Education, poverty and inequality in South Africa1
Transcript of Education, poverty and inequality in South Africa1
Education, poverty and inequality in South Africa1
Paper to the conference of the Centre for the Study of African Economies on Economic growth and poverty in Africa
Oxford, March 2002 by
Servaas van der Berg University of Stellenbosch
Abstract: South African poverty and inequality are strongly rooted in the labour market. Despite the continuing relevance of race, labour market race discrimination has declined as cause of inequality compared to other factors also often correlated with race (e.g. education and location). Moreover, if cognisance is taken of large differentials in educational quality, the residual earnings differentials ascribable to labour market race discrimination may well be small. This emphasises the need to concentrate on the one factor amenable to policy, education. The paper uses census and survey data to show that quantitative educational attainment differentials in South Africa (years of education) have been substantially reduced. Qualitative differentials remain larger, though. Despite massive resource shifts to black schools, overall matriculation results actually deteriorated in the post-apartheid period. Thus the school system contributes little to supporting the upward mobility of poor children in the labour market. The persistence of former racial inequalities is reflected in extremely poor pass rates in mainly black schools (the majority of schools), but with high standard deviations. Regressions of matriculation pass rates from school level data show that racial composition of schools remains a major explanatory factor besides socio-economic background (as measured by school fees set by school governing bodies) and educational inputs (measured by teacher-pupil ratios and teacher salaries as proxy for qualifications and experience). The remarkable differentials in performance amongst black schools cannot be accounted for by socio-economic background or teaching resources, pointing to the importance of school management and to the malfunctioning of large parts of the school system being largely a problem of x-inefficiency rather than allocative efficiency. The optimistic policy interpretation is that additional fiscal resources can play only a limited role, so addressing the issue need not be costly; the pessimistic view is that good administration and management are even more scarce than fiscal resources.
1. Introduction Education enhances the earnings potential of the poor, both in competing for jobs and earnings
and as a source of growth and employment in itself. As Kanbur (1998: 20) puts it, �The
distribution of physical and human capital emerges from the theoretical and empirical
1 This paper draws on some of the author�s ongoing research (only some of it published), inter alia Van der Berg 2001a; 2001b; 2001c; van der Berg et al. 2002.
2
literature as the key to distributional consequences of growth, and as the determinant of
growth itself.�
South Africa is a country with a notoriously skew distribution of income and consequently
high poverty levels for an upper-middle-income developing country. A comparison with Latin
America, a continent sharing South Africa�s middle-income status and high inequality, is
instructive. A large part of Latin American inequality �relates to the difference between the
top 10 percent of the population and the rest� (Inter-American Development Bank (IADB)
1998: 1), but this applies even more to South Africa: The per capita income of the richest
decile exceeds that of the second richest decile by 60% in the USA, by 160% in Latin
America, and by 208% in South Africa2. And like in Latin America, in South Africa, too,
�much of this gap between the top 10 percent and the rest reflects the � slow and unequal
progress in improving the level and quality of schooling� (IADB 1998: 2).
This paper is concerned with this human capital inequality, described by Simkins (1998:4, 8-
11) as one of �apartheid�s footprints in the sand of poverty and inequality�. The premise is that
South African poverty and inequality are strongly rooted in the labour market, thus the origins
of labour market earnings differentials require attention. Despite the continuing relevance of
race for identifying the poor and analysing labour market outcomes, labour market race
discrimination has declined as cause of inequality compared to other factors also often
correlated with race, such as education, location (urban/rural), family size and composition.
Moreover, the educational returns literature needs to be interpreted in the South African
context even more circumspectly than usual, given considerable variations in educational
quality amongst schools and over time usually ignored in earnings functions that may account
for much of the residual earnings differentials usually ascribed to labour market
discrimination. This emphasises the need to concentrate on the one factor amenable to policy
intervention, education, to reduce earnings inequality through more equal educational
attainment and quality.
This paper explores some of the distributional consequences of education, using census and
survey data to show that quantitative educational attainment differentials in South Africa
(years of education) have been substantially reduced. Qualitative differentials remain larger,
2 The South African data are based on expenditures, not income; incomes may even show greater inequality.
3
though. The next section (Section 2) looks at the relation between education, inequality and
poverty in the South African context, whilst also presenting an overview of the educational
situation in South Africa. Despite massive resource shifts to black schools, overall
matriculation results actually deteriorated in the post-apartheid period. Thus the case will be
made that the school system contributes little to supporting the upward mobility of poor
children in the labour market. After a brief discussion of resource allocation in education,
Section 3 applies regression analysis to school level data to show that racial composition of
schools remains a major explanatory factor of matriculation pass rates, besides socio-
economic background (as measured by school fees set by school governing bodies) and
educational inputs (measured by teacher-pupil ratios and teacher salaries as proxy for
qualifications and experience). The conclusions follow in Section 4.
2. Education, inequality and poverty in South Africa
2.1 Racial inequalities in educational attainment
Schooling inequality between races as reflected in years of education completed (educational
attainment) is still large (Table 1), but has been substantially reduced in the past decades. For
instance, Lam (1999) shows the decline in inequality in years of education completed between
two birth cohorts separated by 30 years, reproduced in Table 2. Schooling variance declined
even amongst blacks, while schooling inequality between the different races also greatly
declined. Blacks in the cohort born in 1920 had a mean backlog of 8.0 years of education
compared to whites; those born in 1950 still a 6.0 year backlog, the 1960 cohort a 4.6 year
backlog, and the 1970s cohort a backlog that had been reduced to only 3.2 years.
Table 1: Years of education completed of adult population (20 years or more) by race and location, 1996
Urban Rural Total Black 9.0 7.9 8.5 Coloured 8.9 6.8 8.6 Indian 10.4 9.7 10.4 White 11.9 11.7 11.9 Total 9.7 8.1 9.2 Source: Census 1996
4
Table 2: Educational inequality for two SA cohorts, 1995
Cohorts 55-59 Cohorts 25-29 Mean 5.77 9.05 Standard deviation 4.51 3.60 Coefficient of variation 0.78 0.40 Gini 0.44 0.21
Source: Lam 1999: Table 2
2.2 Educational inequality and labour market earnings
In South Africa, an expanding literature shows educational inequality to be a determining
factor in earnings distribution. In a recent study, Bhorat and Leibbrandt (2001) show that
education affects the propensity of blacks to participate in the labour force, their probability of
being employed and their earnings, with returns to secondary education being particularly
high. Work by inter alia Moll (2000), Mwabu & Schultz (1996), Fallon & Lucas (1998) and
Hofmeyr (1998) broadly confirms the importance of education for blacks, the largest but also
poorest race group.
