Financial Diaries Project 2003-2004

To create the sampling frame for the Financial Diaries, the researchers echoed the method used in the Rutherford (2002) and Ruthven (2002), a participatory wealth ranking (PWR). Within South Africa, the participatory wealth ranking method is used by the Small Enterprise Foundation (SEF), a prominent NGO microlender based in the rural Limpopo Province. Simanowitz (1999) compared the PWR method to the Visual Indicator of Poverty (VIP) and found that the VIP test was seen to be at best 70% consistent with the PWR tests. At times one third of the list of households that were defined as the poorest by the VIP test was actually some of the richest according to the PWR. The PWR method was also implicitly assessed in van der Ruit, May and Roberts (2001) by comparing it to the Principle Components Analysis (PCA) used by CGAP as a means to assess client poverty. They found that three quarters of those defined as poor by the PCA were also defined as poor by the PWR. We closely followed the SEF manual to conduct our wealth rankings, and consulted with SEF on adapting the method to urban areas.

The first step is to consult with community leaders and ask how they would divide their community. Within each type of areas, representative neighbourhoods of about 100 households each were randomly chosen. Townships in South Africa are organised by street - with each street or zone having its own street committee. The street committees are meant to know everyone on their street and to serve as stewards of all activity within the street. Each street committee in each area was invited to a central meeting and asked to map their area and give a roster of household names. Following the mapping, each area was visited and the maps and rosters were checked by going door to door with the street committee.

Two references groups were then selected from the street committee and senior members of the community with between four and eight people in each reference group. Each reference group was first asked to indicate how they define a poor household versus those that are well off. This discussion had a dual purpose. First, it relayed information about what each community believes is rich or poor. Second, it started the reference group thinking about which households belong under which heading.

Following this discussion, each reference group then ranked each household in the neighbourhood according to their perceived wealth. The SEF methodology of wealth ranking is de-normalised in that reference groups are invited to put households into as many different wealth piles as they feel in appropriate. Only households that are known by both reference groups were kept in the sample.

The SEF guidelines were used to assign a score to each household in a particular pile. The scores were created by dividing 100 by the number of piles multiplied by the level of the pile. This means that if the poorest pile was number 1, then every household in the pile was assigned a score of 100, representing 100% poverty. If the wealthiest pile was pile number 6, then every household in that pile received a score of 16.7 and every household in pile 5 received a score of 33.3. An average score for both reference groups was taken for the distribution.

One way of assessing how good the results are is to analyse how consistent the rankings were between the two reference groups. According to the SEF methodology, a result is consistent if the scores between the two reference groups have no more than a 25 points difference. A result is inconsistent if the difference between the scores is between 26 and 50 points while a result is unreliable is the difference between the scores is above 50 points. SEF uses both consistent and inconsistent rankings, as long as they use the average across two reference groups - this would mean that 91% of the sample could be used. However, because only used two reference groups were used, only the consistent household for the final sample selection was considered.

To test this further,the number of times that the reference groups put a household in the exact same category was counted. The extent of agreement at either end of the wealth spectrum between the two reference groups was also assessed. This result would be unbiased by how many categories the reference groups put households into.

Following the example used in India and Bangladesh, the sample was divided into three different wealth categories depending on the household's overall score. Making a distinction between three different categories of wealth allowed the following of a similar ranking of wealth to Bangladesh and India, but also it kept the sample from being over-stratified. A sample of 60 households each was then drawn randomly from each area. To draw the sample based on a proportion representation of each wealth ranking within the population would likely leave the sample lacking in wealthier households of some rankings to draw conclusions. Therefore the researchers drew equally from each ranking.