Inequality, poverty and prospects for redistribution

Abstract

This paper demonstrates that poverty and inequality trends can diverge. It then discusses inequality trends and shows that, despite measurement issues, there is consensus that inequality is very high and has been rising over much of the post-transition period. Due to rising inequality within all groups, and particularly the black population, and lower inequality between race groups, within-group inequality has become the dominant form of inequality. That does not, however, detract from the fact that inequality between groups is still very large. High income inequality largely stems from inequality in access to wage income, due more to wage inequality than to unemployment. A Gini coefficient for wage income amongst the employed of above 0.60 effectively sets a floor to overall income inequality. The high wage premium to educated workers derives from a combination of a skills shortage at the top end of the educational spectrum, driving up their wages, and a surfeit of poorly-educated workers competing for scarce unskilled jobs dampening unskilled wages; if the unemployed were to find jobs, it would be in this bottom part of the wage distribution, and consequently this would not much reduce wage inequality. A continuation of the historical pattern whereby only a small segment of the population obtained good schooling would leave the structures underlying the large wage premium unaltered. The time frame for substantial inequality reduction is thus necessarily a long one, while poverty reduction efforts should not wait for this to occur.

Keywords:

• inequality
• income distribution
• poverty
• South Africa

[T]here is not one distribution of income but many: income is distributed across racial groups, income classes, present and future generations, and so on. Moreover, a given distribution is not a one-dimensional magnitude: it has as many dimensions or components as there are relevant ‘classes’. (Bromberger, 1982:166)

1. Introduction

Income inequality is a matter of great concern in South Africa. But so, indeed, is poverty. This paper focuses mainly on the former, but places emphasis on the relationship between the two, which is too little understood, thus also leading to policy discussions that take too little cognisance of the different policies and time frames required to address these two phenomena. The paper also sets out to show not only trends in inequality, but also how persistent it is likely to be, given its strong roots in labour market (wage) inequality and the inordinately high wage Gini coefficient of 0.60 amongst wage earners. This, it is argued, is likely to remain very stubborn given how strongly it is related to the returns to education and the quality thereof. Thus prospects for a reduction in income inequality are inauspicious except over a very long time frame – but that does not mean that poverty cannot be addressed more directly in the meantime.

Although South Africa is an upper-middle-income country judged by economic factors (such as gross domestic product per capita or the structure of the economy), its social indicators such as life expectancy, infant mortality or quality of education more closely resemble those of a lower-middle-income or even a low-income country. This reflects the fact that resources and human capital are highly inequitably distributed in South Africa. A relatively small group of high-income earners sharply increase average incomes, but can have little impact on average life expectancy. In 1995 (before the full impact of AIDS), South African life expectancy at birth was only 63–10 years less than that of Panama, a country of comparable income, and four years less than the Philippines, a country with one-third of South Africa's per-capita income (World Bank, 1997). White South Africans enjoy living standards comparable with the average in the richest countries in the world, while the poorest one-fifth of South Africans (quintile 1) have average purchasing power similar to that of the average citizens in the very poorest countries (although they do enjoy other benefits from living in a middle-income country). Averages can be misleading, but these do illustrate that the living circumstances of the poorest and richest South Africans are worlds apart. The shared economy and policy environment make inequality potentially inflammatory. The racial dimension of inequality necessarily thus receives much attention, although the opening quote reminds us that inequality has many different dimensions.

In the public mind, and not least amongst policy-makers, there appears to be confusion between trends in distribution and trends in poverty. Also, few people apart from researchers specialising on issues of poverty and inequality measurement have a good perspective on what is known about inequality trends. For this reason the paper first provides a brief overview of issues regarding inequality measurement generally and in the South African context, and then discusses a conceptual framework for considering the relationship between poverty and inequality. The widespread conflation of the two phenomena means that it is quite common not to distinguish between policies aimed at poverty reduction – such as social grants and job creation – and those intended to deal with inequality, such as educations and the returns associated with it. Even sophisticated analyses such as that by the National Planning Commission often fail to distinguish between the time frames for addressing poverty and those needed to reduce inequality. A central message of this paper is that even medium prospects for redistributing income are limited, but that this does not imply that poverty reduction needs to wait or cannot occur in a much shorter time frame.

The conceptual distinction between poverty and inequality is followed by a review of the literature on quantitative dimensions of inequality. This provides a perspective of high and stubborn inequality, although the between-group dimension has been somewhat muted since the transition. Decomposing inequality into its components, one finds that it is driven largely by wage patterns in the labour market. These labour market inequalities, it is concluded, will not improve unless there is a considerable improvement in the quality of education received by the bulk of the population. In a labour market where a premium on skills is likely to remain and where the demand for unskilled and uneducated or poorly educated workers remains slack, prospects for reducing income inequality will remain dim. Government interventions in the labour market cannot overcome this and may well have large negative side-effects, while social spending offers no solution to further mitigating the impact of massive inequality in the economy and labour market.

2. Inequality measurement

Amartya Sen, in Inequality Re-examined, contends that ‘ … every normative theory of social arrangement that has at all stood the test of time seems to demand equality of something – something that is regarded as particularly important in that theory’ (Sen, 1992:13; original emphasis). Equity concepts differ in terms of in what space (dimension) equality is perceived. One can make a strong argument for equality of opportunity rather than equality of income as the space in which equality should be sought. Also, in a Rawlsian sense, social justice demands greatest attention to reducing poverty, not reducing inequality. However, given South Africa's history of inequality (Terreblanche, 2002) and a newly enfranchised population, perceived equity has come to be identified with addressing the legacy of high income inequality.

