Emergent Black Affluence and Social Mobility inPost-Apartheid South Africa

Rulof BurgerDepartment of Economics

University of Stellenbosch

rulof@sun.ac.za

Ronelle BurgerDepartment of Economics

University of Stellenbosch

rburger@sun.ac.za

Servaas van der BergDepartment of Economics

University of Stellenbosch

svdb@sun.ac.za

Development Pol icy Re search Unit No vember 2004

Work ing Pa per 04/87 ISBN 1-920055-02-9

Abstract

The na ture and ex tent of black affluence in South Africa pro vides an in dicator of the impact of

efforts to erad icate the rem nants of apartheid-era ra cial dis crimination in the South African

ed u ca tion sys tem and la bour mar ket. Most stud ies ex am in ing so cial mo bil ity and in equal ity in

South Africa have looked at the bottom of the in come dis tribution, in vestigating changes in

the se verity and also the ra cial in cidence of pov erty. This paper ex plores the same topic by

studying the top of the in come dis tribution.

Firstly, the paper at tempts to iden tify the fea tures that dis tinguish the affluent and spe cifically

the black affluent from the rest of the pop ulation with a de scriptive analysis. In the sec ond

section logit and multinomial logit mod els are used to con sider the im pact of ge ography,

house hold char ac ter is tics and the age, ed u ca tion and oc cu pa tion of the house hold head on

the like li hood of be ing af flu ent. The pa per in ves ti gates how af flu ence pre dic tors vary be tween

different race groups. The third and last sec tion of the paper is de voted to a typology of the

af flu ent.

The paper shows a dra matic in crease in black affluence. Also, the anal ysis here con firms

many of the tra ditional views of so cial mo bility. The paper finds a strong as sociation be tween

ge og ra phy, de mo graphic pro file and af flu ence that is ro bust across pop u la tion groups. The

em pir i cal ev i dence cited is con sis tent with con vex re turns to ed u ca tion and a sub stan tial role

for qual ity of ed u ca tion.

Our typology of the affluent shows that race is a strong de fining char acteristic for the clus ters

identified in 1995 and 2000. Encouragingly, the evidence in dicates that there is a large young

racially in tegrated group emerg ing among the affluent. These house holds have in come

levels be low the av erage for the affluent. House hold heads be longing to this group, how ever,

have ed u ca tional at tain ment lev els ex ceed ing the av er age for the af flu ent sug gest ing that

income might rise to match or ex ceed av erage lev els for the affluent as the group matures.

Acknowledgements

This paper is part of a larger re search pro ject on In equality and Social mo bility in South Africa,

funded in part by the NRF. This pub lication was spon sored by the Sec retariat for Institutional

Support through Eco nomic Re search in Af rica (SISERA). Their gen erous fi nancial

assistance in this regard is acknowledged.

Development Policy

Research Unit

Tel: +27 21 650 5705

Fax: +27 21 650 5711

Information about our Working Papers and other

published titles are available on our website at:

http://www.commerce.uct.ac.za/dpru/

Table of Contents

Introduction.................................................................................................................1

1. Who are the affluent?...................................................................................................2

1.1 Identifying the affluent......................................................................................2

1.2 Racial dimensions of affluence........................................................................4

1.3 Affluence and geography................................................................................7

1.4 Affluence and household size..........................................................................................8

1.5 Affluence and the gender of the household head..............................................10

2. The Determinants of Affluence..........................................................................12

3. Typology of the affluent.....................................................................................................23

Conclusion........................................................................................................................27

Ref er ences..........................................................................................................................28

Addendum A: 1995 OHS/IES and the 2000 LFS/IES....................................................30

Concerns relating to field work, data entry and coding....................................................30

Concerns relating to sampling...........................................................................................31

Introduction

It is vital to con sider to what ex tent the rem nants of apartheid-era ra cial dis crimination has

been ef fec tively elim i nated in the South Af ri can ed u ca tion sys tem and la bour mar ket. Most

studies ex amining so cial mo bility and inequality in South Africa have looked at the bottom of

the in come dis tri bu tion, in ves ti gat ing changes in the ra cial in ci dence and se ver ity of pov erty.

This paper ex plores the same topic by studying the top of the in come dis tribution.

The op por tu ni ties of fered in the new po lit i cal dis pen sa tion have in creased the num ber of the

black “in siders”. Due to this trend it is ex pected that intra-racial in equality will be come an

in creas ingly im por tant con trib u tor to wards over all in equal ity in South Af rica.1 By focusing on

the black mem bers of this group of affluent, this paper endeavours to im prove the

understanding of the emergent forces shap ing so cial mo bility and inequality in South Africa.

This paper will be di vided into three sec tions, with the first two sec tions in vestigating the

char ac ter is tics of the af flu ent and the de ter mi nants of af flu ence and the last sec tion ven tur ing

a typology of the affluent. Through out the paper the fo cus re mains on the re lationship

between race and affluence, with a particular em phasis on iden tifying the unique fea tures of

black af flu ence.

For the anal ysis we use the 1995 Oc tober House hold Sur vey/Income and Ex penditure

Survey (OHS/IES) and the 2000 La bour Force Sur vey/Income and Ex penditure Sur vey

(LFS/IES).2 In both these sur veys, but es pecially the last sur vey, there is sufficient ev idence

of care less field work and sloppy data en try and cod ing to war rant con cern about the

reliability of the data sets. Many of the iden tified prob lems can not be cor rected by clean ing or

editing pro cedures. Due to sam pling frame ad justments and the con cerns about the re liability

of the 2000 IES, the com parability of the two sur veys can not be guar anteed. Thus the

analysis avoids iden tification of trends across the two sur veys. Instead the two time pe riods

are mostly used to test the stability of re lationships be tween vari ables. Re sults were care fully

examined for dis tortions at tributable to the pe culiarities of the data and the authors were

deliberately conservative in the conclusions they drew from the analysis.

The iden tification of the affluent is ex plored in the first sec tion of the paper.

1

1 Van der Berg & Marincowitz (1999) concluded that “the next quarter of a century will probably see the continuedadvancement of the black elite and the gradual growth and consolidation of the black middle class”.

2 See Addendum A for a more detailed discussion of the 1995 OHS/IES and the 2000 LFS/IES and also theproblems associated with these surveys.

1. Who are the affluent?

1.1 Identifying the affluent

Our efforts to iden tify the affluent will to a large ex tent be based on ad aptations of the

literature on the iden tification of the poor. The con ventional ap proach to iden tifying the poor is

to rank house holds ac cording to a particular in come or wel fare in dicator, and to then se lect a

cut-off point to sep arate the poor from the non-poor (Hentschel and Lanjouw 1996: 1).

Income or ex penditure is usu ally cho sen as the in dicator of wel fare (Glewwe 1998: 3) –

although other in dicators, such as education or nutrition, can also be used.

If affluence is in terpreted as be ing de fined solely in terms of in come or ex penditure, the

identification of the affluent will be a sim pler task than the iden tification of the poor. In the case

of pov erty, in come or ex penditure is merely a ready sub stitute for a multi-di mensional con cept

in di cat ing de pri va tion of a va ri ety of fac tors in clud ing se cu rity, nu tri tion and ac cess to

em ploy ment.

