Geospatial Impact Evaluations
Measure real impact
Geospatial Impact Evaluations measure intended and unintended impacts of development programs. Leveraging readily available data like satellite observations or household surveys, GIE methods establish a reliable counterfactual to measure impact - at a fraction of the time and cost of a "traditional" randomized control trial (RCT).
Like RCTs, GIEs can estimate the net effect of a specific program by comparing similar units where the only difference was an intervention, or treatment. Unlike RCTs, GIEs use precise geographic data to establish this counterfactual retroactively, eliminating the need to assign program participants into randomized treatment and control groups within the program design.
Learn more efficiently
GIEs can be completed in a fraction of the time and financial cost of an RCT by eliminating the need for customized data collection in treatment and control groups before, during and after the program.
See portfolio-wide insights
GIE methods are also flexible tools that can either be used to evaluate individual projects or project portfolios. Whereas RCTs are often implemented in narrowly bounded settings, GIEs can be used with data for an entire country (or even multiple countries), which makes it possible to draw conclusions about impacts and cost effectiveness that are broadly generalizable.
Ascertain long-term impact, even in inaccessible places
Additionally, GIEs can be implemented remotely, retrospectively, and affordably, opening up new opportunities to measure long-run programmatic impacts, which is especially useful to evaluators working in conflict and fragile state settings.
Learn more in our #CGDTalks Video
Geospatial impact evaluations (GIEs) rely on subnationally geo-referenced intervention, outcome, and covariate data and quasi-experimental methods of causal inference to measure the intended (or unintended) impacts of development programs.
GIEs seek to mimic the conditions of a randomized control trial (RCT) with observational data. RCTs are powerful because they create conditions under which one can reliably ascertain that individuals’ participation in a program was not correlated with their outcomes. GIEs create similar conditions, but without randomly assigning individuals to treatment and control groups.
For our in-depth exploration of GIEs, please see A Primer on Geospatial Impact Evaluations Methods, Tools, and Impact Evaluations.
Using counterfactuals to find impact
The key is making comparisons across individuals who are sufficiently similar to one another and experiencing changes that are otherwise similar. The best way to make such comparisons is to identify comparison individuals who are geographically close to the program participants, but unlikely to be affected by the program’s presence. Doing so requires geographically precise data on programmatic interventions and their intended (or unintended) outcomes. Such data are rapidly expanding in number, scope, periodicity, and availability.
Rather than using randomization to identify counterfactual cases, GIEs seek to achieve a similar result through one of three methods:
- Strategic subsampling of observational data to identify treatment and control cases that are nearly identical but for the presence or absence of the intervention (e.g. propensity score matching)
- Comparing the pre- and post-intervention change in the outcome of interest for a treatment group relative to a control group (e.g. difference-in-differences, fixed effects)
- Exploiting the discontinuity around a geographic cutoff that is “as-if random” (where the treated cases and control cases on either side of the cutoff are extremely similar across pretreatment covariates)
The effects of trade, aid, and investment on China's image in Latin America
Vera Z. Eichenauer, Andreas Fuchs, Lutz Brückner
Linking Local Infrastructure Development and Deforestation: Evidence from Satellite and Administrative Data
Christian Baehr, Ariel BenYishay, Bradley Parks
Childhood health and the changing distribution of foreign aid: Evidence from Nigeria's transition to lower-middle-income status
Carrie B. Dolan, McKinley Saunders, Ariel BenYishay
A convolutional neural network approach to predict non‐permissive environments from moderate‐resolution imagery
Seth Goodman, Ariel BenYishay, Daniel Runfola
Impacts of a large-scale titling initiative on deforestation in the Brazilian Amazon
Benedict Probst, Ariel BenYishay, Andreas Kontoleon, Tiago N. P. dos Reis