1980s• Land plots allocated to settlersOur Questions • What is the impact of malaria on
settlement? • What is the impact of malaria on land use?
Machadinho Resettlement Project
Machadinho
Site Photo (source: M. Castro)
Legend
Plots
Urban
^ South_Entry
Suburban Center
Machado
River
Road and Type
Newtype
Collector Road
Access Road
Penetration Road
Forest Reserve
Data Source:Center for Regional Development and Planning Federal Univeristy of Minas Gerais, Brazil
Machadinho Resettlement Area
Settlement Infrastructure• Infrastructure constructed between 1982 and
1984, 1200 km2 total area.• Plots were laid out based on topographic
features—steam at the back, with front access to a road.
• Urban and sub urban areas designated• Roads and laid out in advance (3 classes).
Placed to avoid flooding in rainy season.• Forest reserves maintained - right of use to
indigenous rubber tappers.
Allocation of plots • Designated for landless small farmers• 1740 plots, each about 45 hectares Settlement
oversubscribed.• Settlers randomly allocated to plots July/
August 1984. • Settler gets right to use plot –lapses if plot not
farmed.• In theory no trade in plots allowed – in
practice some swaps and trades carried out.
House built in 1985, all made of palm thatch.
House built in 1985. The roof is made of plastic, and thesealing of the whole house is precarious.
House built in 1985. There is no door closing the house.
Source: M. Castro
Settlers in Machadinho in 1985
Main urban area of Machadinho in 1985
Main urban area of Machadinho in 2001
Farming
• Slash and Burn– Cut down vegetation, wait to dry, and burn
• Typically poor soil quality – Burn fertilizes soil– Soil quality declines with use
• Main crops– rubber, coffee, cocoa
Malaria Ecology• Malaria
• Both Plasmodium falciparum and Plasmodium vivax
• River is preferred mosquito habitat• Anopheles Darlingi • Forest fringe for sun/shade• Stagnant water when river falls• Mosquito range up to 7km
Malaria and People
• New settlers lacked natural immunity and knowledge and were very susceptible to malaria
• Indigenous rubber trappers were asymptomatic but were a natural reservoir for malaria parasites
• Frontier pattern of malaria • Clearing forest increases malaria initially – more
fringe - full clearance removes fringe. • Malaria rate peaked in 1986, de Castro et al (2006)
Machadinho Land Use Literature: mainly case studies
• Malaria can lead to settlers abandoning a plot, Martine (1990).
• Many settlers live in town to avoid malaria and for job opportunities, Sawyer (1993).
• High levels of malaria and poor soil quality led to many failures among farmer-settlers in the long run and the emergence of large cattle ranges. de Castro and Singer (2005).
Data• Household Surveys
– For plots occupied and lived on – Malaria, self reported symptoms, episodes per
month/person– Demographic and socioeconomic indicators– Data for 1986 and 1987 are used in the study
• Map of Settlement Area– Plot geography, distances
• Satellite images – Area of plot cleared, crop cover, water cover.
Satellite Images• Remote sensing data on land use
– Acreage and percentage of plot deforested – Acreage and percentage of plot cropped– Data available for year 1985 and 1986
• Remote Sensing Data
Satellite images: M. Castro (PNAS, 2006)
Variables • Measured variables with ArcGIS
– Distance to river, Distance to nearest urban or suburban center, Distance to south entry, Distance to nearest stream, Nearest road type, Adjacent to river or forest reserve, Plot area
• Survey data– Malaria rate, Household characteristics: education, age
structure, number of people live on the plot, number of planters, number of chainsaws.
• Satellite images– Water cover, area cleared, area cropped
Summary Statistics: PlotsTable 1: Summary Statistics
Plot Characteristics Obs MeanStd. Dev. Min Max
Distance to the Machadinho river (=7km if over 7km) 1734 4.75 2.36 0.2 7
Plot within 200 meters of the Machadinho river 1734 0.08 0.27 0 1
Plot within 200 meters of the forest reserve 1734 0.25 0.44 0 1
Distance to the nearest urban or suburban centers (km) 1734 4.38 2.04 0.3 11.6
Nearest road is major road (1=yes; 0=no) 1734 0.13 0.34 0 1
Nearest road is sub-major road (1=yes; 0=no) 1734 0.35 0.48 0 1
Distance to the nearest stream (km) 1734 0.46 0.26 0 1.58
Soil quality index 1670 0.16 0.03 0.07 0.44
Plot area (hectare) 1734 45.41 10.36 16.1 124.1
Distance to the south entry (km) 1734 15.20 7.80 0.44 29.48
Summary Statistics: householdsTable 1: Summary Statistics
Household Characteristics by Plot, 1986 and 1987 Obs Mean St.dev. Min Max
Plot is occupied (1=yes; 0=no) 3468 0.37 0.48 0 1
Self-reported malaria rate 1286 0.28 0.26 0 1
Household head's education 1281 1.66 1.91 0 7
Household head wife's education 1083 1.59 1.79 0 7
Proportion of people on the plot younger than 5 1279 0.13 0.17 0 1
Proportion of people on the plot between 5 and 15 1279 0.26 0.23 0 1
Proportion of people on the plot over 65 on a plot 1279 0.01 0.09 0 1
Total number of people living on the plot 1299 5.17 3.01 1 18
Number of chainsaws 1299 0.53 0.55 0 3
Number of planters 1299 0.50 0.54 0 2
Summary Statistics: Land Use
Land Use by Plot, 1985 and 1986 obs meanSt.