In study ing pov erty, it is often ar gued that expenditure will be a more re liable in dicator of

welfare than in come, since the con sumer’s pref erence to smooth con sumption is likely to

render it less vol atile than income over the short term (Ravallion, 1992: 13). How ever, when

studying affluence, in come might be a pref erable mea sure of wel fare. We would ex pect

income to be less vol atile at the top end of the in come dis tribution and one can as sume that in

most cases rel atively affluent in dividuals will be more likely to ac curately re call their monthly

income than their monthly ex penditure.

Based on Woolard and Leibbrandt’s find ing (2001: 53) that in the South Af rican con text even

sub stan tial ad just ments for house hold struc ture are vir tu ally in con se quen tial for the

identification of poor house holds, we will use per ca pita house hold in come as mea sure of

welfare, mak ing no cor rections for household composition.

In their com par i son of dif fer ent pov erty mea sures, Woolard and Leibbrandt’s (2001: 46) note

that the ex act point at which a poverty line is drawn will al ways be “some what ar bitrary and

often highly con tentious”. The same ap plies to our “line of affluence”. How ever, there are two

considerations that pro vide some guid ance. Firstly, when se lecting a line of affluence, it is

vital that this group should in clude enough ob servations to en able statistical anal ysis – ide ally

also enough ob ser va tions of the sep a rate pop u la tion groups to al low in ves ti gat ing the

affluent house holds of different pop ulation groups in iso lation. Con versely, if this line is too

low and in cludes too many ob servations, the term “affluence” can lose its meaning.

Bearing these con siderations in mind, the affluent is de fined as the rich est 15 per cent of

households – as mea sured by per ca pita in come. The “line of affluence” is there fore drawn at

a per ca pita in come of R22 500 per year (1995 prices). In the 1995 OHS/IES (which is used

as our pri mary data set), this group in cludes 4 456 ob servations, 913 of whom are black.3

2

3 Throughout the text population group will refer to the population group membership of the household head. Also,when we refer to gender, age and education this will be the gender age and education of the household head.

After taking house hold weights into ac count, the group rep resents 16.2 per cent of the

pop u la tion.

A sec ond line at the level of an an nual in come per ca pita of R36 000 is used to iden tify the

very affluent. This group was de fined so as to rep resent the top half of the affluent group. In

the 1995 OHS/IES the very affluent in cludes 2 277 house holds, 326 of whom are black. This

group rep resents 7.7 per cent of the sur vey participants and 8.5 per cent of the pop ulation after

taking sur vey weights into account.

When us ing the sec ond, higher affluence line, those ob servations that fall be low this line, but

lie above the first line of affluence – those house holds who are affluent, but not very affluent –

will be re ferred to as the “merely affluent”. The iden tification of two groups of affluent will also

allow us to test the ro bustness of our re sults to the po sition of our line of affluence.

These lines of affluence are ad justed for in flation and then ap plied to the LFS/IES 2000.4

Table 1.1 shows a 1995/2000 com parison of the break down of affluent house holds per

population group. Even at these fairly mod est lines of affluence, the num ber of very affluent

coloured house holds in 1995 and 2000, and affluent In dian house holds in 2000 are such that

it can be con sidered im prudent to make inferences from these samples.

Table 1.1: The breakdown of affluent households by population group

1995 2000

Population group Affluent Very Affluent Affluent

Black 913 326 1 885

Coloured 242 88 345

Indian 251 125 162

Since our fo cus is on so cial mo bility, it is use ful to split the re maining house holds (the

non-affluent) into two groups: the mid dle-class and the poor. The poor will be de fined as the

poorest 40 per cent of house holds, which im plies us ing a R3 650 per ca pita an nual in come as

a pov erty line.

3

4 The survey’s income data is adjusted upwards by 38 percent to match National Account totals. See Addendum Afor a more detailed description of the problems with the IES 2000.

Table 1.2 sum marises the in come group def initions as de scribed in this sec tion and as it will

be used through out this paper.

Table 1.2: Income classifications

Per capita income

(1995 prices)

Percentage of

population in

IES 1995

Household classification

Less than R 3 650 40 % Poor Non-affluent

Between R 3 650and R 22 500

36% Middle-class

Between R 22 501and R 36 000

16% Merely Affluent Affluent

More than R 36 000 8% Very Affluent

1.2 Racial dimensions of affluence

Considering South Africa’s po litical past, it is not sur prising to find that there is a strong ra cial

dimension to affluence. In 1995 71 per cent of the coun try’s affluent (79 per cent of the very

affluent) was white. De spite rep resenting 70 per cent of the pop ulation, blacks only

comprised 22 per cent of the affluent (15 per cent of the very affluent).

By 2000 the com position of the affluent had changed con siderably. The white share of

affluence had shrunk to 47 per cent while the black share of affluence rose to 41 per cent.

Table 1.3 also shows that the per centage of In dians in the affluent group re mained largely

unchanged while the col oured share of affluence more than doubled.

Table 1.3: Proportion of the affluent belonging to each population group

1995 2000

Population

groupAffluent Very Affluent Affluent

Black 22 % 15 % 41 %

Coloured 3 % 2 % 8 %

Indian 4 % 3 % 4 %

Table 1.4 shows that even though we ob serve a rise in the pro portions of house holds who are

affluent for each pop ulation group, the size of the pop ulation shift as de scribed in Table 1.5

seems im plau si bly large and pos si bly at trib ut able to prob lems with the sur veys. Ac cord ing to

the sur vey the pop ulation share of blacks moved from 70 per cent to 78 per cent be tween 1995

and 2000 while the pop ulation share of whites shrunk from 19 per cent to 11 per cent over the

same pe riod. It is thus likely that the sur vey trends may overrepresent in creases in black

affluence and underrepresent in creases in white affluence.

4

Table 1.4: Proportion of each population group classified as affluent

1995 2000

Population

groupAffluent Very Affluent Affluent

Black 5.1 % 1.9 % 9.0 %

Coloured 6.7 % 2.5 % 16.8 %

Indian 21.9 % 10.8 % 28.9 %

White 59.9 % 35.0 % 74.9 %

Total 16.2 % 8.5 % 18.1 %

Table 1.5: Population group breakdown of households

Population group* 1995 2000

Black 70.0 % 77.8 p%

Coloured 8.3 % 8.1 %

Indian 2.6 % 2.5 %

White 19.2 % 11.4 %

*Race shares do not add to 100 percent due to “Un specified” category

5

Table 1.6 shows that the mean age of the black affluent house hold head is lower in 2000 than

in 1995, while the re verse trend is ob served for white affluent house hold heads over the same

period, sug gesting that inter-racial shifts in affluence have been more pro nounced among

youn ger co horts.

