CHAPTER 6: A COMPARISON OF HOUSEHOLD SOCIO-ECONOMIC STATUS
6.2 Country Comparison over Time – Simple Sum
105 (relatively wealthiest) cluster between 2000 (2.8) and 2006 (3.2) and an increase in the mean SES score for the fifth cluster from 2000 (18.65) to 2006 (23.60). The differences in mean SES between clusters were more similar for the 2000 analysis, suggesting that household SES in Uganda was more evenly distributed in 2000 than in 2006.
For Tanzania, from 2004 to 2007, there was a decrease in the proportion of households allocated to the lowest level of SES as well as a decrease in the per cent of households falling into the relatively richest category. These changes were relatively small. The mean SES score for both the first (poorest) and fifth (wealthiest) clusters decreased slightly from 2004 to 2007 as did the minimum SES score for the total sample. From these observations it could be concluded that overall the level of SES decreased from 2004 to 2007 for households in Tanzania. The differences in mean SES score between the clusters were relatively alike for the two time periods, suggesting that the distribution of SES was similar in 2004 and 2007.
The discussion above demonstrates how the index of SES combined with k-means cluster analysis could be used to monitor changes in household SES status over time. This SES measurement tool, however, only gives an indication of adjustments in the proportion of households falling into the relative categories of SES over time and does not give any indication of the actual level of SES represented by the clusters. The estimated SES scores are relative and their values are not directly comparable between countries: the actual levels of SES represented by the clusters for one country are not necessarily the same for another and the scores do not indicate actual levels of poverty or wealth.
106 using the simple sum method of weight estimation. The results are presented in Table 6.2 as the per cent of households allocated to each cluster by country and year.
Table 6.2: Cluster sizes (per cent of total sample) by country and year, based on the estimated household SES scores using the simple sum index and k-means cluster analysis with five clusters
Note: Relative household wealth increases from cluster one to cluster five.
Source: Author‟s calculations
For four of the five countries analysed, the trends in the changes of the per cent of households allocated to the first and fifth clusters and the location of the largest cluster differ between the CATPCA and simple sum method; only for Kenya were they similar.
For Kenya, both methods showed a decrease in the per cent of households allocated to the relatively poorest cluster from 2003 to 2008 and an increase in the proportion of households allocated to the relatively wealthiest cluster. The largest cluster was the first cluster for 2003 and the second cluster for 2008 based on the classifications by both of the methods. The same conclusions for Kenya can be drawn from the results based on both the CATPCA and simple sum methods: a lesser per cent of the population was extremely poor in 2008 than in 2003;
however, the poorest were worse off in 2008. The distribution of SES across households was more even in 2003 than in 2008 for both methods.
For Egypt, the results from the household classification based on the simple sum method showed that the per cent of households in the first cluster decreased by 0.2 percentage points from 2005 to 2008, whereas the results from the CATPCA-based classification showed an increase of 2.1 percentage points in the per cent of households in the poorest cluster between the two years of analysis. Both methods showed a decrease in the per cent of households allocated to the fifth cluster from 2005 to 2008. For the simple sum method, there was a decrease in the mean SES score for the relatively poorest and relatively wealthiest clusters
Country Year Total (%)
1 2 3 4 5
2005 3.9 15.6 56.6 18.8 5.1 100.0
2008 3.7 15.8 58.9 18.4 3.3 100.0
2003 54.9 31.8 5.9 6.7 0.7 100.0
2008 32.3 33.2 28.4 5.3 0.8 100.0
2001 20.5 59.4 15.1 3.7 1.2 100.0
2006 42.8 45.2 8.7 2.3 1.0 100.0
2000 47.5 34.2 13.0 4.0 1.3 100.0
2006 22.0 53.1 16.3 6.1 2.5 100.0
2004 29.2 59.9 6.9 3.4 0.6 100.0
2007 23.3 66.6 7.0 2.8 0.3 100.0
Cluster (%)
Egypt Kenya Mali Uganda Tanzania
107 which suggests, respectively, that the level of extreme poverty worsened and the level of extreme wealth decreased; assuming that the sum of a household‟s assets are a reflection of its wealth, which is unlikely to be accurate. The CATPCA results showed a reduction in both extreme poverty and wealth between 2005 and 2008.
Discrepancies between the results of the two methods could be found, for Mali, in that the CATPCA method results indicated a possible worsening in extreme poverty – as shown by a reduction in the mean SES score for cluster one between 2001 and 2006 as well as a lower minimum SES score for 2006, whereas the simple sum method results implied a potential reduction in the level of extreme poverty. However, comparing the SES scores from the two years was not entirely reliable as the variables included in the respective indices varied for the two time periods due to changes in the Demographic and Health Surveys over time.
Similarly, the results from the two methods were not in agreement for Uganda: the CATPCA method results showed an increase in the per cent of households allocated to the first cluster from 2000 to 2006, whereas the simple sum method results indicated a decrease in the per cent of the extreme poor. For Tanzania, both methods showed a decrease in the proportion of households allocated to the first cluster and a decrease in the per cent of households falling into the fifth cluster. However, the CATPCA method indicated that the first cluster was the largest for both years, whereas the simple sum method results showed the second cluster to be the largest for both time periods.
It is clear from the previous discussion that the changes in the proportion of households allocated to the different levels of SES differ depending on the method of index construction.
In this case, using the CATPCA method produced different, and often opposite, trends in the movement of households between levels of SES over time to the simple sum method. While the simple sum method offers a quicker, easier means of constructing an index of SES it uses less information and this omission of information clearly affects the classification results.
This outcome supports the conclusion from Chapter 4 that the method of index construction does affect the household classification outcomes; additionally it also affects the trends in the movement of households between clusters over time. In other words, not only does the method of index construction affect the actual cluster sizes, it also affects the direction of changes in the cluster sizes over time. More research and attempts to validate either method are required before a conclusion can be drawn as to which of the methods – CATPCA or
108 simple sum – is more appropriate as a means of assessing household SES as an indicator of household resilience.