Race, Poverty and Deprivation in South Africa

Abstract

The aim of this paper was to explain why poverty and material deprivation in South Africa are significantly higher among those of African descent than among whites. To do so, we estimate the conditional levels of poverty and deprivation Africans would experience had they the same characteristics as whites. By comparing the actual and counterfactual distributions, we show that the racial gap in poverty and deprivation can be attributed to the cumulative disadvantaged characteristics of Africans, such as their current level of educational attainment, demographic structure and area of residence, as well as to the inertia of past racial inequalities. Progress made in the educational and labour market outcomes of Africans after apartheid explains the reduction in the racial poverty differential.

1. Introduction

South Africa stands out as a country with one of the largest racial divisions in the world due to European colonisation and the apartheid regime that followed independence, which officially ended in 1994. South Africa is indeed a racially diverse country: in 2008, nearly 80% of the population had heterogeneous African ancestry, with an additional 9% being people of mixed race (coloured). Whites accounted for another 9%, with the remaining 2.5% having Asian or Indian origins. However, the distribution of resources is extremely unequal across these groups, with whites reporting about eight times the average per capita income and expenditure levels of Africans. This stark inequality indicates little progress since the official end of legal racial segregation as the differential was slightly higher (about 10 times) in 1993.¹ This racial divide has remarkable implications in terms of poverty and deprivation by population group.

The previous literature has devoted extensive attention to poverty in post-apartheid South Africa.² Even though findings on poverty trends remain contested, an apparently increasing consensus agrees that poverty was aggravated in the early periods after the transition, and then improvements in more recent years were the result of the construction of a safety net through the social grant system (Leibbrandt et al., 2010). Among the many features that these studies have outlined in South African poverty, the differential in poverty levels across racial groups stands out as one of the most important. Hoogenveen and Özler (2006) and Özler (2007) proposed lower and upper bound monthly poverty lines based on the cost of basic needs at R322 and R593 in 2000, which we updated to R514 and R946, respectively, in 2008. According to our estimates using the National Income Dynamics Study 2008 (NIDS, 2008), the per capita household income of about 57% of Africans and 28% of coloured people fell below the lowest of these thresholds, in contrast with that of 9% of Asians/Indians and only 1.5% of whites. Using the upper bound poverty line, the percentages of poor people increased to 77, 49, 27 and 7%, respectively. This implies that the corresponding poverty rates for Africans are, respectively, 38 and 11 times higher than those of whites.³ The racial differentials in poverty of other countries that are well known for their racial inequalities are dwarfed by the scale observed in South Africa. For example, the poverty rates among those of African descent in Brazil and the USA are, respectively, about 2 and 3 times higher than those of whites (Gradín, 2009, 2012a).⁴

Similarly, we find that the differentials by race are also large when we move our interest towards direct measures of deprivation. After calculating a composite index based on multiple dimensions (using principal component analysis), Klasen (2000) reported a deprivation rate of 67% for Africans in contrast with only 0.6% for whites in 1993. Bhorat et al. (2006) have shown that the access of poor South Africans to basic services substantially increased in the early years of the post-apartheid period (from 1993 to 2004). However, in 2008, the differences by race in deprivation regarding several dimensions were still large. For example, according to our own calculations,⁵ 30% of Africans in 2008 lived in traditional or informal dwellings, while two-thirds lacked piped water inside their homes, compared with 0.5 and 5.5% of whites, respectively. Regarding home equipment, while 6, 7 and 18% of whites lived in households that did not own a refrigerator, television or radio, these percentages shifted to 47, 34 and 32% in the case of people of African origin. The differential is also large in terms of the accumulation of deprivation. Less than 2% of whites lacked all three of these appliances at home, in contrast with 12% of Africans. Likewise, 45% of Africans reported having insufficient (less than adequate) healthcare coverage, more than doubling the level of 19% for whites in a similar situation.

The aim of this paper was to investigate the reasons that these differentials in well-being remain so large. More specifically, we will measure the extent to which they result from Africans having poorer human capital or sociodemographic endowments. Then, the differentials would come from a compositional effect and represent inequality across those attributes. Alternatively, the differentials could be a consequence of those attributes having a different impact on Africans' well-being.

Disentangling which part can and which cannot be explained by human capital and sociodemographic endowments is relevant as they are both important but have different natures. Differences that come from a compositional effect indicate that the bad performance of disadvantaged groups is driven mostly by their unequal access to education, family planning or the labour market or by the fact that they live in more deprived areas. The part that cannot be explained suggests that the disadvantage more likely stems from schooling, labour market participation or location having a different impact on poverty and deprivation within these groups, which could be caused by the prevailing discrimination in the labour market, different perceived quality of education or different degree of vulnerability due to unobserved factors. The causes associated with the former are more directly solved through redistributive policies at different levels than those coming from the latter, which tend to be more structural. The latter could be in fact the result of inertia of past inequalities through the intergenerational transmission of poverty/deprivation.

The identification of the factors more closely associated with the racial gap in well-being could also be of help in ascertaining the racial implications of any public policy, even if it is not directly aimed at reducing racial inequities, such as conditional transfers seeking a larger attachment of poor children to schooling or of adults to the labour force and development policies addressed at specific regions or communities. The larger the contribution of past inequalities to explain current racial differentials in poverty and deprivation, the slower will be the expected reduction in that differential in the near future.

The structure of the paper is as follows. In the next section, we describe the data and methodology. We then undertake an empirical analysis and finally summarise the paper's main contributions.

2. Data and methodology

2.1 Data

For the analysis, we used two different nationally representative samples of all private households in South Africa with information on households' living conditions. One is the first wave of the NIDS (version 3) from 2008. This data set, provided by the Southern Africa Labour and Development Research Unit (SALDRU, University of Cape Town), includes rich information on an array of dimensions—such as income, expenditure, home appliances owned, neighbourhood, educational level and health status—for 28,250 individuals living in 7,302 households.⁶ The other is the Project for Statistics on Living Standards and Development (PSLSD, 1993), which sampled 43,687 individuals living in 8,809 households, undertaken by SALDRU in collaboration with the World Bank during the 9 months before the country's first democratic elections at the end of April 1994.⁷ Information from both samples was made as comparable as possible, even if the former provided richer information regarding some relevant issues.

2.2 Measuring poverty and deprivation

To measure financial poverty, we computed various indices of the Foster et al. (1984) family (FGT), based on total monthly household income divided by the number of household members.⁸ We used Hoogenveen and Özler's (2006) and Özler's (2007) lower and upper bound absolute poverty lines in 2000 prices (R322 and R593) updated to R514 and R946, respectively, in 2008, and deflated to 1993 prices to R198 and R365, respectively.⁹ For a robust analysis, we also measured poverty with the same poverty lines, using per capita total household expenditure as a well-being indicator.¹⁰ Let P(y) be a member of the FGT family of poverty measures. If y is income, z the poverty line and N the number of individuals in the population,  

(1)

For ⁠, the index is the head-count ratio (or poverty rate); for ⁠, the average normalised poverty gap; and for ⁠, the average normalised squared poverty gap. The first case accounts only for poverty incidence, while the other two add sensitivity to poverty intensity and inequality among the poor.

To take into account the multidimensional nature of racial differentials in well-being, direct measures of material deprivation were also computed across twenty-two attributes reflecting different well-being dimensions: (i) needs insufficiently met (coverage less than adequate compared with household needs in food, housing, clothing, healthcare and schooling); (ii) lack of ownership of a motor vehicle and several home appliances (e.g., radio, television, VCR/DVD, computer, electric/gas stove, microwave, refrigerator/freezer and washing machine); and (iii) exclusion from access to different basic services (e.g., formal dwelling, piped water, flush toilets, electricity, landline telephone, cellular, rubbish collection and street lighting).

As a first step, we used the percentage of population in each group excluded in each of these attributes in their households. This is a flexible way of looking at possible differences among this heterogeneous set of dimensions. Let be a dummy variable, taking the value 1 if the ith individual (i = 1, …, N) is deprived in the jth attribute (j = 1, …, J) and 0 otherwise. Then, the proportion of the population deprived in this attribute is determined by the following:  

(2)

As a second step, we summarised the extent of exclusion for each person from this set of attributes, constructing an individual composite indicator of material deprivation:  

(3)

where can be interpreted as the marginal contribution to the individual indicator of being deprived in the jth attribute, compared with not being deprived. One can obtain these weights in many ways. The literature provides no conclusion regarding the best approach. In our empirical analysis, we estimated them using a multiple correspondence analysis (MCA) for the joint sample of Africans and whites over the set of dummies and then the (standardised) scores (k = 0,1) associated with each category ⁠.¹¹ This composite indicator of deprivation takes values between 0, not deprived with respect to any attribute, and 1, deprived in all of them. It is the linear combination of the original variables providing the largest possible correlation, or explaining the largest share of variability (inertia).

Among the different alternatives that are available in the literature to deal with multidimensional poverty, this approach can be considered a counting-based multidimensional index using the union approach. In this case, one individual is deprived if is deprived in at least 1 out J dimensions (deprivation line). Then, for each deprived individual, the index counts the (weighted) number of dimensions in which a person is deprived, with weights estimated by MCA (or similarly, defined as the inverse of the frequency of each item). Thus, the average of the deprivation indicator corresponds to the index M₀ proposed in Alkire and Foster (2011) under these assumptions.¹²

To measure the extent of larger incidence of severe deprivation among Africans compared with whites, in the absence of a natural deprivation line (unlike poverty), we followed an inter-distributional approach in line with Le Breton et al. (2011).¹³ Thus, we first estimated the corresponding cumulative distribution functions (CDFs) of deprivation among Africans and whites, say and ⁠. Then, we plotted both CDFs to represent the inter-distributional concentration (discrimination) curve for each ⁠, where denotes the quantile (right inverse) function attached to the distribution F. That is, indicates the proportion of Africans with deprivation equal to or below the quantile for whites. Then, parallel to poverty analysis, we used different thresholds (here quantiles t = 0.99, 0.95, 0.90, … at the top of whites' distribution) to compute the proportion of Africans experiencing deprivation above those cut-offs, ⁠. Then we compare each of these proportions with the corresponding theoretical value associated with whites, 1−t.¹⁴ Thus, the vertical line between the 45° line and the corresponding concentration curve for Africans, ⁠, is used as a measure of the disadvantage of this group (i.e., higher deprivation) with respect to whites, for each possible cut-off.¹⁵

2.3 Possible explanations for the racial gap in poverty and deprivation

There are many possible explanations for why poverty and deprivation is so high among Africans in South Africa compared with whites.