Table 1.6: Mean and standard deviations for household per capita income, age andeducational attainment of household head per population group for affluent

Variable Black Coloured Indian White

Per capita

income (1995

prices)

1995 Mean 41 414 39 159 48 727 55 794

Std Dev 35 767 28 326 39 063 63 502

2000 Mean 30 262 32 463 40 432 51 550

Std Dev 25 421 21 727 33 249 55 186

Age of

household

head

1995 Mean 40.2 41.7 43.4 45.6

Std Dev 11.3 12.0 12.9 14.1

2000 Mean 39.7 42.0 44.4 47.3

Std Dev 10.5 13.0 11.9 15.3

Years of

educational

attainment of

household

head

1995 Mean 10.9 11.7 12.1 12.5

Std Dev 3.7 2.8 2.7 1.8

2000 Mean 9.7 11.7 12.0 12.8

Std Dev 4.3 2.6 2.9 2.1

6

1.3 Affluence and geography

Ta ble 1.7 con sid ers the re la tion ship be tween af flu ence and ge og ra phy. Gauteng and the

Western Cape are the only prov inces with a higher pro portion of affluent res idents than the

national average in both 1995 and 2000. The Eastern Cape and Limpopo have the low est

proportion of affluent in habitants. It is likely that the strong de cline in the pro portion of

Gauteng res idents who are affluent are due to prob lems with the sur vey: ei ther at tributable to

adjustments in the sam pling frame or al ternatively, un reliable field work in Gauteng.5

Table 1.7: Proportion of each province’s population classified as affluent

Province Percentage of households

1995 2000

Affluent Very Affluent Affluent

Gauteng 34.1 19.3 25.3

Western Cape 23.0 13.0 30.4

Northern Cape 12.1 6.3 20.7

Kwazulu-Natal 11.9 6.0 13.2

Free State 11.2 5.2 20.3

Northwest 10.9 4.6 15.8

Mpumalanga 9.7 4.6 15.3

Limpopo 7.8 3.4 6.4

Eastern Cape 7.7 3.5 9.8

Total 16.2 8.5 18.1

7

5 See the addendum for a more thorough discussion of problems relating to the two surveys

Table 1.8 shows that affluence is more prev alent in ur ban ar eas. More than 20 per cent of

urban res idents are affluent, while less than 5 per cent of those in ru ral ar eas are affluent.

Table 1.8: Proportion of urban/rural dwellers classified as affluent

Proportion who are affluent

1995 2000

Urban 25.53 24.39

Rural 4.35 3.70

Total 16.2 18.1

1.4 Affluence and household size

Cross-tab u la tions in di cate that af flu ent house holds gen er ally have smaller fam i lies.

According to table 1.9, non-affluent fam ilies had 4.70 mem bers on av erage in 1995,

compared to affluent fam ilies who had 2.73 mem bers on av erage.

Table 1.9: Household size for the affluent

Average number of

members in household

1995 2000

Non-affluent 4.70 4.20

Affluent 2.73 2.41

Very Affluent 2.55 2.41

Table 1.9 also sug gests that there has been a move to wards smaller fam ilies. The av erage

household size de clined from 4.39 house holds mem bers in 1995 to 3.85 house hold

members in 2000. The av erage non-affluent fam ily now had 4.20 members – still

substantially higher than the 2.41 av erage fam ily size of affluent house holds. The de crease in

household size has been slightly larger for affluent house holds, who ex perienced a 14

percent re duction in house hold size com pared to 11 per cent for the non-affluent.

Disaggregating the affluent house holds into different pop ulation groups sug gests that the

effect is mainly driven by the change in house hold size of affluent blacks – as il lustrated by

Table 1.10.

8

Table 1.10: Average household size for the affluent by population group

It is in sightful to further disaggregate black house holds by house hold size. For both affluent

and very affluent black house holds there was a large shift to wards sin gle-member

households. This shift may be partly attributable to prob lems with the sur vey’s sam pling

frame in 1995, which re sulted in undercounting of black sin gle-member house holds.

Table 1.11: Black affluent households, by household size

9

Average household size

1995 2000Populationgroup Affluent Very Affluent Affluent Very Affluent

Black 2.59 2.51 1.87 1.95

Coloured 3.04 2.52 3.08 2.85

Indian 3.51 3.39 3.30 2.98

White 2.73 2.53 2.69 2.54

Total 2.73 2.55 2.41 2.41

Percentage of affluent households

1995 2000Household

size Affluent Very Affluent Affluent Very Affluent

1 38.3 40.2 61.2 56.2

2 18.3 18.9 14.7 17.0

3 15.4 14.4 10.8 12.4

4 14.2 13.5 6.7 8.4

5 6.9 6.3 4.5 4.1

6 3.5 4.0 1.2 0.9

7 1.7 2.2 0.6 0.7

8 1.0 0.4 0.2 0.1

9 0.5 0.2 0.0 0.0

10 0.1 0.0 0.1 0.2

1.5 Affluence and the gender of the household head

The neg a tive re la tion ship be tween fe male-headed house holds and pre dicted in come is

well-es tab lished in the lit er a ture on pov erty de ter mi nants. As the lit er a ture would pre dict,

Table 1.12 shows that fe male-headed house hold were less com mon amongst affluent than

non-af flu ent house holds.6 How ever, the table also shows that the pro portions of

female-headed house holds were very sim ilar for affluent and non-affluent house holds in

1995 and, sur prisingly, by 2000 the very affluent had a higher pro portion of female-headed

households than the “merely” affluent. The Lowess graphs dis play the same trend, but in

more de tail. The ta ble and graphs ap pear to in di cate that the neg a tive sta tis ti cal as so ci a tion

between in come and the prob ability of ob serving a fe male-headed house hold might only hold

true for house holds be low a cer tain in come level.

Table 1.12: Proportion of female household heads in different income groups

1995 2000

Non-affluent 0.343 0.428

Affluent 0.150 0.187

Very Affluent 0.146 0.193

Total 0.312 0.384

10

6 By using Pearson’s chi-squared test, the hypothesis, that affluence is independent of the gender of the householdhead, can be rejected in both periods at a level of confidence exceeding 99,9 percent.

Figure 1: Lowess graph showing pro portion of fe male-headed black house holds by householdincome per ca pita in 1995

Figure 2: Lowess graph showing pro portion of fe male-headed black house holds by householdincome per ca pita in 2000

11

Further anal ysis is re quired to dis tinguish the separate effects and com pare the rel ative

weight of the different pre dictors of affluence. In sec tion two the char acteristics dis cussed in

section one are used as ex planatory vari ables in a lo gistic re gression mod elling the like lihood

of a house hold be ing affluent.

2. The Determinants of Affluence

The model uses pre dictors of low in come as the pre dictors of affluence, ask ing whether the

ab sence of char ac ter is tics as so ci ated with pov erty nec es sar ily in creased the like li hood of

af flu ence.7

The sec ond and fourth col umn of Table 2.1 sum marise the re gression re sults for a lo gistic

re gres sion for af flu ence in clud ing ge og ra phy, race, house hold size, and the gen der,

educational at tainment and age of the house hold head.8 A sec ond model is es timated with

white-education in teraction vari ables in cluded as regressors. The rest of the sec tion is

devoted to a dis cussion of the re sults of these two mod els, first ap plied to all affluent

households and then only to black affluent households.

12

7 Working on the KwaZulu-Natal Income Dynamic Study, Keswell (2001) emphasised the problems with anaggregated analysis of socio-economic dynamics. He showed that social mobility is a heterogeneous processcharacterised by non-linearities.

8 To test the sensitivity of the results for the selected affluence line, coefficients were estimated when the line ofaffluence is shifted both 2 percentage points up and 2 percentage points down from the 15 percent affluencecut-off. All the model’s significant variable coefficients remain within the 95 percent confidence interval when thedefinition of the dependent variable is altered.

