Income and Social Grants - Children living in poverty
Income and Social Grants - Children living in poverty
Author/s:  Katharine Hall
Date: April 2012
Definition
This indicator shows the number and proportion of children living in households that are income poor. Three poverty lines are used: an 'upper' poverty line, a 'lower' poverty line and an 'ultra-low' poverty line equivalent to US$2-a-day. These values are set in the year 2000 and are increased each year in line with inflation.
  • Upper poverty line: R593 per person per month in 2000 prices (R1060 in 2010)
  • Lower poverty line: R322 per person per month in 2000 prices (R575 in 2010)
  • $2-a-day poverty line: R163 per person per month in 2000 prices (R290 in 2010)
Per capita income is calculated by adding all reported income for household members over 15 years, then adding all income from social grants, and dividing the total household income by the number of household members. Income is known to be under-reported generally, and particularly in the General Household Survey. Social grants are also severely under-reported in the GHS. Child poverty is therefore likely to be over-estimated.
Data
Data Source
  • Statistics South Africa (2004 - 2011) General Household Survey 2003 - 2010. Pretoria, Cape Town: Statistics South Africa.
  • Analysis by Katharine Hall, Children’s Institute, University of Cape Town.
Notes
  1. Children are defined as persons aged 0 – 17 years.
  2. Population numbers have been rounded off to the nearest thousand.
  3. Poverty line is set at R350 per month in 2000 Rands, inflated using CPIX for July of each year. The real value of the per capita poverty line is R402 in 2002 and R569 in 2008.
  4. Income is calculated as total reported earnings for household members over 15 years, plus value of social grants received by household, and divided by household size.
  5. Sample surveys are always subject to error, and the proportions simply reflect the mid-point of a possible range. The confidence intervals (CIs) indicate the reliability of the estimate at the 95% level. This means that, if independent samples were repeatedly taken from the same population, we would expect the proportion to lie between upper and lower bounds of the CI 95% of the time. The wider the CI, the more uncertain the proportion. Where CIs overlap for different sub-populations or time periods we cannot be sure that there is a real difference in the proportion, even if the mid points differ. CIs are represented in the bar graphs by vertical lines at the top of each bar.
What do the numbers tell us?
One way of identifying how many children are living without enough resources to meet their needs is to use a poverty line and measure how many children live under it. As money is needed to access a range of services, income poverty is often closely related to poor health, reduced access to education, and physical environments that compromise personal safety. A lack of sufficient income can therefore compromise children’s rights to nutrition, education, and health-care services, for example.
 
International law and the South African Constitution recognise the link between income and the realisation of basic human rights, and acknowledge that children have the right to social assistance (social grants) when families cannot meet children’s basic needs. Income poverty measures are therefore important for determining how many people are in need of social assistance, and evaluating the State’s progress in realising the right to social assistance.
 
No poverty line is perfect. Using a single income measure tells us nothing about how resources are distributed between family members, or how money is spent. But this measure does give some indication of how many children are living with severely constrained resources.
 
South Africa has very high rates of child poverty. In 2010, three quarters of children lived below the upper poverty line (R1060 per person per month), 60% were below the lower poverty line (R575 per month) and 35% were below the ultra-poverty line of R290 per month. lived in households below this poverty line. Income poverty rates have fallen consistently since 2003. Significant decreases in child poverty are evident irrespective of the poverty line used, and occur across all provinces except the Northern Cape.This poverty reduction is likely to be partly the result of a massive expansion in the reach of the Child Support Grant over the same period.

There are substantial differences in poverty rates across the provinces: Using the lower poverty line, over 70% of children in Limpopo and the Eastern Cape are poor. Gauteng and the Western Cape have the lowest child poverty rates – calculated at 38% and 32% respectively.

There are glaring racial disparities in income poverty: while two thirds (67%) of African children lived in poor households in 2010 (using the 'lower' poverty line), only 4% of White children lived below this poverty line, and poverty rates for Coloured and Indian children were 31% and 14% respectively.
 
While other indicators span the period from 2002 onwards, the poverty analysis uses 2003 as its baseline. This is because the General Household Survey did not capture information on social grants in its first year, and so income from grants could not be included in household income for 2002.
 
Technical notes
The General Household Survey asks a set of questions to establish whether household members over 15 years are economically active. For those who are economically active and report their earnings, these amounts are standardised to monthly values.

For those who report earnings in income bands rather than discrete amounts, each income bracket is split into deciles for those who indicated an income in that bracket, and a uniform distribution of income is assigned within each income bracket decile, for those who indicated an income in that bracket.

