inheriting the future: intergenerational persistence of
TRANSCRIPT
Working Paper SeriesNumber 71
Southern Africa Labour and Development Research Unit
byJustine Burns and Malcolm Keswell
Inheriting the Future: Intergenerational Persistence of Educational status in KwaZulu-Natal, South Africa
About the Author(s) and Acknowledgments
Justine Burns: School of Economics, University of Cape Town, Rondebosch 7700, Cape Town (email:[email protected])Malcolm Keswell: Southern Africa Labour and Development Research Unit &School of Economics, University of Cape Town, Rondebosch 7700, Cape Town (email: [email protected]).
We thank two anonymous referees for very helpful comments.
Recommended citation
Burns, B. and Keswell, M. (2011). Inheriting the Future: Intergenerational Persistence of Educational status in KwaZulu-Natal, South Africa. A Southern Africa Labour and Development Research Unit Working Paper Number 71. Cape Town: SALDRU, University of Cape Town
ISBN: 978-1-920517-12-0
© Southern Africa Labour and Development Research Unit, UCT, 2011
Working Papers can be downloaded in Adobe Acrobat format from www.saldru.uct.ac.za.Printed copies of Working Papers are available for R15.00 each plus vat and postage charges.
Orders may be directed to:The Administrative Officer, SALDRU, University of Cape Town, Private Bag, Rondebosch, 7701,Tel: (021) 650 5696, Fax: (021) 650 5697, Email: [email protected]
Inheriting the Future:
Intergenerational Persistence of Educational status
in KwaZulu-Natal, South Africa
Justine Burns and Malcolm Keswell∗
November 14, 2011
Abstract
This paper examines the changes in the educational attainment of threesuccessive generations in South Africa: grandparents, parents and children.Many of the results accord with widely known facts, such as the educationalpenalty faced by individuals who are African or who live in rural areas or infemale-headed households. Similarly, the larger impact of mothers educationon child outcomes relative to fathers education accords with previous work,although it is interesting that this gender difference is only sizeable and signif-icant for relationships between the second and third generation. Key findingsin this paper include the fact that persistence has increased with subsequentgenerations.
JEL Keywords: Education and Inequality; Intergenerational Mobility
JEL Codes: I24, J62
∗Burns: School of Economics, University of Cape Town, Rondebosch 7700, Cape Town (email :[email protected]); Keswell: Southern Africa Labour and Development Research Unit &School of Economics, University of Cape Town, Rondebosch 7700, Cape Town (email : [email protected]). We thank two anonymous referees for very helpful comments.
1 Introduction
Economists increasingly acknowledge that history matters. This idea is prevalent in
the convergence literature, where initial conditions affect a country’s long run growth
rate and convergence to the steady state (Barro and Martin, 1992), as well as in
the literature concerning lock-in and path dependence in technological development
(Arthur, 1989; David, 1985). But history in the form of inherited status also matters
for individual outcomes. For example, genetic endowments are transmitted across
generations, and affect health and cognitive outcomes for subsequent generations.
Similarly, a large body of international literature documents the intergenerational
correlation between parents and their offspring in the domain of earnings, education
and occupation (Solon, 1992; Featherman and Hauser, 1978; Mulligan, 1999; Corak
and Heisz, 1998).
Very few South African studies have examined the intergenerational transmission
of status, be it income status, educational attainment or even occupational choice
between parents and their children. In part, this is owing to data limitations as
few South African data sets exist that allow one to make meaningful comparisons
of parental and child attributes.1 Arguably, understanding the nature and extent of
inherited status in outcomes is vital in devising programmes to redress previously
legislated inequalities. Indeed, analysing access to opportunities by measuring the
nature and extent of social and economic mobility may provide a better measure
of changing opportunities than the more traditional measures of inequality. (Bird-
sall and Graham, 2000). If inequality reflects discrimination against certain groups
or results from linguistic, cultural or historical handicaps that ensure the intergen-
erational transmission of poverty, then mobility, measured over lifetimes and even
generations, will be constrained. Inequality may also reflect persistent differences
in the capacity of individuals and households to exploit markets or to achieve equal
access to education, employment or property rights.
From a policy perspective, inheritance of status is an important dimension. Poli-
cies that are to successfully deal with adverse outcomes need an understanding of
the underlying processes by which they are generated. Recent policy shifts in South
Africa reflect a growing belief in the supremacy of the market to achieve optimal out-
comes, and to redress past inequalities. Equal access to employment and education
1A recent exception is the National Income Dynamics Study data. As additional waves areadded to this panel data set, analysts will have a better basis of exploring these intergenerationalcorrelations.
2
are deemed sufficient to erase the painful past. And yet, there is growing discontent
with the slow pace of change, despite the dramatic shifts to equalise spending on
education across race groups. In part, this is because policy makers have failed to
realise that market-oriented policies on their own are insufficient to overcome the
handicaps imposed (or benefits afforded) by inherited status. To the extent that in-
herited status is important in explaining outcomes in successive generations, policies
to account for inherited inequalities are vital.
Using data from the KwaZulu Income Dynamics Study (KIDS), we proceed by
examining changes in the educational attainment of three successive generations
in South Africa: grandparents, parents and children. The sample is limited to
Indians and Africans, and it is apparent that successive generations have made
significant advances in the number of years of schooling attained, although these
gains are more significant for the second generation than the third. We find that
the intergenerational correlation in education status is higher between the second
and third generation than between the first and second generation. Furthermore,
when the education of both parents is included in a regression, there are differences
in the magnitude of the estimated coefficient by the gender of the parent. However,
this gender difference is only sizable and significant for relationships between the
second and third generation (mothers and fathers to daughters and sons) and not
for relationships between the first and second generation.
2 Measuring Educational Mobility
Much of the literature on intergenerational mobility has focused on the transmission
of economic status, as measured by income, between parents and their children. A
standard Galton regression is estimated as follows:
lnwc = β1 + β2 lnwp + β3agec + β4age2c + β5agep + β6age
2p + ε
where w denotes the logged welfare indicator of interest, and the c and p sub-
scripts refer to child and parent respectively. In this specification, the standardised
coefficient provides an estimate of the intergenerational correlation coefficient for
the welfare indicator. A larger value of β2 indicates a stronger correlation between
the outcomes of parent and child, suggesting high persistence in status across the
two generations. Under the (implausible) assumption that parental education is
exogenous in this regression, a policy interpretation would be that giving parent’s
3
more education leads to higher educational outcomes for their children. Both in-
terpretations would imply that child outcomes are strongly conditioned on parental
attributes. For present purposes, we remain agnostic about the precise interpreta-
tion given to β2 and wish only to document the extent of the correlation between
parent’s educational status and that of their children, leaving aside the question of
the various channels through which transmission might take place.2
While most economic mobility studies have focussed on intergenerational in-
come coefficients, and the role played by education in income mobility, there are
good reasons to focus on education as an outcome in its own right. First, educa-
tional opportunity is likely to be major mechanism through which intergenerational
social mobility is affected. Not only is education likely to have a positive impact on
the chances of upward occupational mobility, but it also raises the opportunity for
upward income mobility. Key studies by Blau and Duncan (1967) and Featherman
and Hauser (1978) argue that the educational attainment of American men is the
main known determinant of their occupational status, and that the educational and
occupational status of fathers affects their sons occupational attainment primarily
via the sons education. Even in studies where intellectual ability has been included
to control for unmeasured heritable traits (ability bias ) and the influence of assorta-
tive mating, the influence of fathers education on sons education has maintained its
relative position as the most important of parental-background influences (Sewell,
1980).3
Secondly, to the extent that one is concerned with income mobility, intergen-
erational schooling correlations may provide an upper bound to the true earnings
correlation since parental tastes and wealth influence childs schooling much more
than their adult earning capacities (Behrman et al, 1980, Haveman, 1995).
Thirdly, a focus on educational mobility as an outcome in its own right may be
especially important in South Africa, given the evidence of a strong negative corre-
lation between the level of education and poverty status. Moreover, since current
2A key limitation of the KIDS data is that it lacks credible options for controlling for the endo-geneity of parental schooling. Current work coming out of the NIDS project (not yet published)is beginning to addresses the causal question. Our concern here is to learn something about theconditioning role of parental education at a key moment in the history of South Africa: the pe-riod immediately following the post Apartheid period, using a sample of individuals that arguablywould not have been exposed to many of the innovations in school reform that were beginning tobe introduced in the late 1990’s.
3While concerns over biased estimates arising from unobserved ability bias are important,Bowles and Gintis (Bowles, 2001) argue that the genetic inheritance of traits contributing tothe cognitive skills measured on IQ and related tests explain less than one twentieth of the inter-generational transmission of economic status.
4
data limitations rule out the possibility of examining the intergenerational persis-
tence of income status, we have to look elsewhere in making inferences about the
importance of income mobility. 4 It is standard in the literature to focus on inter-
mediate outcomes (like schooling) in the absence of better information on income.