It is not straightforward to measure progress in reducing inequality. One reason for this arises from the fact that inequality means comparisons within a population, and, depending on who is to be compared with whom, a wide variety of measures of inequality can be used, with varying strengths and weaknesses. The axioms typically used to assess the qualities of distributional measures (Litchfield, 1999; Fields, 2000; Deaton, 1997; World Bank, 2004) make it clear that most inequality measures do not meet the criteria set for good distributional indicators.

A second reason why it is difficult to measure progress in reducing inequality relates to the sheer difficulty of obtaining accurate data for making valid comparisons. Discussing income distribution in South Africa, Charles Simkins states:

Measurement is crucial. There are several sources of information, all of them incomplete and defective to some degree. … The art of measurement lies in painstaking reconciliation of information (which is sometimes years in arrears), supplemented by the use of judgement. Up to now, the margin of error has been much wider than one would like, making estimates controversial. (Simkins, 2000:13)

Indeed, measurement difficulties in the inequality field are great. To give but two South African examples, first the proportion of national accounts current income directly captured in the census income rose from 42% in 1996 to 65% in 2001 and to 91% in the 2007 Community Survey, rendering comparisons across these censuses/surveys highly suspect (Yu, 2009:46). Furthermore, while the recall method was used in earlier surveys to obtain food expenditure values, the Income and Expenditure Survey 2005 (IES2005) and also the Income and Expenditure Survey 2010/11 (IES2010/11) changed to a weekly diary that had to be kept for four weeks, giving rise to respondent fatigue that was especially noticeable amongst higher income respondents. Consequently, recorded food expenditure fell from 18.3% to 9.6% of all expenditure over a five-year period, caused by a combination of an unlikely 14% recorded decline in real food expenditure, while recorded aggregate consumption expenditure increased by 64% (Yu, 2008:16, Table 10).

In recent years there has been an avalanche of new data sources for estimating trends in poverty and inequality. The trouble now lies less in data availability and more in validity and comparability issues, resulting from differences in samples (including sampling frames, sample attrition and non-responses), definitional changes in the survey instruments, and differences in how field workers interpret and apply certain definitions from survey to survey.

3. Inequality and poverty: A conceptual framework

It is common to talk about inequality and poverty in the same breath, as if reducing the one necessarily also means reducing the other. This need not be the case, as this section will illustrate, with the focus here simply on the poverty headcount ratio rather than more distributionally sensitive poverty measures.

Consider Figure 1, which shows hypothetical income distributions, with the log of per-capita income on the horizontal axis and the density of population on the other. A kernel density curve of this sort is to be read just like a histogram: the heights of the curve shows where the largest proportion of the population is concentrated. Most income distributions are usually approximately log-normally shaped, thus producing a normal (bell-shaped) distribution when the log of income is shown on the horizontal axis.

Figure 1: Relationship between poverty and inequality for a hypothetical log-normal income distribution

Source: Yu (2012:255, Figure 6.4).

When income distribution becomes more unequal, the distribution widens, as is illustrated by the change from the narrower initial curve in Figure 1 to the thicker continuous curve, with more people having very high incomes and also more having very low incomes. At the same time, however, the distribution for this thicker curve also lies more to the left on the graph, as the mean income is maintained but the mean log of income declines to keep average (unlogged) income unchanged.² The Gini coefficient in the original graph (the curve represented by the thin line) was 0.40, and it was increased to 0.60 to generate the thick line. The curve represented by the thick dotted line illustrates what would happen if average incomes were also to improve: If all incomes were to increase by the same proportion (i.e. if the income distribution remained unchanged), the whole density distribution would move to the right, as the new line shows.

Now let us also consider poverty. The poverty line is drawn at a level that originally places 25% of the population in poverty. The proportion of the population that is below the poverty line would increase greatly to 48% when the Gini coefficient increases (i.e. when we move from the thin line to the thick line). However, economic growth that keeps distribution constant would move the whole curve to the right, as in the shift from the thick line to the thick dotted line. The net effect of worsening distribution and economic growth would depend on the relative sizes of the effects: how much worse did distribution become, and how much growth was experienced? In this particular example, the dotted line is drawn such that the poverty headcount rate would again return to its original level of 25%. Given the starting levels of inequality and the large shift in inequality illustrated in these graphs, from a Gini coefficient of 0.40 to 0.60, and a poverty line set at the 25th percentile of initial incomes, the rise in average incomes required to reduce poverty to its initial level would be 108%; that is, more than a doubling of income would be required. The impact on poverty of large increases in inequality is thus considerable. However, substantial economic growth, as has occurred in most modern economies in the long run or in rapidly growing economies such as China in shorter periods, almost always dominates the effect of even sharply rising inequality; that is, it reduces poverty even when distribution worsens. One can see in the figure that even massively rising inequality would not shift more people below the poverty line if average incomes rise enough to shift the income distribution far enough to the right.

It is useful, finally, to consider what would have happened if the poverty line was initially drawn at a much higher level; that is, considerably towards the right of the mode (centre) of the distribution. In such a case, greater inequality may actually move some people above the poverty line; that is, inequality may be good for poverty reduction. The part of the population above the poverty line increases both when inequality increases and when there is economic growth. This is one reason why it is wise not to set poverty lines too high: with a poverty line set above the modal income (the centre of the normal curve), greater inequality can appear to be good for poverty reduction. (Of course, a less smooth distribution – that is, where the density curves have humps – may mean that smaller distributional changes combined with economic growth could have somewhat perverse effects.)

In the South African situation, economic growth clearly has had beneficial effects for poverty reduction after the transition. However, worsening income distribution could have had an opposite effect on poverty. Although the net effect of growth and distribution on poverty was beneficial after 2000, according to most analysts (see discussion below) worsening income distribution in the first part of the transition reduced the benefits of growth.

