RUTA TURISTICA

The aim of this section is to examine the difference in results using the money-metric approach, namely a poverty line of R3 864 per annum adopted by Woolard and Leibbrandt and the non-money-metric approach using fuzzy sets. With regard to the fuzzy sets approach, no benchmark poverty line is given as to what the deprivation index should be to distinguish the poor from the non-poor. For this reason a relative poverty line needed to be determined. The relative poverty line distinguishes the poorest 40% of the population. Thus for each of the surveys, i.e. Census 1996, Census 2001 and Community Survey 2007, it was necessary to determine the deprivation index that separates the poorest 40% from the rest of the population. For 1996 this deprivation index was 0.5891, for 2001 the value was 0.5484, and for 2007 the deprivation index was 0.4129.

The following cross-tabulations illustrate the proportion of individuals characterised as poor in both the fuzzy sets approach using a relative poverty line and in the money-metric approach using the R322 per month poverty line by Woolard and Leibbrandt. The income

75

variable used is the post-SRMI income variable by Yu (2009), which addressed the issues of the original problematic income variable by Stats SA, such as the high proportion of zero and unspecified income households. This may provide an indication as to whether the approaches derive the percentage of the poor quite differently.

In the table below the results obtained for the Census 1996 and Census 2001 are fairly similar. Using Census 1996 and 2001, of those identified as poor in the non-money-metric fuzzy sets approach, 85% were also poor in terms of the money-metric approach. About 14% considered poor under the non-money-metric approach were considered non-poor using the money- metric approach. With respect to the Community Survey 2007, of those identified as poor under the non-money-metric approach, 69.79% were also poor in terms of the money-metric approach and 30.21% considered poor under the non-money-metric approach were considered non-poor using the money-metric approach.

Table 5.14: Proportion of the poor and non-poor using money-metric and the non-money-metric approach, 1996- 2007

Non-money-metric approach Census 1996

Non-poor Poor Total

Non-poor 061.11% 014.51% 042.48% Poor 038.89% 085.49% 057.52% Money-metric approach Total 100.00% 100.00% 100.00% Non-money-metric approach Census 2001

Non-poor Poor Total

Non-poor 058.43% 014.56% 040.88% Poor 041.57% 085.44% 059.12% Money-metric approach Total 100.00% 100.00% 100.00% Non-money-metric approach CS 2007

Non-poor Poor Total

Non-poor 069.44% 030.21% 053.73%

Poor 030.56% 069.79% 046.27%

Money-metric approach

Total 100.00% 100.00% 100.00%

Source: Researcher’s own calculations based on Census 1996, Census 2001 and CS 2007 data.

The demographic characteristics of the poor are illustrated below using both approaches. The black race group is once again the dominant population group amongst the poor. This percentage is roughly 4-5% more using the fuzzy sets approach. In terms of gender, females are also a greater percentage of the poor under both approaches. The percentage increased over the years with the money-metric approach, but declined somewhat between 2001 and 2007 using the non-money-metric approach. The province variable provides some interesting results. The provinces that perform the worst with respect to both approaches are Limpopo, the Eastern Cape and KwaZulu-Natal. However, the difference between approaches for each

of the years is also greatest for these provinces. A greater percentage of the poor occurs using the non-money-metric approach. This indicates that non-income deprivation is greatest in these provinces and much can be done to alleviate poverty through service delivery. The results for educational attainment are fairly similar with low levels of education, namely that a greater proportion of the poor is reflected in both approaches.

Table 5.15: Demographic characteristics of the poor, money-metric vs. non-money-metric (fuzzy sets) approach, 1996-2007

Money-metric approach (per capita income, 2000 prices. Poverty line: R3 864 per annum by

Woolard and Leibbrandt)

Non-money-metric approach (fuzzy sets approach. Relative poverty line: deprivation index that

distinguishes the poorest 40%) Census 1996: 0.5891 Census 2001: 0.5484 CS 2007: 0.4129 Census 1996 Census 2001 CS 2007 Census 1996 Census 2001 CS 2007 Race Black 92.86% 93.56% 94.40% 98.26% 98.37% 98.10% Coloured 5.70% 5.54% 4.71% 1.36% 1.57% 1.75% Indian 0.61% 0.52% 0.53% 0.04% 0.02% 0.08% White 0.58% 0.37% 0.36% 0.04% 0.03% 0.07%