For those who are economically active but did not provide a discrete income amount or indicate an income bracket (unspecified/refused), the median income for men and women in each population group is allocated. The medians are calculated separately for each year.

The method for assigning income is derived from that used by Daniele Bieber and adapted by Debbie Budlender.

Total household income from earnings iscalculated as the total earnings for all household members over 15 years. Total household income from social grants is calculated by allocating the grant amounts for that year for each type of grant reported to be received by household members. Total household income is derived by adding total income from earnings and grants.

Three poverty lines are set in 2000 Rand values. This are inflated using CPIX reported by Statistics South Africa at July each year. Per capita income is calculated by dividing total household income equally by the number of household members.

There are many limitations to working with poverty lines, and this method almost certainly results in an over-estimation of the poverty rate because both income and social grants are under-reported in the General Household Survey.

There are numerous poverty lines to choose from.1 These three poverty lines were selected because they can be linked to poverty lines commonly used by economists in South Africa2. The upper and lower lines are derived from the work of Hoogeveen & Ozler3. The $2-a-day poverty line is an international poverty line used by the World Bank, the OECD and other international groups.

Strengths and limitations of the data
The data are derived from the General Household Survey2, a multi-purpose annual survey conducted by the national statistical agency, Statistics South Africa, to collect information on a range of topics from households in the country’s nine provinces. The survey uses a sample of 30,000 households. These are drawn from Census enumeration areas using multi-stage stratified sampling and probability proportional to size principles. The resulting estimates should be representative of all households in South Africa.
 
The GHS sample consists of households and does not cover other collective institutionalised living-quarters such as boarding schools, orphanages, students’ hostels, old age homes, hospitals, prisons, military barracks and workers’ hostels. These exclusions should not have a noticeable impact on the findings in respect of children.
 
Changes in sample frame and stratification
The current master sample was used for the first time in 2004, meaning that, for longitudinal analysis, 2002 and 2003 may not be easily comparable with later years as they are based on a different sampling frame. From 2006, the sample was stratified first by province and then by district council. Prior to 2006, the sample was stratified by province and then by urban and rural area. The change in stratification could affect the interpretation of results generated by these surveys when they are compared over time.
 
Provincial boundary changes
Provincial boundary changes occurred between 2002 and 2007, and slightly affect the provincial populations. Comparisons on provincial level should therefore be treated with some caution. The sample and reporting are based on the old provincial boundaries as defined in 2001 and do not represent the new boundaries as defined in December 2005.
 
Weights
Person and household weights are provided by Statistics South Africa and are applied in Children Count – Abantwana Babalulekile analyses to give estimates at the provincial and national levels. Survey data are prone to sampling and reporting error. Some of the errors are difficult to estimate, while others can be identified. One way of checking for errors is by comparing the survey results with trusted estimates from elsewhere. Such a comparison can give an estimate of the robustness of the survey estimates. For this project, GHS data were compared with estimates from the Statistics South Africa’s mid-year estimates, as well as the Actuarial Society of South Africa’s ASSA2003 AIDS and Demographic model.
 
