FAQs about the PPI’s New Construction Methodology
In 2017, IPA updated the methodology behind the development of the PPI. This page addresses questions that you may have related to this change. Read the full announcement about the PPI's new and improved construction methodology on our blog.
1. Will each country’s PPI be created with the new methodology going forward?
Yes. All PPIs released after July 2017 are created using the new methodology.
2. Why has the PPI’s construction methodology been changed?
We have made enhancements to the current methodology used to develop PPIs to achieve three objectives:
-
Greater model stability — Several PPI users work in specific regions, states, or provinces within a country. Their beneficiary groups may not be nationally representative. We have therefore redesigned the PPI model to select indicators that are strongly correlated with poverty both nationally and within different regions (for which data is available separately in the national survey). This ensures that the new PPI’s results are more stable and robust when applied to specific regions. We will also be providing estimates of the PPI’s accuracy when applied separately to the sub-national areas for which inference is supported by the survey design.
-
Reduced subjectivity — The selection of the broader candidate set of indicators from the national survey still involves judgement. However, the ten indicators that are most strongly correlated to poverty are then selected from this larger group by a fully automated algorithm. This reduces the possibility of errors due to subjectivity, and also sets us up to scale the PPI more efficiently and cost-effectively.
-
Institutionalization of the scorecard development process — In order to achieve long-term sustainability of the tool, it was important to ‘institutionalize’ the scorecard development process at IPA. The new methodology / code can be run in-house and by multiple team members, reducing risk and costs over time.
3. Has the new methodology been reviewed?
Yes. We have circulated a methodology paper and actively solicited and received feedback on it from experienced technical practitioners with doctorates in econometrics, computer engineering and statistics. In addition, the code used to implement the new methodology has been carefully reviewed by a statistical learning expert.
In December 2017, our paper on the new methodology, titled Household poverty classification in data-scarce environments: a machine learning approach, won the "Best Paper Award" at the NIPS (Neural Information and Processing Systems) 2017 Conference during its "Machine Learning for the Developing World" workshop.
4. How does the PPI’s transition to a new methodology impact my current data collection process?
Unlike in the earlier version of the PPI, each response option for the same indicator can now have different points depending on the poverty line used. Therefore, once you transition to the new PPI —
-
If you are currently using pen and paper to administer the questions and manually computing PPI scores in the field, you should ensure that your field staff carries the version of the PPI scorecard with points allotted for the poverty line(s) of interest to you. The “Scorecards+Look-up Tables” document on the country page will clearly indicate points allocated to each indicator for each poverty line.
-
If you are currently using automated data collection software to compute your PPI scores for you, ensure that it is updated so that points are allocated to each indicator correctly depending on the poverty line under consideration.
Please remember that this change will only affect the countries for which the PPI is being updated going forward, and does not impact existing PPIs. If you have any questions about how this change impacts your specific data collection process, please reach out to the PPI Help Desk.
5. Why do points allocated to the same indicator vary by poverty line?
An individual PPI indicator’s ability to distinguish between poor households (with consumption expenditure below the poverty line) and non-poor households (with consumption expenditure above the poverty line) may not be the same across all poverty lines. For example: Let us assume that an iron is a relatively “basic” asset owned by most households in a country. Ownership of an iron is a useful predictor of poverty at lower poverty lines, because information on this variable helps to more accurately distinguish poor from non-poor households at these lower poverty lines. However, its predictive power may not be as strong when you consider the “higher” poverty lines, as some households both above and below these lines may own this basic asset. Under the new PPI, points allocated to an indicator (which reflect its relative importance to the final PPI score), will vary depending upon which poverty line is being used. This enhances the predictive power of the PPI. In the example above, under the new PPI construction methodology, points allocated for ownership of an iron may be higher for a lower poverty line, and lower for a higher poverty line.
6. How does this transition impact my current data analysis process?
The old PPI allocated a single poverty likelihood value for a bucket of 5 PPI scores. For example, all PPI scores in the range 10-14 had the same poverty likelihood value associated with them. With the new PPI, every individual PPI score is associated with a separate poverty likelihood value. If you are using an automated module to compute poverty likelihoods, be sure to adjust it such that it is able to associate individual PPI scores with poverty likelihood values.
7. Why has each PPI score been associated with a poverty likelihood value?
Associating each individual PPI score with a unique poverty likelihood value allows you to report changes in poverty likelihoods for smaller changes in PPI scores.