Since educational attainment tends to have a finite time horizon, with most indi-
viduals completing their education by their early twenties, it is substantially easier
to estimate intergenerational correlations in education status for parents and their
children, since information on educational attainment is routinely collected in cross-
sectional household surveys.5
Numerous international studies document the association between family back-
ground, parental schooling and the schooling of children. (Behrman (1997) and
Behrman and Knowles (1999) are good survey articles). Mulligan (1999), for ex-
ample, finds that across eight sets of estimates , the intergenerational correlation
coefficient on education attainment ranges from 0.14-0.45, averaging at 0.29.6 This
suggests fairly high mobility (or low persistence) in educational status across gen-
erations, with subsequent generations making large gains in educational attainment
relative to their parents. In addition, these studies almost always find a significant
positive association between child’s schooling and parental education, with mother’s
education being about 10% more important than father’s education at the median
of estimates that include both (Haveman, 1995; Hill, 1987; Schultz, 1993; Case,
1991).7
3 Data
The data used in this analysis comes from the first two waves of the KwaZulu Income
Dynamics Study (KIDS), a panel data set covering approximately 1100 households,
and 11400 individuals. The first round of data was collected in 1993 under the aus-
pices of the Project for Statistics on Living Standards and Development (PSLSD),
which was the first ever nationally representative demographic and socio-economic
survey to be conducted in South Africa. In 1998, a resurvey (which excluded white
4To document intergenerational income correlations between parents and their children requiresincome data for parents and children at the same stage in their respective earnings lifecycles, andto date, this type of data doesnt exist in South Africa.
5Of course, this argument only applies to co-resident household members and this is an impor-tant caveat that must be borne in mind.
6These estimates are based on data for the US, Germany, Malaysia, and Kalamazoo, Michigan.7A notable exception is the study by Behrman and Taubman (1985) that estimated that in the
USA, the impact of fathers education on offspring education is larger than mothers education.
5
and coloured households) was conducted in the province of KwaZulu-Natal only,
providing panel data for Africans and Indians in this province.
There are a number of possible data problems that arise when using cross-
sectional data, most notably measurement error. The effect of measurement error
is to exaggerate the dynamics, since not all of the observed intertemporal variation
in the welfare indicator is due to mobility. In estimation, this is the problem of
errors in variables (Greene, 1997; Solon, 1992.) One way to minimise this error is
to exploit the presence of panel data and average the welfare indicator (in this case,
educational attainment and age) across time periods, which is the approach adopted
here.
The sample used in this analysis is limited to Africans and Indians, aged 21
years and above, who had completed their schooling. We examine the educational
attainment of three successive generations in South Africa: grandparents, parents
and children. We denote grandparents as the first generation, parents (mother or
father) as the second generation and children (daughter or son) as the third genera-
tion. Grandparents, on average, are between the age of 64 and 66, putting them at
school in the 1940s, while second generation adults are, on average, between the age
of 41 and 45, putting them at school in the 1960s, a time period where educational
access for non-whites was rapidly expanded. Third generation individuals, who are
30 years old on average, would have attended school in the late 1970s and early
1980s (Table 1).
4 Empirical Results
4.1 Descriptive Statistics
Successive generations in South Africa have obtained, on average, more years of
education than previous generations. However, the gain in the number of years
of education made by parents relative to grandparents is far larger than the gains
made by children relative to their parents (Table 1 and Figure 1). This difference
in educational attainment may be at least partly attributable to the time period in
which the individual attended school, with second generation individuals attending
school during the time of rapid expansion of educational access to non-whites in
the 1960s. There do not appear to be significant gender differences in terms of
the average years of education obtained, and in this regard, South Africa stands in
contrast to many other countries in Africa (Thomas, 1996).
6
These features confirm previous work by Thomas (1996) on South Africa, and
Peil (1990) on trends in Africa more broadly. Thomas (1996) argues that among
black South Africans, educational attainment of those born in the 1950s and 1960s
increased significantly more rapidly than those born earlier. However, the distri-
bution of this growth has not been uniform, with the least educated having been
excluded from the rise in schooling over the last five decades. Thomas (1996) argues
that black South Africans at the top of the education distribution have benefited
most. Black men in the top quartile of the education distribution completed seven
more years of education over these five decades, while education of those in the
bottom quartile rose by only 1.5 years. Peil (1990), in her analysis of educational
systems in Africa, argues that expanding educational systems in Africa have in-
creasingly allowed offspring to attain higher education than their fathers, with this
tendency being higher amongst older offspring. Many younger offspring are found
to have no more education than their fathers, and in part, this is attributed to
the selective targeting of household resources for education towards some offspring
rather than others.
4.2 Intergenerational persistence in education status
Table 2 presents simple intergenerational correlation coefficients in educational at-
tainment between individual parents and their children, providing a measure of the
extent of educational persistence across generations. Following the standard Galton
approach, a series of regressions were run for each parent-child pair, each taking the
form:
lnEc = β1 + β2 lnEp + β3agec + β4age2c + β5agep + β6age
2p + ε
where E denotes logged education and c and p denote child and parent respec-
tively.8 To begin, only one parent is included in any one regression.
In each regression, the coefficient on parents education was highly significant.
8The actual regression tables are provided in the Appendix for reference purposes. As is stan-dard in the literature, theses types of models are often estimated in log-linear form so that β2
can be interpreted as an elasticity. The log-transformation of course, is only defined for positivevalues. As a large number of individuals reported zero years of education, the education variablewas transformed by adding one to the years of education reported for each individual before logtransforming the variable. In principle, several other transformations are possible that would ap-proximate the log-transformation without recording the zero values to ones, such as the inversehyperbolic sine transformation. But these alternatives often do not amount to major qualitativedifferences in the empirical estimates, so we limit our approach to the log-transformation we whichconsider to be the more standard approach.
7
From the standardised coefficients in these regressions, it is evident that educational
persistence is higher between second generation parents (mothers and fathers) and
their children (sons and daughters) than between first generation parents (grand-
parents) and second generation parents.9 This accords with the evidence presented
earlier that the gain in the number of years of education made by parents relative to
grandparents is far larger than the gains made by third generation children relative
to their parents.10
Furthermore, there appears to be very little difference in the magnitude of the
correlation coefficient by gender of the parent, that is, the correlation between a
mother or fathers education status and their sons (or daughters) are similar. How-
ever, the degree of association between parents and their daughters is higher than the
degree of association between parents and their sons, suggesting lower educational
mobility for third generation girls than boys.
In short, these results suggest that inherited status in education has become
increasingly important over time; that is, the extent of educational mobility is lower
for third generation children. In part, this has to do with the opening up of the
education system to Africans, albeit to poor quality education, in the early 1950s
with the advent of Bantu education, which allowed second generation individuals
far greater educational opportunity than grandparents. Moreover, the higher persis-
tence in educational status between second and third generation individuals suggests
that the reforms of the late 1970s were largely inadequate at substantially increas-
ing the attainment levels of subsequent generations of Indians and Africans. These
results also suggest that inherited status is more important for daughters than sons,
given that third generation girls experiencing lower educational mobility than boys.
The simple correlation coefficients of Table 2 do not control for possible interac-
tions between paternal and maternal education, nor do they control for the influence
of factors such as household income or location. Table 3 summarises the intergener-
ational correlation coefficient on educational status from a series of regressions that
included the education status of both parents , along with additional controls for
race and location.11
9We base our comparisons on the standardised coefficients in order to take account of the dif-ferences in inequality (as proxied by differences in the standard deviation) in educational outcomesacross the three generations.
10This would imply lower persistence i.e. because second generation individuals made relativelylarge gains in educational attainment relative to their first generation parents, one would expect alower correlation between their final educational outcomes and their parents. This translates intohigher educational mobility for second generation individuals.
11The reason for including the education of both parents in the regression is to examine whetherthe education status of one parent has a differential effect on the status of their child, controlling for
8
The magnitude of the correlation coefficients between first and second genera-
tion parents and children is smaller than those between second and third generation
parent-child pairs, again confirming the relatively higher educational mobility for
second generation individuals during the 1960s. Moreover, when the education of
both parents is included in the regression, there are differences in the magnitude of
the correlation coefficient by the gender of the parent, but these are only significant
for the relationships between second and third generation individuals. Gender dif-
ferences in the impact of parental education on child educational outcomes are small
and insignificant between the first and second generation. In stark contrast, gender
differences in the impact of parental education on child outcomes between the sec-
ond and third generation are large and significant at the 10% level.12 Controlling for
parental and offspring age but no other variables, the correlation coefficient between
mothers and their sons is 10 points higher than between fathers and sons, while the
correlation coefficient between mothers and their daughters is almost double that of
fathers and their daughters. These differences remain (although the size of the cor-
relation coefficients decreases marginally) once additional controls for race, location
and gender of the child are added to the regression. This suggests the differential
importance of mother’s education status for that of her children between the sec-
ond and third generation.13 This might arguably be attributable to the gradual
relaxation of influx control laws and the intensification of the migrant labour sys-
tem, which facilitated the employment of African men on the mines during the time
period when third generation children would have been attending school. These
external changes left women to see to the education and nurturing of children in
the household, and thus, it is perhaps not surprising that their educational status
becomes increasingly important in influencing child attainment levels.
their spouses educational attainment. Moreover, this strategy has the added benefit of conditioningout the role played by assortative mating. See section 5 for more on this point.
