Cleaning Up the Murky World of Geocoded Data
While geocoded data represent a tremendous boon for aid transparency and aid effectiveness research, there are limitations to what we can learn from them.
In a recent Brookings article, entitled “How Effective is the World Bank at Targeting Sub-National Poverty in Africa? A Foray into the Murky World of Geocoded Data,” Laurence Chandy, Natasha Ledlie and Veronika Penciakova, discuss the use of geocoded data to target aid at the sub-national level. Highlighting the World Bank’s Mapping for Results and IFPRI's (International Food Policy Institute) Harvest Choice data collection initiatives, the article explores the allocative efficiency of aid with respect to poverty at the first order administrative (i.e., province, state or governorate) level.
AidData staff, students, and faculty also spend a lot of time collecting high-resolution subnational aid information to assess the targeting efficiency of aid, and along the way we have learned that is critical to map aid from a variety of sources, rather than a single donor, to fully understand aid distribution in any given country.
Using a wealth of aid data from multiple donors in Malawi, we can build upon the analysis in the Brookings piece by examining how donors jointly distribute aid within a single country. The following map shows the locations of aid activities for all bilateral and multilateral donors in Malawi down to the second administrative level (i.e., district).
In order to preliminarily assess the efficiency of aid targeting, we assume an ideal scenario where total aid is distributed sub-nationally according to the proportion of poor people residing in each district. For example, a district with 5% of Malawi’s poor should receive 5% of Malawi’s aid receipts. We use poverty headcounts and population data from Malawi’s Third Integrated Household Survey (IHS3) 2010-2011, to calculate these proportions and compare them to geocoded aid disbursements collected by CCAPS and AidData.
Using this benchmark, we can identify how far the actual allocation of aid deviates from this “ideal”. If the difference for a given district is zero, then one might argue it is receiving resources appropriate to their portion of Malawi’s overall poverty burden. A significant positive or negative difference might reflect that a district is getting more or less than their fair share. The actual results show a relatively well-targeted, yet imperfect distribution of national aid resources.
Poverty rates provide a poor proxy for targeting in Malawi because a large portion of the poor live in populated areas with lower poverty rates. This touches on Chandy’s dilemma – should we target aid to areas with higher numbers of poor people, higher proportions of poor people, or both? Are we more concerned with impacting as many lives as possible or reducing pockets of highly concentrated poverty?
While these geocoded Malawi data represent a tremendous boon for aid transparency and aid effectiveness research, there are limitations to what we can learn from it. Donor project documents do not disaggregate total project funding amounts among project activity locations. Absent better reporting from donors, when project activities occurred in multiple districts, we assumed an equal distribution of resources across all locations. Yet, this is an imperfect description of where resources actually hit the ground and doesn’t convey differential aid impact per dollar spent across sectors and environmental contexts.
We agree with Chandy that targeting should not be equated with proximity to the poor. Such a metric does not take into account other relevant contextual data -- on disease rates, past performance of development projects, the quality of local governance, climate change vulnerability, etc. -- that should presumably also inform aid allocation decisions. Geocoded project information supports more nuanced analysis, but merely generating more data is insufficient. In order for donors to be held accountable, the development community must respond to the increasingly availability of geocoded data by coalescing around a set of robust methodologies for assessing the quality of subnational aid targeting efforts.
This post was written by Michael G. Findley, Josiah Marineau, Reid Porter, Jeanette Cunningham Rottas, and Kelly Steffen. Michael G. Findley is an Assistant Professor of Government at the University of Texas-Austin Josiah Marineau, Reid Porter, Jeanette Cunningham Rottas, and Kelly Steffen are Research Fellows at UT-Austin's Innovations for Peace and Development (IPD) research team.
The views expressed here are those of the authors alone, and do not necessarily reflect the views of the institutions to which the authors belong.