Poverty, shocks and school disruption episodes among adolescents in KwaZulu-Natal, South Africa

Abstract

While conventional explanations of drop-out and grade repetition acknowledge the role of socioeconomic factors, this paper uses data collected in a KwaZulu-Natal study of adolescents to investigate the explicit contribution of poverty and shocks to school disruption episodes. The asset-vulnerability framework developed by Moser and others is used to develop a poverty-based theory of school disruption. Evidence against such a theory is also put forward. The results indicate that the poverty-based theory accounts in part for school disruption. Poverty is predictive of school disruption, female adolescents are particularly vulnerable to drop-out episodes, and adolescent pregnancy emerges as an important influence. However, household shocks do not seem to predict school disruption. Programmes that offer incentives for school attendance and improving school quality are put forward as policy options for South Africa.

Keywords:

• drop-out
• grade repetition
• poverty
• shocks

1. Introduction

The effects of policies instituted by a succession of apartheid governments continue to dog post-apartheid transformation in South Africa, particularly as regards the education system and schooling outcomes. Anderson et al. (2001:41), looking at 1995 data, find that while mean schooling has risen for Africans, there are still large racial discrepancies in the proportions completing primary schooling and even larger ones for completion of secondary schooling. Census 2001 data show that while grade enrolment is high overall – from the age of eight about 95 per cent of children are attending school, irrespective of sex – there are racial differences. Among 13 year olds, 54.3 per cent of Africans had completed at least grade six as their highest level of education compared to 88.8 per cent of whites (Statistics South Africa [StatsSA], 2005:52–3). In particular, adolescent pregnancy is a major cause of interrupted and discontinued education in South Africa (May et al., 1998).

However, according to Anderson et al. (2001), the African disadvantage in schooling is the result not primarily of students dropping out of school early, but rather of a slower rate of grade advancement that begins in early grades. This is of some concern since Eisemon (1997) describes repetition rates as a powerful indicator of the performance of an education system. According to Krige et al. (1994), repeating a grade is a demotivating and negative experience for a child, as well as a drain on household resources. A large number of repeaters in a grade also means there will be many children older than the norm for the class, which may cause both educational and social problems. Maharaj et al. (2000:14) highlight another dimension of the grade retention phenomenon in South Africa, the fact that African children complete primary education at a slower pace than Indian and white children.

Youth development in general and the causes of school disruption in particular have not received substantial research attention. This paper uses data collected in KwaZulu-Natal, South Africa, to investigate the role that poverty and shocks play in school disruption episodes. First, we review selected research findings on school disruption in South Africa. Next, we substantiate the theoretical basis for this analysis by introducing the asset-vulnerability framework developed by Moser (1996). We delineate a poverty-based theory of school disruption and then present evidence that contradicts this theory. We then provide details of the dataset used in this study, and follow this with a descriptive analysis of shifts from the household to the adolescent level, focusing on the issues of dropping out and dropping behind. Thereafter, we describe the multivariate analyses. In conclusion we consider possible means of supporting school attainment among adolescents.1

2. Poverty-based theories of school disruption

Income and consumption approaches to poverty analysis have been criticised for their limited ability to account for complex external factors that affect the poor (Moser, 1998). As a way of overcoming this limitation, Moser (1996) developed a classification of assets known as an ‘asset-vulnerability framework’. Insecurity is defined as the exposure to risk, and vulnerability is the resulting possibility of a decline in the well-being of individuals, households and communities in a changing environment. Various types of vulnerability can be associated with each asset, which include labour, human capital and household relations. The more assets people are able to draw on in the right combination, the greater their capacity to protect themselves against external shocks; the fewer assets available, the greater their insecurity (Moser, 1998; Devereux, 1999). The poor not only have few assets but are constrained in their ability to effectively accumulate, protect and make use of the assets that they do have (Carter & May, 1999).

Ownership of assets has also been linked to a higher probability of school attendance (Grootaert & Patrinos, 1999) and thus to the further accumulation of assets in the form of human capital. If such assets are not present, it is difficult for the household to protect its members against external shocks or crises, which means that children may be forced to leave school as part of a household coping strategy.

An obvious tension exists between the contributions that adolescents can make to family income and the importance attached to investments in their education. As the household matures, children move from being net consumers to net producers, yet the pace of this natural transition process may be rapidly increased when a shock occurs (Devereux, 1999). In developing countries, those charged with decision-making in poor households may be compelled by their economic circumstances to rely on adolescents to contribute to household welfare, through employment in the labour force or by undertaking household tasks so that adults are able to spend more time in employment or self-employment (Grootaert & Patrinos, 1999).

Moreover, because opportunity costs for investments in boys and girls differ, household decision-makers may allocate food, provide health care, and invest in education differently according to gender (Kimmel & Rudolph, 1998). Therefore, a shock that affects the household as a whole may have different effects on different household members (World Bank, 2000). Internationally, studies support the notion that drop-out rates for girls are higher than those of boys, and that these tend to increase when economic conditions worsen (United Nations Development Fund for Women [UNDFW], 2000). May et al. (1998) show that in the event of a financial crisis girls may be more at risk of being taken out of school than boys, as female children tend to be prejudiced in terms of furthering their education, since they may eventually marry into another household.