Between 1980 and 1993, South African earnings inequality decreased between race groups,
whilst it increased within race groups (Moll 2000). The net result was to leave overall earnings
inequality largely unchanged, as Table 3 shows. Improved black educational attainment
probably played only a minor role. Moll rather ascribes widening within group earnings
inequality to the removal of labour market discrimination, with some blacks benefiting from
new opportunities for upward occupational mobility, while the earnings premium formerly
enjoyed by even poorly educated whites declined once they lost the protection they had
historically enjoyed. Thus even though education was not directly responsible for changes in
earnings, its distribution determined who could benefit from the new opportunities for blacks
in the labour market.
5
Table 3: Inequality of monthly earnings by race, 1980 & 1993
1980 1993Gini coefficient 0.52 0.51Coefficient of variation 1.19 1.12L-statistic * 0.49 0.50L-statistic: Blacks 0.14 0.28L-statistic: Indians 0.21 0.28L-statistic: Coloureds 0.25 0.35L-statistic: Whites 0.21 0.28Within-group inequality 0.17 0.29Between group inequality 0.32 0.21
* The L-statistic (mean logarithmic deviation) is an additively decomposable measure of inequality that ranges from 0 (complete equality) to infinity. If utility has a logarithmic form, L measures the difference between maximum social welfare with a given income (the ideal state of distribution) and the actual social welfare (Moll 1998: 4). It is calculated as the mean of the natural logarithms of earnings, minus the natural logarithm of mean earnings. Source: Mol 1998: Tables 1 & 2
Research has shown the importance of education not only for earnings but also for labour
force participation and employment. Ferreira and Litchfield (1998: 32) report that between
one-quarter and one-third of income differentials between households in Chile can be ascribed
to differences in the educational attainment of the household head; in South Africa this
proportion is lower (about 16% in 1995), yet still very important. Figure 1 provides an
alternative presentation of the influence of level of education of the household head on
household poverty. For levels of education lower than matriculation (completion of secondary
education), there is clear poverty dominance, with more education of the household head
always being associated with less household level poverty. But at higher education levels
(matric or more), where poverty is far less prevalent, other factors (e.g. race, location,
household size and composition, or education of other household members) intervene to
reduce the role of educational attainment of the household head in household-level poverty.
6
Fig 1: Cumulative density curves of per capita earnings by education of household head, 1995
2.3 Quality differentials in education
One should not forget, though, that the quality of education still varies considerably. This is
again not unique to South Africa: In some Latin American countries, "the poor receive an
inferior quality of schooling� (IADB 1998: 53), with the result that ��individuals from the
lower deciles receive a primary education whose quality (measured in terms of income
generation capacity) is 35 percent lower than that of the next decile above.� (IADB 1998: 54).
Although the old dividing lines of race have blurred in education, with many black pupils now
attending formerly white schools, data for 1977 for 7 provinces summarised in Table 4 shows
that most black pupils (96%) were still in schools which were predominantly (more than 70%)
black. About 5% of the 400 000 pupils in mainly white schools were black, whilst in �mixed�
schools (where no race group constituted more than 70% of pupils), 40% were black. (Van der
Berg 2001b). But there is great quality diversity in mainly black schools, and as a group most
formerly black schools still perform much worse than white schools, as reflected in matric
pass rates (a topic discussed in more detail in Section 3.2). Judging by the high matriculation
failure rates, lenient promotion policies in black schools may cause educational attainment at
levels below matriculation to give an inflated impression of educational standards reached, as
reflected in cognitive levels mastered.
7
Table 4: Schools, pupil-teacher ratios and pupils by school "race type" in seven
provinces, 1997 (Eastern Cape and Mpumalanga not included)
Mainly black
schools
Mainly coloured schools
Mainly Indian schools
Mainly white
schools
Mainly "other" schools
Mixed schools
Total
Schools 13 234 1 015 127 660 260 779 16 075Pupil/Teacher
ratio 35.1 31.8 25.7 25.6 31.4 28.5 33.4
Pupils: 5 897 116 538 393 99 122 400 105 178 254 488 511 7 601 501Of whom: Black 5 876 133 24 172 15 023 21 603 4 197 073 6 134 008 Coloured 10 848 512 718 1 339 11 024 12 70 156 606 097 Indian 4 658 800 82 604 3 557 0 111 592 203 211 White 3 696 413 133 363 346 1 104 211 471 800 Other (race
specified) 1 781 290 23 575 178 237 5 479 186 385
Note: A school is classified in a particular �race type� of more than 70% of pupils are from that particular race groups. Schools where not race dominated in this way were classified �mixed�. Whilst �other� schools are those (mainly in the Western Cape) who refused to divulge race information in the survey by the national Department of Education..
Quality differentials are also reflected in the quality of the matriculation itself, in terms of the
standard at which matric is passed as well as the subject choice. Few mainly black schools
provide an adequate background in Mathematics or Science. Only 45% of all matriculation
candidates wrote Mathematics in 1997 (with a marked male bias); only 21% passed it, and
most only attempted Standard Grade Mathematics, a standard far below what is conventional
in developed countries. Even in the Western Cape, the province with the best matriculation
results, only 24% of matriculation candidates attempted Mathematics at the Higher Grade, and
only 20% passed it. Nationally, the proportions who wrote and passed Science were even
lower at 25% and 16% respectively. Only 50% and 42% of teachers teaching Mathematics and
Science have studied these subjects beyond secondary school level (Edusource 1999: 5).
Another indication of inequality in educational output at higher standards can be gleaned from
data for the Western Cape. As Western Cape pass rates are almost uniformly high (almost
80% of all candidates pass matric), differences in pass rates between schools are relatively
low, as Table 5 shows. However, if more onerous levels of school performance (university
8
exemptions or percentage A-aggregates achieved3) are evaluated, inequality increases
considerably.