Table 2.1: Logistic regression of affluence

Note: The reference household is black, urban and living in the Eastern Cape with a male household head. Although

not reported here, the regression controlled for province of residence

*Significant at 10 percent level, ** Significant at 5 percent level, *** Significant at 1 percent level

13

(-2.14)(-3.16)

-0.0304**-0.0220***White-Education

interaction Squared

(1.79)(2.26)

0.58*0.36**White-Education

interaction

(18.59)(16.80)(13.87)(11.92)

0.0341***0.0315***0.0301***0.0231***Education of household

head squared

(-9.01)(-7.56)(-3.70)(-1.20)

-0.24***-0.21***-0.14***-0.04Education of household

head

(-8.31)(-8.00)(-10.07)(-9.79)

-0.0013***-0.0013***-0.0013***-0.0013***Age of household head

squared

(10.10)(9.79)(11.35)(11.09)

0.15***0.15***0.14***0.14***Age of household head

(-11.62)(11.47)(-8.56)(-8.26)

-0.83***-0.81***-0.63***-0.61***Female household head

(-15.19)(15.09)(-15.68)(-15.56)

-0.62***-0.62***-0.62***-0.63***Number of dependants

(7.64)(7.49)(7.73)(7.26)

1.23***1.20***1.26***1.18***Proportion workers

(-15.56)(15.55)(-3.53)(-3.76)

-1.23***-1.23***-0.26***-0.28***Rural

(-0.14)(22.94)(1.11)(29.95)

-0.252.31***1.022.16***White

(5.76)(5.91)(8.86)(9.17)

0.90***0.91***1.08***1.08***Indian

(4.90)(4.92)(3.96)(4.03)

0.56***0.56***0.42***0.42***Coloured

(-14.43)(-4.75)(-19.70)(-19.50)

-5.51***-5.50***-7.47***-7.51***Constant

0.50240.50140.54540.5438Pseudo R2

25 82125 82128 34928 349Observations

Model 2

2000

Model 1

2000

Model 2

1995

Model 1

1995

(-2.14)(-3.16)

-0.0304**-0.0220***White-Education

interaction Squared

(1.79)(2.26)

0.58*0.36**White-Education

interaction

(18.59)(16.80)(13.87)(11.92)

0.0341***0.0315***0.0301***0.0231***Education of household

head squared

(-9.01)(-7.56)(-3.70)(-1.20)

-0.24***-0.21***-0.14***-0.04Education of household

head

(-8.31)(-8.00)(-10.07)(-9.79)

-0.0013***-0.0013***-0.0013***-0.0013***Age of household head

squared

(10.10)(9.79)(11.35)(11.09)

0.15***0.15***0.14***0.14***Age of household head

(-11.62)(11.47)(-8.56)(-8.26)

-0.83***-0.81***-0.63***-0.61***Female household head

(-15.19)(15.09)(-15.68)(-15.56)

-0.62***-0.62***-0.62***-0.63***Number of dependants

(7.64)(7.49)(7.73)(7.26)

1.23***1.20***1.26***1.18***Proportion workers

(-15.56)(15.55)(-3.53)(-3.76)

-1.23***-1.23***-0.26***-0.28***Rural

(-0.14)(22.94)(1.11)(29.95)

-0.252.31***1.022.16***White

(5.76)(5.91)(8.86)(9.17)

0.90***0.91***1.08***1.08***Indian

(4.90)(4.92)(3.96)(4.03)

0.56***0.56***0.42***0.42***Coloured

(-14.43)(-4.75)(-19.70)(-19.50)

-5.51***-5.50***-7.47***-7.51***Constant

0.50240.50140.54540.5438Pseudo R2

25 82125 82128 34928 349Observations

Model 2

2000

Model 1

2000

Model 2

1995

Model 1

1995

The dummy vari ables rep re sent ing the dif fer ent pop u la tion groups are all sig nif i cant (and

par tic u larly so for whites),9 in dicating that race is an im portant pre dictor of house hold

affluence. Be tween 1995 and 2000 the mag nitude of all the white dum mies in creased, while

the sig nificance value de clined, which could be in dicative of a less uni form im pact of race on

the prob a bil ity of be ing af flu ent.

The change in the im pact of race on affluence is il lustrated in Fig ure 2.1, which var ies the

pop u la tion group for the se lected ref er ence group.10

As ex pected, cet eris pa ri bus, col oured

or In dian house holds have a sub stantially higher like lihood of be ing affluent than black

households. The fig ure also shows that white house holds re main con siderably more likely to

be affluent than any of the other pop ulation groups.

Figure 2.1: Probability of being affluent by population group

Note: Here the reference group is a three-member, ru ral house hold living in KwaZulu-Natal with and one work ing house hold mem ber and a

40-year old, male house hold head with a Grade 6 ed ucation.

The house hold head’s years of ed ucation and years of ed ucation-squared are both also

included as ex planatory vari ables. For both sur veys, the ed ucation-squared terms are

14

0

0.05

0.1

0.15

0.2

0.25

0.3

Black Coloured Indian White

Pro

bab

ilit

yo

fb

ein

gaff

luen

t

1995

2000

9 Unless specified otherwise, significance will refer to a 5 percent level of significance.

10 Since the reference group only determines the initial value from which the population group of the household headis varied, and is independent of the magnitudes of the coefficients, the reference groups were chosen merely inorder to obtain comparable initial values between survey years which is useful for graphical analysis of logitcoefficients.

pos i tive, in di cat ing that the re la tion ship be tween years of ed u ca tion and the like li hood of

being affluent is con vex. This is in line with Keswell’s find ing (2001:16) that “having more

education does not seem to make much difference un less the ed ucation ob tained is

sub stan tial (be yond 10 years)”.

Sur pris ingly, the ed u ca tion terms are sig nif i cantly neg a tive for both sur veys. The com bined

effect of these two vari ables is that the prob ability of be ing affluent de clines with ad ditional

years of ed ucation at low levels of ed ucational at tainment. Figure 2.2 il lustrates the expected

prob a bil i ties of be ing af flu ent, where the level of ed u ca tional at tain ment is var ied for the

specified ref erence group. Hav ing a house hold head with a ter tiary ed ucation rather than

secondary or pri mary ed ucation can also be seen to rad ically im prove the prob ability of be ing

affluent. The turning points are at low lev els of ed ucational at tainment: be fore Grade 2 in

1995 and just after Grade 3 in 2000.

Figure 2.2: Probability of being affluent by level of educational attainment

Note: The reference household is a black, four member household living in urban KwaZulu-Natal with one worker

and a 35 year old, male household head.

15

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Education level of household head

Pro

bab

ilit

yo

fb

ein

gaff

luen

t

1995

2000

There are at least two dif fer ent ex pla na tions for the neg a tive ed u ca tion co ef fi cient at low

education levels. Firstly, the neg ative co efficient for years of ed ucation could sim ply be due to

misspecification of the func tional form. If the true un derlying effect of ad ditional ed ucation on

be ing af flu ent is neg li gi ble at low lev els of ed u ca tional at tain ment and there af ter it starts to

increases at an ex ponential rate, the spec ification used in our model (i.e. in cluding years of

education and years of ed ucation squared) can yield a neg ative co efficient for years of

education. This ex planation would be con sistent with the fitted shape of the education curves

in Figure 2.2.