Poverty rates vary greatly across South African provinces (Leibbrandt and Woolard, 1999), and regional divergence in income seems to persist over time, with relatively poor regions more likely to remain poor and the richest regions acting as local growth poles, with location, trade, education and the gold mining industry driving this evolution (Bosker and Krugell, 2008). There are also large inequalities across provinces and between rural and urban areas in the access to basic services such as piped water, toilet and formal dwelling (Noble et al., 2006; Barron et al., 2009). Thus, the first possible explanation for racial disparities in well-being is blacks being overrepresented in the poorest areas of the country, which seriously undermines their economic opportunities. This may be an important factor in South Africa because the initial spatial distribution by race and migration were historically determined by government intervention. Residential segregation policies based on racial classifications started in colonial times and were accentuated by the apartheid legal system (i.e., the Bantu Authorities and Group Areas Acts, see, e.g., Christopher, 1992). As a result, Africans were confined to rural areas, notably in the homelands, and into urban suburbs. This can help to explain black–white differences in poverty and—especially—material deprivation, in which the access to several services depends on community development. This factor, however, is expected to lose some relevance after the end of legal restrictions to internal migration combined with economic development and growing urbanisation. Indeed, right after apartheid ended, new internal migration patterns emerged, now driven by economic factors, with migrants searching for higher expected wages (Choe and LaBrent Chrite, 2009), moving to provinces with higher GDP and lower reported crime (Bouare, 2001). According to Statistics South Africa's (2006) report on migration and urbanisation, the proportion of black people living in urban areas increased about 10 percentage points (from 40 to 50% between 1991 and 2001, with whites stabilised about 90%). However, this report did not find any significant increase in the rate of migration of blacks over time, and post-apartheid migration seems to have been largely temporary (Posel, 2004).

Another factor that may influence the racial gap in well-being is demographics because Africans are more likely to have larger families. Despite the continuing reduction in fertility rates among all racial groups that occurred in South Africa at least since the 1960s, and despite them being lower than in other countries in the region, Africans, especially in rural areas, report higher fertility rates than whites: 3.11 compared with 1.88 (Moultrie and Timæus, 2003). This higher fertility of Africans has been associated, among other things, with higher incidence of teenage pregnancy; among Africans, 17.8% of 15- to 19-year-old women were pregnant compared with 2.2% among whites. There is also a lower use of contraceptives—58.6 versus 79.8% (Swartz, 2002).¹⁶

An education system characterised by racially segregated schools and under-resourcing of schools for blacks was another legacy of apartheid. This created a double gap in both years of schooling and educational achievement (e.g., Van der Berg, 2007), which can have a significant impact on the earnings differential by race. Indeed, the gap in attained education is still large, with whites having levels comparable with those of developed countries and Africans being closer to the developing world, but it has been substantially narrowed in the past decades thanks to increasing resources allocated to the education of blacks (see, e.g., Chisholm, 2004). Nevertheless, intergenerational education mobility of blacks was lower than that of whites (Nimubona and Vencatachellum, 2007). But it is the gap in quality, starting at primary schools, that generates more concern today, with South African schools generally performing worse than neighbours in the region despite their larger amount of resources (Van der Berg, 2007).¹⁷ This unequal quality of education is one of the factors that can explain that differences in the returns to education accounted for about 40% of the white African wage differential in 2001/02, whereas in 1993, this effect was virtually zero, thus at least partially reversing the benefits of equalisation of schooling attainment (Keswell, 2010).

The other factor that Keswell (2010) suggested that could be responsible for the racial gap in the returns to education was the persistence of previous occupational segmentation. It is precisely the unequal performance of blacks and whites in a recognised dysfunctional labour market that is the likely cause of a large earnings differential. The labour market in South Africa was also segmented across racial lines during apartheid, and despite several improvements and affirmative action initiatives thereafter, the situation has not been completely reversed. Africans tend to report lower participation rates and higher unemployment compared with whites. The former also work in less-skilled occupations and get lower wages.¹⁸ This gap in employment and earnings is only partially accounted for by their lower human capital endowments. Kingdon and Knight (2004a) found that 8 out of the 34 percentage points of the unemployment gap between Africans and whites in 1994 could not be explained by their observed characteristics, and there is also evidence of persistent wage and occupational discrimination by race that survived the end of apartheid (Allanson et al., 2000, 2002; Erichsen and Wakeford, 2001; Rospabé, 2002). Another characteristic of the labour market that persists is a large skill mismatch between the demand and supply sides due to the large pool of unskilled workers created by the apartheid in townships and homelands, combined with the technological shift in the economy towards capital-intensive activities (Arora and Ricci, 2005). In the context of a sluggish economy, affected by the shrinkage of the non-mineral tradable sector since the early 1990s, this situation has produced unemployment rates that are extremely high according to international standards, and with special incidence among the most disadvantaged groups, who, unlike other developing countries, find it difficult to get into the informal sector (Kingdon and Knight, 2004b).

Further, in a country with such a segregative history as South Africa, a more dynamic perspective should be addressed. Growing up in a poor family generally increases a person's chances of experiencing poverty during adulthood through different channels (i.e., Hoelscher, 2004; Magnuson and Votruba-Drzal, 2009). For example, low parental investment or financial stress may, later in life, increase poor children's bad social behaviour and reduce their academic achievement. This is an important issue given the low intergenerational mobility that can be expected in the South African context. Obviously, some current characteristics, such as education, will be correlated with family background, thus capturing part of the effect of the latter factor on the differential in poverty by race. But two households with similar currently observed characteristics could have different economic outcomes on the basis of their families having different economic backgrounds. This is the case of South Africa, where, based on the literature, one would expect a low family background to be associated with lower quality of education. This would in turn increase the explained poverty differential. Subsequently, ignoring past inequalities could lead to an underestimation of the proportion of the racial differential in poverty that is explained, as well as to an overestimation of the contribution of some current characteristics (those correlated with family background).

The larger the proportion of the poverty differential explained by past inequalities, the slower the expected reduction in this differential because the reduction will be mainly driven by convergence in current characteristics, as illustrated by what happened after apartheid. That is, not accounting for this factor could result in a naïve or overly optimistic view of how much improving Africans' situation would reduce poverty differentials.

In summary, the legacy of apartheid and colonisation has left Africans with several drawbacks that make them more likely to be poor, such as living in rural areas or in the poorest provinces, higher fertility, less education, and poorer labour market outcomes, even if it is difficult to determine which of them is more relevant than the others. The end of apartheid and the implementation of affirmative action helped Africans to catch up with whites in many of these aspects, but inequalities are still large and some more subtle forms of discrimination, such as the increasing gap in the returns to education, are playing a more outstanding role. Somehow, current inequalities are expected to be in part explained by past rather than current low endowments of Africans.

In our empirical analysis, we considered a set of variables accounting for most of these potentially explicative factors for racial differences in well-being. We initially organised current household characteristics in the NIDS sample into five groups.¹⁹ First, geographical location accounts for province of residence and a dummy indicating whether the household lives in a rural area.²⁰ Second, we used a set of demographic variables. These include the characteristics of the head of household, such as marital status (i.e., married; single living with partner; widow(er)/divorced/separated; and never married), sex, age interval (i.e., below 25 years old, between 25 and 55, or above 55) and migration status (i.e., whether migrated during the last 5 years; internal migrant, immigrant from abroad or non-immigrant). The number of children and adults in the household was included as the main determinant of family needs. The third group accounts for household members' attained educational level (i.e., the number of years of schooling of the household head and the average for adults in the household, and their corresponding squared values) as the main determinant of their labour market opportunities. The fourth group measures household's labour market attachment. It includes the head of household's labour force status (i.e., employed in regular work, employed in casual work, unemployed, self-employed or not economically active) and occupation (at one digit), the proportion of adults in the household in each labour status and occupation category and the household's adult dependency ratio, defined as the proportion of adults receiving earnings or pension benefits. The information in the PSLSD sample was organised in a very similar way but with some restrictions.²¹ To explore the role of past inequalities to explain current inequalities, we included as an additional potential factor explaining poverty differentials by race a sixth group of variables accounting for family background attained educational level of the mother and father of the household head, only available in NIDS.²²

2.4 Methodology: counterfactual analysis

We first estimated different poverty and deprivation measures by race and then decomposed the racial gap resulting from comparing Africans with whites into the explained (characteristics effect) and unexplained (coefficients effect) parts. This is the aggregate decomposition. Further, we ran a detailed decomposition of the characteristics effect by quantifying the contribution to the gap by the different potential explicative factors mentioned above: geographical location, demographic structure, labour market performance, education and family background. To complete these decompositions, we estimated a counterfactual distribution in which members of the disadvantaged group (Africans) were given the relevant characteristics of the affluent group (whites), using the adaptation of a propensity score technique (DiNardo et al., 1996) in Gradín (2012b). The differential between poverty/deprivation measures of whites and Africans provided the unconditional racial poverty/deprivation gap. The difference between poverty/deprivation in the observed distribution for Africans and in its counterfactual represented the explained (characteristics) effect, while the difference of poverty/deprivation between the counterfactual distribution and that of whites provided a measure of the conditional differential, or unexplained/coefficients effect. Below is a more in-depth explanation of the procedure.

Each individual observation was drawn from some joint density function f over (y, x, g), where y indicates the vector of per capita household income (alternatively expenditure or deprivation in any dimension), x is a vector of observed household characteristics and g identifies whether the individual is white (the reference group, g = w) or African (g = b). The marginal distribution of income for each group g is given by the density  

(4)

This can be obtained as the product of two conditional distributions, where  

(5)

In other words, the actual income density for Africans or whites is determined by the marginal income density of members of the group having each combination of characteristics (a high level of education, living in Cape Town, and so on) times the proportion of group members having this set of characteristics.

Then, we defined the counterfactual income distribution as the distribution of y that would prevail if Africans kept their own conditional income distribution (the probability of having an income level given their characteristics) but had the same characteristics (marginal distribution of x) of whites. We produced this counterfactual distribution by properly reweighting the actual income distribution of Africans:  

(6)

Based on Bayes's theorem, the reweighting scheme can be expressed as the product of two ratios:  

(7)

where the ratio is given by the share of Africans and whites that belongs to each race in the pooled sample (and can be ignored because it is a constant) and the ratio is estimated using a logit model for the probability of being white conditional on x in the pooled sample of whites and Africans. In other words, these weights increased the contribution to the index of interest made by Africans with characteristics more similar to those of whites and decreased the contribution of those with greater dissimilarity.