3. Evidence against a poverty-based theory of school disruption

While there is substantial evidence for a poverty-based theory of school disruption, it is also important to draw attention to literature that shows that school disruption is not linked to poverty and shocks. Some vulnerable households may balance their risk against the maintenance of human capital, and may choose to keep their children in school rather than send them out to work. All over the world, participants in the ‘Voices of the Poor’ study (World Bank, 2000) mention child labour as an undesirable coping mechanism. Moreover, in Moser's urban study (1996) conducted in some developing countries, children who work do not necessarily drop out of school but, against all odds, manage to keep studying.

Among both rural and urban poor South Africans, education is consistently seen as the highest priority need and the most effective route out of poverty, and therefore often protected at great cost. Since the principal asset of the poor is labour time (i.e. time spent working), education increases the productivity of this asset by increasing its value. Financial assistance, particularly in the form of paying for school fees, is a frequently mentioned mode of support offered by social and kin networks in the South African context, and gaining access to education is seen as a way for the household as a whole to benefit (May et al., 1998). A rural study of 30 extremely poor households with malnourished children in the Eastern Cape corroborates these findings: half of the households that had sold assets in the preceding 12 months did so to cover educational expenses (Sogaula et al., 2002).

With unemployment rates of 24 per cent (using the narrow definition) and 38 per cent (using the broad definition), and less than 40 per cent of the working age population actually working (Klasen & Woolard, 2001:2), a poverty-based theory of school disruption seems less likely. Negative job prospects may instead provide an incentive to stay in school longer.

With regard to the gender dimension of the poverty-based theory of school disruption, there is also evidence to the contrary. South Africa is one of five countries out of 34 where the relative disadvantage of girls has been eliminated, and where there is in fact a reverse gender gap (UNDFW, 2000). In 1997 the absolute level of girls' net enrolment in secondary school in South Africa was 97 per cent (UNDFW, 2000:69). In fact, among the poorer quintiles girls had higher primary and secondary enrolment rates than boys (World Bank, 1995). The 2001 Census shows that among 13 year olds grade completion is higher for girls than for boys: 65.3 per cent of girls had completed grade six or higher compared with 52.4 per cent of boys (StatsSA, 2005:53).

4. The data

On balance, evidence for South Africa seems to undermine the poverty-based theory of school disruption. Data from the study of Transitions to Adulthood among Adolescents in Durban (Rutenberg et al., 2001) allow us to investigate these issues more directly. The focus of the Transitions study was not on the issue of household shocks, and the data were not tailored to these purposes. However, because rudimentary shock data were gathered, it provides a unique opportunity to look at how adolescents are affected by poverty and to contribute to the debate on the issue of schooling and poverty in South Africa.

The data used in this paper were collected in September and October 1999. Two administrative areas in the province of KwaZulu-Natal – the Durban Metropolitan and Mtunzini Magisterial Districts – were purposively selected to ensure the sample covered a variety of urban, transitional and rural regions in the province. The sample was 77 per cent urban (the Durban Metropolitan District and the urban part of the Mtunzini Magisterial District) and 23 per cent rural (the Mtunzini Magisterial District).

A modified multi-stage cluster sample approach was drawn for this probability study, and 120 census enumeration areas were randomly selected from a sampling frame of all such areas in the two districts. A total of 2007 structured interviews were conducted with households that contained adolescents between the ages of 14 and 22 years in 118 of the selected segments. In all, 3096 individual interviews were completed with adolescents in this age group in these households. For the purposes of this analysis, 1974 of these household interviews and 3013 of the adolescent interviews were used, where information on both could be linked.

Over three-quarters (76 per cent) of these adolescents were African, 16 per cent were Indian, 6 per cent were white and 2 per cent were coloured.2 Just over half (55 per cent) of the adolescents were female, the average age was about 18 years (mean = 17.5), and 72 per cent were currently in school.

Per-capita poverty lines were constructed, and statistics were computed for per-capita poor households (n = 1087) and per-capita ultra-poor households (n = 598). Although not a perfect measure, total monthly expenditure is widely regarded as preferable to income as a measure of household material well-being (Carter & May, 1999). Total household expenditure, comprising household food expenditure, non-food expenditure and expenditure on infrequent items, was used to calculate whether a household was poor or ultra-poor. Households with a per-capita expenditure of less than R250 a month were defined as per-capita poor, on the basis of an unofficial poverty line used in a recent StatsSA report (2000). Following Carter & May (1999), lower poverty lines that were half the amount of the upper poverty lines were constructed, and ultra-poor households were defined as living on less than R125 per capita per month. While the asset-vulnerability framework favours the use of proxies for wealth other than income or expenditure, the Transitions data make preferable the use of expenditure information to define poverty. However, infrastructural variables were used in the multivariate analyses in addition, in line with the asset-vulnerability framework.

Using these definitions, 57 per cent of households were poor, which is reasonably close to the estimated poverty rate of 50 per cent for KwaZulu-Natal (StatsSA, 2000). Descriptive statistics indicate that the poor, and particularly the ultra-poor, had strikingly less access to services, shelter, education and employment than the non-poor. They were more likely to be African, female headed and living in larger households.

4.1 The experience of shocks in households in KwaZulu-Natal

On the whole, a substantial proportion (41 per cent) of all households reported that they had experienced at least one type of shock during the reference period. Demographic shocks occurred most often (29 per cent of households), with the death of a household member being most common (19 per cent), followed by the injury or illness of a household member (15 per cent). Economic shocks affected 17 per cent of households, among which the loss of a job by a household member (14 per cent) was most frequently reported. Asset-livelihood shocks (i.e. shocks that affected household assets or livelihoods) affected 13 per cent of households, most commonly theft, fire or the destruction of property (10 per cent). A fifth (23 per cent) of households experienced one shock only, while 12 per cent encountered two shocks.