Table 5: Inequality of three measures of educational outcomes at matriculation level between schools in the Western Cape, 1997
Passes University Exemptions
A-aggregate
Mean 80.6% 23.0% 2.6% Standard deviation 22.5% 22.0% 5.0% Coefficient of variation 0.28 0.96 1.94 Gini coefficient 0.15 0.56 0.80 Source: Own calculations from Western Cape Education Department data
Fig. 2 shows literacy and numeracy test scores for 1993 for blacks and whites aged 12 to 18,
where questions have been set at approximately Grade 7 (age 12) level (see also Fuller et al.
1995). Not even the performance of whites is very encouraging, but what is particularly
alarming is that blacks perform far worse on both tests � despite the fact that educational
levels attained by blacks and whites differ relatively little at this age. The unsatisfactory black
performance in matric appears to be partly the delayed effect of lower cognitive achievement
levels at earlier ages. Although blacks aged 12 to 18 in 1995 had attained on average 78% of
white years of education, in 1993 their literacy scores were 55% of white levels, and their
numeracy scores lagged even further at 46% of white levels.
3 University exemptions is a measure of the quality of the matriculation pass obtained; such an exemption has usually acted as entrance barrier to universities. A-aggregates are average distinction levels, obtained by a very small proportion of candidates.
9
Fig. 2: Literacy and numeracy test scores of black and white teenagers (12-18 years),
1993 (scores converted to percentages; ratios relative to white scores)
Literacy and numeracy scores of black and white teenagers (12-18 years), 1993
(scores on 8 point scale converted to percentages; ratios relative to white scores)
38.3%
27.7%
55.1%
45.6%
78.0%
69.5%
60.7%
0%
20%
40%
60%
80%
100%
Literacy score Numeracy score Literacy ratio Numeracy ratio Years educationcompleted: Ratio
BlackWhite
In a previous study (Van der Berg et al. 2002), the same 1993 data from the Living Standards
and Development Survey were used to try to explain combined numeracy and literacy test
scores of teenagers. The regressions (see Appendix) showed that educational attainment does
matter for numeracy and literacy performance, but some other factors also have an influence,
including race (Indians considerable outperform the rest, whilst blacks do about 10% worse
than whites and coloureds of similar education and socio-economic background), location,
parent education, and economic status (as measured by per capita household expenditure). The
regressions explain 39% of the variation in test scores, whereas educational attainment alone
can only account for 21%. Clearly, pupil home background influences not only how well
students progress through school, but also the quality of how well they learn, as measured by
these test scores. Both their educational attainment and their educational quality influence how
well they later fare in the labour market. The regressions imply that black teenagers have a
backlog in test scores compared to whites and coloureds of similar educational levels
equivalent to almost three years of education, and of six years compared to Indians, even when
standardising for home language.4 This illustrates the inability of the former black school
system to provide the educational quality required to integrate most school leavers into a
4 Using the same data for a different purpose, Case and Deaton�s found an even far bigger backlog of black teenagers, equivalent to almost 10 years of education completed: �(F)our additional years generate one additional correct answer on the tests� (Case & Deaton, 1999: 26; cf. also Table 7).
10
modern economy � the yardstick by which, from an economic viewpoint, the educational
system must be measured.
Considering these quality differentials, racial wage differentials for persons with similar
education and experience may result less from labour market discrimination, as is usually
presumed in earnings functions, than from pre-labour market discrimination in school quality.
Measured wage discrimination had declined from the 1970s. As early as 1989, black wages
were barely 15% lower than those of whites in a similar job grade and of the same gender, a
substantial reduction from the 43% differential of 1976 and the 22% of 1985 (McGrath
1990:97), i.e. there was progress towards the "rate for the job" (though this says nothing about
differential access to particular jobs). Using other data sources, Moll (1998:1 & Table 10)
came to a similar conclusion: total discrimination fell from 20% of the Black wage in 1980 to
12% in 1993. Work by Chamberlain (2001) confirms that a substantial proportion of the
unexplained residual in earnings between race groups may be due to differentials in
educational quality rather than conventionally measured labour market discrimination.
2.4 Educational inequalities amongst blacks
Amongst blacks, educational inequality largely follows the lines of income: more affluent
households are better able to support their children through school, implying increasing
stratification within black society. Children from the top two black deciles progress
considerably better through the school system than their poorer counterparts and only at age
15 start falling behind whites. Case & Deaton (1999: 21) conclude that private resources were
a major factor determining differential black educational outcomes under apartheid. �Pupils in
better-off Black households do better in their education, and we find no parallel for Whites.
That the education of Blacks but not Whites is constrained by financial resources is further
supported by the fact that many Blacks who are not in school (but not Whites) � report lack of
resources as the reason.� (Case & Deaton 1999: 28). Furthermore, greater recent access to
formerly white schools for more affluent blacks may have accentuated qualitative educational
differentials amongst blacks.
Data from the 1996 census show mean earnings of full-time employed black workers for
whom the educational level of a parent is known (i.e. children of the head of household still
resident in the household) to be substantially higher where the household head has at least
matriculated. But is this perhaps solely due to more educated parents having more educated
11
children, i.e. to differential attainment? Table 6 shows mean income of such children who
have completed at least matric. Differentials remain substantial: In some way the better
education of the parent (household head) translates into higher earnings for children even
compared to other young workers who also have matriculated, but where the parent had less
education. However, it is not clear whether this measures the quality of education, or some
other non-observed aspect of human capital transmitted from parents to children. Such premia
do decline, though, to about 9% in cases where the children have graduated.
Table 6: Mean monthly earnings of full-time employed black children of head of
household by education and whether head of household has matriculated, 1996
Mean monthly income by education
Matric Matric + diploma or certificate
Matric + other non-
degree
Degree
Head of household matriculated R1 731 R2 658 R2 849 R3 388 Head of household not matriculated R1 380 R2 285 R2 164 R3 104 Premium for head being matriculated 25.4% 16.3% 31.6% 9.1% Note: Cases where the worker reported being in full-time employment but reporting no income were excluded. This had only a very minor effect on the premia. Source: Own calculations from Census 1996 10% Sample.