In parallel with the conventional Oaxaca (1973)–Blinder (1973) procedure, widely used in labour economics to estimate wage discrimination, we used the counterfactual distribution for the following decomposition of the differential between whites and Africans for any poverty index P:  

(8)

The superscripts b, w and x indicate whether poverty was measured for Africans, whites or the counterfactual distribution (conditional on x). P(y) is a poverty index. Thus, the first term in the previous equation represents the part of the poverty differential by race explained by characteristics (or characteristics effect), while the second is the unexplained part (or coefficients effect).

In the detailed decomposition, we wanted to quantify the impact on the poverty/deprivation differential of changes in a single covariate (or set of covariates) x_j instead of the whole vector. For that, we used the Shapley decomposition, which results from averaging over all possible sequences of factors (Chantreuil and Trannoy, 2012; Shorrocks, 2012). The resulting individual effects would be path-independent and add up to the overall effect.²³

Using the same procedure described in this section, we could construct a counterfactual distribution for the J vectors of the dummy variables describing deprivation across the population. Then, the differentials in the proportions of African and white populations deprived with respect to each attribute, or according to the composite indicator, could be decomposed accordingly.²⁴

3. Poverty and deprivation by race in South Africa

In presenting our empirical analysis, we will first provide the results for income poverty and then discuss the main differences when using material deprivation as a well-being indicator.²⁵

3.1 Income poverty differential by race

Racial segregation in South Africa left a legacy of huge differences in poverty across ethnic groups. As the first three rows of Table 1 illustrate, about 71 (87)% of Africans were poor in terms of income in 1993 according to the lower (upper) bound poverty line, compared with 2 (4)% of whites. Fifteen years after the termination of apartheid, poverty incidence using the same thresholds (in real terms) was substantially reduced among Africans, especially more severe poverty, while poverty among whites remained constant (lower bound) or even increased (upper bound). Thus, the differential in poverty rates fell slightly, but still remained high in 2008: 57 (77)% of Africans were poor according to the lower (upper) bound threshold, compared with about 1.5 (7)% of whites in a similar situation. This means that Africans were still thirty-eight (eleven) times more likely to be poor than whites in 2008, compared with forty-two (twenty) times in 1993. Poverty intensity and inequality among the African poor were reduced parallel to poverty incidence in post-apartheid South Africa, as can be inferred from the fact that poverty reductions among Africans were higher using indices accounting for not only incidence but also intensity and inequality (FGT(1) and FGT(2), respectively, see Table A3).

Table 1:

Racial Income Poverty Gap between Africans and Whites in South Africa, FGT(0) (Lower and Upper Poverty Lines), 1993–2008

	Lower poverty line				Upper poverty line
	2008		1993		2008		1993
	FGT(0)	% Differential	FGT(0)	% Differential	FGT(0)	% Differential	FGT(0)	% Differential
Whites	1.5		1.7		6.7		4.3
Africans	57.0		71.0		76.6		86.6
Differential	55.5		69.3		69.9		82.3
Counterfactual	7.3		2.0		20.6		5.3
Unexplained	5.8	10.4	0.3	0.4	13.9	19.9	0.9	1.1
Explained (all characteristics)	49.7	89.6	69.1	99.6	56.0	80.1	81.4	98.9
Geographic	15.0	27.0	8.2	11.8	10.8	15.4	4.6	5.6
Province	2.4	4.3	4.2	6.1	−1.4	−2.0	2.1	2.5
Rural	12.6	22.7	4.0	5.8	12.2	17.4	2.5	3.1
Demographic	13.3	23.9	10.9	15.7	14.4	20.5	11.4	13.9
Head's marital status	1.3	2.4	−0.4	−0.5	2.1	3.0	−0.2	−0.3
Head's immigration	−1.4	−2.5	0.9	1.3	−2.8	−4.0	0.9	1.0
Head's sex	2.4	4.3	−0.6	−0.9	3.8	5.5	−0.8	−1.0
Head's age	−1.8	−3.2	−0.9	−1.3	−4.3	−6.2	−1.4	−1.7
Number of children	10.1	18.1	6.1	8.8	11.1	15.9	5.8	7.0
Number of adults	2.7	4.9	5.7	8.3	4.5	6.4	7.3	8.9
Education	14.8	26.7	31.0	44.7	24.7	35.3	37.0	44.9
Labour	6.7	12.0	19.0	27.4	6.2	8.8	28.4	34.5
Labour status	4.6	8.3	−5.5	−7.9	2.0	2.8	−6.4	−7.8
Occupation	2.1	3.7	24.5	35.3	4.2	6.0	34.8	42.3

Source: Own construction using PSLSD, 1993 and NIDS, 2008.

The main contribution of the present work is, however, a quantification of how much this high poverty (and its reduction) among Africans, compared with whites, can be attributed to the unequal distribution of characteristics by race in South Africa.

3.2 Explained income poverty differential by race in 2008

3.2.1 Aggregate effect

Our first main finding was that a large share of the differential in income poverty by race can be explained by the higher prevalence among Africans of those characteristics most strongly associated with poverty. In general, the proportion explained was larger with the lower than with the upper bound poverty line and increased as we incorporated sensitivity to intensity and inequality among the poor in the poverty index. Thus, extreme poverty was better explained by characteristics than moderate poverty. Table 1 illustrates the results of income poverty for the counterfactual distribution (row 4) and the corresponding aggregate decomposition of the racial differential in poverty into the unexplained and explained parts (rows 5 and 6). We first discuss the results for 2008. We will present an analysis of the trend in the next subsection.

More specifically, 90 (80)% of higher poverty among Africans in 2008 can be attributed to their current characteristics using the lower (upper) bound poverty threshold, with the share rising to 93 and 95 (84 and 88)% in the cases of FGT(1) and FGT(2) (see Table A3). The above proportions among Africans would have been about 7 (21)% of the population had their characteristics been similar to those of whites (counterfactual). Consequently, we estimated the conditional differential in poverty rates with whites to be 6 (14) percentage points. This would be entirely the result of household characteristics having a different impact on the likelihood of being poor, depending on the race. This could be a consequence of unobservable attributes, the different quality of some characteristics (e.g., attained educational level) or direct labour market discrimination, among other reasons. Note that these conditional poverty differentials were still large compared with those of other countries with well-known black–white differences, such as the USA (about 4 percentage points estimated for 2006 in Gradín, 2012a) or Brazil (2 percentage points in 2005 according to Gradín, 2009).

3.2.2 Detailed effect

After measuring the aggregate effect, we identified which of those potential factors described in the previous section are more associated with the racial poverty differential and quantified their contribution. The results are shown from row 7 to the end of Table 1. Focusing first on the case of severe poverty (lower poverty line), education, demographic characteristics and geographical location (the first level of disaggregation of the detailed effect), each accounted for a significant share of 24–27% of the differential, with labour-related factors relegated to explaining (globally) only an additional 12%. Thus, no unique source of those discussed in Section 2.3 accounted for the differential in poverty rates based on race. Rather, higher poverty among Africans seems to be the result of the accumulation of several disadvantages, mostly pre-labour market endowments. The most salient single factor (the second level of disaggregation of the detailed effect) associated with the racial poverty gap was Africans dropping out of school earlier. Despite the progress made after apartheid, years of schooling still explained 27% of the higher poverty incidence with respect to whites (or equivalently, almost 15 percentage points). The second most significant factor was Africans living in rural areas to a greater extent (23% of the differential, or 13 percentage points), followed by their families having more children (18%, or 10 percentage points). Among the labour factors, labour status of household members explained 8% (5 percentage points) of the differential, and their occupation another 4% (2 percentage points). Thus, increasing attachment to school, combined with improved family planning, employment and rural development policies, would likely have the most significant impact on reducing the severe poverty gap based on race.

Some factors made a (small) negative contribution. That is, with values for these characteristics similar to those for whites, Africans would have even higher poverty rates than they actually have. This is the case for age (African household heads are slightly younger on average than whites) and migration (they have lower migration rates).²⁶

The use of two poverty thresholds allowed us to check whether the explicative factors were similar for severe and for moderate poverty. The results for the upper bound poverty line, compared with the lower, showed (the four columns on the right in Table 1) the following: (i) the substantially larger relevance of education, which explained 35% of the differential (25 percentage points of the poverty rate differential instead of 15); (ii) the lower importance of geographical location, now explaining (globally) only 15% of the differential (11 percentage points, compared with 15), especially driven by the negative contribution of the province of residence; (iii) to a lesser extent, some demographic factors, mainly number of adults, also explain more than before; and (iv) labour factors explain a similar amount of percentage points, although a lower percentage of the differential. Thus, in relative terms, education replaced location in explaining higher poverty rates among Africans as we pushed the poverty threshold upwards.²⁷

When it comes to including intensity and inequality in the measure of poverty (shifting from FGT(0) to FGT(1) and FGT(2)), the results were quite similar except for the lower role played by education and the corresponding higher relevance of the other factors (Table A3). This reinforces the idea that education is less associated with higher income poverty among Africans at the bottom of the distribution (whose members contribute more to poverty intensity and inequality than those near the poverty line). Consequently, the more decisive role of education for the upper bound poverty line was maintained but to a lower extent with FGT(1) and FGT(2).²⁸

3.3 Explaining the income poverty trend in post-apartheid South Africa

As mentioned above, poverty among Africans was higher right before the end of apartheid in 1993, so the differential with whites was also larger by about 14 (12) percentage points with the lower (upper) bound poverty line (Table 1). Looking at the decomposition of the racial differential for each year, we observed that the explained part was notoriously reduced during the observed time span, by 19 (25) percentage points from 69 (81) to 50 (56), which indicates that the reduction was driven by a convergence in characteristics. The detailed decomposition in Table 1 shows that this considerable reduction in poverty among Africans between 1993 and 2008 was mainly the result of the progress they made by increasing their years of schooling and by filling jobs in more skilled occupations (for a given level of education), thus catching up with whites. That is, our results suggest that equalisation in attained education and a less racially segregated labour market that took place after the end of apartheid were responsible for the main reduction in the racial poverty gap. Among the demographic factors, only the reduction in the number of adults in African households had a small positive contribution. The more rapid urbanisation of whites and their reduction in the number of children in their households, however, went in the opposite direction, helping to curb the diminishing poverty differentials by race.²⁹

Indeed, the contribution of education to higher poverty rates among Africans was virtually halved from 31 to 15 percentage points using the lower bound, thus being able to explain by itself the entire observed reduction in the poverty rate differential. The reduction in the racial poverty gap associated with education in the case of the upper bound was more limited, from 37 to 25 percentage points, but still able to explain the total reduction in the differential. Indeed, Africans 15–55 years old increased their years of education from 7.2 to 9.1, compared with the increase among whites from 11 to 12.4. Similarly, Africans' occupation played a fundamental role in 1993, contributing significantly to the racial poverty differential that year, even after controlling for education and location (of 24 and 35 percentage points for the lower and upper thresholds, respectively). This role vanished almost entirely in 2008 (to 2 and 4 percentage points, respectively). The change in occupational classification makes the comparison difficult. However, in 2008, the sum of those reporting managerial, professional and technical occupations accounted for 12% of employed African adults (45% for whites), compared with only 6.9% in the closest occupations in 1993 (47.3% of whites).³⁰ The reasons for this increase in the access of Africans to high-skilled occupations would probably need a more in-depth research. Apart from being the direct consequence of their increasing education, the literature suggests the end of legal segregation (i.e., job reservations for whites) or the implementation of affirmative action initiatives as possible additional factors.

