ENVIRONMENTAL AND NUTRITION OUTCOMES - Food Security, Food Prices and Climate Variability - page 168

(percent forest cover and proximity to a protected area (PA)) and biodiversity change (decadal
forest cover loss or gain) using multivariate methods was assessed. In this study, the impact of
weather and seasonality was removed from the analysis by using a “dummy” variable based
on the NDVI for the year/month that the data were collected.
Demographic and Health Survey (DHS) information
Demographic and Health Surveys are the gold standard source of comparative quantitative
data on population, health and nutrition indicators across developing countries. They are
nationally and sub-nationally representative household surveys with large sample sizes that
provide detailed information on these topics by interviewing eligible respondents in each
selected household (women age 15-49 and men age 15-59). The data also include informa-
tion on household and other socio-economic characteristics. DHS data are collected from
probability samples selected using a stratified two-stage cluster design. DHS observations are
weighted to be representative at the national level, according to urban/rural residence, and at
the provincial level (departments, states). More than 300 DHS surveys have been imple-
mented in over 90 countries since the inception of the USAID-funded project in 1984.
Since the mid-1990s, the DHS project has collected geographic information in most sur-
veyed countries at the level of the cluster. The latitude and longitude of each household
cluster allows the connection of environmental information to the survey results. During
fieldwork activities to conduct the DHS surveys, the global positioning system (GPS) coord-
inates for the approximate center of the populated area surveyed (cluster centroid) are col-
lected using hand-held GPS units. During data processing, GPS coordinates are displaced to
ensure that respondent confidentiality is maintained. The displacement is randomly applied so
that rural points contain a minimum of zero and a maximum of five kilometers of positional
error. Urban points contain a minimum of zero and a maximum of two kilometers of error.
A further 1 percent of the rural sample points are offset a minimum of zero and a maximum
of ten kilometers. This random shift eliminates the possibility of calculating exact distances
from the cluster to other locations of interest, and requires that a buffer of some type is used
when linking the DHS and landscape/geophysical data (Brown et al ., 2013).
Malawi and ecosystem services
Malawi has very high rates of deforestation, child undernutrition and child mortality, as well
as a high level of dependence of the population on ecosystem services. Malawi has lost almost
600,000 hectares of primary forest between 1990 and 2005, with regional deforestation rates
as high as 3.4 percent per year (Berry et al ., 2009). Over 80 percent of Malawi's population is
rural and highly dependent on natural resources for food, fuel and maintenance of livelihoods
(United Nations, 2011). A total of 47 percent of children are stunted, and the under-ive
mortality rate is 112 deaths per 1,000 live births (DHS, 2011).
The analysis used a multivariate unweighted binomial logistic regression to examine the
correlation between variables of interest (forest cover or decadal change in forest cover) and
our selected outcome variables (severe stunting, dietary diversity, consumption of vitamin-A
rich foods or experience of diarrhea), while controlling for confounding factors for which
data were available: child's age, water source, toilet facility, mother's education, wealth quin-
tile, NDVI and whether or not the family has migrated into or out of the area. Weather and