A ‘poverty line’ set at a high level can also be used to illustrate another phenomenon with which South Africans are familiar: economic growth and worsening distribution moved a larger part of the population above what can be considered a middle-class threshold, which can be represented by a high poverty line above which poverty concerns are not so pressing.

There is thus an important and nuanced relationship between inequality and poverty. Economic growth shifts distributions to the right, with the effect that poverty would decline unless income distribution worsens drastically. If income distribution remains unchanged and economic growth takes place, all would benefit from such growth and poverty would decline, even if inequality remains at a high level. It is possible – and may also have occurred after the turn of the century – that income inequality can actually worsen while poverty is substantially reduced as a result of growth that reaches some of the poor. However, the more growth there is, the less likely it is that growing inequality will prevent its beneficial effects from reducing poverty.

4. Trends in income inequality: Aggregate, inter-racial and intra-racial

In 1971, Spandau wrote that ‘The main distinguishable feature of the distribution of income by race is its relative constancy during the 35 year period 1924/25 to 1960′ (Spandau, 1971:185). This stability in racial income shares continued until about 1970, despite a rising black population share. However, the black share of income increased drastically from 22% to almost 41% in 2007 (Table 1), bringing to an end a period of mainly widening income inequality between whites and blacks. According to the estimates shown in the table, black per-capita incomes had grown from 8% of those of whites in 1970 to almost 12% in 2000. Although estimates from the 2007 Community Survey appear to show the white–black ratio rising again, this may again be due to comparability issues between the specific surveys selected to present here, rather than an underlying trend. Table 1 probably only broadly approximates the exact levels of the racial incomes for the different years, given the widely divergent data sources that had to be reconciled to extract this table, but the broad trend is probably correct for the two largest race groups. What the data seem to indicate is that black incomes have increased steadily at a rate of around 2.2% per year since 1975, and that similar (although more volatile) growth was experienced by the white population. Thus the racial gap has not improved much after the rapid spurt in black incomes between 1970 and 1975. Although black incomes did not continue to grow very rapidly for any period since 1975, sustained more modest growth did allow incomes to double over this full period.

Table 1: Estimates of total and per-capita income, 1970–2000

	1970	1975	1980	1985	1990	1995	2000	2007
All income in millions (2000 Rand values)
Blacks	50 454	82 683	108 131	132 553	170 630	212 294	252 071	346 964
Coloureds	17 513	20 999	23 854	29 436	36 784	44 417	53 111	67 081
Indians	6 298	8 361	10 892	13 635	17 181	21 480	26 396	47 332
Whites	151 963	189 279	211 504	235 605	260 835	277 411	326 307	396 329
Total	226 228	301 321	354 381	411 230	485 429	555 601	657 884	857 706
Share of income (%)
Blacks	22.3	27.4	30.5	32.2	35.2	38.2	38.3	40.5
Coloureds	7.7	7.0	6.7	7.2	7.6	8.0	8.1	7.8
Indians	2.8	2.8	3.1	3.3	3.5	3.9	4.0	5.5
Whites	67.2	62.8	59.7	57.3	53.7	49.9	49.6	46.2
Total	100	100	100	100	100	100	100	100
Per-capita income (2000 Rand values)
Blacks	3 134	4 479	5 107	5 423	6 008	6 704	7 283	9 070
Coloureds	8 184	8 630	8 822	9 855	11 404	12 722	14 126	15 331
Indians	9 595	11 244	13 296	15 113	17 637	20 592	23 938	38 028
Whites	39 217	44 242	46 670	48 370	51 951	53 840	62 360	85 661
Total	9 936	11 626	12 125	12 385	12 903	13 436	14 716	17 684

Sources: Data up to 2000 taken from Van der Berg & Louw (2004), using estimates derived from many sources; for 2007, an author's own estimate was added, based on applying distributional shares derived from the Community Survey 2007, after applying sequential regression multiple imputation. All estimates use racial shares applied to national accounts-adjusted current household incomes.

The income of any group consists of wages (the product of the average wage and the number employed) and other income, comprising income from the other factors of production (capital, land and entrepreneurship) and income from transfers (in South Africa, social grants). Of these income components, the most important for investigating changes in inter-group distribution are wage levels, employment levels (relative to the whole population), and social grants.

The relationship can be written as follows:

and thus:

where Y is the income of a group, W is the mean wage, E I the number employed, Y_o is income from other sources (such as land rent and dividends), and P is population. This relationship can be interpreted in the following way: if the average wage or the number of people employed rises compared with the population of that group as a whole, then income for that group is likely to increase. Other sources of income, largely derived from assets (land and investments), are less likely to change much in a short time span, given that asset accumulation is a slow process. The exception to this is social grants.

Using this framework, it is now possible to investigate the development of inter-group inequality in the post-1970 period. The rapid increase in black wages from the 1970s moderated in the 1990s, and there are now only modest rises in wages, largely commensurate with the rising skill levels and productivity of the labour force. On the other hand, the black population – historically the least educated – had been most affected by the rising unemployment from the 1970s. These two opposing forces restricted black per-capita income growth. Employment growth accelerated somewhat in the 1990s, but could only keep up with the growth of the population of working age (see discussion of the labour market below).

Social grants for the black population also expanded, with a trend towards grant equalisation from the mid-1970s, so that grant values were already fully equalised at the time of the transition. A further massive expansion of social grants after the turn of the century offered an important source of income for poorer people, reaching parts of the population who are poorly linked to the labour market. This made possible a rise in income at the bottom of the distribution that would not have occurred through market forces alone. The real value of grants paid increased by R600 per member of the whole population (in 2000 Rand terms) in the eight years after the turn of the century, which has had a considerable effect at the bottom of the income distribution. However, the effect of social grants on inequality is necessarily modest – even after rapid expansion it only comprises about 3.5% of gross domestic product, and Armstrong & Burger (2009:17) found that social grants reduced an inequality measure (Generalised Entropy 2, one-half of the square of the coefficient of variation) by only 1%.