The global contribution of demographic factors to higher poverty rates slightly increases over time, with a higher contribution of the number of children (from 6 to 10 percentage points using the lower bound) and a lower contribution of the number of adults (from 6 to 3 percentage points).³¹ The contribution of the higher concentration of Africans in rural areas substantially increased between 1993 and 2008 (the share of rural population decreased more clearly for whites, from 8.5 to 2.9%, compared with the relatively smaller reduction from 66.7 to 61.9% among Africans).

In contrast to the reduction in explained poverty differentials, the unexplained or conditional differential in poverty rates increased from virtually nothing in 1993 to 6 (14) percentage points in 2008. This means that the reduction in poverty differentials was not largely due to the opposite effect of these characteristics becoming less protective in terms of keeping Africans out of poverty, compared with whites.³² This suggests the sources of racial poverty differentials shifting from between-group inequalities in the basic characteristics to more subtle forms of racial disadvantage, such as discrimination or differences in the quality of their attributes.³³ The next subsection explores this in more detail, looking at the role of family background in explaining current inequalities.

3.4 The role of family background

After taking into account past inequalities in our model, the entire set of household characteristics explained 90% of the racial differential in poverty levels, regardless of which poverty line was considered, as reported in Table 2. Thus, family background accounted for about 10% of the differential that before remained unexplained using the upper bound poverty line. But even more relevant, the family background turned out to be the main explicative factor, especially when using the upper bound poverty line, at the expense of the other factors that shrunk, mostly the current level of education. Indeed, as Table 2 shows, family background accounted for 13 (24) percentage points of the gap in poverty rates, representing 24 (34)% of that differential.³⁴ Thus, past inequalities in education had similar or higher relevance than that of the other main factors, whose contributions were reduced, such as the number of children, 23 (22)% of the gap; living in rural areas, 24 (19)%; or years of schooling, 10 (15)%.³⁵

Table 2:

Racial Income Poverty Gap between Africans and Whites in South Africa with Family Background, FGT(0) (Lower and Upper Poverty Lines) in 2008

	Lower poverty line		Upper poverty line
	FGT(0)	% Differential	FGT(0)	% Differential
Whites	1.5		6.7
Africans	57.0		76.6
Differential	55.5		69.9
Counterfactual	7.2		13.2
Unexplained	5.7	10.2	6.5	9.4
Explained (all characteristics)	49.8	89.8	63.4	90.6
Geographic	14.4	26.0	11.8	16.9
Province	1.4	2.5	−1.5	−2.1
Rural	13.0	23.5	13.3	19.0
Demographic	11.9	21.4	12.5	17.8
Head's marital status	1.1	1.9	1.8	2.6
Head's immigration	−3.4	−6.2	−5.8	−8.4
Head's sex	2.3	4.2	3.1	4.5
Head's age	−2.6	−4.6	−4.7	−6.8
Number of children	12.9	23.3	15.6	22.3
Number of adults	1.5	2.7	2.5	3.6
Education	5.6	10.1	10.7	15.3
Labour	4.6	8.3	4.3	6.2
Labour status	3.4	6.2	1.3	1.9
Occupation	1.2	2.1	3.0	4.3
Family background	13.3	24.0	24.1	34.4

Source: Own construction using NIDS, 2008.

The fact that years of schooling is the factor with the largest reduction in their explicative power—27 (45)% of the differential before considering family background—indicates that there is a large intergenerational transmission of low education among Africans, and that in fact current education could be to a large extent acting as a proxy of family background.³⁶ The poorer family educational background of Africans is probably one of the worst legacies of apartheid, and because it is expected to be more persistent over time due to a low intergenerational educational mobility, it is also the most difficult to deal with. The equalisation of current characteristics will not be enough to overcome the remaining racial poverty differential in the short run. Thus, ignoring this factor leads to overestimating the contribution of current attained education to explain the level of the racial poverty gap and the contribution of its equalisation to explain the trend over time. In fact, this could explain why the reduction in poverty in South Africa after apartheid was lower than one would expect given the equalisation across races in the relevant current characteristics, such as attained education and occupation. This claims for more structural measures, such as improving the quality of schools for those with the lowest family background to accelerate the process.³⁷

3.5 Material deprivation

Finally, we took into account the growing consensus, stressing that the experience of poverty transcends financial poverty. That is, we adopted a more multidimensional perspective. We measured the racial gap in material deprivation with regard to different aspects, including needs insufficiently met, lack of appliances and lack of access to basic services. This approach could also be seen as a way of overcoming the lack of reliability of reported income as a measure of well-being in developing countries. Table 3 presents the results. First, we measured the percentage of individuals in each racial group that were deprived with respect to each single attribute. In all cases, Africans were deprived in a much higher proportion than whites, with the largest differentials (about 60 percentage points or more) found in the lack of appliances (e.g., washing machine, motor vehicle, microwave and/or computer) and the lack of access to basic services (such as piped water or flush toilets).

Table 3:

Racial Gap between Africans and Whites in Indicators of Material Deprivation in South Africa, NIDS, 2008

Single indicator	Africans	Whites	Differential	Counterfactual	% Differential explained by
					All	Geographic	Demographic	Education	Labour
Access to
Formal dwelling	30.5	0.5	30.1	5.4	83.5	38.8	10.7	27.1	6.8
Piped water	66.8	5.5	61.4	17.9	79.7	46.3	5.6	24.1	3.7
Flush toilet	58.6	0.6	58.0	6.2	90.5	61.0	4.3	24.5	0.7
Electricity	23.2	1.4	21.8	4.5	85.7	53.7	−2.9	32.4	2.6
Landline telephone	94.0	49.0	45.0	84.5	21.0	9.1	5.3	7.0	−0.3
Cell phone	11.6	4.7	6.9	6.0	81.3	−32.9	−29.3	116.7	26.8
Rubbish collection	55.0	4.3	50.7	3.5	101.5	74.3	2.0	25.2	0.0
Street lighting	66.6	11.9	54.7	21.1	83.2	56.2	5.2	21.4	0.4
Insufficient needs
Food	42.8	10.2	32.7	14.0	88.2	13.0	23.5	46.6	5.0
Housing	42.9	10.9	32.0	15.3	86.4	11.9	12.6	36.6	25.3
Clothing	44.5	18.1	26.4	22.5	83.4	10.3	12.7	41.3	19.1
Healthcare	44.6	19.4	25.2	16.0	113.4	20.8	28.6	43.4	20.6
Schooling	32.9	5.6	27.3	8.7	88.5	14.2	29.8	35.6	8.9
Ownership
Radio	32.4	17.6	14.7	23.9	57.2	−6.6	16.1	35.4	12.4
TV	34.4	7.0	27.4	14.3	73.4	34.1	−4.1	34.9	8.5
VCR/DVD	71.5	16.8	54.6	32.0	72.3	28.3	1.1	36.8	6.1
Computer	93.6	33.9	59.7	71.7	36.8	3.4	6.7	22.5	4.2
Electric/gas stove	36.1	9.8	26.3	7.0	110.9	52.1	−1.5	49.2	11.1
Microwave	72.7	14.3	58.4	30.6	72.2	30.9	2.9	32.8	5.6
Fridge/freezer	46.5	5.6	40.9	12.3	83.6	34.8	4.1	36.0	8.8
Washing machine	85.1	10.1	75.0	47.8	49.7	18.0	4.1	20.9	6.8
Motor vehicle	88.1	18.7	69.4	41.4	67.3	12.6	10.6	28.9	15.2
Composite indicator
Average (M0)	0.58	0.13	0.45	0.3	71.3	30.1	6.2	28.4	6.6
Quantiles
t = 0.99	50.4	1.0	49.4	5.4	90.9	49.1	2.3	35.8	3.6
t = 0.95	74.4	5.0	69.4	16.2	83.8	34.7	8.0	33.0	8.2
t = 0.90	87.5	10.0	77.5	40.9	59.6	20.5	6.2	26.7	6.2
t = 0.75	94.6	25.0	69.6	55.0	56.4	16.3	7.1	21.0	12.1
t = 0.50	98.6	50.0	48.6	84.0	30.0	−0.4	12.0	14.2	4.2

Source: Own construction using NIDS, 2008.

Current household characteristics explained the entire racial gap (or most of it), in cases where the population lacked access to basic services, such as rubbish collection or flush toilets (90%); lacked an electric or gas stove; or received inadequate healthcare and food (88%). Characteristics also explained a large share of the gap in deprivation in the rest of cases. There were only a few exceptions in which they explained only half or less: access to a landline phone (21%), a computer (37%) or a washing machine (50%).

The main factors explaining these deprivations varied in each case. Unequal geographical distribution is associated, to a larger extent, with deprivation in terms of access to basic services, such as rubbish collection (74%), flush toilets (61%), street lighting (56%), electricity (54%), or piped water (46%), as well as lacking appliances, such as an electric/gas stove (52%). Education appeared responsible to a larger extent for insufficient provision of food (47%), healthcare (43%), clothing (41%) and schooling and housing (36–37%), as well as for access to a cell phone (117%) or an electric/gas stove (49%) and a radio (43%). Family demographics were also relevant, to a lesser extent than education, for insufficient schooling (30%), healthcare (29%), food (24%) and the lack of a radio (16%). Labour-related factors were relevant only in explaining the lack of a cell phone (27%), a motor vehicle (15%), as well as sufficient housing (25%), healthcare (21%) or clothing (19%).

Why this heterogeneity in the factors associated with the racial gap in deprivation? Geography is mainly related to the development in the community that is known to vary greatly across South African provinces and between rural and urban areas, with Africans being overrepresented in the poorest provinces and in rural areas. Thus, the geographical factor is more closely connected with the lack of access of Africans to most basic services. The well-known racial gap in schooling helps to better explain the sort of deprivation that, other things constant, can be eliminated with more household income (such as basic needs or purchasing appliances), while it is much less effective in the case of deprivation more related to lower community development (such us the access to some services), where more income is of little help (unless it is enough to push the household to move to a richer neighbourhood). Indirectly, education could also reflect different consumption patterns, such as the access to new technologies or certain goods that are expected to be consumed mostly by highly educated people. Finally, demography refers mainly to the increasing household needs (adults and children) and is more associated with basic needs that vary with household size, such as food, schooling or health care.

