«To this end, and as a pilot case, this paper examines in detail the poverty incidence of the target beneficiaries of Katalyst’s prawn and maize ...»
Katalyst Working Paper Series
Poverty profiling using the Progress out of Poverty
Index (PPI) Version V, 1st May 2012
The objective of this paper is to present a clear and transparent methodology for a precise and
assessable formulation of what the concept ‘pro-poor’ means for Katalyst. The study is based
on the Progress out of Poverty Index framework (PPI) which is being actively used by various
multilateral donor organisations and NGOs around the globe. The paper identifies and justifies two cut-off PPI scores (30-34 and 55-59) for the two poverty lines (USD1.25/day [PPP 2005]) and USD2.5/day [PPP 2005]) to be used to assess the pro-poorness of a sector.
To this end, and as a pilot case, this paper examines in detail the poverty incidence of the target beneficiaries of Katalyst’s prawn and maize sectors. The study was conducted in January 2011, covering roughly eleven of Bangladesh’s northern and southern districts. The total sample size was in excess of 300, and results showed that according to the above definition, both sectors are pro-poor, and the maize sector significantly so. The paper also provides a snapshot of the progress Katalyst has achieved in implementing this system over the one year since its introduction. It concludes by discussing the scope and some of the limitations of the present study, and by providing recommendations.
Dhaka, May 1, 2012 Muaz Jalil, Markus Kupper, Hasan Shahriar Table of Contents
2. What is PPI?
3. Survey Design and Methodology
4. Summary Findings
5. Katalyst’s definition of pro-poor: how pro-poor are the prawn and maize sectors?........14
6. Caveats and recommendations
1. Background In early 2008Katalyst formulated its Pro-Poor Growth in Practice strategy paper. The objective of this paper was to show the link between Katalyst’s work and pro-poor growth. It provided a theoretical foundation to elucidate how working based on market development approach can result in pro-poor growth. Using extensive secondary data and field level validation, Pro-Poor Growth in Practice was able to show how Katalyst’s sector portfolio and market development approach target the poor and contribute to poverty alleviation. The paper successfully showed that it was ‘profitable’ cropping patterns which played the decisive role in deciding the incidence of poverty among small farmer households. Significantly, it also established the fact that the sectors in which Katalyst work (such as fish, maize, vegetable, and prawn) are profitable, and thus provides the best opportunity for small farmers to progress out of poverty.
With respect to validating Katalyst’s portfolio, the strategy paper was successful. However, in terms of its broader objective – to answer how pro-poor Katalyst sectors actually were – it was less precise. Questions remained as to whether Katalyst’s approach and work were actually targeting the poorer segments of its sectors. The present paper constitutes a step towards answering those questions. Although limited in its scope by the number of sectors chosen – only the two core sectors, prawn and maize, are covered – the paper provides specific guidelines for replicating this study across all Katalyst sectors. It is hoped that it may thus provide a stepping stone towards a more rigorous and comprehensive understanding of how the poor involved in Katalyst activities are targeted. It is not intended as a means of reviewing whether Katalyst’s work has resulted in pro-poor growth, but rather in providing an accurate identification of poverty distribution among its target beneficiaries (within the chosen sectors). It is hoped that such an exercise will result in increased accuracy in pro-poor targeting of Katalyst activities in the future, and at the same time provide requisite information for external parties, for whom such information may be useful.
Section 2 outlines the theoretical foundation of and methodology employed in this paper.
Section 3 discusses the survey methodology, and section 4 deals with summary findings and supplementary analysis. Section 5 deals with defining what pro-poorness means for Katalyst, and addressing the degree to which the maize and prawn sectors are pro-poor. The final section examines some of the caveats of the study and provides a roadmap for rolling it out across all Katalyst sectors.
2. What is PPI?
This section explains the Progress out of Poverty Index (PPI). It draws heavily on the Chen and Schreiner paper (2009) and interested readers are encouraged to consult the reference of the present paper for further information 1). The PPI is a simple and accurate tool which measures poverty levels of households and individuals, and assists organisations working in the field of poverty alleviation to improve their performance. It uses ten verifiable indicators (such as “What is the main construction material of the walls [of your house]?” and “Does the household own a television?”) to obtain a score that correlates closely with results of other, exhaustive poverty status surveys. The PPI scorecard for Bangladesh is based on data from the 10,080 households in the Household Income and Expenditure Survey (HIES 2005) conducted by the Bangladesh Bureau of Statistics, which provides the latest and largest household dataset in the country to date (HIES 2010 has yet to be published). Indicators are selected on the basis of being inexpensive to collect, easy to answer quickly, simple to verify and that they correlate closely with poverty. All points in the scorecard are non-negative integers, and total scores range from 0 (most likely to live below a poverty line) to 100 (least likely to live below a poverty line).
The use of scorecards for poverty targeting is nothing new and literature abounds with such methodologies. Gwatkin et al, Stifel and Christiaensen (2007), Zeller et al (2006), Sahn and Stifle (2003 and 2000), and Filmer and Pritchett (2001) use principle component analysis for establishing their scorecards. Wodon (1997) uses five indicators and the 1991 HIES dataset.
Haslett and Jones (2004) use ‘poverty mapping’ and the 2000 HIES dataset in order to estimate poverty rates in Bangladesh at union level; however, their objective was to help governments design pro-poor policies rather than to devise better ways of targeting poor. Similar efforts were made by Kam et al (2004) who applied a nutritional poverty line, which used calorific value rather than income. Zeller, Alcaraz and Johannsen (Zeller, Alcaraz and Johannsen, 2004) developed a very similar PPI-type scorecard for Bangladesh (although using older and smaller datasets. However, they differ in a number of significant aspects, which are discussed in Chen and Schreiner (2009). Other tools have been developed, such as by IRIS (2007), Cortez et al (2005) but they all differ from PPI in their objective, methodology and in terms of ease of use.