12We conduct indirect t-tests of the hypothesis that the coefficient on paternal schooling isthe same as the coefficient on maternal schooling. The differences in these coefficients in theregressions between paternal grandparents and fathers, as well as the regressions between maternalgrandparents and mothers are not statistically significant. However, in the regressions betweenparents and daughters, the computed t-statistic of 1.74 exceeds its critical value of 1.65 for the10% level of significance. In the regressions between parents and children, the computed t-statisticof 2.13 exceeds its critical value of 1.96 for the 5% level of significance. The tables reported inappendix A show the corresponding restricted and unrestricted models estimated in each case.
13These results run contrary to the findings of Behrman and Taubman (1985) who estimatethat in the USA, the impact of fathers education on offspring education is larger than motherseducation. However, the results do accord with their finding that whereas the fathers effect isevenly spread between the sexes, the mothers effect is tilted in favour of daughters. Lillard andWillis (1994) report similar results.
9
It is again evident in Table 3 that educational mobility for third generation girls
is substantially lower than for boys. A necessary next step would be to interrogate
the causal implications of these findings in order to be able to determine whether
policies directed towards improving educational outcomes of women are likely to
have a greater effect on child outcomes that those targeted to men. We leave this
question to future research.
Finally, these simple OLS estimates assume a constant mobility coefficient across
the entire educational distribution. This implies that the strength of persistence in
educational status is uniform irrespective of whether ones parents have relatively
low or high educational attainments. Yet, this need not be the case, and to the
extent that persistence in educational attainment status is higher in the tails of
the distribution, this might be suggestive of an educational poverty trap. Stronger
persistence in the tails of the distribution would imply that children of parents with
low educational attainment are themselves more likely to experience low educational
attainment, while the converse holds true for children with highly educated parents.
The quantile regression results in Table 4 suggest that this is indeed the case when
examining persistence in outcomes between second and third generation parent-child
pairs.
5 Discussion and Conclusion
This paper has examined the changes in educational attainment across three gener-
ations of South Africans. Successive generations of South Africans have experienced
increases in average attainment, although the rate of increase in attainment appears
to have slowed for third generation children. Over time, the educational status of
mothers has come to play a larger role in affecting child educational outcomes rela-
tive to fathers, possibly as a result of influx control and the migrant labour system,
although this gender difference is only sizeable and significant for relationships be-
tween the second and third generation. These results also suggest that educational
mobility is lower for daughters than sons. While these results provide an historical
snapshot of the evolution of educational attainment over time in South Africa, there
are a number of caveats worth making.
10
5.1 Concerns about Sample Selection and Attrition
The analysis uses the first two waves of KIDS data. In both waves, we observe
only the educational attainment of co-resident grandparents, parents and children.
There is of course a selection issue here: non co-resident members of any three
generation dynasty are not in our sample, but this is the best that can be done with
this particular survey.
As should be clear from Table 1, we use the panel structure of the KIDS data
to correct for measurement error in reported schooling, by averaging the reported
schooling for the sub-sample of individuals with completed schooling across the
two waves. This practice is fairly standard in the literature (see for example Solon,
1992). In the event that the data were missing in either of the two waves, we use the
information that was provided in the remaining wave. Of course, for members of the
household that were not co-resident in at least one wave, no information would have
been provided. This could account for why the sample of mothers is much larger
(1886 observations) compared to the sample of fathers (1014 observations). The
fact that fathers might be absent could be due, in part, to the effects of the migrant
labour system, but it could also be because children are borne out of wedlock and
are thus raised ostensibly by the mothers. However, the potential bias is likely to
run in the same direction, as the father is absent in both instances. Ultimately, we
have no satisfactory method of dealing with this problem. The best that can be
done with the given data is to estimate the models with and without controlling for
father’s schooling. These estimates are shown in Tables 2 and 3. Since we make no
claim that our results generalizes to the larger population of fathers, this approach
is justifiable.
With that caveat in mind, our preferred estimates are those where both chan-
nels of intergenerational transmission are controlled for. This is common practice
in the literature because that part of the intergenerational correlation in schooling
attainment explained by unobserved ability that operates through assortative mat-
ing is fully accounted for with this approach. Standard practice seems to favour
controlling for this source of endogeneity rather than worrying about selection bias
in the sample (again, under the proviso that the interpretation is restricted to the
sample of co-resident parents and children). These estimates, shown in Table 3, are
consistent with our hypothesis that assortative mating causes an upward bias in the
rate of mobility, since the estimates of mobility in this instance are much larger than
those shown in Table 2.
11
In terms of attrition, as explained above, as long as we observe the educational
attainment in at least one of the two waves, attrition is of no consequence, as the
sample of households for which there is missing data on both parents in both waves
is vanishingly small. Moreover, as Maluccio (2000) reports, educational attainment
is uncorrelated with whether a household is likely to drop out of the sample in 1998.
5.2 Interpreting the effect of mother’s education
It is not straightforward to interpret the finding that the effect of mothers’ education
is stronger for girls than for boys. From the descriptive statistics, it is clear that
there are no major gender differences in educational attainment, though here we are
limited to univariate comparisons. However, in the regressions run to produce the
coefficients presented in Table 3, where additional controls for gender are included,
there is no statistical association between gender and the log of child’s education
(i.e., the coefficient on Child is female is not statistically significant). So if there
is no significant difference in the educational attainment of girls versus boys, what
accounts for the higher persistence from parents to daughters as compared with
parents to sons?
One possibility is that part of the differential is driven by omitted parental
ability operating through assortative mating. However, in the regressions where
we control for the educational attainment of both parents, the results of which are
summarised in table 3, the extent of persistence drops across the board, and the gap
in persistence between daughters and sons (the final two columns in table 3) widens.
So while assortative mating operates directly on persistence, the differential effect
by gender cannot be explained by this channel.
Another possibility is that the unobserved ability of both parents that do not
operate in concert (e.g., genes that girls inherit solely from mothers) confounds the
estimated effects of parental schooling. Unfortunately, the KIDS data does not con-
tain any plausible options for dealing with this problem. The usual approaches (dif-
ferencing the schooling attained by twin parents; restricting the sample to adopted
children; constructing IVs through schooling reforms) are simply not viable with
this data.14 Having said that, it is not all that clear that adequately controlling for
14The sample of twin parents living in the same household would be vanishingly small in a dataset like KIDS so this approach would not work for practical reasons. The idea behind only lookingat educational outcomes of adopted children would not be plausible in South Africa, where it isvery likely that the majority of “adopted” children in a given household would be under the careof a genetically related individual – most often a grandmother. Finally, while there might besome possibilities to exploit educational policy innovations under Apartheid, the largest increases
12
the endogeneity of parental schooling will eliminate the gender differentiated rates
of intergenerational persistence we observe. This certainly seems to be the case for
the small but growing literature on the USA and Europe that have attempted to
make these corrections, though in these studies, the effect of paternal education
is stronger than that of maternal education. The interpretation of these findings
is that policies that give fathers more education have a larger positive prediction
for offspring education than policies that foster improved maternal education. In
our context, notwithstanding the caveat of remaining endogeneity, we hypothesise
that policies that foster greater schooling attainment among women are likely to
have a greater impact on the educational outcomes of daughters than sons. This is
consistent with the work of Thomas (1996) and Duflo (2003). However, a proper
test of this hypothesis would require better controls for the endogeneity of parental
education, which is beyond the scope of this paper and must therefore be left for
future research.
in educational attainment happened in the 1990’s which also coincided with the most significantpolicy shifts that took place over the last generation. The cohort of school leavers in the KIDSdata would not have been exposed to these policy innovations. This would also be true for all butthe very youngest cohort of school leavers in every other household survey that could be used toinvestigate intergenerational issues.