As a second step, we constructed for each individual a composite indicator defined as the weighted average of deprivation in each attribute, with weights estimated using MCA, as described in the previous section. This indicator measured the degree of accumulation of different forms of deprivation in the same individuals, accounting for 86% of the variability (principal inertia) of the original variables.³⁸ The last six rows of Table 3 display these results jointly with the average of the indicator (M₀).

On average, deprivation among Africans was 58% of the maximum level (all people are deprived in all attributes) compared with 13% in the case of whites.³⁹ To compare the distribution of this indicator for Africans and whites, we computed the percentage of Africans with a level of deprivation higher than that for whites at different percentiles of the whites' distribution, ⁠. Half of the African population experienced deprivation above the 99th white percentile (compared with 1% of whites, by design), and this proportion increased to 74% at the 95th percentile, reaching 99% at the median of the whites' distribution. Thus, the differential between both groups, ⁠, can be used as a measure of the disadvantage of Africans with respect to whites, similar to what was done in the poverty analysis. Further, replacing with the counterfactual ⁠, we check how this disadvantage is corrected after conditioning on characteristics. Figure A1 in the appendix plots the corresponding concentration curves for the actual and counterfactual distributions. In this context, is just the vertical distance between the diagonal and the corresponding curve. The results for the average deprivation composite indicator showed that 71% of the racial gap was explained by characteristics, namely geographical and educational factors, in a similar proportion (30 and 28%), but this masked the different role that these factors played at different levels of the distribution of deprivation discussed below.

The higher deprivation of Africans at the 99th percentile could mostly (90%) be attributed to their poorer household characteristics, but this share decreased sharply as we moved from more severe to more moderate deprivation (that is, from more to less accumulation of deprivation): 84% (95th), …, 30% (50th). Therefore, only the most severe deprivation was explained by the unequal distribution of characteristics by race. The share explained for the 99th (95th) percentile is in fact similar to the case of the lower (upper) bound financial poverty threshold. The main difference between material deprivation and poverty, however, came from the main contributors to the racial gap. The geographical factors turned out to be much more relevant in explaining extreme material deprivation than in the case of poverty, 49 (35)% of the gap for the 99th (95th) percentile. The predominance of geographical factors for the deepest deprivation is related to the previous results in which this factor was shown to be crucial in gaining access to basic services that tend to be unavailable in the less developed areas. The contribution of this factor decreased sharply for lower percentiles (virtually zero at the median). The second most important factor in explaining the gap in extreme deprivation levels by race was the household's educational level, which explained 36% of the gap at the 99th percentile. Its relevance also decreased with lower levels of deprivation, but less sharply than that of location: the contribution of both factors was similar for the 95th percentile, but the household's educational level became the main factor for lower percentiles. On the one hand, higher education, as well as labour, is expected to affect deprivation by facilitating access to higher income and meeting basic needs such as food, housing, clothing or the purchase of some home appliances. But both are also expected to affect consumption patterns and then make more likely the acquisition of new technologies.

The inclusion of family background as an explicative factor (Table 4) substantially increased the percentage of the gap explained by characteristics by reducing or eliminating the effect of unobservables for most dimensions. Deprivation in most attributes was explained to a great extent (80% or more) by characteristics with only a few exceptions (only about 40% for lack of landline phone and computer and 73% for lack of washing machine). Additionally, educational family background turned out to be a crucial factor, replacing education in most cases. It was the main reason for higher deprivation of Africans in meeting sufficient needs or in the ownership of several appliances, and the second—after geographical location— regarding the access to basic services.

Table 4:

Racial Gap between Africans and Whites in Deprivation Indicators in South Africa with Family Background, NIDS, 2008

Single indicator	Counterfactual	% Differential explained by
		All	Geographic	Demographic	Education	Labour	Family background
Access to
Formal dwelling	4.8	85.5	41.2	7.6	13.8	−5.5	28.5
Piped water	85.7	85.5	41.8	2.5	13.1	−1.7	29.8
Flush toilet	94.7	92.0	48.5	4.6	13.3	−3.7	29.3
Electricity	97.8	96.2	39.3	−0.8	18.2	−4.8	44.2
Landline telephone	25.2	42.8	−0.8	25.8	3.0	−10.9	25.6
Cell phone	96.5	118.6	−14.1	−32.5	57.3	−12.5	120.4
Rubbish collection	96.4	101.4	63.8	1.9	14.6	−6.9	27.9
Street lighting	80.6	86.4	44.5	0.8	14.7	−4.4	30.8
Insufficient needs
Food	8.1	106.5	12.0	16.8	27.1	7.4	43.3
Housing	9.6	104.0	9.4	14.8	18.3	21.0	40.5
Clothing	11.9	123.4	7.7	14.7	22.2	19.1	59.8
Healthcare	11.4	131.9	11.5	22.8	23.2	22.4	51.9
schooling	3.8	106.5	12.8	25.0	16.9	8.4	43.4
Ownership
Radio	89.4	147.7	−4.3	40.3	27.9	18.3	65.5
TV	92.9	99.5	28.1	0.0	18.8	−0.2	52.9
VCR/DVD	84.1	101.8	20.7	4.9	21.3	9.6	45.2
Computer	31.1	41.4	−1.7	12.8	13.4	−8.4	25.2
Electric/gas stove	94.8	117.6	39.8	1.0	27.2	6.1	43.4
Microwave	76.7	84.6	23.7	1.0	21.6	−0.3	38.5
Fridge/freezer	92.4	95.0	26.1	5.7	21.8	3.8	37.6
Washing machine	69.4	72.6	13.5	9.5	11.0	8.1	30.5
Motor vehicle	67.2	79.7	8.5	8.9	17.8	12.7	31.8
Composite indicator
Average (M0)	20.3	84.8	23.5	8.2	16.2	2.0	35.0
Quantiles
t = 0.99	2.7	96.4	38.8	3.7	19.5	−2.4	36.8
t = 0.95	12.6	89.0	26.9	7.1	17.2	4.7	33.1
t = 0.90	21.6	84.2	19.6	8.8	14.5	6.4	35.0
t = 0.75	38.5	80.1	15.8	7.0	13.4	8.7	35.1
t = 0.50	75.5	47.2	−8.6	22.4	10.3	−5.8	28.9

Source: Own construction using NIDS, 2008.

Similar results were found in the case of the composite indicator. Family background raised the share of the racial gap explained by characteristics on average and at all percentiles. Characteristics generally explained most of the gap, between 80% at the 75th percentile and 96% at the 99th (but still 47% at the median). The qualitative roles of education and geographical location discussed above were preserved, but with smaller shares. Family background explained 35% of the gap in average deprivation, more than the other factors. At the 99th percentile, geographical location and family background were the most important factors, but while the relevance of location still decreased for lower levels of deprivation, family background had a similar explicative role above the 75th percentile, thus becoming the most important factor.

4. Conclusions

People of African descent in South Africa face higher poverty and deprivation rates than whites. These racial differentials are high, even compared with those in other countries known for their high racial inequalities, such as Brazil and the USA. In this paper, we have investigated the extent to which the high racial poverty and deprivation differentials in South Africa are explained by inequalities in the distribution of characteristics across races. To do so, we have estimated a counterfactual distribution in which Africans were given the characteristics of whites.

Our results showed that the higher levels of African financial poverty and extreme material deprivation could mostly be explained by the accumulation of past and present disadvantaged characteristics. No factor took prominence in explaining the racial gap in poverty levels. Rather, the accumulation of mainly pre-labour market disadvantages among Africans produced higher poverty.

The overrepresentation of Africans in poor rural areas is one of the factors more strongly associated with higher poverty, and to a greater extent, with their higher material deprivation, especially regarding access to basic services. This factor has become more important after apartheid because despite the progress made, important spatial inequalities persist, and migration, no longer administratively controlled, has been mostly temporal, probably influenced by a dysfunctional labour market that reduced the opportunities for Africans in urban areas.

The demographics have also contributed to higher poverty among Africans. The larger number of children in African households compared with whites, due to higher fertility rates of African women especially in rural areas, made them struggle to meet their needs, especially healthcare, schooling and food. Despite the reduction in the average number of children per household in both groups, this factor has increased over time its contribution to the poverty gap. This could be explained by an increasing association between the number of children in a household and poverty (ceteris paribus).

If a factor were likely to be the main contributor to the higher income and material deprivation of Africans, it would be education. We have shown that indeed education was also one of the most important factors to explain current higher poverty rates and higher material deprivation, especially unmet needs and the lack of several basic home appliances, mainly associated with low income due to lower labour market opportunities. This is a persistent source of inequality that has shown important progress after apartheid and is one of the main sources of reduction in poverty differentials between Africans and whites. However, the literature has pointed to an increasing role, after the end of racial segregation, of the racial gap in the quality of education, with Africans performing worse than those in some poorer neighbour countries. This is consistent with our finding that the importance of unobservables to explain poverty differentials by race has increased significantly after apartheid. Additionally, the inclusion of educational family background (years of schooling of household head's parents), which could be a good proxy for lower quality education and other unobservables, is the most important factor in explaining poverty and deprivation across several dimensions, mainly reducing the role of years of schooling but also of unobservables. This indicates that low intergenerational mobility in education is expected to make progress slower in reducing poverty and deprivation differentials.

Finally, it is well known that Africans face a dysfunctional labour market with chronically high unemployment rates and high racial segregation across occupations. After having controlled for other factors, such as education and geographical location, labour market outcomes were of great relevance after apartheid, but became much less important later. Improvement in the access of some Africans to more highly skilled jobs is the likely cause. This was likely the result of a more integrative labour market environment for Africans with the end of legal segregation (i.e., job reservations for whites) and the implementation of affirmative action initiatives.

Figure A1:

Inter-distributional Concentration Curves in Material Deprivation for Africans and Counterfactual Distributions. Source: Own construction using NIDS, 2008.

Acknowledgements

I acknowledge the financial support from the Spanish Ministerio de Educación y Ciencia (Grant ECO2010-21668-C03-03/ECON) and Xunta de Galicia (Grant 10SEC300023PR).

1

These are our estimations using NIDS, 2008 and PSLSD, 1993, respectively. See the next section for details.

2

Among others, see May (2000), Statistics South Africa (2000), Van der Berg and Louw (2004), Meth (2006), Agüero et al. (2007), Özler (2007), Seekings (2007), Argent et al. (2009), Leibbrandt et al. (2009a, 2010) and Van der Berg et al. (2008).