The more important types of shocks recorded in the South African Participatory Poverty Assessment (May et al., 1998) match those reported in the Transitions survey: death, illness, loss of employment and fire. The high proportion of households that noted the death of a household member could be due to the high incidence of HIV/AIDS in KwaZulu-Natal (Shisana & Simbayi, 2002).

The Transitions survey confirms too that the poor are most exposed to a wide array of risk, and that living with risk is part of life for poor people (World Bank, 2000:135). Poor households were more likely to experience shocks than the non-poor, with 48 per cent of ultra-poor households having experienced a shock in the previous 24 months, compared to 45 per cent of the poor and 37 per cent of the non-poor.

According to the World Bank (2000), the accumulation of different shocks is a source of significant stress for households. Indeed, in the Transitions study, poor households experienced significantly more of each type of shock than those that were not poor. In addition, significantly more of the poor and the ultra-poor, when compared with the non-poor and the non-ultra-poor respectively, experienced the interaction of shocks.

4.2 Coping strategies used in response to shocks

Just over half (51 per cent) of the households that experienced one or more shocks did not report any form of reaction as a strategy for managing the impact of these shocks. It is possible that this may be more likely among poorer households and might introduce a systematic bias. Of those households that experienced at least one shock, and did respond, over two-thirds (69 per cent) used an economic response to at least one shock (sold their assets or used their savings, borrowed money from a moneylender, used insurance or used an informal financial association, known as a stokvel). Half of all the households that responded to a shock used reciprocal relationships and social networks as a coping strategy. Only 3 per cent of the households that responded to a shock took one or more of their children out of school. From the level of the household this coping response was extremely infrequently used.

4.3 School disruption

In the following sections the focus shifts to the adolescent level, and the issue is defined in terms of school disruption episodes. Two definitions of school disruption were used: ‘dropping out of school’ and ‘dropping behind in school’. By definition these events are more likely to occur among older adolescents since they accumulate over time. Moreover, dropping out is not a terminal condition and the definition includes adolescents who have temporarily dropped out and may return to school at a later date.

In the Transitions study the full educational history of all adolescents aged 14 to 22 years in the sampled households was recorded in a calendar format. Each respondent was asked to state the grade he or she was attending at each age. In the calendar, all full years of education, repeated grades, full and partial years of school absence, and reasons for any interruption were recorded. Since the household respondent was required to state whether any shocks had occurred during the 24 months prior to the interview, the education history was confined to the same period. Disruption episodes were defined as having occurred within the 2-year period so as to link the shock and disruption information. Furthermore, 452 adolescents who had completed grade 12 at the start of this period were excluded from the analysis, since their educational histories could not be linked to shock information. Those who were not in school but had not completed their schooling were also included in the analysis.

A school disruption episode was registered if during the current year and/or during the two calendar years preceding the survey the adolescent reported either that they had not completed a grade, or that they had left school after completing the year of schooling without having completed grade 12, or that they had repeated one or more grades, or a combination of these possibilities.

4.4 Dropping out of school

Completing grade 12 is regarded as an important achievement, the lack of which jeopardises future economic and human development. Basic education is guaranteed for all children in South Africa's Constitution, and non-payment of school fees should not result in a child being excluded from school. Additional schooling costs, in the form of school uniforms, books and transport, can, however, place a burden on poor households. Since schooling is not free, a negative income shock could lead to school disruption. However, the data do not enable us to look at the role school costs play in school disruption.

A school drop-out was defined as being in one of grades 1 to 12 at the start of the period under review, and as having left school at some stage during this time. Adolescents defined as not having dropped out of school either repeated a grade and were included in the second definition of disruption, or experienced a continuous progression from grade to grade without a disruption episode. Adolescents who, for example, completed grade 10 and then moved on to complete a diploma, without returning to school, were defined as not having experienced a disruption episode, as this movement was regarded as having arisen from the adolescent's choice of education.

The findings show that 16 per cent of the adolescents dropped out of school. Girls (18 per cent) were more likely than boys (14 per cent) to experience a drop-out episode. Moreover, 19 per cent of Africans, as opposed to 9 per cent of coloureds, 9 per cent of Indians and 8 per cent of whites, dropped out of school. As expected, the average age of those who dropped out was higher (18.4 years) than those who had not (16.7 years). Further, 21 per cent of adolescents in rural areas experienced a school drop-out episode, compared with 14 per cent of those in urban areas, and the difference is significant.

An assessment of reasons given for leaving school before matriculating revealed a number of gender differences. Over a third (38 per cent) of female adolescents left school because they fell pregnant. Thirteen per cent of girls were currently pregnant, while 46 per cent had been pregnant at some stage. Fifty-nine per cent of the latter had experienced a drop-out episode, and 41 per cent had repeated at least one grade. Among those who had never been pregnant, 24 per cent had experienced a drop-out episode and 60 per cent had repeated a grade.