Unspecified 0.26% n/a n/a 0.31% n/a n/a

100.00% 100.00% 100.00% 100.00% 100.00% 100.00% Gender Male 49.35% 45.31% 43.39% 47.02% 45.72% 46.89% Female 50.65% 54.69% 56.61% 52.98% 54.28% 53.11% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% Province Western Cape 4.92% 5.39% 5.48% 1.55% 2.10% 2.16% Eastern Cape 19.68% 18.25% 17.95% 24.71% 22.50% 21.80% Northern Cape 2.10% 1.76% 1.60% 0.75% 0.82% 0.93% Free State 7.35% 6.80% 5.95% 4.59% 5.13% 3.90% KwaZulu-Natal 22.50% 23.60% 23.81% 27.24% 26.31% 26.44% North West 9.25% 8.88% 8.21% 9.13% 9.03% 8.74% Gauteng 9.57% 12.09% 13.65% 3.99% 6.16% 7.87% Mpumalanga 8.02% 7.80% 7.99% 6.95% 7.46% 7.81% Limpopo 16.60% 15.43% 15.37% 21.08% 20.51% 20.35% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% Education No schooling 42.08% 39.72% 27.63% 49.73% 47.74% 33.98% Incomplete primary 23.92% 24.52% 30.07% 23.68% 24.35% 30.54% Incomplete secondary 30.27% 29.41% 36.23% 24.17% 23.76% 30.92% Matric 3.21% 5.29% 5.11% 2.11% 3.53% 3.69% Matric + cert/dip 0.43% 0.87% 0.69% 0.26% 0.51% 0.54% Degree 0.09% 0.20% 0.28% 0.04% 0.12% 0.32% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00%

Source: Researcher’s own calculations based on Census 1996, Census 2001 and CS 2007 data.

77

Probit regressions were run across the surveys to determine whether the results differ significantly using the two different approaches and which people are more likely to be poor in each survey. The Table 5.16 below summarises the results and it is possible to examine whether any of the following explanatory variables are more likely to be associated with higher deprivation when using a different approach. The derivation of the explanatory variables is the same as in Section 5.3.

The results outlined below in Table 5.16 for the years 1996, 2001 and 2007 using both money-metric and non-money-metric approaches indicate that all explanatory variables are found to be significant at the 1% level.

Table 5.16: Deprivation in South Africa using money-metric and non-money-metric approaches

Dependent variable: Poverty status (1: poor, 0: non-poor)

Money-metric approach Fuzzy sets approach

Census 1996 Census 2001 CS 2007 Census 1996 Census 2001 CS 2007 Poverty line R3864 p/a R3864 p/a R3864 p/a 0.5891 0.5484 0.4129 Independent variable Coefficient Coefficient Coefficient Coefficient Coefficient Coefficient

Gender: Female 0.2658 0.3157 0.3706 -0.0030 -0.0409 -0.0315

Race: Black 1.1442 1.3727 1.2649 1.4182 2.1670 1.9005

Race: Coloured 0.5700 0.7450 0.6966 0.2956 1.1478 0.9716

Race: Indian 0.0682 0.1801 0.3791 -1.1120 -0.3162 -0.0928

Province: Eastern Cape 0.4551 0.4261 0.4376 1.1018 1.0026 1.2479 Province: Northern Cape 0.5463 0.4197 0.3371 0.1552 0.1987 0.4432 Province: Free State 0.4469 0.3416 0.3072 0.1417 0.2232 0.2125 Province: KwaZulu-Natal 0.2382 0.2203 0.2228 0.9207 0.7120 0.9424 Province: North West 0.1554 0.1610 0.1972 0.4830 0.4854 0.6886 Province: Gauteng -0.1829 -0.1264 0.0098 -0.3785 -0.2362 0.0468 Province: Mpumalanga 0.2342 0.1836 0.2377 0.3234 0.3786 0.6165

Province: Limpopo 0.3991 0.3804 0.4388 1.0082 1.0705 1.4120

Age in years -0.0124 -0.0193 -0.0143 -0.0173 -0.0412 -0.0489

Age in years-squared 0.0000 0.0000 -0.0001 0.0000 0.0002 0.0003 Education spline: Incomplete primary -0.0568 -0.0618 -0.0322 -0.0872 -0.0889 -0.0909 Education spline: Incomplete secondary -0.1329 -0.1269 -0.1042 -0.1347 -0.1320 -0.1275 Education: Matric -0.2533 -0.2585 -0.2524 -0.3184 -0.3070 -0.3342 Education: Matric + Cert/Dip -0.8458 -0.7520 -0.7545 -0.7346 -0.7464 -0.7909 Education: Degree -1.0107 -1.0274 -1.1371 -1.0981 -0.9612 -0.9412

Household size 0.1169 0.1480 0.1252 0.0141 0.0158 0.0153

Labour status: Employed -0.6729 -0.7214 -0.6027 -0.8910 -0.7190 -0.5452 Labour status: Unemployed 0.1766 0.0708 0.3099 0.1672 0.0496 0.0576

Constant -0.2583 -0.3348 -0.6978 -0.5192 -0.6515 -0.2372

Pseudo R-squared 0.3805 0.3968 0.3005 0.4185 0.3691 0.3581

Number of observations 33022467 42615642 46738626 33022467 42615642 46738626

All statistics are significant at the 0.01 level.

Source: Researcher’s own calculations based on Census 1996, Census 2001 and CS 2007 data.