3

These are our estimations using NIDS, 2008. The situation does not change significantly when expenditure reported in the same survey measures well-being in South Africa. Expenditure poverty among Africans was still about twenty-five times higher than that among whites with both thresholds.

4

Estimates were obtained using the official poverty line in the case of the US 2007 Current Population Survey and the 50% of the median (120 reals) in the case of Brazil's 2005 Pesquisa Nacional por Amostra de Domicílios. For a more detailed comparison of income distributions in Brazil, the USA and South Africa (using the 2005/06 Income and Expenditure Survey), see Gradín (2012b).

5

Based on NIDS, 2008.

6

We use NIDS post-stratification weights, which are design weights calibrated to match the 2008 midyear population estimates (see Wittenberg, 2009). About 88% of households had a zero non-response rate. About 7% of the sample of adults did not respond to the individual questionnaires. Derived variables in the survey include imputed values for such non-response (see Leibbrandt et al., 2009b for details).

7

The survey was designed as a self-weighting sample, but different problems led to the under-representation of whites and certain areas in the sample, so sample weights were constructed at the level of the old provincial/homeland boundaries and race (DataFirst, 2011).

8

Income was obtained by aggregating all forms of income from the adult questionnaire. This includes income (reported or imputed) from the labour market, government investments, implied rental income, remittances and subsistence agriculture and excludes items of a capital nature, such as inheritance, retrenchment payments, retirement gratuities, lobola/bride payments, gift income, loan repayments, sale of household goods income and ‘other’ income. In the case of PSLSD, we took the closest definition of total monthly income.

9

After applying a conversion rate of R4.25 per dollar (Leibbrandt et al., 2010), both lines correspond respectively to 121 and 223 PPP dollars in 2008.

10

This includes food and non-food expenditure, household rent (or imputed rent for those households that are not paying a rent) and full imputations in the case of NIDS (and the closest definition available for the PSLSD).

11

See Asselin (2009) for a detailed discussion of the use of MCA in the measurement of multidimensional poverty. We speak here of deprivation, not poverty, because we deliberately used only dichotomous variables, although using multiple categories of the variables instead does not significantly change the results. In particular, note that the distance function between profiles used by MCA, the chi-square metric, weights the Euclidean distance by the inverse of the relative frequencies. This makes exclusion from more common attributes contribute more to individual deprivation than exclusion from rare attributes. This is in line with other views in the literature, such as the approaches followed by Desai and Shah (1988) or Tsakloglou and Papadopoulos (2002). For example, the latter use the normalised proportion of non-deprived population as the weight for each attribute. Indeed, replacing weights in equation (3) by would produce a new individual indicator highly correlated with ours (about 96% in our empirical analysis). So this and our approach are very close.

12

That is, the product of the head-count ratio (proportion of deprived people) and the average intensity of deprivation among the deprived (weighted sum of dimensions along which they are deprived). Given that all individual single-deprivation indicators are of the 1/0 type (deprived versus not deprived), M₀ = M₁ = M₂, as defined by Alkire and Foster (2011).

13

See Yalonetzky (2012) for a discussion of this in relation with other alternative approaches to measure interdistributional inequality.

14

Obviously, researchers can choose one from among several alternatives to compare white and African distributions. We can use the average deprivation (also computed in our empirical analysis) or construct FGT-type indices of deprivation. In the absence of a natural ‘deprivation line’ and for the sake of simplicity, we adopted here this approach to explain the larger incidence of deprivation among Africans (reference group) using alternative thresholds indexed to the distribution of whites (comparison group), which is the distribution that remains constant after the counterfactual analysis (thus, our deprivation thresholds are the same for the original and counterfactual distributions).

15

Note that due to discontinuities in the CDFs, the measure is not zero when whites are used as both the reference and the comparison groups.

16

‘The high rate of teenage pregnancies has far reaching consequences, especially for the Africans and coloureds who are the poorest and most disadvantaged groups in the country. The majority of these pregnancies are neither planned nor wanted. The father of the child seldom acknowledges or takes responsibility for the financial, emotional and practical support of the child. The mother often leaves school, thus ending her opportunities for personal development, making her vulnerable to poverty, exploitative sexual relationships and violence as well as low self-esteem. Because of the youth of the mother, her child is particularly vulnerable to perinatal mortality. Once born, the child is usually brought into a situation where he or she cannot be supported emotionally or financially’ (Department of Social Development, 2000, p. 45).

17

Analysing students' mathematical and reading performance in Grade 6, Spaull (2011) concluded that ‘South Africa is still a tale of two schools: one which is wealthy, functional and able to educate students, while the other is poor, dysfunctional, and unable to equip students with the necessary numeracy and literacy skills they should be acquiring in primary school.’

18

Using the same data set we use here, Leibbrandt et al. (2010) reported that participation rates have increased 25.6 percentage points between 1993 and 2008 among Africans, compared with a reduction of 6.4 percentage points among whites. But this means that it is still 10 percentage points lower among the former (53.9 compared with 64.3%). The standard unemployment rate was 27% among Africans compared with 10% among whites (10 percentage points of increase for the same period, compared with 6.9 percentage points for whites). The inclusion of discouraged workers makes this situation even worse. Additionally, average monthly wage was R2,576 for Africans compared with R11,240 for whites (22.4% increase for the former between 1993 and 2008, compared with only 4% for the latter). About 33.4% of African workers were in the informal sector (5% of whites).

19

In some cases, a category for observations with missing values was also included to avoid the loss of information.

20

We considered eight categories for province of residence in the NIDS sample, after having combined Free State and North West into one category due to sample size problems.

21

More specifically, the demographic information differs in that marital status distinguishes whether there was a spouse and if he or she was present, deceased or absent. Immigration status only accounted for migration during the past 5 years. Note also that the provincial organisation in South Africa changed after 1994, and thus, in the PSLSD, we considered four categories: Cape, Transvaal, Orange Free State and the rest of the country.

22

These variables contain a large number of missing observations. For this reason, a category accounting for them was included in each case.

23

See Sastre and Trannoy (2002) for a formalisation of the procedure to compute the Shapley decomposition. In this paper, the Shapley decomposition was implemented in two stages. First, we computed the contribution of each group of factors (e.g., location) to the overall poverty differential. Then, we computed the individual contribution of each specific factor (e.g., province and rural area) to the total group's contribution.

24

The individual composite indicators of deprivation in the counterfactual distribution were computed using the same weights w^j estimated with the original distribution.

25

The logit regressions used to construct the counterfactual distributions are shown in Table A2.

26

See Table A1 for average values of explicative variables between whites and Africans.

27

This is in line with Gradín (2012b), who pointed to the increasing relevance of education (at the expense of the other factors) in explaining the black–white income differential for higher percentiles in South Africa using the 2005/06 Income Expenditure Survey.

28

The risk of expenditure poverty for both whites and Africans was higher, compared with income poverty, and so was the differential, 61 (77) percentage points with the lower (upper) poverty line. The percentage of poverty in 2008 that was explained by characteristics was roughly similar to that of income, but the reasons differed. The main difference was the much more important role played by education in explaining the differential in poverty rates with the lower bound poverty line (24 percentage points, or 39% of the gap). This is because the educational level attained was larger for those African adults identified as suffering from severe poverty with income but not with expenditure (7.7 years on average) than for those in the reverse situation (5.9 years). Thus, the association between Africans' higher severe poverty and lower educational level was stronger when using expenditure rather than income.

29

Similar results to those discussed in this subsection can be found using the FGT(1) and FGT(2) indices (see Table A3).

30

These figures refer to all observations. Restricting to only non-missing observations, the increase was from 9.2% in 1994 to 14.5% among Africans in 2008 (47.2 to 53.2% for whites). This seems to be consistent with other more comparable sources (although restricted to a shorter period). Using data from the South African Labour Force Survey (http://www.statssa.gov.za), there was an increase from 10.3 to 13.3% from March 2000 to March 2007 (46.1 to 54.3% for whites).

31

Households in both groups reduced the average number of children between 1993 and 2008 (whites, from 1.07 to 0.76; Africans, from 2.93 to 2.23). Apparently, the relatively higher risk of poverty among households with children in 2008 explains this increase in the contribution of this factor to the racial poverty differential.

32

Obviously, ascertaining which factors changed their impact the most (detailed coefficients effect) would be quite interesting, as we have done with the characteristics effect. However, the disaggregation of the coefficients effect involves additional technical difficulties. There is no clear procedure for it using our methodology. It could be done following, for example, Yun's (2004) approach, consisting of estimating poverty regressions for both groups; yet the small number of poor whites observed, especially in the NIDS data set, discouraged us from doing so.

33

The decline in expenditure poverty incidence for Africans between 1993 and 2008 was more limited, compared with that of income, especially for severe poverty (lower bound)—from 68 to 64% of the population—but also for the upper bound poverty line—from 88 to 80%. Thus, the reduction in the differential in poverty rates was smaller with expenditure of 6 (9) percentage points with the lower (upper) bound poverty line. As in the case of income, this was due to a reduction in the explained poverty gap—9 (25) percentage points—that was partially compensated for by an increase in the unexplained part, from 1 to 4 (from 2.5 to 18) percentage points. The impact of more years of schooling among Africans in reducing these differentials was more limited than in the case of income: 5 (11) percentage points. In fact, in the case of expenditure, labour market attachment (mainly due to occupation) turned out to be much more important in explaining the reduction in poverty, with 10 (20) percentage points.

34

Results were computed after removing a few observations to avoid a disproportionately large weight in the counterfactual distribution.

35

Consequently, the main difference in explaining racial differentials for moderate as opposed to severe income poverty was the larger contribution of family background, and to a lesser extent, attained education (and lower contribution of the remaining factors).

36

Household characteristics also explained most of the racial gap in expenditure poverty rates (100 and 88% of the lower and upper bounds). In this case, family background on its own explained 35 (39)% of the gap, far above the contribution of education and living in rural areas—20 (17)% each—or number of children, 17 (14)%.

37

The percentage of the racial gap in income poverty explained by characteristics was also about 90% or even higher when using FGT(1) and FGT(2), with family background contributing 21–27% to the gap. Education played a similar role, explaining 9–11% (see Table A4).

38

The remaining 14% is accounted for by three other residual dimensions, not used for constructing the index, which are orthogonal with the main one, primarily explaining some rare profiles. The (negative) correlation of the individual composite indicator or deprivation with income and expenditure is 47 and 50%, respectively.