dev. min max
Water area (fixed) 1734 0.05 0.05 0 0.33
Proportion of land deforested 3468 0.10 0.09 0 0.69
Proportion of land cropped 3468 0.01 0.02 0 0.27
Malaria Prevalence 1986
Malaria Prevalence 1987
Percentage deforested in 1985 Percentage cropped in 1985
Clearance and Cropping 1985
Percentage
0-5%
5-10%
10-20%
20-30%
30-100%
Percentage deforested in 1986 Percentage cropped in 1986
Clearance and Cropping 1986
Percentage
0-5%
5-10%
10-20%
20-30%
30-100%
The Impact of Malaria on Land Occupancy
Simultaneous EquationStructural Model
1, 0it itY if y Plot Occupied if
m: malaria ratey:latent variable for occupancyx: plot variablesr: is distance to river
h:household variabless: distance to south entryd: time dummiesi: plot/household* Identification
*x i r i hit it tt t itim hx d vr
( | , , , ,* )it it i i it tm it ix i s i h t t ity E m x r h d v hx s d u
Removing Household Characteristics
• Household characteristics go into the error term – valid because of randomization of plot allocation
Note that error also includes difference in conditional expectation of malaria with and without household characteristics but this isuncorrelated with plot characteristics
'x i r ii t it tt t im vx r d
( | , , , ) 'it it i i t ix i s i t tm t ity E m x r d vx s d u
Correlation Matrix Between Plot Fixed Characteristics and Household Characteristics
Distance to river
Adjacency to river
Adjacency to forest
reserve
Distance to nearest
urban/suburban center
Nearest road is
major road
Nearest road is sub major
road
Distance to the nearest
stream
Soil quality index
Plot areaPlot water
area percentage
Distance to south entry
Household head's education 0.0086 0.0371 0.0113 0.0632 0.007 -0.0142 -0.0311 0.0124 0.0523 -0.0468 -0.0335
Household head wife's education
0.0281 0.0245 -0.0209 0.0474 -0.0156 -0.0232 -0.0347 -0.0108 0.0475 -0.044 0.0006
Proportion of people on the plot younger than 5
0.0999 -0.0449 0.0283 0.0108 0.0025 0.0424 -0.011 -0.0385 0.0287 0.0183 0.0137
Proportion of people on the plot between 5 and 15
0.0093 0.0269 0.0066 -0.0108 -0.0224 0.0375 0.0605 -0.0513 -0.0315 0.0017 -0.0088
Proportion of people on the plot over 65 on a plot
-0.0858 0.1100* 0.0195 -0.0141 -0.0176 -0.0176 -0.0197 -0.0331 0.0074 0.0402 0.0462
Total number of people living on the plot
0.0306 0.0302 0.0178 -0.0741 -0.0429 0.1184* 0.0988 -0.0498 -0.0269 -0.0331 -0.0277
Number of chainsaws a plot has
0.0225 0.015 -0.0263 -0.0185 0.0033 0.0597 0.0296 -0.0033 0.0419 -0.0268 0.0323
Number of planters a plot has
-0.0435 0.0382 -0.0028 -0.0432 0.0442 0.0278 0.0625 0.026 -0.0439 -0.0925 -0.0086
Reduced Form• Substituting for expected malaria
x i i itrit t tx r dm
() ()( )x m x i i s imi t m t t ir tt x ry s d u
( )m rm
r
Plot Occupied 1, 0it itY if y
• Standard Heckman selection model
Identification• We need an “instrument” for malaria in the
settlement equation. Identifying assumption is that distance to river (up to 7 km) is correlated with malaria exposure but does not directly affect settlement. We correct for being adjacent to the river.
• We need an “instrument” that affects settlement but not malaria. We use distance to the south entry and connection with the outside world.