2.5 Earnings and the demand for skills
Educational differentials in association with the demand for labour determine returns to
education.5 Little is known about the evolution over time of South African returns to
education, not even to speak about what these would have been in the absence of apartheid-
based labour market interventions. Bhorat & Hodge (1999) have shown that South African
labour demand patterns reflect a growing demand for higher skilled labour and declining
demand for low-skilled workers. International experience also indicates that without an
acceleration in the availability of such skills, educational premia are likely to remain high. In
the USA, only an enormous expansion in secondary schooling after 1910 made possible a
reduction in the returns to education until the 1950s, wherafter returns again rose as skills
demand outstripped their supply because of �skill-biased technological change� (Goldin &
Katz 1999: 25; for more recent evidence, see also Murphy & Welch 1994). Thus reducing
5 Returns to education result from the interaction between the supply and demand for human capital, and the latter is related to the economic growth path, which is itself changing. Moreover, insofar as education itself may determine growth, there is an endogeneity problem that cannot be resolved in forecasting the evolution of earnings.
12
labour market inequality would require substantial improvement in the supply of skills
through more and better quality education. The needs of the economy in terms of the type
rather than the level of educational output should also be considered, and may be an
unobserved variable that affects the returns to education (e.g. the importance of Mathematics
for further training).
3. Implications for education policy: More resources, or better resource use?
The foregoing all points to the importance of improving the educational attainment and quality
of education (as measured by cognitive tests) of the poor in order to reduce earnings inequality
and poverty. In terms of resource use in education, two alternative routes are possible:
applying more resources, or using resources better.
3.1 Fiscal resources for education
South Africa allocates, by international standards, a large share of its national resources to
education; its public education spending ratio of about 7 per cent of GDP is amongst the
highest in the world (without even considering the skills training levy). Moreover, education
spending has already increased rapidly. Shifting substantially more fiscal resources to
education does not appear to be a viable proposition. Moreover, larger financial flows to
education in the past five years did not in fact increase real resources for education
commensurately, as fiscal resource shifts were overshadowed by wage increases for teachers;
the pupil/teacher ratio declined only marginally from 33.7 to 32.7 from 1996 to 2000 (South
Africa 2001: 35). In contrast to the international experience6, teacher salaries outpaced the
growth of per capita GDP. Cutbacks in educational personnel in some richer provinces were
barely matched by increases in personnel in educationally more poorly endowed provinces,
despite much larger fiscal shifts. As Donaldson perceptively remarked some years earlier, "...the constraint at work � is not (only) finance, but the limited real resources available to the economy. Competent teachers, nurses, doctors and community workers are scarce, as is the capacity to produce books, medical supplies, and building materials. So the growth and improved distribution of social services must be viewed as the growth and improved distribution of the inputs required for delivering these services." (Donaldson 1993: 147)
6 Lee and Barro (1997: 17/18) provide some evidence of the relative decline of teachers� wages in international context:
�The ratios of estimated real salaries of primary school teachers to per capita GDP have typically declined over time; from 1965 to 1990, the value dropped from 2.5 to 2.2 in the OECD, from 4.9 to 3.6
13
The strong bargaining power of teacher unions allowed them to raise their real salaries
substantially. Unlike their public sector contemporaries, black teachers did not benefit from
the wave of black advancement following democratisation, as there were few opportunities for
promotion within teaching7, thus their frustrations were often vented in wage bargaining. Thus
fiscal resources for education increasingly had to be directed to personnel spending, leaving a
growing dearth of non-personnel spending: from 1995/6 to 1997/8, real personnel expenditure
increased by 20% while non-personnel expenditure in schools declined by 17% (South Africa
1998: 27). Yet despite more aggregate spending, both the number of successful matriculants
and university exemption remained stagnant or even declined, whilst the school-age
population was growing, implying that declining proportions of the school-completing age
were attaining matriculation or exemptions.
As the growth in pupil numbers still exceeds the growth rate of the economy due to continued,
though declining, growth in the school-age population and longer retention of pupils in the
school system, a government team investigating the medium term expenditure framework
concluded that there will be a major funding problem in education in coming years, unless
! more funds are allocated to education � which they regard as fiscally infeasible, and which
internationally has been shown not always to improve educational outcomes (Gupta et al.
1999: 4);
! pupil-teacher ratios rise even further � which is unacceptable to government, teacher
unions and parents alike;
! teacher salaries decline in real terms � which is strongly opposed by the teacher unions;
! some combination of the above occurs (South Africa 1998).
Thus there is limited scope for substantially increasing fiscal resources per pupil in coming
decades, so the question of resources largely revolves around issues of resource allocation
versus resource efficiency. If the malaise of the South African educational system lies in x-
inefficiency rather than in allocative inefficiency, reallocating resources between levels of
education would bring little gain, and it is not even clear which level of education most
requires additional resources. There is perhaps a stronger case for shifting more financial
in the overall group of developing countries, and from 7.4 to 1.7 in the CPEs (centrally planned economies).�
7 De Villiers (1996: 288-9) reports that more than 90 per cent of teachers will not receive more than one promotion in a lifetime of teaching
14
resources to non-personnel teaching resources; personnel spending is so dominant that even a
small relative shift could greatly increase the availability of classroom resources.
3.2 Efficiency in educational resource use
However, the clearest need appears to be in utilising existing resources better, even in their
present application. The major inefficiencies are in what used to be the black school system,
by far the largest part of the system, where the quality of learning in schools is often abysmal.
The Culture of Learning, Teaching, and Service (COLTS) campaign launched in 1996 �was
the first more or less official recognition of the fact that efficiency and work effort problems,
rather than funding by itself, were at the heart of the problems in the education sector� (South
Africa 1998: 35). Former President Mandela, and later also the new President and the new
Education Minister, publicly put the blame for mediocre education results on lack of teacher
discipline.