39

Tables A5 and A6 reports basic information about the MCA. The square correlation of dummy categories with the indicator was on average 0.85, with the largest values (above 0.95) for formal dwelling, DVD and microwave and the lowest values (between 0.6 and 0.7) for needs met insufficiently. The largest contribution was then made by the lack of a washing machine, microwave, vehicle, computer, DVD or piped water (between 0.040 and 0.055), and the lowest by the lack of a cellular phone or radio (0.001–0.005).

Appendix

Table A1:

Regressors: Average Values and Standard Errors of Continuous Variables

NIDS, 2008	Africans		Whites			Africans		Whites
	Mean	Standard error	Mean	Standard error		Mean	Standard error	Mean	Standard error
Western Cape	0.038		0.263	0.441	Household head (cont.)
Eastern Cape	0.155		0.043	0.204	Married (ref.)	0.407		0.721	0.449
Northern Cape	0.012		0.021	0.143	Single with partner	0.105		0.034	0.180
Free State/North West	0.224		0.106	0.308	Divorced/widow(er)	0.218		0.187	0.390
Kwazulu Natal	0.149		0.089	0.285	Never married	0.270		0.058	0.234
Gaunteg (ref.)	0.208		0.342	0.475	Immigrant (missing)	0.057		0.078	0.268
Mpumalanga	0.081		0.103	0.304	Non-immigrant (ref.)	0.719		0.491	0.500
Limpopo	0.133		0.032	0.177	Internal immigrant	0.196		0.343	0.475
Rural area	0.619		0.029	0.168	Immigrant from abroad	0.028		0.088	0.284
Number of children	2.230	(1.96)	0.758	0.945	5-year migrants	0.135		0.294	0.456
Number of adults	3.392	(2.10)	2.413	0.955	Years of schooling (missing)	0.024		0.038	0.190
Dependency ratio	0.578	(0.30)	0.334	0.302	Years of schooling	6.504	(4.69)	12.270	3.587
Average years of schooling	5.426	(2.96)	9.621	3.254	Labour status (missing)	0.122		0.212	0.409
% Adults in household					Not economically active (ref.)	0.330		0.186	0.389
Not economically active (ref.)	0.351	(0.34)	0.250	0.331	Discouraged unemployed	0.032		0.021	0.143
Discouraged unemployed	0.050	(0.14)	0.022	0.099	Strictly unemployed	0.092		0.028	0.165
Strictly unemployed	0.134	(0.23)	0.056	0.160	Formal employee	0.304		0.405	0.491
Formal employee	0.233	(0.31)	0.336	0.328	Self-employed	0.075		0.125	0.330
Self-employed	0.050	(0.16)	0.086	0.210	Casual employed	0.045		0.023	0.150
Casual employed	0.039	(0.13)	0.030	0.118	No occupation (or missing) (ref.)	0.635		0.462	0.499
Manager	0.008	(0.06)	0.064	0.178	Manager	0.012		0.092	0.290
Professional	0.027	(0.12)	0.113	0.222	Professional	0.034		0.132	0.339
Technician	0.006	(0.06)	0.071	0.176	Technician	0.006		0.057	0.232
Clerk	0.021	(0.11)	0.065	0.173	Clerk	0.022		0.063	0.243
Service worker	0.041	(0.14)	0.031	0.121	Service worker	0.047		0.026	0.158
Skilled farmer	0.014	(0.09)	0.004	0.041	Skilled farmer	0.021		0.005	0.070
Craft trade worker	0.045	(0.15)	0.076	0.187	Craft trade worker	0.062		0.127	0.333
Operator	0.030	(0.12)	0.016	0.081	Operator	0.047		0.023	0.149
Elementary occupation	0.087	(0.20)	0.014	0.080	Elementary occupation	0.113		0.013	0.113
Household head					Household head's parents
Female	0.520		0.231	0.422	Years of schooling (missing)	0.270		0.296
24 years old or less	0.053		0.034	0.181	Years of schooling (missing)	0.332	(0.47)	0.317	(0.47)
25–55 years old (ref.)	0.642		0.654	0.476	Years of schooling—mother	2.290	(3.76)	10.810	(2.99)
56+ years old	0.305		0.312	0.464	Years of schooling—father	2.003	(3.64)	11.088	(3.89)
PSLSD, 1993
Cape	0.062		0.259		Household head (cont.)
Transvaal (ref.)	0.204		0.553		Spouse present (ref.)	0.592		0.860
Orange Free State	0.059		0.072		Deceased spouse	0.248		0.037
Rest of the country	0.675		0.116		Absent spouse	0.099		0.038
Rural area	0.667		0.085		No spouse	0.061		0.065
Number of children	2.932	2.27	1.068	1.17	5-year migrants	0.069		0.214
Number of adults	4.631	2.51	2.817	1.21	Years of schooling (missing)	0.014		0.005
Dependency ratio	0.632	0.29	0.316	0.28	Years of schooling	4.524	4.12	11.826	3.50
Average years of schooling	4.187	2.40	8.393	3.33	Labour status (missing)	0.174		0.014
% Adults in household					Not economically active (ref.)	0.335		0.117
Not economically active (ref.)	0.449	0.29	0.255	0.31	Discouraged unemployed	0.038		0.004
Discouraged unemployed	0.068	0.17	0.012	0.07	Strictly unemployed	0.022		0.012
Strictly unemployed	0.042	0.13	0.014	0.08	Formal employee	0.335		0.739
Formal employee	0.209	0.27	0.514	0.35	Self-employed	0.063		0.100
Self-employed	0.038	0.13	0.070	0.19	Casual employed	0.033		0.013
Casual employed	0.028	0.10	0.028	0.10	No occupation (or missing) (ref.)	0.633		0.237
Professional	0.021	0.09	0.152	0.27	Professional/technical	0.026		0.241
Manager	0.002	0.02	0.109	0.22	Manager	0.004		0.162
Clerical/sales	0.020	0.09	0.140	0.23	Clerical/sales	0.026		0.102
Transport	0.016	0.08	0.014	0.08	Transport	0.040		0.027
Service	0.048	0.14	0.040	0.13	Service	0.061		0.044
Farming	0.017	0.10	0.004	0.05	Farming	0.033		0.008
Artisan	0.012	0.07	0.056	0.15	Artisan	0.021		0.108
Foremen	0.006	0.05	0.017	0.08	Foremen	0.012		0.040
Operator	0.024	0.10	0.012	0.07	Operator	0.039		0.025
Labourer	0.070	0.17	0.003	0.03	Labourer	0.105		0.004
Household head
Female	0.321		0.096
Age (missing)	0.031		0.002
24 years old or less	0.013		0.030
25–55 years old	0.563		0.795
56+ years old	0.393		0.173

Source: Own construction using PSLSD, 1993 and NIDS, 2008.

Table A2:

Logit Regressions of the Probability of Being White (versus African)

NIDS, 2008	Coefficient (1)	Standard errors	Coefficient (2)	Standard errors	PSLSD, 1993	Coefficient (3)	Standard errors
Western Cape	2.02	0.35	1.50	0.37	Cape	1.53	0.20
Eastern Cape	−0.89	0.48	−1.13	0.54	Orange Free State	−1.13	0.23
Northern Cape	0.85	0.42	0.52	0.44	Rest of the country	−2.20	0.22
Free State/North West	0.35	0.31	0.19	0.38	Rural area	−1.12	0.22
Kwazulu Natal	−0.95	0.42	−1.38	0.40	Number of children	−0.28	0.08
Mpumalanga	0.11	0.39	0.35	0.36	Number of adults	−0.33	0.07
Limpopo	−0.39	0.50	−0.33	0.45	Dependency ratio	−2.59	0.48
Rural area	−3.00	0.30	−2.73	0.30	Years of schooling	0.11	0.11
Number of children	−0.61	0.21	−0.84	0.24	Years of schooling2	0.00	0.01
Number of adults	−0.31	0.10	−0.19	0.09	% NEA	−3.00	0.44
Dependency ratio	−2.41	0.65	−3.33	0.72	% Discouraged unemployed	−6.00	0.85
Years of schooling	−0.12	0.23	−0.24	0.26	% Strictly unemployed	−5.84	0.89
Years of schooling2	0.01	0.01	0.01	0.01	% Formal employee	−8.34	2.00
% NEA	0.20	0.64	0.81	0.62	% Self-employed	−5.56	0.82
% Discouraged unemployed	0.07	1.06	0.23	0.94	% Casual employed	−7.22	2.12
% Strictly unemployed	−0.41	0.84	0.32	0.93	% Professional/technical	2.68	1.96
% Formal employee	−2.60	0.84	−2.61	0.92	% Manager	10.57	2.31
% Self-employed	−3.23	0.98	−3.02	1.10	% Clerical/sales	5.83	1.96
% Casual employed	−2.07	1.10	−3.24	1.40	% Transport	3.86	2.12
% Manager	4.42	2.16	1.88	1.33	% Service	1.93	1.85
% Professional	0.44	1.05	1.01	1.11	% Farming	2.35	2.10
% Technician	4.06	1.56	3.62	1.98	% Artisan	4.30	2.04
% Clerk	1.09	1.02	0.84	1.18	% Foremen	2.86	2.28
% Service worker	−0.68	1.17	−0.75	1.21	% Operator	1.50	2.25
% Skilled farmer	0.97	1.64	2.01	1.64	% Labourer	−2.91	2.34
% Craft trade worker	−0.61	1.05	−0.66	1.04	Household head
% Operator	0.19	1.73	0.35	1.50	Female	0.46	0.33
% Elementary occupation	−1.69	1.44	−1.91	1.51	25–55 years old	−1.07	0.39
Household head					56+ years old	−0.55	0.43
Female	−0.57	0.29	−0.56	0.28	Deceased spouse	−1.90	0.36
25–55 years old	−0.96	0.53	−0.69	0.56	Absent spouse	−2.37	0.36
56+ years old	−0.09	0.58	0.41	0.58	No spouse	−1.73	0.32
Single with partner	−2.10	0.46	−2.20	0.41	5-year migrants	1.14	0.21
Divorced/widow(er)	0.04	0.31	0.03	0.32	Years of schooling	−0.21	0.08
Never married	−2.78	0.40	−3.07	0.40	Years schooling2	0.03	0.00
Internal immigrant	−0.16	0.27	−0.28	0.26	Discouraged unemployed	−0.22	0.76
Immigrant from abroad	−0.95	0.43	−1.51	0.65	Strictly unemployed	1.26	0.62
5-year migrants	−0.07	0.28	−0.09	0.32	Formal employee	0.54	2.05
Years of schooling	1.06	0.22	1.02	0.22	Self-employed	1.47	0.44
Years of schooling2	−0.03	0.01	−0.03	0.01	Casual employed	−0.94	2.23
Discouraged unemployed	−0.03	0.99	0.72	0.72	Professional/technical	0.67	2.03
Strictly unemployed	0.25	0.78	−0.12	0.76	Manager	−0.31	2.10
Formal employee	0.77	0.61	0.85	0.83	Clerical/sales	−0.68	2.07
Self-employed	2.09	0.60	1.96	0.74	Transport	−0.85	2.12
Casual employed	−0.05	1.06	0.91	1.49	Service	−0.21	1.98
Manager	−2.14	1.06	−0.72	0.85	Farming	−0.29	1.98
Professional	−0.60	0.72	−1.22	0.89	Artisan	0.67	2.06
Technician	−1.25	0.97	−1.26	1.22	Foremen	0.99	2.13
Clerk	−0.79	0.72	−0.86	0.83	Operator	0.44	2.18
Service worker	−1.09	0.82	−1.40	1.00	Labourer	−0.07	2.17
Skilled farmer	−0.16	1.07	0.35	1.23	Intercept	4.23	0.77
Craft trade worker	0.74	0.72	0.81	0.82
Operator	−1.50	1.23	−1.44	1.28
Elementary occupation	−1.51	0.99	−1.29	1.02
Years of schooling (mother)			0.54	0.19
Years of schooling2 (mother)			−0.02	0.01
Years of schooling (father)			−0.09	0.13
Years of schooling2 (father)			0.02	0.01
Intercept	−3.81	1.57	−5.54	1.72
Pseudo-R2	0.627		0.712			0.740
Wald χ2(39; 61;34)	444.03		391.12			1,200.04
Prob. > χ2	0		0			0
Number of observations	23,582		23,582			39,171