Eighteen per cent of males as opposed to 2 per cent of females reported that they had left school because they needed to work, and 9 per cent of males expressed no interest in attending school as opposed to 3 per cent of females. While there was a sex difference in leaving school to perform domestic duties, it was small – 3 per cent of females and 1 per cent of males. Overall, for both sexes – 27 per cent of males and 30 per cent of females – the cost of school fees was the central reason for leaving school before matriculating. This underscores other arguments (May et al., 1998; Sogaula et al., 2002) that the costs of education remain a significant barrier to accessing education.

As Table 1 shows, those adolescents who experienced a drop-out episode were more likely than those who did not to agree that in their school teachers were drunk and teachers were threatened by students. With regard to connectedness, drop-outs were more likely than non-drop-outs to agree with ‘negative’ comments (such as ‘a lot of crime in my neighbourhood/community’), and less likely to agree with ‘positive’ comments (such as ‘I have many friends at this school’). Interestingly, drop-outs were more likely than non-drop-outs to know someone with HIV/AIDS, and someone who had died of HIV/AIDS. Drop-outs were also more likely to have worked and to have looked for work than their counterparts, and this group spent more hours and more months on average doing work than non-drop-outs.

Table 1: School, community and work characteristics of adolescents who drop out (means and percentages)

Characteristics	Drop-out	No drop-out	p-value
Problems in school
Teachers drunk	16.6	7.8	0.000
Teachers threatened	27.9	21.1	0.038
Bad security	28.7	23.0	0.109
Connectedness in school
Many friends at this school	61.2	75.4	0.000
Teachers care about students	89.2	93.1	0.021
Sexual harassment a problem	12.2	17.4	0.018
Connectedness in neighbourhood/community
Many friends	53.2	67.8	0.000
Adults help other families in trouble	73.6	80.6	0.001
A lot of crime	43.9	35.5	0.001
A lot of violence among youth	25.0	18.4	0.002
Happier if lived in another community	48.5	39.4	0.001
People trust one another	56.9	65.1	0.001
HIV/AIDS
Know anyone with	21.6	16.0	0.035
Know anyone who has died of	19.9	15.2	0.011
Work
Ever worked for cash	31.8	18.8	0.000
Worked in last 12 months	74.6	65.4	0.048
Ever spent time looking for work	27.6	12.4	0.000
Looking for work last 12 months	81.9	72.1	0.041
Mean months spent doing this work^a	5.3	4.4	0.056
	4.2	3.8
Mean hours spent doing this work per week^a	37.2	25.1	0.000
	21.2	21.1
Notes: Standard deviations in italics.
^aThese variables are defined only for those who are working.

Significantly, more of the poor (21 per cent) than the non-poor (9 per cent), and more of the ultra-poor (23 per cent) than the non-poor (12 per cent) had experienced a drop-out episode during the reference period. Almost half (48 per cent) of those who dropped out, as opposed to 42 per cent of those who did not, came from households that had experienced one or more shocks during the reference period, and the difference was significant. Therefore drop-outs are more likely to come from a poor household and from one that has experienced a household shock.

Those adolescent drop-outs who returned to school during the defined time period were included in the analysis in order to determine the proportion of pregnant drop-outs who eventually return to school. Of the 39 female adolescents only two returned to school by the end of this period. Likewise, an analysis of the 1996 Census shows that in every age group (12 to 15; 16 to 25; 26+) women who had given birth were less likely to be studying than those who had not (StatsSA, 2001).

4.5 Dropping behind in school

Instead of dropping out of school, some adolescents may drop behind, and have to repeat one or more grades. Sixty-two per cent of the adolescents repeated one or more grades during the reference period, with no significant difference between females and males. Thirty-seven per cent of girls who dropped behind were currently pregnant or had been pregnant. The younger ones were more likely to repeat than those in the older age groups: 69 per cent of 14 to 16 year olds, 58 per cent of 17 to 19 year olds, and 48 per cent of 20 to 22 year olds were retained in a grade. Noteworthy percentages of all race groups dropped behind: 65 per cent of Africans, 59 per cent of coloureds, 49 per cent of Indians and 57 per cent of whites. In addition, more adolescents in rural areas (66 per cent) than in urban areas (59 per cent) repeated a grade.

As Table 2 shows, those who dropped behind displayed less ‘connectedness’ than those who did not, both at school and in their neighbourhood or community. These adolescents were less likely to have worked or to have spent time looking for work than those who did not drop behind, and they spent less time on average doing work than their counterparts.

Table 2: School, community and work characteristics of adolescents who drop behind (means and percentages)

Characteristics	Repeat	No repeat	p-value
Connectedness in school
Many friends at this school	71.9	77.7	0.003
Happier if attended other school	37.4	31.2	0.003
Connectedness in neighbourhood/community
Happier if lived in another n/c	42.5	38.3	0.039
Work
Ever worked for cash	15.8	29.2	0.000
Worked in last 12 months	59.2	75.2	0.000
Mean months spent doing this work	5	4.3	–
Mean hours spent doing this work	32.9	22.4	–
Spent time looking for work	9.5	23.6	0.000
Looking for work last 12 months	69.3	78.8	0.037

Adolescents who dropped behind were more likely to come from poor households (65 per cent) than from non-poor households (57 per cent) and from ultra-poor households (65 per cent) than from non-ultra-poor households (60 per cent). However, adolescent ‘repeaters’ were not more likely to come from households that had experienced one or more shocks during the reference period. Sixty-three per cent of adolescents from ‘shock’ households repeated a grade, compared with 61 per cent of adolescents from ‘non-shock’ households. Therefore, grade repetition seems to be associated with poverty but not with shocks.