An other pos si ble ex pla na tion is that no ed u ca tional at tain ment might re flect the lack of

access to ed ucation, whereas a low level of ed ucational at tainment shows an ac cess to

education, but also re flects a self-se lection which is in dicative of some per sonal

characteristics (such as a lack of re solve). This might im ply that those with low lev els of

educational at tainment are in deed less likely to pos sess the qual ities often re quired in or der

to be come affluent than those with no ed ucation at all. Since there are no ob servations for

which the house hold head only ob tained ei ther Grade 1 or Grade 2 in 1995 (those lev els of

ed u ca tion for which the re gres sion pre dicts neg a tive re turns), it is im pos si ble to dis tin guish

be tween the two afore-men tioned ex pla na tions. Re-es ti mat ing the 2000 re gres sion with a

separate dummy variable for ev ery different year of ed ucation shows that the prob ability of

being affluent is not sig nificantly different when the house hold head has a Grade 1 ed ucation

but sig nificantly lower when hav ing a Grade 2 ed ucation rather than having no ed ucation.

There is no sig nificant difference be tween a house hold head with no ed ucation and a Grade 3

education. This re sult implies that the second explanation is more likely to be the reason for

the negative coefficient for the years of education variable.

In an at tempt to con trol for qual ity of ed ucation, the sec ond model in troduces an in teraction

effect be tween white and ed ucational at tainment – fol lowing Kingdon and Knight (2002) and

Chamberlain and Van der Berg (2002) in as suming that it is plau sible that qual ity of ed ucation

is sys tematically re lated to the race of the house hold or at least per ceived as such by

em ploy ers. The co ef fi cients of the white-ed u ca tion in ter ac tion and the square of this

interaction effect are sig nificant. In the pres ence of the two new vari ables the white pop ulation

group dum mies be come in sig nif i cant.11

This sug gests that the su perior qual ity of ed ucation

that whites re ceived un der apartheid could be the most important re maining av enue whereby

whites are ad vantaged rel ative to other race groups in the labour market.

Ac count ing for vari a tion in ed u ca tion qual ity re sults in a mar ginal in crease in the ex plan a tory

power of the model, so for the rest of this sec tion the dis cussion will fo cus on the sec ond

model in Table 2.2.12

16

11 It is interesting to note that the same effect is not present for other race groups. When including interactionvariables for education and the Indian and Coloured population groups respectively, the interaction variables werenot significant in any of the cases for either year.

12 There is a marginal increase in both the R2 and the number of observations that are correctly classified as beingaffluent or non-affluent at a probability cut-off value of 50 percent (not shown in Table 2.1).

Figure 2.3 and 2.4 in corporate the im pact of the ed ucational at tainment and white in teraction

co ef fi cient, into the rep re sen ta tion of the re la tion ship be tween ed u ca tional at tain ment and the

likelihood of affluence. The black ed ucation curves look sim ilar to those in Fig ure 2.2, but the

returns to ed ucation for whites are now con cave for both years and higher for all lev els of

education in 1995. An un usual shift took place in the ed ucation curve for whites in 2000. At

educational at tainment levels be low four years the white tail now falls be low that of the black

population group. Since only two of the 1 584 white ob servations have less than four years of

education in 2000, this part of the curve should be disregarded.

Figure 2.3: Probability of being affluent, by population group and educationalattainment (1995)

Note: The reference household is a black, four member household living in urban Limpopo with one worker and a 45 year old, male household

head.

17

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Education level of household head

Pro

bab

ilit

yo

fb

ein

gaff

luen

t

Blacks

Whites

Figure 2.4: Probability of being affluent, by population group and educationalattainment (2000)

Note: The reference household is a black, four member household living in urban Limpopo with one worker and a 45 year old, male household

head.

Con sid er ing the im pact of ge og ra phy, the model shows that the ru ral in di ca tor is sig nif i cant

and re duces the like lihood of affluence in both 1995 and 2000. The effect of the prov ince of

residence is weaker. Only the dum mies for Free State and North west Prov ince are sig nificant

in both years and the co efficient of the Free State dummy switches its sign.

The gen der of the house hold head was also highly sig nificant in both years. Fig ure 2.5

illustrates the effect of having a fe male, rather than a male house hold head and liv ing in a

rural rather than in an ur ban area for both sur veys. The fig ure dem onstrates that gen der was

a more pow erful pre dictor of affluence than liv ing in an ur ban area in 1995, but that the area of

residence was more im portant in 2000.

18

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Education level of household head

Pro

bab

ilit

yo

fb

ein

gaff

luen

t

Blacks

Whites

Figure 2.5: Probability of being affluent by gender of household head and area ofresidence

Note: The reference household is a black, five member household living in rural Limpopo with one worker and a 35 year old household head

with Grade 8 education.

The num ber of non-work ing house hold mem bers and the pro portion of em ployed house hold

members are both strongly sig nificant in 1995 and 2000. The mag nitudes of the co efficients

are also very sim ilar as is il lustrated by Figures 2.6 and 2.7. It can be ob served that the

prob a bil ity of be ing af flu ent de creases dra mat i cally as the num ber of non-work ing house hold

members in the ref erence group in creases. The prob ability of be ing affluent can also be seen

to de crease steadily as the pro portion of house hold mem bers who work decreases.

19

0

0.01

0.02

0.03

0.04

0.05

0.06

1995 2000

Pro

bab

ilit

yo

fb

ein

gaff

luen

t

Urban Male

Rural Male

Urban Female

Rural Female

Figure 2.6: Probability of being affluent by number of non-working household members

Note: The ref erence house hold is a black house hold living in rural Limpopo with one worker and a 50 year old, male house hold head with

Grade 7 ed u ca tion.

Figure 2.7: Probability of being affluent by proportion of household members who work

Note: The ref erence house hold is a five mem ber black house hold living in rural Limpopo with a 50 year old, male house hold head with Grade

7 ed u ca tion.

20

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

1 0.8 0.6 0.4 0.2 0

Proportion of working member households

Pro

bab

ilit

yo

fb

ein

gaff

luen

t

1995

2000

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0 1 2 3 4 5 6 7 8

Number of dependants

Pro

bab

ilit

yo

fb

ein

gaff

luen

t

1995

2000

In both years the age and age-squared vari ables (al lowing for a non-lin ear effect for age) had

a sig nificant im pact on the prob ability of be ing affluent. Age can here be in terpreted as a

proxy for years of ex perience. As both the ory (e.g. Mincerian earn ings func tions) and

empirical stud ies (e.g Bhorat and Leibbrandt (2001: 125)) would sug gest, the co efficient of

the age vari able is pos itive and the co efficient for the age squared vari able is neg ative.

Having an older house hold head in creases the prob ability of be ing affluent at a de creasing

rate un til it reaches its turn ing point in the mid-fifties (54 in 1995, 57 in 2000). After this point,

an older house hold head is re lated to a de creasing prob ability of be ing affluent.

To fo cus more nar rowly on the in come dy namics within the black pop ulation, we es timate a

logistic re gression for the black house holds. Based on the ear lier ob servation that

sin gle-mem ber house holds were seem ingly an in creas ingly im por tant phe nom e non among

blacks, a sin gle house hold dummy variable is added to the regressors we used in the orig inal

full sam ple logit model (Table 2.1). The re sults of this lo gistic regression are presented in

Table 2.2.