Legend
Plots
Urban
^ South_Entry
Suburban Center
Machado
River
Road and Type
Newtype
Collector Road
Access Road
Penetration Road
Forest Reserve
Data Source:Center for Regional Development and Planning Federal Univeristy of Minas Gerais, Brazil
Machadinho Resettlement Area
Distance to River as
IV for malaria
South Entry
Heckman Selection ModelDependent variable Plot
Occupancy Malaria
Rate
Independent variables (2) (1)
Distance to river 0.0156**(0.00501)
-0.0431***(0.00429)
Nearest road is collector (best) road
0.157***(0.0337)
-0.0973***(0.0211)
Distance to the south entry (km)
-0.00462***(0.00128)
Year 87 0.114***(0.0134)
-0.0598***(0.0133)
rho -0.105(0.064)
Wald test of independent equations
2.71
No of observations 3333
Malaria Rate: Reported vs Predicted 0
24
68
Den
sity
0 .2 .4 .6 .8 1self reported malaria rate at plot level
Self-reported malaria rate
02
46
8D
ensi
ty0 .2 .4 .6 .8 1
E(mrate|Zg>0)
Predicted malaria rate conditional on occupancy
Malaria: reported vs predicted
Obs. mean S.D Min Max
Self-reported malaria rate for occupied plots
1286 0.282 0.261 0 1
Predicted malaria conditional on occupancy for occupied plots
1246 0.281 0.099 0.082 0.517
Predicted malaria unconditional on occupancy for all plots
3340 0.326 0.107 0.096 0.582
Self-reported malaria rate in 1986 Predicted malaria rate in 1986
Malaria Rate
0-10%
11-25%
25-100%
Self-reported vs. Predicted Malaria Rate: 1986
Self-reported malaria rate in 1987 Predicted malaria rate in 1987
Malaria Rate
0-10%
11-25%
25-100%
Self-reported vs. Predicted Malaria Rate: 1987
Result: The Impact of Malaria on Land Occupancy
• Point estimates of the effect of malaria on occupancy (t stat 2.98)
• Effect on probability of land occupancy of going from no malaria to 0.326 malaria rate (area average) is 0.12
• No malaria would have raised settlement fraction of plots from 0.37 to 0.49
( ) 0.0150.36
0.043m r
mr
Malaria and Land Use
• Similar structural model• Simpler because land use observed for each
plot – still problem of finding expected malaria
• Tobit Model – Many plots have zero clearance/cropping– Left censored data
Dependent variable Percentage Deforested Percentage Cropped
Independent variables (3) (4)
Distance to Machadinho river -0.00140(0.000716)
-0.000517**(0.000166)
Adjacent to river -0.0129*(0.00567)
-0.00316**(0.00121)
Adjacent to forest reserve -0.0114**(0.00361)
-0.00283***(0.000725)
Nearest road is collector (best) road 0.0524***(0.00583)
0.00687***(0.00118)
Nearest road is access (2nd best) road 0.00892**(0.00341)
0.00129(0.000711)
Distance to nearest stream (km) 0.0143*(0.00665)
0.00345**(0.00133)
plot area (hectare) -0.000645***(0.000143)
-0.0000899**(0.0000303)
Distance to the south entry (km) -0.000696***(0.000200)
-0.000313***(0.0000404)
Year 86 0.0659***(0.00153)
0.0187***(0.000554)
No of observations 3340 3340
Table 4 The Impact of Malaria on Land Use: Tobit Model
The Impact of Malariaon Land Use
• No effect of the river on clearance• Land close to the river is more likely to be
cropped • Plot occupancy uncorrelated with clearance• Occupied plots less likely to be cropped• Plots adjacent to major road more likely to be
cleared and cropped
Land Use by Occupancy 1986
Obs. mean S.D Min Max
Proportionplot cleared
Plot occupied 825
0.135 0.093 0 0.620
Plot not occupied 1190 0.124
0.092 0.001 0.092
Proportion plot cropped
Plot occupied 825 0.017
0.023 0 0.235
Plot not occupied 1190 0.025
0.029 0 0.269
Malaria and Land Use
• Malaria deters settlement and living on a plot• However, people may live in town and clear a
plot in the same way as an occupied plot• Cropping is more prevalent in high malaria /
non-occupied plots– Less malaria among farmers– More access to work in town– More money for capital and seeds
Conclusions
• Malaria deters settlers from living on plots • Land use, clearance and cropping, is not
deterred – commuting to work on the plot is possible
• Pattern may be particular to this settlement area in Brazil– good roads and transport
• Occupancy different from land use