智慧数据 | 高价值论文 | 使用高分辨率贫困地图对社会援助进行地理微观定位 | 地球大数据 | 第11期【中国科讯】

智慧数据是面向具体应用场景，具有“数据要素质量较高、数据维度丰富多样、多源融合的信息网络以及认知理解的语义知识等特征的大数据资源”。

使用高分辨率贫困地图对社会援助进行地理微观定位
Geographic microtargeting of social assistance with high-resolution poverty maps 
Sustainable Isabella S. Smythe and Joshua E. Blumenstock （Columbia University, New York, NY ）

PANS | 2022-08-01| Vol 119,  32(2022)

Hundreds of millions of poor families receive some form of targeted social assistance. Many of these antipoverty programs involve some degree of geographic targeting, where aid is prioritized to the poorest regions of the country. However, policy makers in many low-resource settings lack the disaggregated poverty data required to make effective geographic targeting decisions. Using several independent datasets from Nigeria, this paper shows that high-resolution poverty maps, constructed by applying machine learning algorithms to satellite imagery and other nontraditional geospatial data, can improve the targeting of government cash transfers to poor families. Specifically, we find that geographic targeting relying on machine learning–based poverty maps can reduce errors of exclusion and inclusion relative to geographic targeting based on recent nationally representative survey data. This result holds for antipoverty programs that target both the poor and the extreme poor and for initiatives of varying sizes. We also find no evidence that machine learning–based maps increase targeting disparities by demographic groups, such as gender or religion. Based in part on these findings, the Government of Nigeria used this approach to geographically target emergency cash transfers in response to the COVID-19 pandemic.

URL：https://www.pnas.org/doi/10.1073/pnas.2120025119

Cite：S. Smythe, J. E. Blumenstock, Geographic microtargeting of social assistance with high-resolution poverty maps. Proceedings of the National Academy of Sciences 119, e2120025119 (2022).

Fig 1 Targeting performance of policies at different administrative units. ROC curves show the performance of geographic targeting policies designed at the state (Admin-1), LGA (Admin-2), and ward (Admin-3) levels, where all households in a targeted administrative unit receive full benefits and households in untargeted units receive no benefits. The targeting of an administrative unit is determined based on the average wealth of the unit as calculated from NLSS data. True and false positive rates are calculated based on the portion of true poor households that are targeted, where true poverty status is determined based on the NLSS.

Fig 2 Coverage and correlations of ML and benchmark poverty maps to NLSS-estimated ground truth poverty maps.

Fig 3 Ward-level targeting performance. Curves show performance of geographic targeting for programs of different sizes for three different approaches to targeting: an optimal approach based on the NLSS evaluation data; the ML-based approach based on high-resolution poverty maps; and a survey benchmark based on DHS data, imputing a representative portion of missing values. A shows ROC curves based on whether the NLSS households in targeted wards are poor (Left) or extreme poor (Right). B shows the fraction of program benefits going to the poor (Left) and extreme poor (Right) as the size of the antipoverty program varies.

Fig 4 Comparison of targeting fairness for selected demographic groups (assigned based on the head of household). Under perfect individual targeting, the fraction of transfers going to members of a demographic group would be equal to the fraction of total poor households belonging to that demographic group. A shows the percentage difference between the number of households in each demographic group expected to receive transfers and the number that actually receive transfers, when 10% of the population is targeted. Error bars show bootstrapped 95% CIs. B shows how the fraction of transfers going to sample subgroups varies as a function of program size and as a fraction of total population. Dashed horizontal line indicates the proportion of poor households whose head of household belongs to that demographic group. Results pictured are for ward-level targeting.

推荐人：李   琼

审    核：赵昆华

中国科学院文献情报中心

数据资源部

中国科学院文献情报中心立足中国科学院、面向全国，主要为自然科学、前沿交叉科学和高技术领域的科技自主创新提供文献信息保障、战略情报研究服务、公共信息服务平台支撑和科学交流与传播服务，同时通过国家科技文献平台和开展共建共享为国家创新体系其他领域的科研机构提供信息服务。

点击下方微信名片> 添加关注！

转载授权、合作、投稿事宜请在本文留言！

近期文章推荐

↓↓↓点击标题即可查看