There has also been an important shift in ownership of assets that generate income (dividends and land rent). Nevertheless, direct asset ownership by those formerly excluded from the economic mainstream is still relatively small and asset income does not make a great contribution to aggregate black incomes.

5. Income distribution across the full population

An official report published in 2002 by StatsSA compared IES data for 1995 and 2000, and found only small a small increase in the Gini coefficient from 0.56 to 0.57 (StatsSA, 2002). Hoogeveen & Özler (2006:87), using the same data sources, found evidence of an increase in the incidence of more extreme poverty between 1995 and 2000, but that more moderate poverty remained unchanged. They attributed the small rise in the Gini coefficient to it being most sensitive to changes in the middle of the distribution. Employing an inequality measure more sensitive to changes at the lower end of the distribution, namely the mean logarithmic deviation, Hoogeveen & Özler (2006:73) found greater evidence of a rise in inequality. In an earlier paper these authors indeed argued that the rates of return to education, how people are rewarded for their education in the labour market, increased only for highly educated individuals in urban areas between 1995 and 2000 (Hoogeveen & Özler, 2004).

Leibbrandt et al. (2010a), although utilising the same datasets, focused on the income of individuals aged 18 and older, rather than on households. They found that individual incomes declined from 1995 to 2000 by an average of 40%, which appears inordinately large and, as they noted, was inconsistent with national accounts trends. They also suggested that the main reason for the increase in poverty was a change in the returns to endowments (e.g. how well education was rewarded in the labour market); in particular, they found falling returns to education for black people, contrasting with rising returns to education for whites. They argue that ‘Perhaps the most persuasive explanation of the evidence is substantial economic restructuring of the South African economy in which wages are not bid up to keep pace with price changes due to a differentially slack labour market’ (Leibbrandt et al., 2010a:1). However, this appears to be contradicted by results from the Labour Force Surveys.

Simkins (2004) performed analyses on the 1995 and 2000 IESs after making some adjustments to the data where they appeared to be incorrect or incomplete. His research indicates that inequality increased substantially between 1995 and 2000. Van der Berg & Louw (2004) analysed the post-apartheid income distribution using the IES datasets for 1995 and 2000 and adjusting these to the national accounts, and found a fairly small increase in inequality within race groups.

However, as the IES1995 and IES2000 datasets are particularly problematic for comparing the income distribution across years, we turn our attention to studies that do not rely on the IES datasets for inference.

Leibbrandt and co-authors analysed data from 10% samples of the 1996 and 2001 censuses, focusing on both income and access poverty. According to them, real household income among the more affluent increased at the same time as poverty increased, and thus there was a rise in income inequality, driven mainly by increasing inequality within race groups (Leibbrandt et al., 2006:102–3). The UNDP's 2003 Human Development Report for South Africa (UNDP, 2003) also suggested that inequality was worsening, as the estimated Gini coefficient rose from 0.596 in 1995 to 0.635 in 2002, but they did not indicate their data sources.

Yu attempted to obtain comparability in definitions and data analysis across the three IES surveys (his results are summarised in Table 2) and concluded that ‘ … there was an evident increase of Gini coefficient between IES1995 and IES2000, while the IES2000 and IES2005 Gini coefficient values were very similar, regardless of the income categorization method used‘ (Yu 2008:29).

Table 2: Gini coefficient using annual per-capita income

	COICOP	Post SRMI
IES1995	0.66	–
Census 1996	–	0.69
IES2000	0.71
Census 2001	–	0.76
IES2005 (excluding imputed rent)	0.72	–
IES2005 (including imputed rent of dwelling of 7% per year)	0.72	–
Community Survey 2007	–	0.74

Notes: IES2005 was classified according to the new structure of national accounts of the United Nations. To make comparisons over time, re-classification of some items in accordance with the old Standard Trade Classification or the new COICOP (Classification of Individual Consumption According to Purpose) was necessary for all three IES survey. SRMI = sequential regression multiple imputation.

Sources: Yu (2008:20, Table 12; 2009:44, Table 22).

To derive the Gini coefficients in Table 2 from the censuses of 1996 and 2001 and the Community Survey of 2007, Yu (2009) used sequential regression multiple imputation whereby values were imputed for the many zero reported incomes and missing incomes, after considering other characteristic of those households. He found a strong increase (seven percentage points) in the Gini coefficient between 1996 and 2001. Supporting evidence was found in other studies employing alternative measures: Leibbrandt et al. (2006) found an increase in the Gini coefficient from 0.68 to 0.73 using one method, and from 0.74 to 0.79 using another; Simkins (2004) found that the Gini coefficient for households grew from 0.66 to 0.69; and Ardington et al. (2005) found that the Gini coefficient rose from 0.74 to 0.82. Other confirmation comes from the IES surveys. From 1995 to 2000, there was a rise of almost five percentage points in the Gini coefficient. The period after 2000 shows a less clear trend in most inequality estimates. Between Census 2001 and the Community Survey of 2007, there was a minor (one point) decline in the Gini coefficient. According to the IES, a further one-half of a point rise took place between 2000 and 2005. The All Media Products Survey (AMPS) and October Household Survey (OHS)/Labour Force Survey data point to a stable Gini coefficient for both income (AMPS) and wages respectively.