Such inefficiencies result from a typical principal-agent problem. Educational outputs are
notoriously difficult to monitor, as is teacher effort (input), thus low teacher productivity is
hard to overcome through incentive schemes. The educational authorities have responded by
attempting to shift monitoring to the parent community as the final �principal�. Unfortunately,
however, this policy is least successful in schools where parents themselves have had little
formal education and therefore are hesitant to confront teachers or ignorant about what can be
expected of teachers � precisely those schools where failure rates are greatest. Moreover, lines
of authority are also not always clear and school principals often find it difficult to act against
undisciplined teachers or pupils.
As always when there is a principal-agent problem, one avenue for improvement is through
providing more information. The education authorities have a paucity of information to
analyse the educational situation and their policy options. Presently, the only measure of
educational output available to them is matriculation results, but these still do not identify the
roots of the problem (nor are they properly analysed at the school level). It has been shown
above that literacy and numeracy levels amongst many blacks are already far below par as
early as age 12. Allocating resources based on matriculation results cannot adequately address
a problem which requires much earlier intervention. Where to direct resources cannot be
decided without information on the qualitative performance of different parts of the school
system. This requires large scale and continued efforts at measuring cognitive achievement at
15
different levels, in order to better understand the relationship between home background of
pupils, educational inputs, and enhanced cognitive achievement. Moreover, identifying
schools performing badly in order to take remedial action requires a better understanding of
how schools perform and the causes thereof.
Against this background, school level data may shed light on whether matriculation results
vary systematically with socio-economic background and with educational inputs.8 The policy
issue is whether educational resource shifts can effectively overcome differentials flowing
from socio-economic background or accumulated educational backlogs from the previous
dispensation.
The data consists of matriculation pass rates for 1999 and 2000 for secondary schools who had
matriculation candidates in the latter year, matched to data on 1997 school resources.
Matching the data records was not always possible, as schools were identified in different
ways in the 1997 data set than in the examination data published by the national Department
of Education. The matching process could not be completed for one of the seven provinces
covered in the 1997 data set, leaving data for only six of the nine provinces. Although the final
sample did include a large proportion of schools in all of the six provinces covered, it is
unclear whether missing observations were non-random. The sample size of almost 2 800
covers about half of all schools, matriculation candidates and matriculation passes, and about
70% of those in the provinces included. Matriculation class size in schools included in the
sample was above average for the six provinces included, while the 56% pass rate in the
sample differs little from the actual average in these provinces (58%). Despite variations in the
sampling proportions across provinces (the Free State, Northern Cape and Northern Province
were better presented than average) largely due to the problems of matching the two data sets,
it was decided not to re-weight the data at the provincial level. Thus the results apply to a
fairly but not fully representative sample of about half of all schools and matriculation
candidates nationally.
In addition to the variables in the 1997 data set, the examination data set contains pass rates
for 1999 and 2000, as well as the number of candidates who wrote the examination in 2000.
8 Moreover, progress to matric favours the more affluent populations, and the quality of matric, in terms of grades attained, subject choice and cognitive outcomes, may further differ systematically to the detriment of the poor.
16
The correlation (r=0.85) between the pass rates for 1999 and 2000 indicates moderate but not
great stability in pass rates at the school level. The average pass rate was used in the analysis
to reduce the effect of year to year variations, especially for small schools, and the number of
candidates who wrote in 2000 was used to weight each observation (number of candidates was
unavailable for 1999). The 1997 racial composition of schools provides a fair approximation
of their matric class in 1999 and 2000, and 1997 school fees and educational resources were
taken as proxy for socio-economic status and availability of educational resources over the
school career.
Table 7 shows that pass rates in schools still differ substantially between socio-economic
groups (proxied by school fees), school �race types� and provinces. The performance in the
poorer schools is generally weak, rising from below 44% in the poorest to 96% in the richest
group of schools, and there is far more variation in the performance in poor schools.
Noticeable is that the standard deviation is much lower in the top two school fee groups � pass
rates in more affluent schools are almost uniformly high.
17
Table 7: Number of schools, mean and standard deviation of average matriculation pass rates in sample by school fee group, �race type� and province
Number of schools Mean Standard deviation Total sample: 2 768 55.5% 27.3% School fee group:
<R20 651 43.9% 23.4% R20-R49 1 177 47.8% 22.2% R50-R99 496 54.1% 22.2% R100-R199 82 69.7% 23.2% R200-R999 243 91.9% 15.0% R1000+ 119 97.2% 8.6%
School "race type": Black 2 106 43.3% 20.3% Coloured 138 75.5% 14.2% Other 97 91.8% 13.9% Indian 42 80.5% 12.3% White 179 97.3% 3.8% Mixed 206 79.0% 21.5%
Province: Free State 289 49.1% 29.5% Gauteng 367 62.2% 29.5% Kwazulu-Natal 784 55.7% 25.5% Northern Cape 87 66.2% 25.6% Northern Province 1 002 43.9% 20.2% Western Cape 239 81.8% 19.7% Source: Own calculations
It is again useful to distinguish schools by �race type�.9 Not only are mainly black schools
worse off in terms of mean matriculation passes (43%), but their pass rates vary considerably,
even though standard deviations are smaller when schools are sorted into �race types�, as
experiences are more uniform within each category. The highest standard deviations occur in
mainly black schools and in schools not dominated by a single race group � such schools have
a quite varied experience, unlike mainly white schools, amongst whom the lowest recorded
pass rate was 68%. Table 8 shows that mainly black schools, by far the majority, perform
abysmally. Most such schools have pass rates in the range 20-60%, and ten percent have even
lower pass rates, whilst only 3 out of 179 mainly white schools have pass rates below 80%.
9 As in Table 6, schools with more than 70% of a specific race group were classified as of that �race type�.
18
Table 8: Frequency distribution of schools by �race type� and pass rate range, 1999-2000
Range of average pass rate
Black
Coloured
Indian
Mixed
Other
White
Total
Below 20% 207 3 1 211 20-39% 743 3 11 1 758 40-59% 668 23 2 33 5 731 60-79% 342 57 16 45 11 3 474 80-100% 146 55 24 114 79 176 594 Total 2 106 138 42 206 97 179 2 768 Below 20% 9.8% 0.0% 0.0% 1.5% 1.0% 0.0% 7.6% 20-39% 35.3% 2.2% 0.0% 5.3% 1.0% 0.0% 27.4% 40-59% 31.7% 16.7% 4.8% 16.0% 5.2% 0.0% 26.4% 60-79% 16.2% 41.3% 38.1% 21.8% 11.3% 1.7% 17.1% 80-100% 6.9% 39.9% 57.1% 55.3% 81.4% 98.3% 21.5% Total 100% 100% 100% 100% 100% 100% 100%
At provincial level, the large differences in pass rates between the best performing province
(the Western Cape with a mean pass rate of 82%) and the worst in the sample (the Northern
Province, with a 45% pass rate) are well known. Notably, performances within provinces are
more varied than that within �race type�, as reflected in the standard deviations.