Table 2.2: Logistic regression of affluence for black sample

Note: The reference household is urban, living in the Eastern Cape with a male household head. Although not reported here, the regression

controlled for province of residence.

*Significant at 10 percent level **Significant at 5 percent level *** Sig nificant at 1 percent level

21

1995 2000

Observations 18 522 20 499

Pseudo R2

0.4053 0.4078

Constant -0.02 -8.06***(-3.16) (-14.40)

Rural -0.26*** -1.27***(-3.53) (-12.74)

Workers as proportion 1.26*** 1.13***(7.73) (6.62)

Number of dependants -0.62*** -0.51***(-15.68) (-9.78)

Single member households 0.36** 0.96***(2.26) (9.50)

Female household head -0.63*** -1.04***(-8.56) (-11.36)

Age of household head 0.14*** 0.24***(11.35) (9.58)

Age of household head squared -0.0013*** -0.0021***(-10.07) (-7.60)

Education of household head -0.14*** -0.27***(-3.70) (-8.95)

Education of household head squared 0.0301*** 0.0379***(13.87) (18.09)

In most re spects the re sults re semble that of the lo gistic re gression for the sur vey’s full

sample of affluent. The ru ral dummy is sig nificant and with the expected sign. As in the full

sam ple lo gis tic re gres sion, the North west dummy is pos i tive and sig nif i cant.

Figure 2.8: Probability of being affluent, by household size

Note: The ref erence house hold is a black house hold living in rural Limpopo with a one worker and a 50 year old, male house hold head with

Grade 5 ed u ca tion.

The sin gle mem ber house hold dummy variable is neg ative and sig nificant. This in dicates that

even after cor recting for the cost of an ad ditional house hold mem ber, be ing a sin gle-member

house hold sig nif i cantly in creases the prob a bil ity of be ing af flu ent. Fig ure 2.8 plots the

expected prob ability of be ing affluent for different house hold sizes, show ing that for the black

pop u la tion group the over all re la tion ship be tween house hold size and af flu ence is roughly

similar be tween the two years, with a steep rise in a sin gle mem ber house hold’s like lihood to

be affluent.

This ob servation is in line with the re search of McElroy (1985), Ermisch & DiSalvo (1997),

Card & Lemieux (1997) and lo cally also Klasen & Woolard (2000: 11-14) and Keller (2002:22)

that find that vulnerable in dividuals (e.g. the un employed) are less likely to leave their cur rent

household to set up their own house holds. Ac cordingly, sin gle mem ber house holds are

expected to be a self-se lected group with higher than average lev els of em ployment and

af flu ence.

The re gression shows that the age of the house hold head raises the prob ability of be ing

affluent at a de creasing rate. Re turns to ed ucation are con vex be yond Grade 3 in 1995 and

Grade 4 in 2000.

22

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0 1 2 3 4 5 6 7 8

Number of dependants

Pro

bab

ilit

yo

fb

ein

gaff

luen

t

1995

2000

The next sec tion uses clus ter anal ysis to iden tify nat ural group ings among affluent

households in 1995 and 2000. Due to prob lems with the sur veys, the paper has cho sen to

steer clear of com parisons be tween the two sur vey pe riods in the re gression anal ysis. The

clus ter anal y sis was based on dis sim i lar i ties in eight house hold char ac ter is tics, most of

which were de mographic vari ables. Sur veying of de mographic vari ables is more

straight for ward than que ry ing house holds re gard ing their in come and ex pen di ture; hence

demographic vari ables are gen erally more re liably re ported, thus pro viding grounds for a

comparison be tween the two years in the section below.

3. Typology of the affluent

Cluster anal ysis is uti lised to find group ings with shared traits and char acteristics among the

sample of affluent in the 1995 and 2000 sur vey.13

The ob servations are partitioned into four

non-over lap ping groups or types ac cord ing to dis sim i lar i ties in per ca pita in come and

pre dic tors of af flu ence us ing the Min kow ski dis tance met ric as the mea sure of dis sim i lar ity.14

According to Table 3.1, race is a prom inent iden tifying char acteristic for all four types that

emerge from the anal ysis for 1995. The first two group ings are com parable in terms of the

average age of the house hold head (37 and 38 years re spectively), but the first group is

predominantly white (94 per cent) and has a sub stantially higher av erage per ca pita

household in come and years of ed ucational at tainment than the sec ond group that is

predominantly black (87 per cent). Both groups are fairly re liant on wages as source of

income: it com prises more than 70 percent of household income for both groups.

The third and fourth groups are both largely white, with per ca pita in come, age and

ed u ca tional at tain ment be ing the main dif fer en ti at ing fea tures. The third group’s av er age

household in come of R51 637 is less than a fifth of the fourth group’s av erage house hold

income of R293 328. More than half of the fourth group’s in come is de rived from net profit

while it rep resents only 12 per cent of in come for the third group. Part of the in come gap could

be due to differences in ed ucational at tainment. The difference in the av erage age of the

household head points to an ad ditional explanation for the in come gap. Ac cording to the

model in sec tion 2 the likelihood of affluence is the high est at the age of 46 and de clines after

this point. Given that the fourth group has an av erage house hold head age of 60 years

compared to the 47 years average of the fourth group ing, it is thus not en tirely sur prising that

the house hold in come of the third group is sub stantially higher than that of the fourth group.

23 per cent of the third group’s income is from pensions.

23

13 In both years the number of types was specified as five, but in both 1995 and 2000, one identified type has beenomitted because it consisted of fewer than 10 observations. Consequently, the columns do not add up to the totalsprovided.

14 Variables were standardised to have a zero mean and standard deviation of one.

Table 3.1: Race, per capita income, household size, age and education per type for 1995

Gauteng is strongly rep resented among all four types of affluent, and particularly prev alent

among the first group ing of young whites. The shares of East ern Cape, North West and

Limpopo are no ticeably higher for the group of young blacks than for the other three types.

The share of Western Cape res idents ap pear to vary with the pro portion of whites.

Table 3.2: Provincial residence per type in 1995

Table 3.3 shows the strong ur ban pres ence of type 1 and 3, which com bined in cludes al most

all of the white affluent and 74 per cent of all the affluent. Type 2 and 4 have better ru ral

rep re sen ta tion.

Table 3.3: Rural residency and gender of household head per type in 1995

24

Types

Type asshare of

totalBlack Coloured

shareIndianshare

Whiteshare

Percapita

income Age EducationHousehold

size

1 0.44 0.00 0.02 0.03 0.94 R46 577 36.6 12.9 3.21

2 0.23 0.87 0.08 0.05 0.00 R35 228 37.7 11.0 2.56

3 0.30 0.06 0.02 0.03 0.89 R51 637 60.4 11.8 2.17

4 0.02 0.14 0.00 0.04 0.82 R293 328 47.2 12.9 2.50

All affluent 0.22 0.04 0.04 0.71 R51 886 44.2 12.1 2.74

TypesWestern

CapeEastern

CapeNorthern

CapeFreeState

KwaZulu-Natal

NorthWest Gauteng

Mpumalanga Limpopo

1 0.17 0.05 0.02 0.05 0.13 0.04 0.48 0.04 0.03

2 0.07 0.11 0.01 0.06 0.15 0.11 0.34 0.02 0.14

3 0.19 0.07 0.02 0.06 0.14 0.05 0.40 0.04 0.04

4 0.17 0.10 0.01 0.05 0.12 0.09 0.39 0.02 0.06

All affluent 0.15 0.07 0.02 0.05 0.14 0.06 0.42 0.03 0.06

TypesProportion

UrbanProportion

RuralProportion

MaleProportion

Female

1 0.92 0.08 0.90 0.10

2 0.77 0.23 0.84 0.16

3 0.91 0.09 0.78 0.22

4 0.67 0.33 0.87 0.13

All affluent 0.88 0.12 0.85 0.15

Table 3.4 de tails the re sults of the clus ter anal ysis for the 2000 sur vey. There are sharper

dif fer ences be tween group av er ages for years of ed u ca tion while the race effect is some what

less pro nounced than in 1995, pointing to wards a grow ing role for pro ductivity re lated

characteristics rather than race in de termining the group ings among the affluent.

