This document summarizes methods for predicting and mapping poverty levels using a combination of survey data, geospatial data, and statistical modeling techniques. Household survey data that includes measures of poverty, such as an asset-based Poverty Probability Index (PPI), can be combined with high-resolution geospatial data layers and spatial statistical models to generate predictive maps of poverty levels at fine spatial resolutions, such as 1km pixels. These predicted poverty maps can then be used for targeting interventions, monitoring changes over time, and integrating with other datasets like locations of financial service providers or mobile money usage patterns. The document outlines an example application of these methods to map predicted PPI scores and financial inclusion metrics in Kenya, Uganda, Tanzania,
3. Difficulties with mapping poverty
• Reliance upon census data
– Collected every 10+ years
– Difficult to obtain (especially matched to boundaries)
– Often only available many years after data collection
• Restricted to areal units
• Difficulties in comparing
– between countries
– across time
4. Household surveys and GPS
• National household surveys measuring
poverty indicators
• Asset or consumption based
• Rising popularity of rapid assessment
methods such as PPI
• Increasing use of GPS and availability of data
5. What about areas with no data?
• Extrapolate from locations where we have observations to
locations where we don’t
– E.g: 100 m population predictions and demographics for Africa,
Asia, South America
– E.g:Predicted poverty or literacy rates at 1 km within countries
• Account for uncertainty in those predictions
www.worldpop.org
8. Mapping poverty for GatesFigure 10. Scatterplot of observed versus predicted values for East Africa. The one-to-one line is shown in red.
(a) (b)
Figure 11. (a) Predicted map of the MPI headcount ratio for Kenya, Uganda and Tanzania. This displays the
mean value of the predictive posterior distribution at each 1x1km pixel. Major waterbodies and city names
are also overlaid for context. (b) The precision of the model output for East Africa as measured using the 95%
credible interval. Major waterbodies and city names are also overlaid for context.
(a) (b)
Figure 13. (a) Predicted map of the MPI headcount ratio for Pakistan. This displays the mean value of the pred
1x1km pixel. Major waterbodies and city names are also overlaid for context; (b) The precision of the model outp
95% credible interval. Major waterbodies and city names are also overlaid for co
(a) (b)
9. Example: FinScope surveys in Kenya
• 6449 household
surveys
• Nationally
representative
• Questionnaire on:
– Assets
– Financial literacy
– FSP usage (savings,
credit, mobile money,
investments)
– PPI
10. 10
What is the PPI?
• A poverty measurement tool for
organizations with a mission to
serve the poor
• 10 easy-to-answer questions and
a scoring system
• Provides the likelihood that the
survey respondent’s household is
living below the poverty line
• Country-specific; there are PPIs
for 60 countries
To download the PPI and learn more, visit:
www.progressoutofpoverty.org
Which households are below the poverty
line?
11. GIS data Layers
• Distance to Roads
• Population
• Night-time lights
• Aridity
17. PPI score = 32Population= 2314 +/- 340
Advantage of gridded
predictions
1 pixel = 1 km
1400 living in poverty
18. Disadvantages
• Predictions are only as good as the model and the
data used in the model
• Predictions are averages and may not represent fine-
scale variation
29. What would you do with these maps?
• Monitoring change
– Over time (eg FinScope surveys 2006/2009/2013)
– In response to interventions
• Targeting interventions/products
– Against local financial levels/market penetration
• Integrating user/customer data.
• Future datasets:
– Fine-resolution human settlement layers
30. Summary
• Survey data with GIS can provide predictions
in areas with little or no data
• Better GIS layers can help build better
predictive models
• Quality and design of surveys is crucial
• Rapid surveys will allow for greater coverage
of the population and greater sensitivity to
change
33. Direction for a New Model
A revision to the current model for funding, operations, and governance
is needed to sustain the PPI long-term and solidify the PPI as a global
industry standard.
The benefits of a new model include:
• Better integration with other organizations, and likely faster scale
• Broader input, transparency and greater responsiveness to users
• Long-term stability giving users confidence to invest in adopting the
tool
Grameen Foundation is now working with key PPI stakeholders to
design an Institutional / “Club” Model where a group of international
organizations buy into the PPI and a new administrative home would be
found.
Transition anticipated by March 31, 2016.
To download the PPI and learn more, visit:
www.progressoutofpoverty.org