There is thus agreement about inequality trends, although the levels vary widely depending on the technique employed to deal with some data and measurement issues. This can be seen in Figure 2, which shows a number of different estimates based on published datasets and the estimates of Yu (2008). Levels of the Gini coefficient shown here lie between 0.612 and 0.826, using relatively similar methods of calculating these Gini coefficients.³

Figure 2: Gini coefficients for income, expenditure or consumption derived from various data sources and using alternative techniques

Source: Own calculations from OHS 2005 & LFS 2005b.
Notes: STC = Standard Trade Classification; COICOP = Classification of Individual Consumption According to Purpose; SRMI = sequential regression multiple imputation; NIDS = National Income Dynamics Study; PSLSD = Project for Statistics on Living Standards and Development; OHS = October Household Survey; GHS = General Household Survey; LFS = Labour Force Survey; IES = Income and Expenditure Survey; CS = Community Survey.

There was thus probably a strong upward trend in inequality as measured by the Gini coefficient in the second half of the 1990s, and largely stable inequality since. Clearly, inequality is high, but how high is anybody's guess. As intimated earlier when discussing measurement errors relating to assets, high incomes are probably underestimated. However, low incomes may also be underestimated (preliminary analysis of teacher salaries reported in surveys point to about a 30% undercount relative to what the state pays them), and thus the net effect of mismeasurement on inequality is not clear.

6. Inequality within groups

Estimates derived from varied data sources for intra-group inequality show a fair degree of consistency (Table 3). Inequality seems to have been rising in all race groups.

Table 3: Gini coefficients estimated from various surveys and censuses for intra-group distribution (income or expenditure per capita), 1970–2000

	Blacks	Coloureds	Indians	Whites	Total
1970	–	0.53	0.42	0.43	–
1975	0.49	–	–	–	–
IES1995	0.57	0.52	0.49	0.47	0.66
IES1995 (expenditure)	0.58	0.52	0.49	0.47	–
Census 1996	0.68	0.57	0.53	0.52	–
Census 1996 (post SRMI)	0.62	0.53	0.48	0.46	0.69
IES2000	0.59	0.55	0.51	0.49	–
Census 2001 (post SRMI)	0.65	0.60	0.58	0.57	0.76
Community Survey 2007 (post SRMI)	0.66	0.62	0.61	0.56	0.74

Sources: Van der Berg & Louw (2004); Yu (2009:41, Table 21).

Notes: SRMI, sequential regression multiple imputation.

If the Gini coefficient is rising for all groups, why is it not rising for South Africa as a whole? The answer lies in the fact that income inequality between race groups has been declining for most of this period. It is here where the deficiency of the Gini coefficient, which is not decomposable between groups whose incomes overlap, is an issue. The Gini coefficient is highest amongst the black population, although it is also high in other groups. For all provinces, inequality is high – and in some cases (e.g. Kwazulu-Natal) it is even higher than for the different race groups or for the population as a whole.

This rising inequality within the less privileged groups is partly explained by the fact that the historically largely white middle class has been joined by large numbers of coloureds, Indians and particularly now also blacks, so that the dividing line between the affluent and the rest of the population is no longer race – although race is self-evidently still a major determinant of affluence. However, the poor are still overwhelmingly black and rural.

7. A further note on poverty trends

Although this paper does not deal explicitly with trends in poverty per se, some of these have been mentioned as part of the discussion above, and a brief summary may be in order here. There appears to be a broad consensus that poverty has declined after the turn of the century, although this downward trend may have come to an end with the global recession of 2008. There is disagreement, however, on how rapidly poverty has declined. Again, the source of the disagreement has much to do with the comparability of data sources. Comparability of data sources is probably also the culprit for somewhat mixed evidence on the poverty trends in the latter half of the 1990s, although the weight of evidence seems to indicate worsening poverty in this period.

Long-term poverty movements may perhaps be less sensitive to choice of data sources, as shorter term fluctuations and minor movements in the data have less influence. Two such analyses have recently been undertaken. Leibbrandt and his associates produced a number of papers using the same underlying data, the best known of which is Leibbrandt et al. (2010b). Using data from PSLSD1993, IES2000 and NIDS2008, the study finds that poverty declined slightly between the first two surveys, and remained unchanged since. While these trends for each of the sub-periods is contrary to most other findings (see Yu 2008), the long-term trend of a small decline in poverty over the period as a whole is perhaps better to be trusted.

In an interesting analysis, Gradín (2012) compares poverty in 1993 and 2008, using the same first and last data sources as the study by Leibbrandt et al. (2010b) referred to in the previous paragraph. In particular, Gradín tries to explain the reduction in poverty amongst blacks that he finds with the trend for whites, for whom poverty rose slightly from very low levels. He concludes that a large part of the reason for the diminishing (although still very large) difference in poverty between these two race groups was a reduction in the education deficit experienced by blacks.

8. Decomposing inequality into between-group and within-group components

To decompose inequality into its between-race and within-race components, it is convenient to make use of the Theil index. The only annual data available on income distribution, from the AMPS survey, shows rising overall inequality. Within-group inequality has risen, while the between-group inequality component has declined. In other words, inequality is gradually becoming less race based. Whereas 61% of inequality could in 1993 still be ascribed to inequality between groups, this proportion has now dwindled to 35%. Much of this resulted from black upward mobility stimulated by new opportunities opening up for parts of the black population previously constrained by apartheid-era policies, and the removal of the protection earlier offered to parts of the white population, leading to some downward mobility (Moll, 2000).