13
Table 1: Educational attainment across generations (all individuals age 21 years orolder)
Table 1. Educational attainment across generations (all individuals age 21 years or older)
Female Years of education and age Mean N Male years of education and age Mean N Paternal grandmother education 1.1
(1.91) 776 Paternal grandfather education 1.26
(2.23) 743
Paternal grandmother age 67.94 (12.08)
789 Paternal grandfather age 65.66 (12.77)
761
Maternal grandmother education 1.25 (2.01)
1206 Maternal grandfather education 1.38 (2.36)
1170
Maternal grandmother age 66.23 (12.28)
1221 Maternal grandfather age 64.25 (12.65)
1191
Mother education 4.63 (3.53)
1886 Father education 4.83 (3.64)
1014
Mother age 41.90 (15.42)
2177 Father age 44.89 (13.56)
1124
Daughter education 6.75 (3.30)
986 Son education 6.11 (3.34)
1111
Daughter age 30.08 (7.70)
1171 Son age 31.46 (8.23)
1282
Note: Standard deviations in brackets. Figure 1: Average years of education by generation and gender
Note: Standard deviations in brackets.
14
Table 2: Persistence in education status between parents and children
Table 2: Persistence in education status between parents and children Parent child pair Standardised
Coefficient Adjusted
R-squared df
Maternal grandmother to mother 0.37a 0.35 1109
Maternal grandfather to mother 0.38b 0.36 1074
Paternal grandmother to father 0.30c 0.27 722
Paternal grandfather to father 0.33d 0.29 693
Mothers to sons 0.37e 0.24 867
Fathers to sons 0.39f 0.22 409
Mothers to daughters 0.49g 0.28 800
Fathers to daughters 0.46h 0.21 415
Mothers to child 0.43i 0.26 1673
Father's to child 0.42j 0.22 830
Notes:
Each line of the table represents a different regression. In each separate regression, the dependent variable is the ln of scaled education of the child in the parent-child pair. Each regression includes age controls for the parent and child, and education of the parent. For example, for the results for maternal grandmother to mother, the regression includes controls for ln education of maternal grandmother, the age of the mother, age squared of the mother, age of the maternal grandmother, and age squared of the maternal grandmother. a. See Appendix 2, Table B, Model 1, !4. b. See Appendix 2, Table B, Model 2, !7; c . See Appendix 2, Table A, Model 1, !4.; d. See Appendix 2, Table A, Model 2, !7.; e. See Appendix 2, Table D, Model 1, !4.; f. See Appendix 2, Table D, Model 1, !7.; g. See Appendix 2, Table C, Model 1, !4.; h. See Appendix 2, Table B, Model 2, !7.; i. See Appendix 2, Table E, Model 1, !4.; j. See Appendix 2, Table E, Model 1, !7.
Furthermore, there appears to be very little difference in the magnitude of the correlation co-efficient by gender of the parent, that is, the correlation between a mother or father’s education status and their sons (or daughters) are similar. However, the degree of association between parents and their daughters is higher than the degree of association between parents and their sons, suggesting lower educational mobility for third generation girls than boys. In short, these results suggest that inherited status in education has become increasingly important over time, that is, the extent of educational mobility is lower for third generation children. In part, this has to do with the opening up of the education system to Africans, albeit to poor quality education, in the early 1950s with the advent of Bantu education, which allowed second generation individuals far greater educational opportunity than grandparents. Moreover, the higher persistence in educational status between second and third generation individuals suggests that the reforms of the late 1970s were largely inadequate at substantially increasing the attainment levels of subsequent generations of Indians and Africans. These results also suggest that inherited status is more important for daughters than sons, given that third generation girls experiencing lower educational mobility than boys.
Notes: Each line of the table represents a different regression. In each separate regression,the dependent variable is the ln of scaled education of the child in the parent-child pair.Each regression includes age controls for the parent and child, and education of theparent. For example, for the results for maternal grandmother to mother, the regressionincludes controls for ln education of maternal grandmother, the age of the mother, agesquared of the mother, age of the maternal grandmother, and age squared of the maternalgrandmother.
a. See Appendix A, Table A.6, Model 1
b. See Appendix A, Table A.6, Model 2
c. See Appendix A, Table A.5, Model 1
d. See Appendix A, Table A.5, Model 2
e. See Appendix A, Table A.8, Model 1
f. See Appendix A, Table A.8, Model 1
g. See Appendix A, Table A.7, Model 1
h. See Appendix A, Table A.6, Model 2
i. See Appendix A, Table A.9, Model 1
j. See Appendix A, Table A.9, Model 1
15
Tab
le3:
Per
sist
ence
ined
uca
tion
stat
us
bet
wee
npar
ents
and
childre
n,
contr
olling
for
the
educa
tion
ofb
oth
par
ents
.
10
Tab
le 3
: Per
sist
ence
in e
duca
tion
stat
us b
etw
een
pare
nts
and
child
ren,
con
trol
ling
for
the
educ
atio
n of
bot
h pa
rent
s.
Firs
t gen
erat
ion
to s
econ
d ge
nera
tion
Seco
nd g
ener
atio
n to
third
gen
erat
ion
Dep
. Var
iabl
e: L
n ed
ucat
ion
Pare
nt
Mat
erna
l gr
andp
aren
ts to
m
othe
rs
Pate
rnal
gra
ndpa
rent
s to
fa
ther
s Pa
rent
s to
chi
ld
Pare
nts
to d
augh
ter
Pare
nts
to s
ons
Add
ition
al C
ontr
ols
in
regr
essi
on
B
etaa
Adj
. R-s
q (d
f)
Bet
ab A
dj. R
-sq
(df)
B
etac
Adj
. R-s
q (d
f)
Bet
ad A
dj. R
-sq
(df)
B
etae
Adj
. R-s
q (d
f)
Age
M
othe
r 0.
23
0.39
0.
13
0.31
0.
34
0.32
0.
37
0.29
0.
31
0.33
Fath
er
0.26
10
51
0.27
67
2 0.
21
697
0.21
35
3 0.
20
335
Age
and
chi
ld is
Afr
ican
M
othe
r 0.
24
0.40
0.
19
0.38
0.
34
0.32
0.
36
0.29
0.
31
0.35
Fath
er
0.21
10
50
0.12
67
1 0.
18
696
0.21
35
2 0.
14
334
Age
and
chi
ld l
ives
in
Rur
al
area
M
othe
r 0.
22
0.42
0.
14
0.37
0.
33
0.33
0.
35
0.31
0.
30
0.35
Fa
ther
0.
21
1050
0.
16
671
0.17
69
6 0.
18
352
0.14
33
4
A
ge a
nd C
hild
is fe
mal
e M
othe
r
0.
34
0.32
Fath
er
0.21
69
6
A
ge,
child
is
Afr
ican
and
in
rura
l are
a M
othe
r 0.
22
0.42
0.
17
0.39
0.
33
0.33
0.
36
0.31
0.
31
0.36
Fa
ther
0.
20
1049
0.
10
670
0.17
69
5 0.
19
351
0.13
33
3
A
ge,
child
is
fem
ale,
Afr
ican
an
d liv
es in
rura
l are
a M
othe
r
0.
33
0.33
Fa
ther
0.
17
694
Not
es:.
In e
ach
regr
essi
on, t
he d
epen
dent
var
iabl
e is
the
ln o
f sca
led
educ
atio
n of
the
child
in th
e pa
rent
-chi
ld p
airin
g.
Thes
e co
-eff
icie
nts
corr
espo
nd to
the
regr
essi
ons
in A
ppen
dix
2, T
able
B, M
odel
s 3-
6; b
. The
se c
o-ef
ficie
nts
corr
espo
nd to
the
regr
essi
ons
in A
ppen
dix
2, T
able
A
, Mod
els
3-6;
c. T
hese
co-
effic
ient
s co
rres
pond
to th
e re
gres
sion
s in
App
endi
x 2,
Tab
le E
, Mod
els
3-8;
d. T
hese
co-
effic
ient
s co
rres
pond
to th
e re
gres
sion
s in
A
ppen
dix
2, T
able
C, M
odel
s 3-
6; e
. The
se c
o-ef
ficie
nts
corr
espo
nd to
the
regr
essi
ons
in A
ppen
dix
2, T
able
D, M
odel
s 3-
6.
Not
es:.
Inea
chre
gres
sion
,th
ede
pend
ent
vari
able
isth
eln
ofsc
aled
educ
atio
nof
the
child
inth
epa
rent
-chi
ldpa
irin
g.
aT
hese
coeffi
cien
tsco
rres
pond
toth
ere
gres
sion
sin
App
endi
x2,
Tab
leB
,M
odel
s3-
6.
bT
hese
co-e
ffici
ents
corr
espo
ndto
the
regr
essi
ons
inA
ppen
dix
2,T
able
A,
Mod
els
3-6.
cT
hese
co-e
ffici
ents
corr
espo
ndto
the
regr
essi
ons
inA
ppen
dix
2,T
able
E,
Mod
els
3-8.
dT
hese
co-e
ffici
ents
corr
espo
ndto
the
regr
essi
ons
inA
ppen
dix
2,T
able
C,
Mod
els
3-6.
eT
hese
co-e
ffici
ents
corr
espo
ndto
the
regr
essi
ons
inA
ppen
dix
2,T
able
D,
Mod
els
3-6.