A non-ra cial group of affluent emerges with al most half of the group con sisting of blacks and

36 per cent whites. This type cap tures 45 per cent of the affluent and is young with an av erage

household head age of 34. The group has an average house hold in come per ca pita of R

32 666, which is be low the av erage for the affluent, but house hold heads be longing to this

group have ed u ca tional at tain ment lev els ex ceed ing the av er age for the af flu ent, sug gest ing

that in come might rise to match or ex ceed av erage lev els for the affluent as the group

matures.

Apart from the non-ra cial group, there are two pre dominantly white groups and one mainly

black group. The pre dominantly white groups have the high est and sec ond high est av erage

household in come per ca pita. The non-ra cial group has an av erage ed ucational at tainment

of 12.7 years and the black group’s av erage is 5.7 years. Of the two pre dominantly white

groups of affluent, the more affluent group has a 14 year ed ucation av erage and the less

affluent group’s average is 12.3.

Table 3.4: Race, per capita in come, house hold size, age and ed ucation per type for 2000

According to Table 3.5, 41 per cent of the pre dominantly black group lives in Gauteng with the

bulk of the re maining mem bers spread be tween the Free State, KwaZulu-Natal, the

NorthWest and Mpumalanga. 55 per cent of the highly affluent white group lives in Gauteng.

Table 3.5: Provincial residence per type in 2000

25

TypesWestern

CapeEastern Cape

NorthernCape

FreeState

KwaZulu-Natal

NorthWest Gauteng

Mpumalanga

NorthernProvince

1 0.04 0.03 0.01 0.14 0.10 0.13 0.41 0.10 0.03

2 0.23 0.09 0.02 0.06 0.16 0.03 0.35 0.04 0.02

3 0.18 0.04 0.02 0.05 0.09 0.03 0.55 0.01 0.02

4 0.17 0.08 0.02 0.05 0.14 0.07 0.39 0.05 0.04

All affluent 0.16 0.07 0.02 0.07 0.14 0.06 0.39 0.05 0.03

Types

Type asshare of

totalBlackshare

Colouredshare

Indianshare

Whiteshare

Per capitaincome

(1995-prices) Age EducationHousehold

size

1 0.17 0.96 0.03 0.01 0.00 23 221 44.6 5.7 1.4

2 0.30 0.05 0.06 0.05 0.83 38 411 57.4 12.3 2.9

3 0.07 0.12 0.06 0.03 0.79 136 402 42.5 14.0 1.9

4 0.45 0.48 0.11 0.05 0.36 32 666 34.2 12.7 2.6

All affluent 0.41 0.08 0.04 0.47 40 264 43.6 11.5 2.4

The mainly black group’s low av erage in come (R23 221), low level of average ed ucational

attainment (5.7 years), small av erage house hold size (1.4) and the dom inance of males (92

percent) sug gests that many of this group’s mem bers may be un skilled mi grant la bourers.

Remittances rep resent an av erage of 17 per cent of this group’s ex penditure while

remittances’ share of ex penditure is neg ligible for all other affluent groups. If the av erage

household in come per ca pita was re defined to exclude sup port pro vided to other house holds

or to in clude the ad ditional in dividuals who are de pendent on this house hold’s in come, most

of this group would fall be low the R22 500 per ca pita per year line of affluence.

Table 3.6: Rural residency and gender of household head per type in 2000

TypesProportion

Urban

Proportion

Rural

Proportion

Male

1 0.92 0.08 0.92

2 0.95 0.05 0.79

3 0.94 0.06 0.80

4 0.92 0.08 0.79

26

Conclusion

The paper shows that the prev alence of Black affluence is low, but has grown con siderably.

Blacks’ share of the affluent rose from 22 per cent in 1995 to 41 per cent by 2000. The

proportion of blacks among the very affluent in creased from 15 per cent to 28 per cent over the

same pe riod. How ever, it should be noted that the sur vey may over estimate shifts in the ra cial

composition of affluence dur ing this pe riod due to underenumeration of whites in the 2000

IES/LFS. Adjusting for the sur vey’s over statement of growth in black affluence is un likely to

affect the con clusion that there has been a strong in crease in the so cial mo bility of blacks

since the po litical tran sition in 1994. It is clear that there has been sub stantial prog ress in

erad i cat ing la bour mar ket dis crim i na tion.

The anal ysis of the up per part of the income dis tribution con firms many of the tra ditional

views of so cial mo bility in South Africa which have mostly been ac quired through ana lysing

the lower part of the income dis tribution. The paper finds a ro bust as sociation be tween

affluence and the gen der and age of the house hold head as well as ru ral res idence. These

relationships ap pear to be robust across population groups.

The paper’s re sults are also con sistent with con vex re turns to ed u ca tion and a sub stan tial

role for the qual ity of ed ucation. The per sis tent sig nif i cance of the white ed u ca tion in ter ac tion

effect shows that ra cial differences in the re turns to ed ucation have en dured. Anal ysis of

South African ed ucation data dem onstrates that blacks are still re ceiving an ed ucation that is

in fe rior in qual ity, which strength ens the plau si bil ity of ed u ca tion qual ity as an ex pla na tion for

the con tinued sig nificance of the in teraction effect. Thus, it ap pears that race con tinues to be

an im portant de terminant of the al location of char acteristics that earn a return in the labour

market, thus constraining social mobility.

The typology of the affluent in dicates the race effect is some what less pro nounced in 2000

and while there are sharper dis tinctions be tween group av erages for years of ed ucation,

sug gest ing an in crease in the im por tance of pro duc tive char ac ter is tics in de ter min ing the

delineation of group ings among the affluent. The anal ysis for 2000 shows the emer gence of a

large, young, ra cially in te grated gen er a tion of af flu ent, which can be op ti mis ti cally in ter preted

as ev idence of a move towards a dy namic non-ra cial society.

27

Ref erences

Bhorat, H. and Leibbrandt, M., 2001. Modelling Vulnerability and Low Earnings in the South

African Labour Market. In Fighting Poverty: Labour Markets and Inequality In South

Africa, edited by Haroon Bhorat, Murray Leibbrandt, Muzi Maziya, Servaas van der

Berg and Ingrid Woolard. Cape Town: UCT Press.

Card, D. and Lemieux, T. 1997. Adapting to Circumstances: The Evolution of Work, School,

and Living Arrangements Among North American Youth. NBER Working Paper No.

