In the case of many development interventions, impact depends on the successful targeting of individuals or households most in need. Such targeting relies on accurate information about each household’s level of wealth or poverty, which can be difficult and costly to collect. Household asset surveys are expensive to conduct, while self-reported census data can be highly inaccurate. As a lower-cost alternative, governments and development agencies have increasingly adopted a community-based approach to targeting, which relies on the assumption that a community is able to identify its poorest households.
However, a necessary condition for community targeting to work is that members of the community actually have learned who is rich and who is poor. Because individuals learn through word-of-mouth, it may be the case that the structure of a village social network influences how accurately its members can identify those in greatest need of aid.
This study used data collected in a 2008 study on the islands of Sumatra, Central Java, and South Sulawesi in Indonesia. These islands, the most populous in Indonesia, were selected to broadly represent the country’s geographic and ethnic diversity. Across the islands, researchers selected 640 “hamlets,” or neighborhoods, of approximately 54 households. Among these 640 hamlets, approximately thirty percent were urban and seventy percent were rural, in order to give a representative sample for the original study, which evaluated the effectiveness of community-based targeting.
Researchers analyzed data collected in 2008 on effectiveness of community-based targeting in Indonesia. The original study demonstrated evidence of the community-based targeting approach to generate accurate poverty rankings. Researchers in this study test the degree to which these results can be attributed to established network theory that explains how information gets diffused and aggregated within a community.
This study first analyzes correlations between community characteristics and how information is aggregated within that community. Variation within a village is used to see whether a more central household has better information and whether a household is better able to assess if household A or B is wealthier when the ranking household is closer to A and B in the network.
Researchers then develop a learning model of how households learn about each other’s wealth. Using within-village variation, the authors estimate the parameters of the model. The researchers then simulate data from the model to look at how variation in village network structure corresponds to the degree to which villages on average can correctly rank their members' wealth. They then check that the predictions from the simulations are consistent with the empirical data.
Finally, they apply the model to the original 2008 experiment, which compared the performance of hamlets randomly assigned to either community-based targeting or proxy-means testing (PMT.) The purpose of this is to test if those hamlets predicted by the model to be better at information exchange, based on social network characteristics, actually achieve more accurate targeting than PMT.
Results and Policy Implications
The results of this study substantiate common intuition on how information is aggregated within a community. The analysis confirms that households “better connected” to the community can more accurately rank their neighbors’ relative wealth or poverty. Similarly, community members were able to more accurately rank those households with whom they are more closely connected or regularly interact.
Further, the study demonstrates that how network structure within the community can accurately predict a community’s ability to rank, and therefore target, poverty within their community. For example, communities with higher average connections per household are likely to have a lower error in ranking wealth. Similarly, higher clustering (average percent of an individual’s contacts also connected to each other) and higher fraction of households interconnected to each other (through some chain of connection) are associated with more accurate targeting.
This has significant implications for the targeting of development interventions. The study demonstrates that with information about a community’s network characteristics, we may be able to predict where policies that rely on information diffusion, such as community-based targeting of the poor, will be most effective. However, while these measures of social networks are standard in network theory, they are not readily available for most poor communities. To leverage network statistics to improve targeting programs and reduce costs will require a low-cost method for measuring network characteristics.
Photo: Community leaders gather and talk in Central Kalimantan, Indonesia. Photo by Achmad Ibrahim for Center for International Forestry Research (CIFOR).