4.6 Multivariate analyses

Descriptive statistics are not sufficient to reveal causality, and unobserved factors may mediate an apparent relationship. Therefore, two survey logistic regression models for each type of disruption episode were constructed – one that included shocks as predictors and excluded variables endogenous to shocks, and another that excluded shocks and included endogenous variables. While the data are not ideally suited to multivariate analysis, they point to the type of analysis that should be done, and add to the debate on education in the South African context.

Variables that are endogenous to demographic shocks may be affected by a change in the size or composition of the household, such as the number of children, the average age of the members, the proportion of members educated to grade 10, or the proportion of the members employed. These variables may also be affected by a change in health status, such as a period of illness in the household. Variables that are endogenous to economic shocks, such as the employment of the household head, a pension or other government grant, or per-capita poor, may be affected by a change in household income. Infrastructural variables, mainly ‘homestead characteristics’, are taken to be indices of asset vulnerability, and function as proxies for wealth. Many of the variables were endogenous (e.g. connectedness to school and community), and collinearity between a number of these variables was therefore likely. Where joint significance was found between variables, one of the variables was excluded. In all the models that were run, an ‘education of mother’ variable was initially included. However, it was later excluded as information was available only for those adolescents whose mothers lived in the same household.

The models were weighted by sampling weights and the cluster identifier variable was the enumeration area. The adjusted Wald F test was used after running survey logistic regression models with and without groups of variables to determine whether or not variables were significant as a group. A maximum likelihood logit estimation specifying the enumeration area as the cluster identifier variable was run after the final model had been ascertained, in order to obtain the overall model fit statistics.

In the first model the dependent variable was a school drop-out episode, shocks were included and variables endogenous to shocks were excluded (see Table 3). Two age groups (17 to 19; 20 to 22) were individually significant and jointly significant. Since the older ones were expected to be more likely to drop out of school than the younger ones, the third age group (20 to 22) was left in the model. Having piped internal water and having a flush toilet were both significant individually and jointly. Having piped water was a positive predictor of drop-out, but having a flush toilet was a negative predictor of drop-out. According to Clark, in the South African context water-related variables are exceedingly difficult to understand in relation to poverty because of the bulk infrastructure provision (personal communication, S Clark, Research Associate, Graduate Group in Demography, University of Pennsylvania, Durban, 19 September 2002). Since having piped water was not easily substantiated by theory it was excluded from the model. Overall, the model explains just 9 per cent of the variance in the outcome variable. Since the data and the literature point to an association between grade repetition and school drop-out, it seems obvious to include grade repetition as a predictor variable. However, since the study covered only 2 years, and since an adolescent who was repeating would not have dropped out of school, it was not possible to do so.

Table 3: Shocks included, all variables endogenous to shocks excluded

Adolescent/household characteristics	Odds ratio	Standard error	z-score	p-value	95% confidence interval
Adolescent characteristics
Female (male)	0.425	0.258	1.650	0.102	–0.086	0.936
20–22 years (14–16 years)	1.261	0.137	9.170	0.000	0.988	1.533
Coloured (Black)	0.658	0.851	0.770	0.441	–1.028	2.343
Indian (Black)	–0.031	0.354	–0.090	0.930	–0.733	0.671
White (Black)	0.370	0.460	0.800	0.423	–0.542	1.282
Household head characteristics
Female (male)	0.012	0.136	0.090	0.929	–0.256	0.280
Age	–0.005	0.006	–0.770	0.441	–0.018	0.008
Highest grade passed	–0.016	0.157	–0.100	0.919	–0.328	0.296
Household member characteristics
Adult male present	–0.227	0.168	–1.350	0.181	–0.560	0.107
Homestead characteristics
Number of rooms	–0.048	0.042	–1.140	0.257	–0.131	0.035
Own home	–0.119	0.148	–0.800	0.425	–0.413	0.175
Flush toilet	–0.745	0.325	–2.290	0.024	–1.388	–0.102
Electricity	0.189	0.261	0.720	0.471	–0.328	0.706
Telephone	–0.544	0.346	–1.570	0.118	–1.229	0.141
Traditional house	0.029	0.189	0.150	0.878	–0.346	0.404
Shack	–0.199	0.547	–0.360	0.716	–1.282	0.883
Household shocks
Demographic	–0.002	0.254	–0.010	0.994	–0.505	0.501
Economic	–0.171	0.406	–0.420	0.674	–0.975	0.632
Asset–livelihood	–0.451	0.423	–1.070	0.289	–1.289	0.387
Demographic/asset-livelihood	0.628	0.435	1.440	0.152	–0.234	1.489
Demographic/economic	0.066	0.458	0.140	0.885	–0.841	0.973
Economic/asset-livelihood	0.075	0.627	0.120	0.905	–1.166	1.317
School characteristics
Problems	0.123	0.079	1.570	0.118	–0.032	0.279
Connectedness	–0.025	0.081	–0.310	0.756	–0.185	0.134
Community characteristics
Connectedness	–0.009	0.064	–0.150	0.883	–0.136	0.117
Notes: Reference category for each group indicated in parentheses where applicable. Maximum likelihood logit estimation model (n = 2124). Dependent variable: adolescent school drop-out episode. Prob > chi^2 = 0.000. Log likelihood = –679.221. Pseudo R ^2 = 0.091.