16
Table 4: Quantile Regressions of intergenerational transmission of education status
12
Table 4: Quantile regression analysis of intergenerational transmission of education status Dep var: Ln of child's education Full sample First
quartile Second quartile
Third quartile
Top quartile
Child is African -0.11 ** -0.06 -0.09 ** -0.08 ** -0.65 *
0.04 0.08 0.04 0.03 0.05 Child lives in urban areas 0.14 * 0.11 0.05 0.07 ** 0.13 *
0.05 0.07 0.04 0.03 0.05 Child is female 0.03 0.06 0.06 ** 0.04 ** -0.02
0.04 0.05 0.02 0.02 0.07 Average age of child 0.03 0.06 ** 0.10 * 0.04 * -0.12 *
0.03 0.03 0.02 0.01 0.03 Average age of child squared 0.00 0.00 * 0.00 * 0.00 * 0.00 *
0.00 0.00 0.00 0.00 0.00 Age of mother -0.04 -0.05 -0.03 *** -0.01 0.06 ***
0.04 0.04 0.02 0.02 0.04 Average age of mother squared 0.00 0.00 0.00 ** 0.00 0.00
0.00 0.00 0.00 0.00 0.00 Ln of mother's education 0.28 * 0.35 * 0.16 * 0.09 * 0.37 *
0.04 0.04 0.02 0.02 0.03 Father's age 0.02 0.03 0.02 0.01 -0.09 *
0.04 0.03 0.02 0.01 0.03 Average father's age squared 0.00 0.00 0.00 0.00 0.00 ***
0.00 0.00 0.00 0.00 0.00 Ln of father's education 0.13 * 0.17 * 0.08 * 0.04 ** 0.34 *
0.03 0.04 0.02 0.02 0.03 Constant 1.84 ** 0.76 1.03 *** 1.72 * 3.29 *
0.86 1.08 0.54 0.44 1.23
Adjusted/Pseudo R-squared 0.33 0.26 0.15 0.09 0.29 N 706 *=Significant at 1% level; **=Significant at 5% level; ***=Significant at 10% level 5. Conclusion This paper has examined the changes in educational attainment across three generations of South Africans. Successive generations of South Africans have experienced increases in average attainment, although the rate of increase in attainment appears to have slowed for third generation children. Over time, the educational status of mothers has come to play a larger role in affecting child educational outcomes relative to fathers, possibly as a result of influx control and migrant labour system. Educational mobility higher for boys than girls. Target resources towards education, females in particular. Data can tell us that history matters – inherited status is important, and need active interventions to minimse role of inherited inequalities. Non-linearities. .
∗ = Significant at 1% level; ∗∗ =Significant at 5% level; ∗∗∗ = Significant at10% level
17
Bibliography Arthur, W.B. (1989) “Competing Technologies, Increasing Returns and Lock-in by Historical Events”, The Economic Journal, Vol. 99(394), March. Barro, R.J. and Martin, S.i (1992) “Convergence”, Journal of Political Economy, Vol . 100(20), April. Behrman, J (1999) “Labour markets in developing countries”, Handbook of Labour economics, Vol 3, eds. Ashenfelter, O and Card, D; Elsevier Science. Behrman, J and Taubman, P (1985) “Intergenerational earnings mobility in the United States: Some estimates and a test of Becker’s Intergenerational Endowments Model” , The Review of Economics and Statistics, Vol 67, 1, Feb, pg 144-151 Behrman, J, Zdneck, H, Taubman, P and Wales, T (1980) Socioeconomic Success: A study of the effects of genetic endowments, family environment and schooling; Amsterdam: North-Holland Publishing Co) Behrman, J (1997) “Mother’s schooling and child education: A survey”, Mimeo. UPenn. Birdsall, N and Graham, C. (2000) New Markets, New Opportunities: Economic and social mobility in a changing world, The Brookings Institute, Washington DC. Blau, P and Duncan, D.D. (1967) The American Occupational Structure. New York: Wiley. Bowles, S. and Gintis, H. (2001). "The Inheritance of Economic Status: Education, Class and Genetics," M. Feldman, Genetics, Behaviour and Society. Oxford: Elsevier. Case, A.; and Katz, L. (1991). "The Company You Keep: The Effects of Family and Neighbourhood on Disadvantaged Youths," National Bureau of Economic Research. Corak, M and Heisz, A (1998) “The Intergenerational Earnings and Income Mobility of Canadian Men: Evidence from Longitudinal Income Tax data”, Journal of Human Resources, XXXIV, No. 3. David, P.A. (1985) “Clio and the Economics of QWERTY”, American Economic Review Papers and Proceedings, Vol. 75(2), May. Featherman, D and Hauser, R. (1978). Opportunity and Change. New York: Academic Press. Greene, W.H. (1997) Econometric Analysis. New Jersey: Prentice Hall.
18
Haveman, R.; and Wolfe, B. (1995) "The Determinants of Children's Attainments: A Review of Methods and Findings." Journal of Economic Literature, XXXIII(December), pp. 1829-78. Hill, M and Duncan, G (1987) “Parental Family Income and the Socio-economic Attainment of children”, Social Science Res., 16(1), pp 39-73 Maluccio, J. (2000) “Attrition in the KwaZulu Natal Income Dynamics Study, 1993-1998”, FCND Discussion Paper No. 95, IFPRI, Washington D.C. Mulligan, C.B. (1999). "Galton Versus the Human Capital Approach to Inheritance." Journal of Political Economy, 107(6), pp. S184-S223. Peil, M (1990) “Intergenerational mobility through education: Nigeria, Sierra Leone and Zimbabwe”, International Journal of Educational Development, 10, no.4, 311-25 Sewell, W.H., Hauser, R.M. and Wolf, W.C. (1980). "Sex, Schooling and Occupational Status." American Journal of Sociology, 86, pp. 551-83. Schultz, P.T. (1993). "Returns to Women's Education," E. M. a. H. King, M.A., Women's Education in Developing Countries: Barriers, Benefits, and Policies. Baltimore and London: John Hopkins University Press, 51-99. Solon, G (1992) “Intergenerational Income Mobility in the United States”, The American Economic Review, Vol 82, Issue 3 (June 1992, pg 393-408). Thomas, D (1996) “Education across generations in South Africa”, American Economic Review Papers and Proceedings, May.
19
Appendices
A Complete Regression Tables
20
Tab
leA
.5:
Pat
ernal
Gra
ndpar
ent
toF
ather
(Dep
enden
tV
aria
ble
:L
nof
fath
er’s
educa
tion
)
13
App
endi
x 2
Tabl
e A
: Pat
erna
l Gra
ndpa
rent
to F
athe
r (D
epen
dent
Var
iabl
e: L
n of
fath
er's
educ
atio
n)
Var
iabl
e M
odel
1
Mod
el 2
M
odel
3
Mod
el 4
M
odel
5
Mod
el 6
! "
!
"
! "
!
"
! "
!
"
C
onst
ant
1.75
6
* 1.
329
**
1.
674
**
2.
906
*
1.83
2
* 2.
655
*
(0
.624
)
(0
.597
)
(0
.759
)
(0
.730
)
(0
.724
)
(0
.723
)
Fa
ther
’s A
ge
0.02
0 0.
292
0.
025
0.36
5
0.02
7 0.
398
***
0.00
9 0.
133
0.
017
0.25
7
0.00
8 0.
124
(0.0
16)
(0.0
16)
(0.0
17)
(0.0
16)
(0.0
16)
(0.0
16)
Squa
re o
f Fat
her’
s A
ge
0.00
0 -0
.646
*
0.00
0 -0
.711
*
0.00
0 -0
.730
*
0.00
0 -0
.467
**
0.
000
-0.5
82
* 0.
000
-0.4
53
**
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
Ln
of P
ater
nal G
rand
mot
her's
Edu
catio
n
0.36
1 0.
298
*
0.16
6 0.
137
* 0.
225
0.18
5 *
0.17
3 0.
142
* 0.
212
0.17
4 *
(0
.041
)
(0.0
52)
(0.0
49)
(0.0
49)
(0.0
49)
Pate
rnal
Gra
ndm
othe
r's A
ge
-0.0
12
-0.1
77
-0.0
19
-0.2
66
-0
.025
-0
.348
-0.0
07
-0.0
98
-0
.016
-0
.225
(0
.015
)
(0.0
16)
(0.0
15)
(0.0
16)
(0.0
15)
Squa
re o
f Pat
erna
l Gra
ndm
othe
r's A
ge
0.00
0 0.
184
0.00
0 0.
303
0.
000
0.40
0 **
* 0.
000
0.14
7
0.00
0 0.
281
(0.0
00)
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
Ln
of P
ater
nal G
rand
fath
er's
Educ
atio
n
0.