9. Decomposing inequality by income component

Although the Gini coefficient is not decomposable between different groups where some incomes overlap, it is possible to decompose it by income source. Table 4 shows such a decomposition, distinguishing three income sources that Leibbrandt et al. (2001:23) referred to as ‘ … the key labour market, asset ownership and state welfare processes driving South Africa's inequality’; namely, wages, income from assets and other sources (including dividends and land rent), and transfers from government. This table shows that the largest source of inequality between households lies in the distribution of wages across households. (Such a decomposition does not consider variable household size, however.) Amongst the two-thirds of households that do have wage income (row 1), the Gini coefficient for such income is a high 0.65 (row 4). This reflects the great inequality in wage income between households, both in terms of wage levels earned and the number of wage earners. Considering wage income inequality across all households, the Gini coefficient would have been 0.77 if wages were the only source of income of the whole population (row 5). Altogether, differentials in wage earnings per household statistically ‘explain’ 77.9% of overall inequality (row 8). Classifying income sources slightly differently, Leibbrandt et al. (2010b:34–5) find this last percentage to be 88% in the 1993 PSLDS, 91% in IES2005 and 85% using the 2008 NIDS data, Leibbrandt et al. (2012) find that it was 73% in PSLDS1993 and 71% in NIDS 2008, while Armstrong & Burger (2009:17) find that wages contribute 69% to the Generalised Entropy 2 measure using IES2005.

Table 4: Decomposition of inequality into three main income sources, IES2005

			Wages	Transfers	Other income (e.g. from assets)	Total
1	% of households receiving income source	Pk	66	46	72
2	Mean income from income source (Rand)		35 588	3 397	11 292	51 277
3	Share in total income (%)	Sk	71.3	6.6	22.0
4	Gini coefficient for income source for households receiving such income	Ga	0.651	0.457	0.829
5	Gini coefficient for income source for all households	Gk	0.770	0.753	0.878
6	Gini coefficient correlation with total income rankings	Rk	0.923	0.028	0.734
7	Contributions to Gini coefficient of total income	SkGkRk	0.507	0.001	0.142	0.6501
8	% share in total income		77.9	0.2	21.9

Sources: Author's own calculations from IES2005.

Notes: G_k = P_kG_k + (1 – P_k), where k denotes the income sources (wages, transfers, other income).

Different income sources thus play varying roles in income distribution. While asset income is a source of inequality, social grants have virtually no impact on inequality, but do reduce poverty. Their limited effect on inequality may partly derive from the fact that social grants often enter larger households, which generally have higher aggregate incomes, even if per-capita incomes may be low. As is by now well established (Case & Deaton, 1998; Case et al., 2000; Bertrand et al., 2000; Duflo, 2000; Edmonds et al., 2001; Keller, 2004; Klasen & Woolard, 2009), grants have an important influence on household formation and composition: households tend to form around income, and social grants are often an important source of such income in poor rural areas.

The decomposition analysis shows that the main origin of income inequality lies in the labour market, through the distribution of jobs and wage formation processes. To understand South African inequality better, one must therefore turn to investigating the labour market.

10. The labour market and inequality

An analysis across different household and labour market surveys between 1997 and 2007 (whereafter the StatsSA labour market surveys no longer contained wage information) shows that the magnitudes relating to wages in the decomposition undertaken above (the Gini coefficient for households earning wages, the Gini coefficient for household wage earnings, and the share of households earning wages) all remained very stable over the period.

It is often thought that unemployment lies at the root of the distributional problem in South Africa, and that solving this problem would improve income distribution considerably. This is far from the truth, however: A simulation exercise using IES2000 data shows that more jobs would have a much more beneficial effect on poverty than on inequality: 2.5 million additional jobs would reduce the Gini coefficient by only about 0.033 (assuming that the incumbents would be those first in the job queue, and that they would be compensated according to their productive and other characteristics as derived from Mincerian earnings functions). However, such jobs would reduce the poverty headcount ratio by almost nine percentage points. In contrast, an increase in wages of as much as 30% would only reduce the poverty headcount by about four percentage points, while leaving the Gini coefficient slightly (1.1 percentage point) higher.

Leibbrandt et al. (2009) use the decomposition technique of Glewwe to further analyse inequality in wage earning; this method does consider household size. Surprisingly, they find that more than two-thirds of the inequality in household earnings in Labour Force Survey 2006 ‘is the result of unequal wage income, rather than the fraction of household members that are of working age or who are actually working’ (Leibbrandt et al., 2009:20).

Although it is widely believed that the South African economy did poorly in terms of job creation since the transition (the term ‘jobless growth’ is often used), the bottom panel of Figure 3 shows that the proportion of people employed at each age remained virtually unchanged between 1995 and 2005; that is, job creation kept pace with the growth of the potential labour force in the economically active age groups. What did change was that a larger proportion of this labour force chose to participate in the labour market, implying growing unemployment (see the top panel of the figure). This growth in labour force participation rates has been explained as either the result of greater feminisation of the labour force (Casale, 2004; Casale & Posel, 2002) or as a response to age restrictions imposed in the education system (Burger & Von Fintel, 2009; Burger et al., 2012). The resultant increased competition for jobs naturally affected inequality, because a large part of the population has little skills to offer in the labour market and is least likely to obtain jobs when these are scarce. The top panel of Figure 4 shows the conditional probability of employment from probit regressions (after setting other factors such as gender, experience, and location in terms of province or urban–rural at their mean values). Additional education brings little reward in terms of the probability of employment up to Grade 11, but thereafter it brings large rewards. Matric or higher qualifications thus appear to act as a threshold level of education required to compete effectively for jobs. This may result from the productivity gains associated with higher education, but may also reflect the limited signal that grade attainment gives at levels below matric: many pupils are routinely promoted to higher grades despite having learnt little. Only certification in the form of an externally assessed examination (matric) is really rewarded by employers, who can only observe productivity after a worker has already been engaged. Thus, at lower levels, an additional year of education does not improve the likelihood of employment.