A priori, one would expect matriculation pass rates to be better:
• The higher the school fee, because of the additional resources this brings and because it
reflects socio-economic status.
• The lower the pupil/teacher ratio, a measure of class size.
• The higher teacher salaries (which partly reflect teacher qualifications and seniority).
• In schools where black pupils are not numerically dominant � black schools are still most
likely to reflect the ravages of the previous education dispensation and are most likely to
have continued problems with the culture of learning. Racial composition still reflects past
privilege.
• In historically better-endowed provinces such as Gauteng and the Western Cape, with
good provincial education management, mainly metropolitan pupils, and who did not need
to incorporate former homeland structures.
19
Broadly speaking, regression results were consistent with these expectations, although
provincial dummies were not very informative.10 Regression 1 in Table 9 shows that the
matriculation pass rates of schools were associated with the socio-economic background as
measured by school fees, teaching resources (both the pupil-teacher ratio and the average
teacher salary in a school, a measure of the quality of teachers), provincial location, but also,
disturbingly, by the race category of schools. The exceedingly high coefficient of
determination for a cross-sectional regression of this nature shows that well over half (57%) of
the variation in pass rates is explained by the equation. Moreover, apart from some provincial
dummies, all variables are highly significant, and the signs for all non-provincial variables are
as expected. The coefficient of the dummy for mainly white schools indicates that, holding
constant the level of school fees, pupil/teacher ratios, teacher salary level and province, a
school containing mainly white pupils has a matriculation pass rate 33 percentage points
higher than a similar school with mainly black pupils (black schools is the omitted or
reference value). This is a highly disturbing, but not surprising, finding � the educational
system still is unable to overcome the effect of past patterns of education, dysfunctional
management structures and dismal functioning of black schools. Although extremely large,
this differential needs to be put into historical context: In 1994, the last year for which racial
data were collected on matriculation pass rates, the black pass rate was 49% as against the
97% of whites. Table 6 showed that pass rates in dominantly black schools in this sample were
43% as against 97% in dominantly white schools, a differential of 54 percentage points which,
as the race type dummy showed, remains very large after standardising for school fees,
educational resources and province.
Kwazulu-Natal and the Western Cape both perform significantly better (5 and 8 percentage
points respectively) than similar schools in the Northern Province (the reference province).
Other provincial dummies are not significant .
10 A full log-transformation of the basic model did not improve results. Robust regression tests show the results to be insensitive to the presence of outliers.
20
Table 9: Regressions of matriculation pass rates by school in six provinces, 1999-2000 (t-values shown below coefficients)
Dependent variable: Pass rate (average 1999 & 2000)
Regression 1: All schools
Regression 2: Mainly black
schools
Regression 3: Other schools
School fees per pupil (Rand per annum)
.0129566*** .0341672*** .0114441***
12.188 9.353 12.735
Pupil-teacher ratio -.1353492*** -.1187243** -.3745764*** -3.397 -2.724 -3.360
Average teacher salary .0004347*** .0004122*** .0009125*** 8.072 6.357 8.962
Mainly coloured school (dummy) 21.92395*** - - 10.952
Mainly Indian school (dummy 24.18364*** - - 9.549
Mixed school (dummy) 20.52412*** - - 13.034
Mainly white school (dummy) 32.95964*** - - 17.584
Mainly other school (dummy) 29.14448*** - - 12.763
Kwazulu (dummy) 5.35849*** 5.8295*** -1.061254 5.425 5.137 -0.329
Free State (dummy) -1.605378 -2.724809* 7.582651* -1.317 -1.978 2.151
Northern Cape (dummy) 3.454839 1.813976 5.108072 1.420 0.455 1.368
Gauteng (dummy) -.9128545 -1.866107 1.339477 -0.852 -1.468 0.423
Western Cape (dummy) 8.418335*** 3.10358 10.51278*** 4.758 1.012 3.285
Constant 9.945317* 10.49816 -4.127132 2.048 1.818 -0.393
N 2 768 2 106 662 R2 .5719 .0862 .4214 R2-adjusted .5698 .0827 .4143 Standard error 17.908 19.396 13.259
* indicates .10 level of significance ** indicates .01 level of significance
*** indicates .001 level of significance Replacing school fees by their natural logarithm (implying that their effect is proportional), or
replacing the pupil-teacher ratio and the teacher salary with a single variable, the teacher cost
per pupil11, had little effect on the results and the other coefficients remained relatively
11 The coefficient for school fees is five times as large as that for teacher cost per pupil. This shows that school fees do more than only augment fiscal resources, for two reasons: • School fees are likely to be spent more effectively than other school expenditures, being controlled by school
governing bodies, who are closer to schools� real needs. • School fees are a good proxy for economic status. Where school choice exists in practice (as in metropolitan
21
unaffected, implying that the underlying logic of the model was not greatly affected by the
particular specification used.
Regressions 2 and 3 did not use school race type as explanatory variables, but instead
regressions were applied to mainly black schools and to non-black schools separately, which
still left the samples at a healthy 2 106 and 662 schools respectively. One should not read too
much into the magnitude of coefficients in a model which is not fully specified, thus
comparison between the regressions should not be interpreted too finely. With this caveat, the
results from these two regressions can be summarised as follows:
• Again, both school fees and educational resources had a significantly positive influence on
matriculation pass rates in both regressions.