8. Which poverty lines will be supported by the new PPIs?
The new PPI will support at a minimum 13 poverty lines for all countries. Users can choose the poverty line(s) that is/are most relevant to their population and purposes.
-
Three national poverty lines (The 100% national poverty line (NPL) which is determined by the Government, 150% of NPL, and 200% of NPL)
-
Three 2005 PPP international poverty lines ($1.25, $2.50, and $5.00/day)
-
Three 2011 PPP international poverty lines ($1.90, $3.20, and $5.50/day). These lines are consistent with the World Bank’s expanded group of international poverty lines. Read this blog post for more information. We may add $1.00/day for countries with higher poverty in order to provide more granularity within extreme poverty.
-
Four poverty lines corresponding to the 20th, 40th, 60th, and 80th percentiles of the population.
Depending upon the country, additional poverty lines, such as the food poverty line and higher poverty lines, may be supported.
Where a new PPI is comparable with an earlier version, poverty lines supported by the earlier version will be added.
9. How has IPA eased this transition for me?
Each new PPI now comes with a customized, easy-to-use, Excel-based data analysis tool which automatically computes poverty likelihoods and rates from your survey data. This means that you can simply collect responses to the 10 PPI questions as before, enter these responses into the tool, and it will automatically compute your poverty results. This data analysis tool:
- Comes with a built-in sample size calculator that allows you to determine the required sample size based on the desired confidence level and confidence interval.
- Accepts responses to the 10 PPI questions provided by each household as input, allocates points to each response, and computes a corresponding PPI score for each poverty line under consideration.
- Converts each PPI score to a corresponding poverty likelihood, or probability of the household being below each poverty line to which the PPI is calibrated.
- Averages these poverty likelihoods to provide you with the poverty rate of these households or share/proportion of households below each poverty line.
- Compares poverty rate of these households with official national and regional poverty rates for the country.
- Summarizes poverty results in an easy-to-read tabular format.
- Plots these results into easy-to-read charts, allowing you to quickly visualize your poverty data.
10. What will happen to existing PPIs which have been created using the old methodology?
Existing PPIs which have been created before July 2017 using the old methodology continue to be valid and will remain available via our website and supported by our Help Desk.
Any updates to these PPIs, however, will be made using the new PPI construction methodology. Please refer to the PPI Development Pipeline on our website to see which PPIs are due to be updated.
11. What is the difference between the Simple Poverty Scorecard and the PPI?
From 2005 to 2016, Mark Schreiner was the primary developer of the PPI scorecards, which he referred to as Simple Poverty Scorecards. In 2017, after transitioning from Grameen Foundation to IPA and in order to continually improve the tool and achieve long-term sustainability, the PPI team hired its own technical development staff and a new method was developed for creating PPI scorecards. Thus, up until mid-2017, the Simple Poverty Scorecard and the PPI were the same, but any PPIs released AFTER July 2017 are not the same as Simple Poverty Scorecards produced by Mark Schreiner. The PPI team cannot support any Simple Poverty Scorecards that are not listed or linked to on the PPI website as we were not involved in their development.
12. Will I need to re-register on the PPI website to obtain access to the new PPIs?
No. You can still log into our website using your existing credentials. The new PPIs and supporting documentation will all be available on their respective country pages for download as before.
13. Can I compute change in poverty rates over time with a baseline from a PPI created using the old methodology and an endline from a PPI created using the new methodology?
Theoretically, the answer is yes. However, the PPI team will need to test this for every PPI. As a country’s PPI is updated, we will test the accuracy of estimates of changes in poverty rates over time using a baseline from the earlier version of the PPI, and an endline from the new, updated PPI. We will report these accuracy figures in each PPI’s User Guide and offer our advice and recommendations on whether you can combine poverty estimates provided by the old and new PPIs.
14. Who can I contact if I have any questions related to the new PPI?
Please contact us at ppi@poverty-action.org with any questions or feedback related to the new PPI.
15. Will the PPI continue to remain available as a free tool after this transition?
Yes. A single national PPI scorecard for each country will continue to remain available as a free resource on our website. To access the PPI and supporting documents, simply create an account or use your existing account to log into our website.
16. Will the PPI learning materials and resources currently on your website continue to remain available?
Yes. We will refresh these materials to reflect changes made in the PPI where applicable, and will announce any major changes via our email newsletter and announcements section of our website. To subscribe to our newsletter, click here.