Figure 3: Participation and employment as proportion of the population by age, 1995 and 2005

Source: Author's own calculations from IES2005.

Figure 4: Conditional probability of employment and conditional log of wages by years of education

While the level of education obtained sends a signal to employers and thus influences the probability of employment, the convex returns to education, as found in the Mincerian earnings function for South Africa (Bhorat & Leibbrandt, 2001; Keswell & Poswell, 2004), are evidence of higher productivity being associated with higher attainment. The second panel of Figure 4 shows the conditional log of the wage (i.e. for workers who otherwise have the average characteristics). The slope steepens after matric, indicating the existence of a threshold level beyond which an additional year of education is associated with a greater productivity gain.

The role of education in both employment and wages renders it of central importance to the labour market and thus to income inequality. Analysis of earnings functions has shown that the unexplained part of earnings differentials between race groups is quite large and stubbornly high (Burger & Jafta, 2006). This residual is often regarded as an upper estimate for labour market discrimination. However, in South African circumstances a large part of it may be due to differences in the quality of education received by members of different groups; that is, it may be better seen as pre-labour market discrimination rather than discrimination in the labour market (Du Rand et al., 2011; Burger & Van der Berg, 2011).

What is surprising about the conventional returns to education (i.e. if one ignores the possibility that the returns may simply reflect ability bias rather than a causal effect of education), is how convex they are. Figure 4 shows that there are very low returns for levels of education below matric. This raises serious concerns, as it indicates that employers do not find additional years of education associated with much productivity gains. In contrast, the returns to matric, and especially to degrees, are extremely high. One interpretation is that grade progression is a weak indicator of cognitive gains at school, and that it is only the threshold of matric – the only externally tested grade in the school system – that emits a reliable signal to employers. This signal is even stronger if qualifications beyond matric are added. In this respect, the gains in years of education seen in recent decades may not have reduced the high skills shortage much, as the final hurdle of matric, and particularly matric plus a university exemption, is still too high for many to jump. Whereas the 2011 census data show that more than one-third of the cohort born in 1960 had not even completed primary education, that proportion had fallen to 6% for the cohort born in 1990. But according to census numbers there does not appear to have been similar progress in raising the proportion of the youngest cohorts that has obtained a university degree. The reason appears to be that the cognitive threshold that must be crossed is still too high for most South African school children, given the weak education system. Educational quality thus remains a central concern in labour market outcomes. For this reason, we will return to this issue in the discussion of policy.

The Gini coefficient for wage income amongst the employed of above 0.60 effectively sets a floor to overall income inequality. If we consider other factors that may also affect income inequality, all but one are likely to increase it even above these levels: income from assets (dividends and land rent) largely goes to the rich; poor households are generally larger, which means that the already skew-distributed household incomes are converted into even larger differences in per-capita incomes; and employment is less common amongst the poor, both because of household composition (more children) and higher unemployment. So, all in all, per-capita incomes would be expected, for these reasons, to be even more unequal than the Gini coefficient of 0.60 for wages mentioned above. The only factor that slightly shifts the balances the other way is social grant income, which tends to favour the poor. But it is quite small compared with overall incomes and, as has been indicated earlier, has a far greater effect on poverty alleviation than on income distribution.

One can thus safely assume that wage inequality sets a floor below overall income inequality in South Africa: unless wage inequality can be reduced, overall inequality will remain high, with a Gini coefficient in excess of 0.60. The policy imperative in terms of the reduction of inequality is thus to reduce wage inequality. The policy options will now be considered.

11. The policy options

The complex link between government policy and economic inequality has been well summarised by Bromberger:

 … those distributions of income in which we are primarily interested are determined by immensely complex processes in which government activity interacts with relatively autonomous initiatives and adjustments by ‘the myriad forces of the market’. There does not exist a well-tested, widely-endorsed body of theory to model all of these processes. But it is clear that governments cannot readily control all of them, and there are limits to what governments may be able to do to change distributions. We must avoid assuming that if there is a change, or no change, government policy is responsible. Nor should we assume that government policies are either coherent or necessarily successful. (1982:167)

It is difficult to allocate blame or praise to individual government policies for their effect on distribution, simply because the ‘immensely complex processes’ make it almost impossible to know the full distributional effect of any individual policy. Nevertheless it is clear that the combined effect of the whole array of policies under apartheid was to increase racial inequality. Current inequalities still partly reflect the legacy of the systematic exclusion of the black population from opportunities, the mainstream economy and decision-making.

Some policies are often mentioned in debates around South African distribution. It is worth briefly considering a number of these:

• Job creation and economic growth: Apart from the fact that it is not clear how government policy can best contribute towards creating jobs, due to the complexity of economic interactions that Bromberger mentioned above, the performed simulation which showed that millions of jobs would have little effect on income distribution speaks against this as a means of reducing income inequality. Creating jobs is an important goal, but will not bring down the Gini coefficient much. The reason is also clear: most of those at the back of the job queue have little education and skills to offer in the labour market, and would therefore only receive low wages if they do find jobs. The simulation illustrates that this would have little effect. Yet the importance of growth and employment should not be forgotten, if not for their impact on income distribution, then for their effect on poverty. Moreover, access to jobs for the poor may also contribute to alternative opportunities to learn and become more productive and to building of social networks, which could contribute to reducing the steepness of the educational return structure.

• Social spending: In the decade after the transition, government social spending per person increased by about two-thirds and became much better targeted. This applies particularly to social grants, which constitute an important income source for the poor. Targeting occurs through the means test for social grants, through the fact that poorer people have more children and therefore benefit more from public school spending, and because the rich largely avoid using public health facilities, leaving larger benefits (although a poor quality of service) to those parts of the population who cannot always vote with their feet to leave the public health sector. But fiscal redistribution can never compensate for a highly inegalitarian outcome of market distribution. Moreover, there are fiscal limits to redistribution of this nature and capacity limits in the state apparatus to how much people actually benefit from such redistribution, as the quality of government services is often poor.