• Surprisingly, the coefficient for school fees is almost three times as large in Regression 2
as in Regression 3 � economic status is a very important determinant of matriculation pass
rates in black schools, pointing to strong differentiation within dominantly black parts of
the school system. This is consistent with an interpretation that largely rural black schools,
where communities are generally poorer, perform much worse than black schools in urban
communities, where parents are less poor and often more educated than their rural
counterparts.
• Also surprising is that the coefficient for class size (pupils per teacher) is much smaller in
mainly black schools, precisely those schools in which the historically high pupil-teacher
ratio has usually been seen as particularly detrimental. The coefficient for teacher quality
(as measured by salary) is also somewhat lower for black schools. These results do not
support the widely held view that, at the margin, resource shifts from other schools to
blacks schools would improve overall educational performance, at least not without
additional attention to the efficiency of resource use in these schools.
• Importantly, although school resources and socio-economic background matter, as the
coefficient of determination of 0.08 indicates, the model leaves the very large variation in
pass rates in black schools largely unexplained. In contrast, almost half the much smaller
variation in results in other schools is explained by these variables.
areas or large towns), parents mostly choose the �best� school they can afford for their children. Where distance limits school choice, as in rural areas, school fees are set at a level commensurate with what the overwhelming majority of parents can afford. Thus school fees in both cases reflect the economic status of most parents.
22
• An interesting result concerns the provincial dummies: Kwazulu-Natal�s black schools
fare significantly better than expected, whilst for other schools the Western Cape
significantly outperforms most provinces. The Free State, interestingly, has significantly
lower pass rates in its mainly black schools and higher pass rates in other schools.
Thus race (or what lies behind this variable) still appears to be the principal factor determining
differential matriculation pass rates, followed by economic status (school fees). Even massive
shifts in school level resources since the transition to democracy still leave mainly black
schools performing much worse than white schools, and the regressions indicate that further
shifts of educational resources may have a very limited effect. That black schools fare
systematically worse in terms of pass rates cannot be accounted for by resources or the socio-
economic status of pupils alone. More educational resources for black schools can make only
a small contribution to improving educational outcomes � and therefore their long term
position in the labour market. The pupils who have furthest to catch up are those in rural areas,
where socio-economic status, including education of parents, is weakest, and where good
teachers are hard to come by. Although resources matter, greater resource inputs alone cannot
much improve this situation, without a fundamental reorganisation in how schools function.
The quarter of black schools which bettered the total sample�s average pass rate of 56% have
only slightly better pupil-teacher ratios (32.5 as against 34.0) and slightly better remunerated
teachers (R83 900 versus R82 387) than the rest, but socio-economic status differs more, as
reflected in schools fees of R92 versus R36 per annum. But even this is not a large difference.
Above and below average performance in black schools cannot be explained by differences in
resources, and only to a limited extent by differences in socio-economic status. Moreover, the
greatest improvement in educational chances for black children probably arise for those who
can afford to send their children to historically white schools or to private schools, where pass
rates are far greater. Here too, private resources, as reflected in school fees, matter. But
differences in public educational resources play only a small role in the performance of black
schools. Again, factors associated with the functioning of the school as a productive unit
rather than the availability of resources appear to be crucial to proper performance,
particularly in poor communities.
Thus this analysis supports the contention that school efficiency, and most likely particularly
management, requires most attention to overcome the legacies of past inequalities in
education.
23
4. Conclusion Despite the lasting influence of apartheid, educational access is no longer a major problem in
South Africa, as more than 90% of children of all race groups remain at school until attaining
matric or reaching age 16. The racial gaps in educational attainment (years of education
completed) have also been substantially reduced over the past decades. However, there are
severe problems with the quality of education of a large part of the South African school
system, as reflected in cognitive tests of numeracy and literacy and also shown by
matriculation results. The deficient performance of particularly mainly black schools is a
source of concern, as this shows that reduced earnings inequality may well be more difficult
than rising educational attainment at lower school levels would indicate. Only limited scope
remains for additional resource outlays to redress this malfunctioning of the major part of the
school system. Moreover, the evidence shows that more resources is not the solution to bad
educational performance, as some of the worst performing schools are well-resourced, whilst
some schools perform excellently with limited resources.
This paper has shown that the persistence of former racial inequalities is reflected in extremely
low pass rates in mainly black schools (the majority of schools), but with high standard
deviations, pointing to a varied experience. Regressions show that only a small portion of the
remarkable differentials in performance amongst poor black schools can be accounted for by
socio-economic background or teaching resources, pointing to the importance of school
management. Thus it appears as if the malfunctioning of the bulk of the school system is
largely a problem of x-inefficiency rather than allocative efficiency. The optimistic policy
interpretation is that additional fiscal resources can play only a limited role, so addressing the
issue need not be costly; the pessimistic view is that good administration and management are
even more scarce than fiscal resources.
Given strong economic growth, both poverty and racial inequality may be strongly reduced in
the coming decades, and the trend towards increased inequality within racial groups may be
arrested. But from the perspective of distribution and even as a factor in growth itself, an
improvement in the quality of education amongst South Africa�s poor is likely to be very
rewarding. This requires urgent attention at the highest levels to the functioning of the worst-
performing schools, to ensure continued upward mobility of the largest part of the workforce
and to prevent growth itself becoming endangered.
24
References Bhorat, H & Hodge, J. 1999. Decomposing shifts in labour demand in South Africa. South
African Journal of Economics 67(3): 348-380 Bhorat, Haroon & Leibbrandt, Murray. 2001. Correlates of vulnerability in the South African
Labour Market. Chapter 3 in: Bhorat, Haroon; Leibbrandt, Murray; Maziya, Muzi; Van der Berg, Servaas; & Woolard, Ingrid (eds.). 2001. Fighting poverty: Labour markets and inequality in South Africa. UCT Press: Lansdowne: 74-106
Case, Anne & Deaton, Angus. 1999. School inputs and educational outcomes in South Africa. Quarterly Journal of Economics 114(3): 1047-1084
Chamberlain, Doubell. 2001. Earnings functions, labour market discrimination and quality of education in South Africa. Masters thesis. Stellenbosch: Department of Economics, University of Stellenbosch
De Haan, Arjan; Lipton, Michael; Darbellay, Eliane; O�Brien, David; & Samman, Emma. 1997. The role of government and public policy in poverty alleviation in sub-Saharan Africa. Review of the literature prepared for the African Economic Research Consortium�s collaborative project on Poverty, Income Distribution and Labour Markets. Mimeo. Sussex: Poverty Research Unit
De Villiers, AP. 1996. Effektiwiteit in Suid-Afrika se onderwysstelsel: �n Ekonomiese analise. Unpublished Ph.D. thesis. Stellenbosch: University of Stellenbosch
Donaldson, Andrew R. 1993. Basic needs and social policy: The role of the state in education, health and welfare. In: Merle Lipton & Charles Simkins (Eds.) 1993. State and market in post apartheid South Africa. Johannesburg: University of Witwatersrand Press: 271-320
Edusource. 1999. Edusource Data News 24. Johannesburg: Education Foundation. March. Fallon, Peter & Lucas, R. 1996. South African labor markets: Adjustment and inequalities.