374
0.32
6 *
0.29
4 0.
258
* 0.
135
0.11
8 *
0.18
0 0.
158
* 0.
112
0.09
8 **
(0
.039
)
(0
.048
)
(0
.049
)
(0
.048
)
(0
.048
)
Pa
tern
al G
rand
fath
er's
Age
-0.0
05
-0.0
70
-0
.002
-0
.024
-0.0
02
-0.0
37
-0
.002
-0
.024
-0.0
02
-0.0
33
(0
.013
)
(0
.014
)
(0
.013
)
(0
.013
)
(0
.013
)
Sq
uare
of P
ater
nal G
rand
fath
er's
Age
0.00
0 0.
076
0.
000
0.02
7
0.00
0 0.
041
0.
000
0.03
1
0.00
0 0.
040
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
A
fric
an
-0
.679
-0
.301
*
-0
.488
-0
.217
*
(0.0
75)
(0
.086
)
R
ural
-0
.480
-0
.273
*
-0.2
85
-0.1
62
*
(0.0
58)
(0.0
67)
R 2
0.27
0.
29
0.30
0.
38
0.36
0.
39
Df
722
693
672
671
671
670
Not
es:
Age
of
pate
rnal
gra
ndpa
rent
s in
clud
es d
ecea
sed
indi
vidu
als.
! ‘
s ar
e re
gres
sion
coe
ffic
ient
s w
hen
all
varia
bles
are
nor
mal
ised
. A
ll va
riabl
es a
re a
vera
ged
to
acco
unt
for
mea
sure
men
t er
ror
(Sol
on,
1992
). S
tand
ard
erro
rs a
re i
n pa
rent
hese
s. *
Rep
rese
nts
stat
istic
al s
igni
fican
ce a
t 1%
lev
el,
** r
epre
sent
s st
atis
tical
si
gnifi
canc
e at
5%
, and
***
repr
esen
ts s
tatis
tical
sig
nific
ance
at 1
0% le
vel.
“Afr
ican
” is
a ra
ce d
umm
y, a
nd “
Rur
al”
is a
loca
tion
dum
my.
Ta
ble
B. M
ater
nal G
rand
pare
nt to
Mot
her (
Dep
ende
nt V
aria
ble:
Ln
of m
othe
r's e
duca
tion)
Age
ofpa
tern
algr
andp
aren
tsin
clud
esde
ceas
edin
divi
dual
s.γ
’sar
ere
gres
sion
coeffi
cien
tsw
hen
all
vari
able
sar
eno
r-m
alis
ed.
All
vari
able
sar
eav
erag
edto
acco
unt
for
mea
sure
men
ter
ror
(Sol
on,1
992)
.St
anda
rder
rors
are
inpa
rent
hese
s.“A
fric
an”
isa
race
dum
my,
and
Rur
alis
alo
cati
ondu
mm
y.∗
=Si
gnifi
cant
at1%
leve
l;∗∗
=Si
gnifi
cant
at5%
leve
l;∗∗∗
=Si
gnifi
cant
at10
%le
vel.
21
Tab
leA
.6:
Mat
ernal
Gra
ndpar
ent
toM
other
(Dep
enden
tV
aria
ble
:L
nof
mot
her
’sed
uca
tion
)
14
M
odel
1
Mod
el 2
M
odel
3
Mod
el 4
M
odel
5
Mod
el 6
! "
!
"
! "
!
"
! "
!
"
C
onst
ant
1.74
5
* 2.
056
*
1.74
0
* 1.
978
*
1.89
7
* 1.
951
*
(0
.399
)
(0
.410
)
(0
.479
)
(0
.480
)
(0
.470
)
(0
.474
)
M
othe
r’s
Age
0.
000
0.00
2
0.00
0 0.
007
0.
006
0.09
6
0.00
4 0.
064
0.
003
0.04
4
0.00
2 0.
039
(0.0
10)
(0.0
10)
(0.0
10)
(0.0
10)
(0.0
10)
(0.0
10)
Squa
re o
f Mot
her’
s A
ge
0.00
0 -0
.362
**
0.
000
-0.3
81
**
0.00
0 -0
.426
*
0.00
0 -0
.393
**
0.
000
-0.3
74
**
0.00
0 -0
.369
**
(0.0
00)
(0.0
00)
(0.0
00)
(0.0
00)
(0.0
00)
(0.0
00)
Ln o
f Mat
erna
l Gra
ndm
othe
r's E
duca
tion
0.
423
0.37
4 *
0.
258
0.22
7 *
0.27
1 0.
238
* 0.
250
0.22
0 *
0.25
4 0.
224
*
(0.0
29)
(0
.034
)
(0
.034
)
(0
.034
)
(0
.034
)
M
ater
nal G
rand
mot
her's
Age
-0
.003
-0
.043
0.
003
0.04
2
0.00
4 0.
058
0.
006
0.09
3
0.00
6 0.
093
(0.0
10)
(0
.010
)
(0
.010
)
(0
.010
)
(0
.010
)
Sq
uare
of M
ater
nal G
rand
mot
her’
s A
ge
0.00
0 0.
033
0.00
0 -0
.046
0.00
0 -0
.055
0.00
0 -0
.091
0.00
0 -0
.090
(0
.000
)
(0.0
00)
(0.0
00)
(0.0
00)
(0.0
00)
Ln o
f Mat
erna
l Gra
ndfa
ther
's Ed
ucat
ion
0.40
3 0.
383
* 0.
278
0.26
3 *
0.22
6 0.
214
* 0.
219
0.20
7 *
0.20
9 0.
198
*
(0
.027
)
(0
.031
)
(0
.034
)
(0
.032
)
(0
.034
)
M
ater
nal G
rand
fath
er's
Age
-0.0
12
-0.1
97
-0
.014
-0
.214
-0.0
14
-0.2
15
-0
.012
-0
.183
-0.0
12
-0.1
86
(0
.010
)
(0
.010
)
(0
.010
)
(0
.010
)
(0
.010
)
Sq
uare
of M
ater
nal G
rand
fath
er’s
Age
0.00
0 0.
185
0.
000
0.20
8
0.00
0 0.
209
0.
000
0.16
9
0.00
0 0.
173
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
A
fric
an
-0
.251
-0
.103
*
-0
.071
-0
.029
(0.0
64)
(0
.072
)
R
ural
-0
.296
-0
.165
*
-0.2
72
-0.1
52
*
(0.0
45)
(0.0
51)
R 2
0.35
0.
360
0.39
0.
40
0.42
0.
42
Df
1109
10
74
1051
10
50
1050
10
49
Not
es:
Age
of
pate
rnal
gra
ndpa
rent
s in
clud
es d
ecea
sed
indi
vidu
als.
! ‘
s ar
e re
gres
sion
coe
ffic
ient
s w
hen
all
varia
bles
are
nor
mal
ised
. A
ll va
riabl
es a
re a
vera
ged
to
acco
unt
for
mea
sure
men
t er
ror
(Sol
on,
1992
). S
tand
ard
erro
rs a
re i
n pa
rent
hese
s. *
Rep
rese
nts
stat
istic
al s
igni
fican
ce a
t 1%
lev
el,
** r
epre
sent
s st
atis
tical
si
gnifi
canc
e at
5%
, and
***
repr
esen
ts s
tatis
tical
sig
nific
ance
at 1
0% le
vel.
“Afr
ican
” is
a ra
ce d
umm
y, a
nd “
Rur
al”
is a
loca
tion
dum
my.
Age
ofpa
tern
algr
andp
aren
tsin
clud
esde
ceas
edin
divi
dual
s.γ
’sar
ere
gres
sion
coeffi
cien
tsw
hen
all
vari
able
sar
eno
r-m
alis
ed.
All
vari
able
sar
eav
erag
edto
acco
unt
for
mea
sure
men
ter
ror
(Sol
on,1
992)
.St
anda
rder
rors
are
inpa
rent
hese
s.“A
fric
an”
isa
race
dum
my,
and
Rur
alis
alo
cati
ondu
mm
y.∗
=Si
gnifi
cant
at1%
leve
l;∗∗
=Si
gnifi
cant
at5%
leve
l;∗∗∗
=Si
gnifi
cant
at10
%le
vel.
22
Tab
leA
.7:
Par
ent
toD
augh
ter
(Dep
enden
tV
aria
ble
:L
nof
child’s
educa
tion
)
15
Tabl
e C
. Par
ent t
o D
augh
ter (
Dep
ende
nt V
aria
ble:
Ln
of c
hild
's ed
ucat
ion)
V
aria
ble
Mod
el 1
M
odel
2
Mod
el 3
M
odel
4
Mod
el 5
M
odel
6
!
"
!
"
!
"
!
"
!
"
!
"
C
onst
ant
1.37
7
* 2.
078
**
1.
808
1.82
2
2.