Additionally readers may visit the Progress out of Poverty website maintained by Grameen Foundation at http://progressoutofpoverty.org The PPI scorecard can measure a particular household’s ‘poverty likelihood’, that is, the probability that the household has a per capita expenditure below a given poverty line. This paper uses the USD1.25/day [2005 purchasing power parity 2] (PPP) and USD2.5/day [2005 PPP], which are the internationally-accepted extreme poverty and poverty line respectively. The scorecard can also be extended to reveal the poverty likelihood of a group of households, say for instance within a sector. This can be done by simply taking the average scorecard value for the entire sample household surveyed within the sector. However, if one wishes to identify the likelihood of being poor for a household involved in a certain occupation (e.g. the likelihood of a maize farmer being below a given poverty line), while at the same time maintaining statistical significance for this claim, then sample size becomes critical.
In developing PPI indicators, Chen and Schreiner (2009) started with a listing of 100 potential indicators in the areas of family composition, education, housing, ownership of durable goods, and employment. Indicators are not chosen because of their direct ability to predict poverty but rather on their extent of correlation. For example, ownership of a television is probably more likely to alter in response to changes in poverty than the education of members of the household. After this, an iterative process was initiated using a screening framework, involving both judgmental and statistical approaches. At the end of this process the list was reduced to 10 categorical indicators. The use of non-statistical criteria improves robustness through time and helps ensure that indicators are simple and make sense to users. Finally, logistic regression was used on a subsample of the total HIES (2005) dataset; this allowed the testing of sample robustness of the scorecard model.
The accuracy of the scorecard is contingent upon the stability of the relationship between indicators and poverty. So long this does not change, and the scorecard is applied to households that are representative of the same population from which it was constructed, this calibration process will produce unbiased estimates of poverty likelihoods. Unfortunately, HIES data employed in developing the latest PPI scorecard are over five years old and there is justified reason to assume that the relationship may have changed between some of the variables and poverty likelihood. This is specifically the case for questions 9-10 (Appendix I) PPPs are spatial deflators and currency converters, which eliminate the effects of the differences in price levels between countries, thus allowing volume comparisons of GDP components and comparisons of price levels. They are essentially price relatives that show the ratio of the prices in national currencies of the same good or service in different countries (EUROSTAT-OECD Methodological manual).
which refer to possession of a radio-cassette player and a wristwatch. Technological obsolescence and rapid expansion of telephone connectivity have made the latter extremely rare. Thus most farmers in the survey, while not owning a wristwatch, had a mobile phone which was also used for time keeping.
A question was also asked in relation to cropping patterns and areas cultivated under lease, in order to compare the findings of the present study with those from the 2008 Pro-Poor Growth in Practice paper. Hence it is not surprising that in the present study in case of both the maize and prawn sectors we saw that the greater the amount of land available to be farmed, the lower the likelihood of being poor. In the present study the questionnaire therefore, included seven additional questions beyond the ten standard PPI questions, including one on mobile phone ownership (see Appendix I). However, since the PPI scorecard is based on the original ten questions, the seven additional questions were excluded from the final results. The following section discusses the survey structure and technique employed in this study in greater details.
3. Survey Design and Methodology
The study focused on two Katalyst sectors, namely prawn and maize. Since its objective was to assess the poverty demography of Katalyst’s target beneficiaries, it made sense to select the survey sample from Katalyst’s beneficiaries within those sectors. In choosing the sampling methodology, resource feasibility and practicality were also taken into account. While it would have been preferable to undertake probability sampling, allowing us to make statistical inferences, the resources involved would have been prohibitive. For instance, if the target beneficiary group was in excess of 100,000 (which is the case with most Katalyst sectors), then the sample size dictated by probability sampling would be 400 3 (Israel, 1992). Such a sample selection would be compulsory if Katalyst benefitted all farmers directly, rather than targeting a particular segment through specific channels, as is the case. As Katalyst’s approach entails working with scale agents and their networks (such as distributors, retailers, and CIC centres), This is assuming a 95% confidence interval with 5% precision. In addition, we assume that the attributes being measured are distributed normally or nearly so. If this assumption cannot be met, then the entire population may need to be surveyed (Israel, 1992). This implies for a PPI study to assume that the incidence of poverty (as measured by PPI) is usually distributed among Katalyst target beneficiary groups, which may not be the case.
to undertake a probability sampling the correct methodology would be a stratified sampling 4, and this would increase the required sample size geometrically. This is further complicated by the fact that Katalyst runs multiple interventions within each sector and thus any sample frame should represent that as well. This implies for each sector the requirement for a multistage stratified sampling in order to achieve a sufficient probability 5. Considering that Katalyst intervenes in 17 sectors with an average of over 100 interventions running at any one time, time, financial and human resources for such a study would be prohibitive and simply beyond Katalyst’s scope, even if it were carried out only once during the phase.
Based on this, the MRM team decided to employ a qualitative sampling plan (that is, nonprobability sampling). To be specific, a snowball sampling methodology was applied, whereby service providers benefited by Katalyst interventions were asked to identify groups of farmers and target beneficiaries to whom they provided services. Even though this is not a statistical approach, it is not necessarily inexact. As long as the sample is representative of the target population, there may be a significant correspondence between the attributes being measured and the target population (which in this case is the poverty profile). One also has to take into account the fact that Katalyst’s approach depends on this ‘service provider-target beneficiary’ relationship: embedded information is usually channeled through this delivery mechanism. Thus unless the surveyed service provider (a retailer for example, or prawn postlarvae trader) systemically chose a biased sample of their customers base, a representative sample should ensue, ergo our target beneficiaries.