Source: Own construction using PSLSD, 1993 and NIDS, 2008.

Notes: Some dummies have been added for variables with many missing values. Reference: married male household head, 15–24 years old, non-migrant, formal employee in elementary occupation, in urban Gauteng for NIDS sample (Transvaal for PSLSD sample). Estimated robust standard errors took into account individuals being ‘clustered’ across families.

Table A3:

Racial Poverty Gap between Africans and Whites in South Africa, FGT(1) (Lower and Upper Poverty Lines) in 1993–2008

	FGT(1)								FGT(2)
	Lower poverty line				Upper poverty line				Lower poverty line				Upper poverty line
	2008	% Differential	1993	% Differential	2008	% Differential	1993	% Differential	2008	% Differential	1993	% Differential	2008	% Differential	1993	% Differential
Whites	0.9		1.6		2.1		2.3		0.6		7.4		1.1		3.5
Africans	26.7		41.7		45.7		59.3		16.0		29.0		31.6		45.4
Differential	25.8		40.0		43.6		57.0		15.4		21.7		30.4		41.9
Counterfactual	2.6		0.8		8.9		1.9		1.3		0.4		4.8		1.1
Unexplained	1.7	6.7	−0.8	−2.1	6.8	15.6	−0.4	−0.6	0.7	4.7	−6.9	−32.0	3.7	12.0	−2.4	−5.8
Explained (all characteristics)	24.1	93.3	40.9	102.1	36.8	84.4	57.4	100.6	14.6	95.3	28.6	132.0	26.8	88.0	44.4	105.8
Geographic	7.2	27.9	6.9	17.2	9.5	21.7	6.7	11.7	4.4	28.6	5.5	25.2	7.5	24.7	6.5	15.5
Province	1.1	4.3	4.2	10.5	0.3	0.7	3.7	6.6	0.6	3.9	3.5	16.2	0.7	2.3	3.9	9.2
Rural	6.1	23.6	2.6	6.6	9.1	20.9	2.9	5.2	3.8	24.7	2.0	9.1	6.8	22.4	2.6	6.3
Demographic	6.9	26.7	7.0	17.5	9.9	22.7	9.0	15.8	4.4	28.5	4.8	22.4	7.5	24.5	7.3	17.4
Head's marital status	1.0	3.9	0.2	0.4	1.1	2.4	0.0	0.0	0.8	5.5	0.2	0.9	0.9	3.1	0.1	0.3
Head's immigration	−0.5	−1.9	0.4	1.0	−1.3	−2.9	0.7	1.2	−0.2	−1.5	0.2	0.8	−0.7	−2.4	0.5	1.1
Head's sex	1.0	4.0	−0.4	−1.0	2.2	5.1	−0.6	−1.0	0.6	4.0	−0.3	−1.3	1.4	4.7	−0.4	−1.0
Head's age	−0.6	−2.4	−0.3	−0.6	−2.1	−4.7	−0.6	−1.1	−0.3	−1.7	−0.1	−0.3	−1.1	−3.7	−0.3	−0.8
Number of children	4.9	19.1	4.0	10.1	7.8	17.8	5.0	8.7	2.9	18.7	2.9	13.4	5.6	18.3	4.1	9.9
Number of adults	1.0	4.0	3.0	7.6	2.2	5.0	4.5	8.0	0.6	3.6	1.9	8.8	1.4	4.5	3.3	8.0
Education	6.2	24.0	16.5	41.2	12.6	28.9	24.9	43.7	3.4	22.0	10.9	50.5	7.9	25.9	18.5	44.2
Labour	3.8	14.7	10.5	26.2	4.9	11.2	16.7	29.4	2.5	16.2	7.3	33.9	3.9	12.8	12.1	28.8
Labour status	3.3	12.6	−3.0	−7.6	3.0	6.9	−4.4	−7.8	2.3	15.1	−1.9	−8.9	2.9	9.5	−3.3	−8.0
Occupation	0.5	2.1	13.5	33.8	1.9	4.3	21.2	37.2	0.2	1.2	9.3	42.7	1.0	3.2	15.4	36.7

Source: Own construction using PSLSD, 1993 and NIDS, 2008.

Table A4:

Racial Poverty Gap between Africans and Whites in South Africa with Family Background, FGT(1) and FGT(2) (Lower and Upper Poverty Lines) in 2008

	Lower poverty line				Upper poverty line
	FGT(1)	% Differential	FGT(2)	% Differential	FGT(1)	% Differential	FGT(2)	% Differential
Whites	0.9		0.6		2.1		1.1
Africans	26.7		16.0		45.7		31.6
Differential	25.8		1.2		43.6		3.9
Counterfactual	2.7		15.4		6.6		30.4
Unexplained	1.8	7.0	0.6	3.9	4.5	10.3	2.7	9.0
Explained (all characteristics)	24.0	93.0	14.8	96.1	39.1	89.7	27.7	91.0
Geographic	6.5	25.2	3.7	24.2	9.4	21.4	7.1	23.2
Province	0.8	3.0	0.5	3.1	0.0	0.1	0.5	1.5
Rural	5.7	22.2	3.2	21.1	9.3	21.3	6.6	21.7
Demographic	6.0	23.1	3.8	24.9	9.0	20.6	6.7	21.9
Head's marital status	0.7	2.9	0.7	4.5	1.0	2.3	0.8	2.8
Head's immigration	−1.2	−4.7	−0.6	−3.8	−2.7	−6.2	−1.6	−5.3
Head's sex	0.9	3.6	0.5	3.3	1.8	4.2	1.2	3.9
Head's age	−0.7	−2.6	−0.2	−1.2	−2.2	−4.9	−1.1	−3.8
Number of children	5.6	21.8	3.1	20.2	9.8	22.5	6.7	21.9
Number of adults	0.6	2.2	0.3	1.9	1.2	2.7	0.7	2.4
Education	2.4	9.3	1.3	8.4	5.0	11.5	3.1	10.1
Labour	3.6	14.0	2.8	17.9	3.9	9.0	3.5	11.5
Labour status	3.4	13.3	2.8	18.0	2.6	6.1	2.9	9.5
Occupation	0.2	0.8	0.0	0.0	1.3	3.0	0.6	1.9
Family background	5.5	21.4	3.2	20.7	11.8	27.1	7.4	24.3

Source: Own construction using NIDS, 2008.

Table A5:

MCA: Deprivation Composite Indicator of Africans and Whites in South Africa- Burt/Adjusted Inertias

Dimension	Principal inertia	Percent	Cumulative percent
Dimension 1	0.07608	86.33	86.33
Dimension 2	0.00500	5.67	92.00
Dimension 3	0.00064	0.72	92.72
Dimension 4	0.00054	0.61	93.33
Total	0.08812	100
(22,193 observations)

Source: Own construction using NIDS, 2008.

Table A6:

MCA: Deprivation Composite Indicator of Africans and Whites in South Africa - Statistics for Column Categories in Standard Normalisation (First Dimension)

Categories		Coordinate	Square correlation	Contribution	Categories		Coordinate	Square correlation	Contribution
Formal dwelling	No	0.586	0.966	0.011	Healthcare	No	0.611	0.639	0.01
	Yes	−1.56	0.966	0.03		Yes	−0.849	0.639	0.014
Piped water	No	1.513	0.928	0.041	Schooling	No	0.489	0.658	0.008
	Yes	−0.997	0.928	0.027		Yes	−1.141	0.658	0.018
Flush toilet	No	1.336	0.862	0.039	Radio	No	0.239	0.858	0.002
	Yes	−1.211	0.862	0.035		Yes	−0.538	0.858	0.004
Electricity	No	0.509	0.895	0.009	TV	No	0.704	0.914	0.015
	Yes	−1.931	0.895	0.035		Yes	−1.533	0.914	0.034
Landline telephone	No	2.678	0.936	0.035	VCR/DVD	No	1.6	0.959	0.04
	Yes	−0.324	0.936	0.004		Yes	−0.837	0.959	0.021
Cellphone	No	0.119	0.86	0.001	Computer	No	2.75	0.919	0.044
	Yes	−0.978	0.86	0.005		Yes	−0.401	0.919	0.006
Rubbish collection	No	1.169	0.835	0.031	Electric/gas stove	No	0.744	0.916	0.017
	Yes	−1.189	0.835	0.032		Yes	−1.49	0.916	0.034
Street lighting	No	1.245	0.85	0.028	Microwave	No	1.768	0.954	0.048
	Yes	−0.803	0.85	0.018		Yes	−0.891	0.954	0.024
Food	No	0.676	0.681	0.013	Fridge/freezer	No	0.959	0.929	0.024
	Yes	−1.042	0.681	0.019		Yes	−1.319	0.929	0.033
Housing	No	0.675	0.66	0.013	Washing machine	No	2.306	0.928	0.055
	Yes	−1.035	0.66	0.019		Yes	−0.684	0.928	0.016
Clothing	No	0.679	0.631	0.012	Motor vehicle	No	2.329	0.925	0.048
	Yes	−0.951	0.631	0.017		Yes	−0.557	0.925	0.011

Source: Own construction using NIDS, 2008.