• Black economic empowerment policies and wage controls: Such policies include interventions to change labour market outcomes, and processes to redistribute assets. Affirmative action and minimum wage legislation have a limited and uncertain impact on distribution, given the complexity of predicting outcomes in a general equilibrium framework. If these policies are enforced excessively, labour market efficiency could suffer seriously, and only a small number of people benefit directly from such interventions. The social and political imperatives that may exist for such policies, which have largely a symbolic value, have to be weighed against their distortionary consequences for the economy.

• Education: The large differentials in earnings and access to jobs between the highly educated and the less educated lies at the heart of income inequality, as the decomposition analysis has shown. The high wage premium to educated workers derives from a combination of a skills shortage at the top end of the educational spectrum, driving up wages of the educated, and a surfeit of poorly-educated workers competing for scarce unskilled jobs, thus dampening unskilled wages. A continuation of the historical pattern of education, where only a small segment of the population (in the past racially defined) obtain a proper school education, would leave the structures underlying the large wage premium unaltered. Thus wage inequality will continue. A fundamental restructuring of education to ensure that most South Africans can obtain education of good quality will probably take decades, and it will take decades more before this fully feeds first through the school system and then the labour market. A collapse of the high returns to education is thus not about to happen in South Africa in the foreseeable future. Labour market inequality, which is central to income inequality, will not be reduced substantially unless there is a major transformation in the education system. Access to scarce employment will remain determined largely by education, and jobs will remain scarce because growth will continue to be constrained by a skills shortage.

12. Conclusion and discussion

This article has discussed the quantitative dimensions of income inequality in South Africa and has pointed out that severe data quality and measurement issues make any hard statement on income inequality difficult to substantiate. Nevertheless, a few major conclusions from this paper and from the data can help to inform the policy debate:

• The relationship between poverty and inequality is not straightforward: poverty can indeed be declining while inequality grows. Alleviating poverty must be far higher on any policy agenda than reducing inequality, although South African history makes attention to inequality (and particularly inter-racial inequality) an important concern in its own right.

• Sustained economic growth since the political transition allowed more attention to be paid to poverty alleviation, but income distribution has not improved. In particular, within-group inequality has increased quite strongly in the post-transition period.

• Overall income inequality, as measured by the Gini coefficient, seems to have risen sharply in the latter half of the 1990s and largely to have stabilised since. However, comparability issues between data sources and the difficulty of capturing some sources of income (such as dividends) cast some doubt on these conclusions, as do the large and inexplicable differences in levels of Gini coefficients derived from different data sources, even for the same period.

• Inequality has shifted in nature to becoming less race based than in the past. Within-group inequality has become by far the larger share of overall inequality. This is caused both by the increase in within-group inequality and the reduction of between-group inequality (although not all data sources provide evidence of the latter).

• Decomposition of inequality by income source shows that wage income is the dominant component in overall income inequality. Unemployment, often considered one of the main reasons for poverty and inequality, does indeed much affect poverty, but its effect on inequality is more limited. This wage inequality derives from educational differences, plus perhaps some discrimination.

This then brings us to some implications of the above, following on the discussion of policy alternatives above.

From an equity perspective, reducing income inequality is less important than reducing poverty. However, political and social realities and South Africa's history make inequality a source of concern. Given the historical legacy, however, the focus of inequality reduction will probably remain on inequality between race groups, whilst there will be more tolerance of the growing inequality within the black population. Further growth and consolidation of the black middle class is likely to reduce the social distance between the white and the black middle class and diminish the racial aspect of inequality.

Although fiscal redistribution has had some success through considerable expansion and much better targeting of social spending, there are fiscal and capacity limits to such redistribution, and it would leave the original source of inequality unchanged, unless it can contribute to improving educational quality. Direct transfers through social grants have alleviated poverty, but did not reduce inequality much.

The creation of jobs, important as that may be for poverty reduction, will do little for overall inequality. Neither will interventions in the labour market to artificially change labour market outcomes have much chance of success in the absence of improved education. To reduce income inequality substantially requires a different pattern of wages. This is only possible if education quality is improved greatly for the bulk of the population. A large segment of the labour force competes for the limited number of low-skill low-wage jobs; if the unemployed were to find jobs, it would be in this part of the wage distribution, and consequently such jobs would not reduce wage inequality much. In contrast, at the top end of the wage distribution high wages are paid to the relatively small part of the labour force with both more and better quality education.

Inequality remains deeply imbedded in South African society and will not disappear of its own accord. Interventions are required to reduce income inequality, but most of these interventions (affirmative action and Black economic empowerment, fiscal redistribution) can have only limited effects. The one exception is education, although solutions to the dilemma of poor educational performance and quality are not easy to find. Yet this remains the crucial requirement for the creation of a less unequal society. Whilst reducing poverty requires more jobs, reducing inequality requires better education.

Acknowledgement

The author acknowledges the financial support of the National Research Foundation.

Notes

²Note that the log of the mean is not the same as the mean of the logs, and changing the distribution while the mean of income remains unchanged will usually result in a different mean of the log of incomes. Simply put: , where y_i is the income of the ith household, and the left-hand side of the equation represents the log of mean income and the right-hand side the mean of the log of all incomes.

³Some authors (e.g. the World Bank in its World Development Review) report household income, but the method employed throughout for these figures was to compare all individual (per capita) incomes, thus the weight was derived by multiplying the household weights with the population size, as Deaton (1997) suggests.