142
**
* 2.
127
**
*
(0.5
13)
(1.0
40)
(1.3
12)
(1.3
15)
(1.3
02)
(1.3
00)
Dau
ghte
r’s
Age
0.
018
0.20
2
-0.0
08
-0.0
72
-0
.039
-0
.314
-0.0
38
-0.3
05
-0
.035
-0
.278
-0.0
42
-0.3
40
(0.0
19)
(0.0
45)
(0.0
50)
(0.0
50)
(0.0
49)
(0.0
49)
Squa
re o
f Dau
ghte
r’s
Age
-0
.001
-0
.460
**
0.
000
-0.0
17
0.
000
0.21
1
0.00
0 0.
203
0.
000
0.17
5
0.00
0 0.
224
(0.0
00)
(0.0
01)
(0.0
01)
(0.0
01)
(0.0
01)
(0.0
01)
Ln o
f Mot
her's
Edu
catio
n
0.42
8 0.
489
*
0.30
7 0.
366
* 0.
306
0.36
5 *
0.29
5 0.
352
* 0.
299
0.35
6 *
(0
.028
)
(0.0
52)
(0.0
52)
(0.0
51)
(0.0
51)
Mot
her’
s A
ge
-0.0
06
-0.0
93
0.00
7 0.
095
0.
007
0.08
6
-0.0
01
-0.0
08
0.
002
0.02
9
(0
.018
)
(0.0
44)
(0.0
44)
(0.0
43)
(0.0
43)
Squa
re o
f Mot
her’
s A
ge
0.00
0 0.
218
0.00
0 0.
129
0.
000
0.13
7
0.00
0 0.
227
0.
000
0.19
6
(0
.000
)
(0.0
00)
(0.0
00)
(0.0
00)
(0.0
00)
Ln o
f Fat
her's
Edu
catio
n
0.
332
0.45
8 *
0.15
2 0.
209
* 0.
149
0.20
6 *
0.12
8 0.
176
* 0.
139
0.19
1 *
(0.0
34)
(0.0
46)
(0.0
48)
(0.0
47)
(0.0
47)
Fath
er’s
Age
-0.0
08
-0.1
28
0.
001
0.01
3
0.00
1 0.
017
0.
002
0.03
0
0.00
1 0.
009
(0
.033
)
(0
.041
)
(0
.041
)
(0
.040
)
(0
.040
)
Sq
uare
of F
athe
r’s
Age
0.00
0 0.
157
0.
000
-0.1
68
0.
000
-0.1
70
0.
000
-0.1
87
0.
000
-0.1
72
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
A
fric
an
-0
.021
-0
.011
0.
152
0.08
0
(0.0
96)
(0
.108
)
R
ural
-0
.202
-0
.139
*
-0.2
55
-0.1
75
*
(0.0
67)
(0.0
76)
R 2
0.28
0.
21
0.29
0.
29
0.31
0.
31
Df
800
415
353
352
352
351
Not
es:
Age
of
pate
rnal
gra
ndpa
rent
s in
clud
es d
ecea
sed
indi
vidu
als.
! ‘
s ar
e re
gres
sion
coe
ffic
ient
s w
hen
all
varia
bles
are
nor
mal
ised
. A
ll va
riabl
es a
re a
vera
ged
to
acco
unt
for
mea
sure
men
t er
ror
(Sol
on,
1992
). S
tand
ard
erro
rs a
re i
n pa
rent
hese
s. *
Rep
rese
nts
stat
istic
al s
igni
fican
ce a
t 1%
lev
el,
** r
epre
sent
s st
atis
tical
si
gnifi
canc
e at
5%
, and
***
repr
esen
ts s
tatis
tical
sig
nific
ance
at 1
0% le
vel.
“Afr
ican
” is
a ra
ce d
umm
y, a
nd “
Rur
al”
is a
loca
tion
dum
my.
Age
ofpa
tern
algr
andp
aren
tsin
clud
esde
ceas
edin
divi
dual
s.γ
’sar
ere
gres
sion
coeffi
cien
tsw
hen
all
vari
able
sar
eno
r-m
alis
ed.
All
vari
able
sar
eav
erag
edto
acco
unt
for
mea
sure
men
ter
ror
(Sol
on,1
992)
.St
anda
rder
rors
are
inpa
rent
hese
s.“A
fric
an”
isa
race
dum
my,
and
Rur
alis
alo
cati
ondu
mm
y.∗
=Si
gnifi
cant
at1%
leve
l;∗∗
=Si
gnifi
cant
at5%
leve
l;∗∗∗
=Si
gnifi
cant
at10
%le
vel.
23
Tab
leA
.8:
Par
ent
toSon
(Dep
enden
tV
aria
ble
:L
nof
child’s
educa
tion
)
16
Tabl
e D
. Par
ent t
o So
n (D
epen
dent
Var
iabl
e: L
n of
chi
ld's
educ
atio
n)
Varia
ble
Mod
el 1
M
odel
2
Mod
el 3
M
odel
4
Mod
el 5
M
odel
6
!
"
! "
!
"
! "
!
"
! "
C
onst
ant
1.94
4
* 1.
575
2.55
5
**
3.06
6
* 2.
811
**
3.
037
*
(0
.529
)
(0
.999
)
(1
.209
)
(1
.203
)
(1
.194
)
(1
.199
)
So
n’s
Age
0.
024
0.27
4
0.00
6 0.
054
0.
031
0.28
2
0.03
6 0.
326
0.
039
0.35
1
0.03
9 0.
354
(0.0
17)
(0.0
39)
(0.0
39)
(0.0
39)
(0.0
39)
(0.0
39)
Squa
re o
f Son
’s A
ge
-0.0
01
-0.5
84
* -0
.001
-0
.302
-0.0
01
-0.5
30
-0
.001
-0
.565
-0.0
01
-0.6
00
***
-0.0
01
-0.5
98
***
(0
.000
)
(0
.001
)
(0
.001
)
(0
.001
)
(0
.001
)
(0
.001
)
Ln
of M
othe
r's E
duca
tion
0.
334
0.37
4 *
0.
255
0.31
0 *
0.25
9 0.
314
* 0.
251
0.30
5 *
0.25
5 0.
309
*
(0.0
29)
(0
.049
)
(0
.048
)
(0
.048
)
(0
.048
)
M
othe
r’s
Age
-0
.024
-0
.365
-0
.083
-1
.071
**
* -0
.099
-1
.274
**
-0
.080
-1
.028
-0.0
91
-1.1
68
***
(0
.019
)
(0.0
51)
(0.0
50)
(0.0
50)
(0.0
50)
Mot
her’
s A
ge
0.00
0 0.
481
***
0.
001
1.27
7 **
0.
001
1.46
8 **
0.
001
1.22
0 **
0.
001
1.35
7 **
(0.0
00)
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
Ln
of f
athe
r's E
duca
tion
0.
308
0.39
4 *
0.15
0 0.
200
* 0.
107
0.14
2 **
0.
106
0.14
2 **
0.
095
0.12
6 **
(0
.038
)
(0
.044
)
(0
.046
)
(0
.045
)
(0
.046
)
Fa
ther
’s A
ge
0.
004
0.05
1
0.02
8 0.
412
0.
033
0.48
2
0.02
1 0.
305
0.
026
0.38
5
(0.0
35)
(0.0
44)
(0.0
44)
(0.0
44)
(0.0
44)
Squa
re o
f Fat
her’
s A
ge
0.
000
0.00
6
0.00
0 -0
.548
0.00
0 -0
.612
0.00
0 -0
.438
0.00
0 -0
.515
(0.0
00)
(0.0
00)
(0.0
00)
(0.0
00)
(0.0
00)
Afr
ican
-0.2
98
-0.1
52
*
-0.1
83
-0.0
94
***
(0.0
92)
(0
.112
)
R
ural
-0
.225
-0
.156
*
-0.1
48
-0.1
03
***
(0
.068
)
(0
.082
)
R
2 0.
24
0.22
0.
33
0.35
0.
35
0.36
D
f 86
7
40
9
33
5
33
4
33
4
33
3
Not
es:
Age
of
pate
rnal
gra
ndpa
rent
s in
clud
es d
ecea
sed
indi
vidu
als.
! ‘
s ar
e re
gres
sion
coe
ffic
ient
s w
hen
all
varia
bles
are
nor
mal
ised
. A
ll va
riabl
es a
re a
vera
ged
to
acco
unt
for
mea
sure
men
t er
ror
(Sol
on,
1992
). S
tand
ard
erro
rs a
re i
n pa
rent
hese
s. *
Rep
rese
nts
stat
istic
al s
igni
fican
ce a
t 1%
lev
el,
** r
epre
sent
s st
atis
tical
si
gnifi
canc
e at
5%
, and
***
repr
esen
ts s
tatis
tical
sig
nific
ance
at 1
0% le
vel.