The second ‘drop-out’ model excluded shocks and included variables endogenous to shocks. As Table 4 shows, this model explained 10 per cent of the variance in the outcome variable. The two older age groups were significant individually and jointly and the middle age group variable was therefore excluded. The third and fourth quartiles of the education distribution of the household to grade 10 were jointly significant. These variables were negative predictors of drop-out, and the third quartile was therefore left out of the model. The number of children in the household was a positive predictor of drop-out. This variable and the household size variable were tested for joint significance, but no significance was found. Being poor predicts drop-out, and having a flush toilet is a negative predictor of drop-out. Odds ratios indicated that being in the oldest age group and being poor increase the odds of experiencing a drop-out episode. Adolescents in per-capita poor households were more likely to experience a drop-out episode than non-drop-outs. Therefore, the poverty-based theory of school disruption accounts in part for its incidence.

Table 4: All shocks excluded, all variables endogenous to shocks included

Adolescent/household characteristics	Odds ratio	Standard error	z–score	p–value	95% confidence interval
Adolescent characteristics
Female (male)	1.330	0.320	1.180	0.239	0.826	2.142
20–22 years (14–16 years)	3.601	0.558	8.270	0.000	2.650	4.895
Coloured (Black)	2.338	1.689	1.180	0.242	0.559	9.776
Indian (Black)	1.316	0.463	0.780	0.436	0.656	2.640
White (Black)	1.599	0.824	0.910	0.364	0.576	4.439
Household head characteristics
Female (male)	1.098	0.129	0.800	0.428	0.870	1.385
Age	0.998	0.005	–0.360	0.718	0.988	1.008
Highest grade passed	1.065	0.149	0.450	0.652	0.808	1.405
Employed	1.076	0.257	0.310	0.760	0.670	1.726
Household member characteristics
Average age	1.023	0.014	1.660	0.099	0.996	1.052
Adult male present	0.904	0.171	–0.530	0.595	0.622	1.315
Household size	0.878	0.058	–1.980	0.050	0.770	1.000
Number of children	1.195	0.103	2.060	0.041	1.007	1.419
Proportion educated (1st quartile)
2nd quartile	0.844	0.212	–0.680	0.501	0.512	1.389
4th quartile	0.525	0.155	–2.180	0.031	0.293	0.942
Proportion employed (1st quartile)
2nd quartile	0.951	0.198	–0.240	0.811	0.630	1.437
3rd quartile	0.883	0.246	–0.450	0.657	0.509	1.534
4th quartile	0.916	0.247	–0.320	0.746	0.538	1.562
Homestead characteristics
Number of rooms	1.004	0.056	0.070	0.942	0.899	1.122
Own home	0.847	0.156	–0.900	0.369	0.588	1.220
Flush toilet	0.566	0.154	–2.090	0.039	0.330	0.970
Electricity	1.231	0.343	0.750	0.457	0.709	2.136
Telephone	0.714	0.218	–1.100	0.273	0.390	1.308
Traditional house	0.983	0.216	–0.080	0.936	0.636	1.518
Shack	0.859	0.374	–0.350	0.728	0.363	2.035
Income/expenditure
Pension	1.303	0.232	1.490	0.139	0.916	1.853
Other grant	1.289	0.285	1.150	0.254	0.831	1.997
Per-capita poor	2.589	0.724	3.400	0.001	1.487	4.506
School characteristics
Problems	1.146	0.090	1.740	0.085	0.981	1.339
Connectedness	0.989	0.074	–0.150	0.878	0.852	1.147
Community characteristics
Connectedness	0.989	0.060	–0.190	0.850	0.877	1.114
Notes: Reference category for each group indicated in parentheses where applicable. Maximum likelihood logit estimation model (n = 2150). Dependent variable: adolescent school drop-out episode. Prob > chi^2 = 0.000. Log likelihood = -672.922. Pseudo R ^2 = 0.098.

When the ‘drop-out’ models were run separately for girls and boys, the strength increased for the ‘girl only’ models and decreased for the ‘boys only’ models, while the predictors remained relatively consistent. In both the ‘boys only’ and ‘girls only’ models, adolescents in the second and third age categories (17 to 19 and 20 to 22) were more likely to experience a drop-out episode than younger adolescents. Those in households with flush toilets were less likely to drop out, and those in households with piped internal water were more likely to drop out. White boys were more likely to drop out than African boys, while those who lived in households with fewer rooms and those who experienced problems at school were more likely to drop out. Moreover, in the ‘boys only’ model that excluded shocks and included variables endogenous to shocks, boys in households that had experienced a period of illness of a household member were more likely to experience a drop-out episode, and those in households in which the third and fourth quartiles of the household were educated were less likely to experience a drop-out episode. In the models run for girls only, white girls were less likely to drop out of school than African girls. In the model that excluded shocks, girls in households that were per-capita poor were more likely to drop out. While formally defined shock variables did not feature as predictors of drop-out, boys were more likely to drop out if a household member had experienced a period of illness during the previous 3 months. Finally, being poor was a predictor of drop-out among girls.

The two types of model were also run with grade repetition as the dependent variable, and joint significance was tested for. In both models – see Tables 5 and 6 – the pseudo R ² values were found to be very low, at 3 per cent. As noted, this is most likely attributable to the way the ‘grade repetition’ variable was defined, which made it unsuitable for assessing causality.

