Also in the Article

Dietary diversity. We constructed our dietary diversity measure—IDDS—based on diet information of children younger than 5 years collected through the Demographic and Health Surveys (DHS) program administered by the U.S. Agency for International Development (USAID). The DHS program collected nationally representative data on population demography, health, and nutrition in over 90 countries. We compiled data from DHS across 27 developing countries in Africa, Central and South America, Southeast Asia, and Eastern Europe (Fig. 2) and collected in different years between 2000 and 2013.

To create the outcome variable, IDDS, we followed the food grouping recommended by the Food and Agriculture Organization of the United Nations (21) comprising 14 food groups: cereals; vitamin A–rich vegetables and tubers; white roots and tubers; dark green leafy vegetables; other vegetables; vitamin A–rich fruits; other fruits; organ meat; flesh meat; eggs; fish; legumes, nuts, and seeds; milk and milk products; and oils and fats. We reduced the food groups to 10. We combined other vegetables and other fruits into one group because many DHS do not disaggregate these two food groups. Similarly, we grouped organ meat, flesh meat, and fish into one group. We also removed oils and fats, as many DHS do not have information on this food group. For a given food group, we assigned a score of 1 if a child’s diet over the previous 24 hours included at least one food item belonging to that group and 0 otherwise. We then summed these values over all food groups to create the IDDS. We averaged the IDDS of children within a household, as the unit of analysis is a household.

We defined vitamin A– and iron-rich food following the guidelines of the Food and Agriculture Organization of the United Nations (21). Vitamin A–rich food items are those in the following food groups: vitamin A–rich vegetables and tubers, dark green leafy vegetables, and vitamin A–rich fruits. Iron-rich food items are those included in the combined organ meat, flesh meat, and fish food group.

Exposure to forest. Our forest cover data came from the global MODIS Vegetation Continuous Field products. These products are yearly (2000 to 2010) representations of the Earth’s surface in terms of percent tree cover at 250-m spatial resolution (36).

The communities surveyed by DHS (referred as “clusters” in DHS documentation) were georeferenced. We were thus able to integrate the DHS data with the forest cover and the spatial confounder data. We associated each household in the surveys with the forest cover of the year of the survey, except for the 2011–2013 surveys, for which we used the 2010 forest cover. We selected 3 km within forest edges as a criterion to define forest households because, on average, people in rural developing countries walk about 35 min to come to the closest forest to collect forest products (5). Using the rule of thumb that a person walks about 5 km in an hour, a 35-min walk is about 3-km distance. In addition, pollinators, which are one mechanism through which forests may affect diet (Fig. 1), generally forage within 3 km of nest sites (37, 38). The other criterion that forest covers at least 30% of the area within 5-km buffer of communities is based on studies suggesting that natural forest habitats need to cover at least 30% of a given area to maintain pollination services (39). We used 5-km buffer because the locations of communities in DHS were randomly displaced up to 5 km to protect anonymity of survey respondents. Moreover, because of this displacement, communities located between 3 and 8 km from forest edges could actually be within 3 km of forest edges and thus forest households. Given this uncertainty, we excluded households of communities located between 3 and 8 km from forest edges. We also removed urban communities from our analyses, as forests are mainly located in rural areas. Last, we excluded children under 12 months old, as their diet is dominated by breast milk, particularly in low- and lower middle–income countries (40). Our final data set comprised 43,011 households (11,338 forest households and 31,673 nonforest households).

Confounding characteristics. We identified both site and household level confounders. Forest cover is a site characteristic and, thus, relevant confounders are site characteristics affecting both forest cover and diet. However, households also self-selected themselves whether to migrate or stay in or out forested areas. Therefore, household characteristics that influence both where households choose to live and diet are also relevant confounders. Here, we presented the rationale for choosing each confounder. Fuller description and data sources of the confounding variables are in table S1.

Variables related to returns to agriculture. Confounding site characteristics include site variables related to returns to agriculture because of the major role of agriculture in both forest land conversion (41) and food availability worldwide.

Returns to agriculture are higher for lands with higher agricultural potential and more easily accessed. Therefore, we controlled for variables that capture land agriculture potential (agriculture suitability, slope, and elevation) and access (distance to a road or a city). In addition to its effects on returns to agriculture, access also influences the availability of marketed food.

Livestock production. Livestock production is one of the major drivers of deforestation, including through expansion of pasture lands (42). It contributes to human diet by providing animal-based food. Intensive livestock production can also reduce human dietary diversity by promoting monoculture of crops that can be used for both animal feed and human food (43). We used the variable ruminant livestock density to control for livestock production.

Development. Development is linked to forest cover, particularly deforestation, through complex pathways (43, 44). Development has also been linked to health-related outcomes, including nutrition. Low- and middle-income countries in sub-Saharan Africa and South Asia have higher prevalence of childhood undernutrition than any other regions on the globe, with 42% of children younger than 5 years in East Africa being stunted (2). To control for development, we used GDP of the areas where communities are located. The community GDP was converted to U.S. dollars using purchasing power parity factor to adjust for spatial and temporal variations across communities.

Population. Larger population triggers more deforestation by putting more pressure on forest resources (44). Increasing population size can render an area more attractive to market and thus provide access to marketed food (45). We controlled for population size.

Education. Level of education affects one’s decision to migrate (46). Parents’ education also influences children’s nutrition (20). Thus, we controlled for the levels of education of heads of households.

Proxies for income. Income or wealth is suggested to determine migration (47). Greater income is also associated with improved nutrition (48). The DHS data do not have information on household income. We controlled for a combination of variables that are among those widely used to proxy household income for targeting social programs, namely, education and age of heads of households, household size, and number of children younger than 5 years in a household (49).

Note: The content above has been extracted from a research article, so it may not display correctly.

Also in the Article

Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.

We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.