“Afr
ican
” is
a ra
ce d
umm
y, a
nd “
Rur
al”
is a
loca
tion
dum
my.
Age
ofpa
tern
algr
andp
aren
tsin
clud
esde
ceas
edin
divi
dual
s.γ
’sar
ere
gres
sion
coeffi
cien
tsw
hen
all
vari
able
sar
eno
r-m
alis
ed.
All
vari
able
sar
eav
erag
edto
acco
unt
for
mea
sure
men
ter
ror
(Sol
on,1
992)
.St
anda
rder
rors
are
inpa
rent
hese
s.“A
fric
an”
isa
race
dum
my,
and
Rur
alis
alo
cati
ondu
mm
y.∗
=Si
gnifi
cant
at1%
leve
l;∗∗
=Si
gnifi
cant
at5%
leve
l;∗∗∗
=Si
gnifi
cant
at10
%le
vel.
24
Tab
leA
.9:
Par
ent
toC
hild
(Dep
enden
tV
aria
ble
:L
nof
child’s
educa
tion
)
17
Tabl
e E.
Par
ent t
o C
hild
(Dep
ende
nt V
aria
ble:
Ln
of c
hild
's ed
ucat
ion)
V
aria
bles
M
odel
1
Mod
el 2
M
odel
3
Mod
el 4
M
odel
5
Mod
el 6
M
odel
7
Mod
el 8
! "
!
"
! "
!
"
! "
!
"
! "
!
"
C
onst
ant
1.61
2
* 1.
565
**
1.
776
**
1.
964
**
2.
081
**
1.
777
**
2.
084
**
2.
087
**
(0.3
68)
(0.7
08)
(0.8
70)
(0.8
71)
(0.8
61)
(0.8
70)
(0.8
63)
(0.8
63)
Chi
ld’s
Age
0.
022
0.24
2 **
* 0.
011
0.10
4
0.02
0 0.
173
0.
026
0.22
1
0.02
4 0.
205
0.
020
0.17
2
0.02
4 0.
206
0.
024
0.20
5
(0
.013
)
(0
.029
)
(0
.029
)
(0
.029
)
(0
.029
)
(0
.029
)
(0
.029
)
(0
.029
)
Sq
uare
of C
hild
’s A
ge
-0.0
01
-0.5
24 *
0.
000
-0.2
86
-0.0
01
-0.3
54
-0.0
01
-0.3
94
-0.0
01
-0.3
85
-0.0
01
-0.3
50
-0.0
01
-0.3
86
-0.0
01
-0.3
82
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
Ln
of M
othe
r's E
duca
tion
0.
382
0.43
2 *
0.
284
0.34
2 *
0.28
2 0.
340
* 0.
276
0.33
3 *
0.28
1 0.
339
* 0.
276
0.33
3 *
0.27
2 0.
328
*
(0.0
20)
(0
.035
)
(0
.035
)
(0
.034
)
(0
.035
)
(0
.034
)
(0
.035
)
M
othe
r’s
Age
-0
.014
-0
.208
-0.0
35
-0.4
45
-0.0
41
-0.5
31
-0.0
38
-0.4
86
-0.0
36
-0.4
55
-0.0
38
-0.4
88
-0.0
39
-0.5
01
(0
.013
)
(0.0
32)
(0.0
32)
(0.0
32)
(0.0
32)
(0.0
32)
(0.0
32)
Squa
re o
f Mot
her’
s A
ge
0.00
0 0.
327
***
0.
000
0.66
8 **
* 0.
001
0.74
7 **
* 0.
001
0.70
2 **
* 0.
000
0.67
7 **
* 0.
001
0.70
3 **
* 0.
001
0.71
4 **
*
(0.0
00)
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
(0
.000
)
Ln
of F
athe
r's E
duca
tion
0.32
1 0.
424
* 0.
157
0.21
2 *
0.13
6 0.
184
* 0.
123
0.16
6 *
0.15
8 0.
214
* 0.
123
0.16
6 *
0.12
4 0.
168
*
(0
.025
)
(0
.032
)
(0
.033
)
(0
.032
)
(0
.032
)
(0
.033
)
(0
.033
)
Fa
ther
’s A
ge
0.
000
0.00
6
0.01
4 0.
202
0.
016
0.23
3
0.01
2 0.
178
0.
014
0.20
9
0.01
2 0.
179
0.
013
0.18
6
(0.0
24)
(0.0
29)
(0.0
29)
(0.0
29)
(0.0
29)
(0.0
29)
(0.0
29)
Squa
re o
f Fat
her’
s A
ge
0.
000
0.04
1
0.00
0 -0
.348
0.
000
-0.3
74
0.00
0 -0
.326
0.
000
-0.3
56
0.00
0 -0
.326
0.
000
-0.3
36
(0.0
00)
(0.0
00)
(0.0
00)
(0.0
00)
(0.0
00)
(0.0
00)
(0.0
00)
Afr
ican
-0.1
57
-0.0
81 *
*
-0
.004
-0
.002
-0
.003
-0
.002
(0
.066
)
(0
.077
)
(0
.077
)
R
ural
-0
.214
-0
.147
*
-0
.212
-0
.147
*
-0.2
14
-0.1
48 *
(0.0
47)
(0
.056
)
(0
.056
)
C
hild
is F
emal
e
0.02
8 0.
023
0.03
4 0.
028
(0
.039
)
(0.0
39)
R 2
0.26
0
0.
220
0.32
0
0.
320
0.33
0
0.
320
0.33
0
0.
330
Df
1673
83
0
69
7
69
6
69
6
69
6
69
5
69
4
Notes
:Age
of p
ater
nal g
rand
pare
nts
incl
udes
dec
ease
d in
divi
dual
s. !
‘s a
re re
gres
sion
coe
ffic
ient
s w
hen
all v
aria
bles
are
nor
mal
ised
. All
varia
bles
are
ave
rage
d to
ac
coun
t fo
r m
easu
rem
ent
erro
r (S
olon
, 19
92).
Sta
ndar
d er
rors
are
in
pare
nthe
ses.
* R
epre
sent
s st
atis
tical
sig
nific
ance
at
1% l
evel
, **
rep
rese
nts
stat
istic
al
sign
ifica
nce
at 5
%, a
nd *
** re
pres
ents
sta
tistic
al s
igni
fican
ce a
t 10%
leve
l. “A
fric
an”
is a
race
dum
my,
and
“R
ural
” is
a lo
catio
n du
mm
y.
Age
ofpa
tern
algr
andp
aren
tsin
clud
esde
ceas
edin
divi
dual
s.γ
’sar
ere
gres
sion
coeffi
cien
tsw
hen
all
vari
able
sar
eno
r-m
alis
ed.
All
vari
able
sar
eav
erag
edto
acco
unt
for
mea
sure
men
ter
ror
(Sol
on,1
992)
.St
anda
rder
rors
are
inpa
rent
hese
s.“A
fric
an”
isa
race
dum
my,
and
Rur
alis
alo
cati
ondu
mm
y.∗
=Si
gnifi
cant
at1%
leve
l;∗∗
=Si
gnifi
cant
at5%
leve
l;∗∗∗
=Si
gnifi
cant
at10
%le
vel.
25
The Southern Africa Labour and Development Research Unit (SALDRU) conducts research directed at improving the well-being of South Africa’s poor. It was established in 1975. Over the next two decades the unit’s research played a central role in documenting the human costs of apartheid. Key projects from this period included the Farm Labour Conference (1976), the Economics of Health Care Conference (1978), and the Second Carnegie Enquiry into Poverty and Development in South Africa (1983-86). At the urging of the African National Congress, from 1992-1994 SALDRU and the World Bank coordinated the Project for Statistics on Living Standards and Development (PSLSD). This project provide baseline data for the implementation of post-apartheid socio-economic policies through South Africa’s first non-racial national sample survey. In the post-apartheid period, SALDRU has continued to gather data and conduct research directed at informing and assessing anti-poverty policy. In line with its historical contribution, SALDRU’s researchers continue to conduct research detailing changing patterns of well-being in South Africa and assessing the impact of government policy on the poor. Current research work falls into the following research themes: post-apartheid poverty; employment and migration dynamics; family support structures in an era of rapid social change; public works and public infrastructure programmes, financial strategies of the poor; common property resources and the poor. Key survey projects include the Langeberg Integrated Family Survey (1999), the Khayelitsha/Mitchell’s Plain Survey (2000), the ongoing Cape Area Panel Study (2001-) and the Financial Diaries Project.
www.saldru.uct.ac.zaLevel 3, School of Economics Building, Middle Campus, University of Cape Town
Private Bag, Rondebosch 7701, Cape Town, South AfricaTel: +27 (0)21 650 5696
Fax: +27 (0) 21 650 5797Web: www.saldru.uct.ac.za
southern africa labour and development research unit