Households that are more likely to be poor irrespective of the approach being used, be it the money-metric or non-money-metric approach, are the non-white race groups with blacks having a much larger positive coefficient than the coloured or Indian race groups. This outcome is as expected, with blacks being associated with the worst levels of deprivation compared to any other race group. Those residing in provinces other than the Western Cape (reference group) or Gauteng are also more likely to be poor.

The coefficients for the various education variables across the surveys using both methods are all negative. However, low levels of education – for example, incomplete primary and incomplete secondary – have smaller negative coefficients than higher levels of education. For each additional year of schooling the expected level of deprivation is lower. Thus those with low levels of education are also more likely to be poor. Education is often seen as a means to break the cycle of poverty. Hence governments should increase efforts to improve the education system.

The labour status variable is, as expected, significant. The unemployed coefficients are positive, indicating that once again those who are unemployed have a higher expected level of deprivation. Households that have an employed head are less likely to be poor, as shown by the negative coefficient. This gives impetus to improving labour market conditions as well as addressing the structural nature of unemployment to ultimately address issues such as poverty.

The age variable indicates that, as expected, households headed by those who are either very young or old are more likely to be poor. Household size also bears a positive and significant relationship with being poor. One would expect that the larger the household size the greater the number of dependants in that household.

Gender is of particular interest. Using the money-metric approach, females were found to be more likely to be poor across all three surveys, as their significant and positive coefficients illustrate. However, using the fuzzy sets approach with the relative poverty line, the coefficients are negative. This may be an indication that the government’s efforts to alleviate poverty for the most vulnerable groups are successful, as the deprivation index is more of a service delivery index, whereas the money-metric approach may reflect that women are still “money poor”. Although efforts are made to improve labour market outcomes, women are

79

still discriminated against and have lower earnings than their male counterparts47. A similar

result is found with the race variable, where the Indian race group has different signs for the different approaches. As expected, non-white race groups are shown to be more likely to be poor using the money metric approach, with the Indian race group having the smallest positive coefficient. However, using fuzzy sets the relation is negative. Thus the only intuitive response may be that, with respect to access to services, dwelling type and so forth, Indians are better off than the white race group.

Certain coefficients may have the same sign using both approaches, but the size of coefficients differs. This is also an interesting result. This is particularly evident in the province variables. As mentioned, the coefficients are positive for both approaches in all provinces except Gauteng, and the Western Cape, which is the reference group. For the Free State and Northern Cape the coefficients are smaller under the fuzzy sets approach. This may perhaps be an indication of good service delivery. However, for provinces such as Limpopo, the Eastern Cape and KwaZulu-Natal, households residing in these provinces are more likely to be poor under the fuzzy sets approach, as the larger coefficients illustrate. This has implications for the government’s efforts to alleviate conditions for the poor through improved basic service delivery. The government’s efforts would be better targeted to these provinces.

5.5 Conclusion

Chapter Five set out to critically evaluate and discuss the results of the fuzzy sets approach to poverty analyses. The chapter began by discussing poverty and deprivation in South Africa by province, race and gender. On the whole, all provinces improved in terms of the mean deprivation index across the surveys. However, the legacy of the past still persists, with Limpopo and the Eastern Cape being the worst performing provinces. Poverty still has a clear racial dimension, with blacks being the worst off. However, they did experience the highest decline in mean deprivation. Deprivation is also associated with gender, with females still being more deprived. The slightly more rapid decline of their mean deprivation index may be an indication of the effectiveness of the government’s targeted efforts. The mean deprivation by employment status of the household head, the educational attainment, area type and age of head was also evaluated.

47 _{See Burger and Yu (2006) as well as Woolard and Woolard (2006) for further discussion on wage trends and} earnings inequality.

For a more detailed depiction of the provinces, the best and worst performing magisterial councils, municipalities and district councils were shown. Furthermore, a comparison was made between the deprivation index and the problematic household income variable. It was seen that even at R0 and low-income levels, a certain proportion of households does have a low deprivation index.

In Section 5.3 an OLS regression analysis was done. A multivariate analysis was necessary to see whether any of the explanatory variables were more likely to be associated with deprivation. Thus it was possible to determine the impact of demographic, education, location and labour status variables on deprivation. A comparison between the money-metric and non- money-metric approach was attempted in Section 5.4. A poverty line of R3 864 per annum as proposed by Woolard and Leibbrandt was used as well as a relative poverty line to distinguish the poorest 40% using the fuzzy sets approach. Probit regressions were also run to ascertain whether there is a stark difference in results when using two different approaches.

A key result to emerge is the difference in poverty trends over the 1996 – 2007 period. Most studies reviewed in Chapter Three that used the money-metric approach showed that poverty trends were upward in the 1990s, before a downward trend took place in the 2000s. This was irrespective of the survey data used. The non-money-metric poverty trends derived in this chapter, however, show a continuous downward trend over the period. The overall mean deprivation in South Africa has declined since 1996.

81