Table 5: All shocks included, all variables endogenous to shocks excluded

Adolescent/household characteristics	Odds ratio	Standard error	z- score	p-value	95% confidence interval
Adolescent characteristics
Female (male)	0.983	0.151	–0.110	0.914	0.726	1.332
20–22 years (14–16 years)	0.678	0.090	–2.920	0.004	0.521	0.882
Coloured (Black)	1.334	0.487	0.790	0.432	0.647	2.749
Indian (Black)	0.471	0.116	–3.050	0.003	0.288	0.768
White (Black)	0.540	0.189	–1.760	0.080	0.270	1.078
Household head characteristics
Female (male)	0.899	0.099	–0.970	0.335	0.724	1.117
Age	0.998	0.005	–0.360	0.717	0.989	1.008
Highest grade passed	1.107	0.127	0.890	0.375	0.883	1.389
Household member characteristics
Adult male present	1.277	0.223	1.400	0.164	0.904	1.805
Homestead characteristics
Number of rooms	1.023	0.024	0.940	0.350	0.975	1.072
Own home	1.039	0.125	0.320	0.750	0.819	1.317
Flush toilet	1.227	0.201	1.250	0.215	0.887	1.698
Electricity	1.081	0.235	0.360	0.720	0.703	1.664
Piped internal water	0.608	0.124	–2.440	0.016	0.406	0.911
Telephone	1.216	0.293	0.810	0.419	0.755	1.958
Traditional house	1.163	0.168	1.050	0.297	0.874	1.547
Household shocks
Demographic	1.153	0.187	0.870	0.384	0.835	1.591
Economic	1.040	0.218	0.190	0.851	0.687	1.576
Asset-livelihood	1.362	0.411	1.020	0.309	0.749	2.476
Demographic/asset-livelihood	0.709	0.290	–0.840	0.402	0.316	1.593
Demographic/economic	0.694	0.244	–1.040	0.302	0.346	1.394
Economic/asset-livelihood	1.225	0.489	0.510	0.612	0.556	2.703
School characteristics
Problems	0.972	0.036	–0.770	0.446	0.903	1.046
Connectedness	1.017	0.057	0.300	0.763	0.910	1.136
Community characteristics
Connectedness	1.151	0.086	1.870	0.063	0.992	1.336
Notes: Reference category for each group indicated in parentheses where applicable. Maximum likelihood logit estimation model (n = 2193). Dependent variable: adolescent school grade repetition episode. Prob > chi^2 = 0.000. Log likelihood = –1351.006. Pseudo R ^2 = 0.024.

Table 6: All shocks excluded, all variables endogenous to shocks included

Adolescent/household characteristics	Odds ratio	Standard error	z-score	p-value	95% confidence interval
Adolescent characteristics
Female (Male)	1.009	0.160	0.060	0.953	0.738	1.381
20–22 years (14–16 years)	0.702	0.107	–2.330	0.022	0.519	0.949
Coloured (Black)	1.500	0.487	1.250	0.215	0.788	2.855
Indian (Black)	0.578	0.149	–2.130	0.036	0.347	0.963
White (Black)	0.712	0.253	–0.960	0.340	0.352	1.438
Household head characteristics
Female (Male)	0.941	0.105	–0.550	0.585	0.755	1.173
Age	0.997	0.005	–0.630	0.532	0.987	1.007
Highest grade passed	1.144	0.142	1.090	0.279	0.895	1.463
Employed	1.123	0.204	0.640	0.524	0.784	1.61
Household member characteristics
Number of members	1.021	0.035	0.610	0.546	0.955	1.091
Number of children	0.922	0.045	–1.670	0.097	0.838	1.015
Adult male present	1.256	0.235	1.210	0.227	0.866	1.82
Average age	0.972	0.010	–2.770	0.006	0.953	0.992
Proportion educated grade 10 (1st quartile)
2nd quartile	1.369	0.253	1.700	0.092	0.950	1.973
3rd quartile	1.453	0.284	1.910	0.059	0.986	2.141
4th quartile	1.612	0.504	1.530	0.130	0.867	2.996
Proportion employed (1st quartile)
2nd quartile	0.989	0.135	–0.080	0.938	0.755	1.298
3rd quartile	0.965	0.166	–0.210	0.836	0.686	1.356
4th quartile	0.819	0.143	–1.150	0.254	0.580	1.156
Homestead characteristics
Number of rooms	1.023	0.030	0.750	0.454	0.964	1.085
Own home	1.024	0.121	0.200	0.842	0.811	1.293
Traditional house	1.154	0.171	0.970	0.334	0.861	1.548
Flush toilet	1.091	0.191	0.500	0.618	0.772	1.544
Piped internal water	0.617	0.121	–2.450	0.016	0.418	0.911
Electricity	1.018	0.228	0.080	0.935	0.653	1.588
Telephone	1.192	0.274	0.760	0.447	0.756	1.878
Income/expenditure
Pension	1.460	0.196	2.810	0.006	1.119	1.905
Other grant	1.124	0.230	0.570	0.568	0.750	1.686
Per capita poor	1.203	0.166	1.350	0.181	0.916	1.58
School characteristics
Problems	0.964	0.034	–1.020	0.310	0.898	1.035
Connectedness	1.019	0.059	0.330	0.742	0.909	1.144
Community characteristics
Connectedness	1.152	0.093	1.760	0.082	0.982	1.351
Notes: Reference category for each group indicated in parentheses where applicable. Maximum likelihood logit estimation model (n = 2150). Dependent variable: adolescent school grade repetition episode. Prob > chi^2 = 0.000. Log likelihood = –1309.558. Pseudo R ^2 = 0.032.