Informal discussion paper No.12 on aspects of the economy of South Africa. Washington, D.C.: World Bank
Ferreira, Francisco H. G. & Litchfield, Julie A. 1999. Calm after the storms: Income distribution in Chile, 1987�1994. World Bank Policy Discussion Paper 1960. Washington, D.C.: World Bank
Fuller, Bruce; Pillay, Pundy; Sirur, Neeta. 1995. Literacy trends in South Africa: Expanding education while reinforcing unequal achievement? Mimeo. Cape Town: Saldru
Goldin, Claudia & Katz, Lawrence F. 1999. The returns to skills in the United States across the twentieth century. NBER Working Paper 7126. Cambridge, Mass.: National Bureau of Economic Research
Gupta, Sanjeev; Verhoeven, Marijn; & Tiongson, Erwin. 1999. Does higher government spending buy better results in education and health care? IMF Working Paper. Washington, D.C.: International Monetary Fund
Hofmeyr, J and Lucas, R. 1998. The rise in union wage premia in South Africa. Institute for Economic Development Discussion Paper 83. Boston: Boston University
Inter-American Development Bank (IADB). 1998. Facing up to Inequality in Latin America. Economic and Social Progress in Latin America 1998-1999 Report. Washington, D.C.: Johns Hopkins University Press
Kanbur, Ravi. 1998. Income distribution and development. World Bank Working Paper WP98-13. Mimeo. Washington, D.C.: World Bank
Lam, David. 1999. Generating extreme inequality: Schooling, earnings, and intergenerational transmission of human capital in South Africa and Brazil. Mimeo. Ann Arbor: University of Michigan
25
Lee, Jong-Wha & Barro, Robert J. 1997. Schooling quality in a cross section of countries. NBER Working Paper 6198. Cambridge, MA: National Bureau of Economic Research.
Londoño, Juan Luis. 1996. Poverty, inequality, and human capital development in Latin America, 1950-2025. World Bank Latin American and Caribbean Studies: Viewpoints. Washington, D.C.: World Bank
Moll, Peter. 2000. Discrimination is declining in South Africa but inequality is not. Studies in Economics and Econometrics 24(3): 91-108
Murphy, Kevin M. & Welch, Finis. 1994. Industrial change and the rising importance of skills. In: Danziger, Sheldon & Gottschalk, Peter. 1994 (Eds.). Uneven tides: Rising inequality in America. New York: Russel Sage Foundation: 101-132
Mwabu, G. and Schultz, T. 1996. Education returns across quantiles of the wage function; Alternative explanations for the returns to education by race in South Africa. American Economic Review 86(2): 335-339
Saldru. 1994. South Africans rich and poor: Baseline household statistics. Cape Town: Saldru Simkins, Charles. 1998. On the durability of South African inequality. Macarthur Foundation
Working Paper. Princeton: Macarthur Network on Inequality and Poverty South Africa, Budget Office, Department of Finance. 1998. 1998 medium term expenditure
review: Education. Pretoria: Department of Finance South Africa, National Treasury. 2001. Intergovernmental Fiscal Review 2001. Pretoria:
Government Printer Van der Berg, Servaas. 2001a. Social policy to address poverty. Chapter 7 in: Bhorat, Haroon;
Leibbrandt, Murray; Maziya, Muzi; Van der Berg, Servaas; & Woolard, Ingrid (eds.). 2001. Fighting poverty: Labour markets and inequality in South Africa. UCT Press: Lansdowne: 171-204.
Van der Berg, Servaas. 2001b. Resource shifts in South African schools after the political transition, Development Southern Africa, September 2001: 309-325
Van der Berg, Servaas. 2001c. An analysis of the impact of resource inputs and socio-economic status in South African education using multiple data sets. Paper to the Bi-annual Conference of Economic Society of South Africa. Johannesburg, September
Van der Berg, Servaas; Wood, Louise; & Le Roux, Neil. 2002. Differentiation in black education. To appear in Development Southern Africa
26
Appendix: Regressions of test scores of South African teenagers 13-18 on literacy and
numeracy tests, 1993 Regression 1: Test score of teenagers (out of 14) regressed on years of education completed, a
metropolitan dummy, two race dummies (one for black, one for Indian), parent education, and the natural log of per capita household expenditure
Regression 2: Test score of teenagers (out of 14) regressed on years of education completed, a metropolitan dummy, two race dummies (one for black, one for Indian), parent education, the natural log of per capita household expenditure, and whether the home language was English (some test questions were in English)
(t-values shown below the coefficients in small font) Dependent variable: Test score out of 14 Regression 1: Regression 2 Years of education completed .3957*** .3968*** 9.52 9.65
Metro .54677** .5502** 2.32 2.36
Race dummy: Black -1.5015*** -1.1641*** -5.28 -3.99
Race dummy: Indian 2.5689*** 1.4319** 3.90 2.04
Parent Education .0994*** .09769*** 3.35 3.33
Natural logarithm of household expenditure per capita .5600*** .4834*** 3.61 3.13
English home language - 1.6814*** - 4.29
Constant .0311 .0565 0.04 0.06
n 819 819 R2 .382 .396 R2-adjusted .377 .390 Standard error 2.69 2.67
* indicates .10 level of significance ** indicates .05 level of significance
*** indicates .01 level of significance