PPI Construction
The development and update of every PPI is done by IPA’s PPI team and includes the input of a variety of stakeholders throughout the process. This page briefly describes the process for creating a PPI utilizing the new construction methodology. For more technical details, please refer to this paper.
This is done by first narrowing down the survey questions to a smaller sub-set of approximately 100 questions which meet the following criteria as far as possible.
- Inexpensive to collect, easy to answer quickly, and simple to verify.
Example: “Of what material is the roof of the residence made?” - Liable to change over time as poverty level changes.
Example: “Does every child between the age of 6 to 12 attend school?”
This sub-set of questions is then run through a statistical model which:
- Selects ten questions based on the frequency with which they are identified as being useful predictors of poverty, across several different geographic sub-samples of the survey data (where such data is available). This ensures that the results are more accurate when the PPI is applied in specific sub-national areas of a country.
- Estimates weights to each response associated with the ten questions while constraining their size to minimize the influence of unusual observations, so they can be accurately applied out-of-sample.
- Normalizes these weights so that the total score is between 0 and 100. These adjusted weights form the basis for the scorecards.
- Associates each score with a poverty likelihood (or probability), based on the relationship in the survey data. This relationship between the scores and the probability of poverty is captured by the look-up tables.
Each PPI scorecard is published with a Design Documentation Memo or User Guide. Read this document for your country’s PPI for important details on its construction, poverty lines, accuracy, benchmarking, and instructions for using it.