In both models the two older age groups (17 to 19 and 20 to 22) were jointly significant, and the ‘middle’ age group (17 to 19) was therefore left out of the model. In both models, having piped internal water and living in a shack were negative predictors of repeating a grade. These variables were jointly significant, and in each model the ‘shack’ variable was removed. In both models adolescents in the oldest age group (20 to 22) were less likely to repeat a grade than adolescents aged 14 to 16, and Indian adolescents were less likely than Africans to repeat a grade. Adolescents living in households with piped internal water were less likely to repeat a grade.

In the model that excluded shocks and included variables endogenous to shocks (see Table 6), the average age of the household was a negative predictor of grade repetition. Therefore, as the average age of the household increased, the likelihood of a grade repetition episode decreased. Having piped internal water negatively predicted grade repetition. In line with the poverty-based theory of school disruption, adolescents in households in which a state pension was received were more likely to experience grade repetition. Adolescent level findings evidenced in these grade repetition models confirmed findings in the grade retention literature. These models indicate that socioeconomic status may play a role in adolescent grade repetition, but the impacts of significant predictors are modest. Logistic regression models that predict an adolescent drop-out episode account for more of the variance in the outcome variables, and show relatively convincingly that certain adolescent and homestead characteristics, including being poor, predict adolescent school drop-out episodes.

Two models were run using whether an adolescent had experienced both a grade repetition episode and a drop-out episode in the reference period as a dependent variable. In both types of models adolescents in the two older age groups (17 to 19 and 20 to 22) were more likely to experience both types of disruption episodes than those aged 14 to 16 years. Those living in traditional houses and those living in houses with piped internal water were also more likely to experience school disruption. In the model that excluded shocks and included predictors endogenous to shocks, adolescents in per-capita poor households were more likely to experience both types of disruption episode than those in non-poor households. Adolescents in households where the fourth quartile of the household were employed were less likely to experience a disruption in their schooling than those in which the first quartile of the household only were employed.

Finally, it is worth noting that in all the models that included shocks and excluded variables endogenous to shocks, shock and shock interaction variables were found to be unimportant and not significant. It is likely that null results may in fact represent poor measurement of the variables in question, rather than null effects.

5. Conclusion

The findings from the Transitions study (Rutenberg et al., 2001) suggested that adolescents from poor households are more likely to experience school disruption episodes than those from non-poor households. Girls were more vulnerable to drop-out episodes than boys, and poverty predicted drop-out among girls. In contrast, boys who experienced problems in the school environment and those who came from households where a period of illness had been experienced were more likely to drop out of school. Adolescent pregnancy was not treated as a shock and perhaps should have been, since it emerged as an important influence. None of the formally defined shock variables or shock interaction variables were predictive of school disruption, although we are mindful of the methodological challenges that all data have. The data suggest that poor households do not resort to removing children from school as a coping strategy in response to shocks, and that shocks are not associated with a higher probability of school disruption. Therefore, school disruption is associated with poverty, but not with shocks.

Since school disruption is associated with poverty, we need to ask how school attendance and progression among the poor can be supported, to ensure that human capital is developed, and poor adolescents and households can move out of poverty. The Poverty and Inequality Report of 1998 recommends that poor families be assisted by programmes that augment their assets, expand their existing coping or household management strategies, or facilitate new opportunities (May, 1998).

The implementation of an income grant linked to a school incentive scheme that targets adolescents is one option for consideration in South Africa. This would have short-run benefits for the poor in terms of income received, and long-term impacts in the development of human capital. Such grants may also assist the return to school of adolescent girls who have experienced a pregnancy but who have not continued with their education, and those children unable to attend school because of the fees or additional school expenses. The benefits of such incentive-based income grants include reduced drop-out rates, advanced progression through grades and a reduced reliance on child labour (Behrman et al., 2001).

Clearly, a policy prescription that focuses on keeping children in school is of little value if the learning environment fails to build human capital successfully. Case & Yogo (1999) found that the quality of schools in a respondent's magisterial district of origin had a large and significant effect on the rate of return to schooling for African men. Moreover, school quality significantly affects educational attainment and the probability of employment. Striving to improve the quality of schooling is therefore a fundamental first step.

Acknowledgements

This publication is made possible through support provided by the US Agency for International Development through the Horizons Project (under the terms of co-operative agreement No. HRN-A-00-97-00012-00), the FOCUS on Young Adults Project (co-operative agreement No. CCP-3-73-A00-6002-00), the MEASURE/Evaluation Project (co-operative agreement HRN-A-00-97-00018-00), and by a Mellon Foundation grant to the University of KwaZulu-Natal. The opinions expressed herein are those of the authors and do not necessarily reflect the views of the funding organization. The authors have benefitted from discussions with Sam Clark, Ali Karim, Bob Magnani, John Maluccio and Benjamin Roberts.

Notes

An earlier version of this article was presented at the Demographic Association of Southern Africa Conference, Data Definition and Dissemination, School of Government, University of the Western Cape, 26–27 September 2002. The paper is also available as a School of Development Studies Working Paper (No. 35).

¹In South Africa a large proportion of schoolgoers are in fact young adults. The term ‘adolescent’ is used for convenience in this article to cover both adolescents and young adults.

²The racial classification used by StatsSA (2000) has been adopted here.