6142.. Available from www.nber.org

Chamberlain, D. and Van Der Berg, S. 2002. Earnings Functions, Labour Market

Discrimination and Quality of Education in South Africa. Stellenbosch Economic

Working Papers 2/2002. Available from http://www.sun.ac.za/econ/w_paper_list.htm

Diamond, C.A., Simon, C.J. and Warner, J.T. 1990. A Multinomial Probability Model of Size

Income Distribution. Journal of Econometrics 43: 43-61.

Ermisch, J. and Di Salvo, P. 1997. The Economic Determinants of Young People’s Household

Formation. Economica 64: 627-644.

Glewwe, P. 1988. The Distribution of Welfare in Cote d’Ivoire in 1985. Living Standards

Measurement Study Working Paper 29. Washington D.C.: World Bank

Hentschel, J. and Lanjouw, P., 1996. Constructing an Indicator of Consumption for the

Analysis of Poverty. Principles and Illustrations with Reference to Ecuador. Living

Standards Measurement Study Working Paper 124. Washington D.C.: World Bank

Kingdon, G. and Knight, J. 2002. Quality of schooling and the race gap in Labour market

outcomes in South Africa. Paper to the Conference of the Centre for the Study of

African Economies: Oxford, March 2002.

Keller, S. 2002. Household Formation, Poverty and Unemployment – the Case of Rural

Households in South Africa. Paper 7 of the FASID country background report on

Poverty and the Role of Rural institutions in a globalising South African economy.

Keswell, M. 2001. Intergenerational Mobility: A Study of Chance and Change in

Post-Apartheid South Africa. DRPU/FES Conference on Labour Markets and

Poverty in South Africa, 15 to 16 Novembers. Available from

http://www.commerce.uct.ac.za/DPRU/Conference2001.htm

Klasen, S. and Woolard, I., 2000. “Surviving Unemployment without State Support:

Unemployment and Household Formation in South Africa”. The Institute for the Study

of Labour (IZA), Discussion Paper No. 237. Available from www.iza.org/

publications/discussionpaper2000.html

28

McElroy, M.B., 1985. The Joint Determination of Household Membership and Market Work:

The Case of Young Men. Journal of Labour Economics 3 (3): 293-314.

Leibbrandt, M., Bhorat, H. and Woolard, I. 1999. Understanding Contemporary Household

Inequality in South Africa. DPRU Working Paper, May 1999.

Leibbrandt, M., and Woolard, I. 1999. Household Incomes, Poverty and Inequality in a

Multivariate Framework. DPRU Working Paper, December 1999.

Ravallion, M., 1992. Poverty Comparisons: A Guide to Concepts and Methods. Living

Standards Measurement Study Working Paper 88. Washington D.C.: World Bank

Woolard, I. and Leibbrandt, M., 2001. Measuring Poverty in South Africa. In Fighting Poverty:

Labour Markets and Inequality In South Africa, edited by Haroon Bhorat, Murray

Leibbrandt, Muzi Maziya, Servaas van der Berg and Ingrid Woolard. Cape Town:

UCT Press.

Van Der Berg, S. and Marincowitz, G. 1999. Pov erty and af fluence: Trends in and pros pects

for black in comes. Chap ter for the Institute for Fu tures Re search, Stellenbosch,

June 1999.

29

Addendum A: 1995 OHS/IES and the 2000 LFS/IES15

Concerns relating to field work, data entry and coding

There are sev eral prob lems in the In come and Ex penditure sur vey (IES) 2000 re lating to the

individual ob servations. There are 308 cases where the sur vey ei ther gives a miss ing value

for the house hold’s food ex penditure or re ports zero ex penditure. The sur vey also shows a

strong in crease in the pro portion of house holds that paid no in come tax (from 49 per cent in

1995 to 77 per cent in 2000) which does not seem plau sible. On the in come side there are

also prob lems with some cat egories: the sharp de cline in occupational per quisites, share of

income at tributable to the house hold as a whole and other spec ified and un specified in come

may point to wards a de terioration in the quality of field work in 2000. The def i ni tions and

categories in the sur vey were de signed so that in come and ex penditure should add up to the

same to tal, but there are wor ry ing dis crep an cies be tween in come and ex pen di ture to tals in

the 2000 sur vey. Only 3.6 per cent of the ob servations in the 1995 sur vey had a difference

between to tal in come and to tal ex penditure that exceeded 30 per cent, com pared to 13.7

percent of the ob servations for the IES 2000 survey.

There are also some sim ple ad dition mis takes in the 2000 sur vey: some of the

rec re ation/en ter tain ment and hous ing sub to tals were mis cal cu lated and grain was dou ble

counted in to tal ex pen di ture. For several ob servations the com ponents of in come do not add

up to to tal in come.

There is ev idence that sug gests that the data qual ity might be particularly poor for Gauteng.

This prob lem might have an im pact on the re liability of the con clusions of this paper be cause

a sub stantial share of the emergent black affluent lives in Gauteng.

In both 2000 and 1995 there are prob lems with the matching of in dividuals and house holds

between the IES and the re lated LFS, which could be at tributed to field work, cod ing or data

entry mis takes. When trying to merge the 2000 IES and LFS, 103 732 cases match

successfully, but there are 1 639 cases unique to the LFS dataset and 421 cases unique to

the IES. Of the matched cases, there are 268 cases for which the race vari able from the two

datasets does not match, 839 for which gen der does not match, and 1 263 cases for which

age does not match (only 178 of these cases had an age difference of one year, which one

can prob ably safely ig nore). There are 2087 cases where one or more of these vari ables

(race, gen der, age) do not match be tween the two datasets. Al together 8984 in dividuals are

members of house holds for which one or more of these vari ables do not match across the two

datasets, leav ing only 96 808 in dividuals in house holds where there are no match ing

problems of some sort (91.5 per cent of 105 792 cases in the two datasets, or 92.9 per cent of

the 104 153 cases in the IES per son dataset).

30

15 This summary of the problems in the IES 2000 draws on comments from Laura Poswell, Charles Simkins andIngrid Woolard.

Concerns relating to sampling

There is also ev idence of overnumeration of the white pop ulation in 1995 and

underenumeration of the white pop ulation in 2000. The 1995 sam pling frame also

undercounted single black house holds in 1995 and this was cor rected in 2000.

The sam pling and weight ing prob lems in the sur veys re sult in dis crepancies be tween the

demographic and in come data given by external data sources and the weighted to tals of the

IES. The na tional ac counts say that real house hold in come per ca pita rose 7 per cent from

1995 to 2000, while com parisons be tween the IES 1995 and 2000 in dicate that in come per

capita had fallen. Also, the 1995 IES over estimates the pop ulation size and the 2000 IES

un der es ti mates the pop u la tion size.

The prob lems with the sur vey’s field work and sam pling frame re sult in sub stantial

discrepancies be tween Na tional Ac counts and the 2000 IES’ es timation of total in come. In an

attempt to cor rect for this dis crepancy, in come was ad justed up wards by 38 per cent.

Because the mea sure of affluence used in 2000 was de fined rel ative to affluence in 1995, the

adjustment to IES only aimed to match in come changes in the two IES sur veys to changes in

National Ac counts to tals over the pe riod.

31