< back to all forums
 
no image

Using PPI scores for comparative analysis

Jonathan Finighan
• Firetail
• United kingdom
• 09/02/13
• 2 Comments
 

Hi there,
I'm using the PPI score to compare households participating in a project in terms of their level of interaction with the project service - a mobile phone based extension service for smallholder farmers. So one of the metrics I am comparing along different lines in the sample (users' gender, occupation, crop types, and poverty level with respect to an international poverty line of US$1.25/day) is average calls made to the service over a specified period. 
Is it correct to calculate a group of households' PPI scores and then, using secondary data sources to estimate the average poverty level of the entire rural population for the country in question (e.g. 67%, using the US$1.25 line again), categorise your sample into those who are "above" and "below" the poverty line, based on their estimated % likelihoods of falling below the poverty line? I guess you could phrase the distinction between these categories more correctly as "households who are less/more likely than the average household to fall below the poverty line." Is that a correct approach? I have seen this method of using the PPI scores to categorise households into "below and above the poverty line" was used to compare levels of mobile phone ownership/access between the two groups, and I'm interested to know if this is appropriate use of the PPI. 
In my case, a project in Tanzania, using the PPI with a random sample suggests that the poverty level for participating households is 36.6%, but the poverty level for the entire country is considerably higher 67.9% (using World Bank data for the same poverty level); however, splitting households into two groups as per the above method ("more/less likely than the average to be poor...") implies only 14% of participating households fall in the "poorer" category.  
Any guidance on this would be most appreciated!
Many thanks,
Jonathan

2 Comments
Frank Ballard says (09/03/2013):
Hi, Jonathan. Thank you for this great post. I can see two questions here that need addressing. The first concerns proper data analysis and the second concerns discrepancies in poverty rates you found and those from the World Bank. To address the first, I would recommend flipping the analysis -- instead of categorizing households as poor or not poor and then looking at the usage rates of the extension service for either category, start by separating the population by average calls made to the service (0 per month, 1-3 per month, 4-6 per month, etc.) and then determine the poverty rate of those segments. It will be apparent immediately whether poverty level is correlated to up-take of the mobile service. For the second, when benchmarking PPI data, use poverty rates on page 57 of Tanzania’s PPI Design Documentation, found on the Tanzania PPI page or this link: http://www.progressoutofpoverty.org/node/1709/download . This will improve comparability. Additionally, benchmarking is most effective when PPI data can be compared to similar rates. This is because poverty rates vary widely across a country. If your operations are not nationally representative, then comparing your poverty figures to the national poverty rate may not be helpful. The document I referenced above provides poverty rates for Mainland Tanzania, Dar es Salaam, other urban areas, and rural areas. Rates for these categories should improve your analysis. Please let us know if you have further issues. One final note, just to be sure, I want to recommend converting household poverty scores to poverty likelihoods before performing any analysis. All poverty rates should be found by averaging poverty likelihoods, not by averaging poverty scores and then finding the poverty likelihood for that average poverty score. Please let us know if you have any other questions. Also, we would welcome the opportunity to learn the results of your research – sounds interesting! Best regards, Frank
 
Sunday Mailumo says (09/03/2013):
Hi. I used the PPI to analyse my data recently. Its really a good tool to measure poverty moreso, the difference from the expenditure approach method is so inconsequential when used to measure poverty incidence or prevalence. However, may I know whether it is possible to 1) Calculate a poverty line using the PPI, 2) Is it possible to calculate the poverty gap or depth and severity? If yes, how is it done? Sunny
 
 
< back to all forums