Analyses of the seven surveys from 2002 to 2008 suggest that over- and under-estimation may have occurred in the weighting process:
  • When comparing the weighted 2002 data with the ASSA2003 AIDS and Demographic model estimates, it seems that the number of children aged 0 – 9 years was under-estimated in the GHS, while the number of children aged 10 – 19 was over-estimated. The pattern is consistent for both sexes. The number of very young males aged 0 – 4 years appears to be under-estimated by 15%. Girls in this age group have been under-estimated by 15.8%. Males in the 10 – 14-year age group appear to be over-estimated by 5.7%.
  • Similarly in 2003, there was considerable under-estimation of the youngest age group (0 – 9 years) and over-estimation of the older age group (10 – 19 years). The pattern is consistent for both sexes. The results also show that the over-estimation of males (9%) in the 10 – 19-year age group is more than double the over-estimation for females in this age range (3.8%).
  • In the 2004 results, it seems that the number of children aged 7 – 12 years was over-estimated by 6%, as well as the number of persons aged 13 – 22 years. The number of very young children appeared to have been under-estimated. The patterns of over- and under-estimation appear to differ across population groups. For example, the number of White children appears to be over-estimated by 14%, while the number of Coloured persons within the 13 – 22-year age group appears to be 9% too low.
  • In 2005, the GHS weights seem to have produced an over-estimate of the number of males within each five-year age group. The extent of the overestimation is particularly severe for the 10 – 14-year age group. In contrast, the weights produce an under-estimate of the number of girls – the error seems greatest in respect of the younger age groups. These patterns result in male-to-female ratios of 1.06, 1.13, 1.10 and 1.09 respectively for the four age groups covering children (ie 0 – 4, 5 – 9, 10 – 14 and 15 – 19 years).
  • The 2006 weighting process yielded the same results as in 2005. The one exception is that the under-estimation of females is greatest in the 5 – 9 and 15 – 19-year age groups. This results in male-to-female ratios of 1.03, 1.10, 1.11 and 1.12 respectively for the four age groups covering children.
  • The 2007 weighting process produced an over-estimation for boys and an under-estimation for girls. The under-estimation of females is in the range of 3 – 5% while the over-estimation is in the range of 1 – 7%. This results in male-to-female ratios of 1.07, 1.06, 1.08 and 1.08 respectively for the four age groups covering children.
  • Overall, assuming the ASSA2003 Aids and Demographic model to be the ‘gold standard’, it appears that the GHS2008 over-estimates both male and female populations under the age of 19 years, except for 0 – 4- year-old females. The extent of over-estimation for boys is in the range 0 – 7%. It is particularly severe for boys aged 10 – 14 years. Over-estimation is in the range of 2 – 5% for girls aged five years and above. For girls aged 0 – 4 years, the ASSA2003 model suggests that these may have been under-estimated by about 1%. The GHS2008 suggests a sex ratio of 1.03 for children aged 0 – 4 years, which is higher than that of the ASSA model and Statistics South Africa's mid-year estimates.
The apparent discrepancies in the seven years of data may slightly affect the accuracy of the Children Count – Abantwana Babalulekile estimates. Since 2005 the male and female patterns vary in respect of a particular characteristic, which means that the total estimate for this characteristic will be somewhat slanted toward the male pattern. A similar slanting will occur where the pattern for 10 – 14-year-olds, for example, differs from that of other age groups. Furthermore, there are likely to be different patterns across population groups.
 
Disaggregation
Statistics South Africa suggests caution when attempting to interpret data generated at low level disaggregation. The population estimates are benchmarked at the national level in terms of age, sex and population group while at provincial level, benchmarking is by population group only. This could mean that estimates derived from any further disaggregation of the provincial data below the population group may not be robust enough.
 
Reporting error
Error may be present due to the methodology used, ie the questionnaire is administered to only one respondent in the household who is expected to provide information about all other members of the household. Not all respondents will have accurate information about all children in the household. In instances where the respondent did not or could not provide an answer, this was recorded as “unspecified” (no response) or “don’t know” (the respondent stated that they didn’t know the answer).
References and Related Links
1 Woolard I & Leibbrandt M (2006) Towards a poverty line for South Africa: Background note. Cape Town: Southern Africa Labour and Development Research Unit, UCT

2 see,for example, Leibbrandt M, Woolard I, Finn A & Argent J (2010). Trends in South African Income Distribution and Poverty since the Fall of Apartheid. OECD Social, Employment and Migration Working Papers, No.101. OECD Publishing.

3 Hoogeveen J & Ozler B (Eds.) (2006) Poverty and Inequality in post-Apartheid South Africa: 1995-2000. Cape Town: HSRC Press.  

4 Statistics South Africa (2003-2009). General Household Survey 2002-2008 Metadata. Cape Town, Pretoria: Statistics South Africa.

Barnes H (2009). Child poverty in South Africa: A Money Metric Approach using the Community Survey 2007. Pretoria: Department of Social Development.

Hall K & Wright G (2010) A profile of children living in South Africa in 2008. Studies in Economics and Econometrics, 34(3): 45-68.

Ravallion, M (2010). Poverty lines across the world, Policy Research Working Paper 5284. Washington: World Bank Development Research Group.

Streak J, Yu D, & van der Berg S (2009). Measuring child poverty in South Africa: Sensitivity to the choice of equivalence scale and an updated profile. Social Indicators Research, 94(2), 183-201.



 

Author: Katharine Hall

Definition
This indicator shows the number and proportion of children living in households that are income poor. Three poverty lines are used: an 'upper' poverty line, a 'lower' poverty line and an 'ultra-low' poverty line equivalent to US$2-a-day. These values are set in the year 2000 and are increased each year in line with inflation.
  • Upper poverty line: R593 per person per month in 2000 prices (R1060 in 2010)
  • Lower poverty line: R322 per person per month in 2000 prices (R575 in 2010)
  • $2-a-day poverty line: R163 per person per month in 2000 prices (R290 in 2010)
Per capita income is calculated by adding all reported income for household members over 15 years, then adding all income from social grants, and dividing the total household income by the number of household members. Income is known to be under-reported generally, and particularly in the General Household Survey. Social grants are also severely under-reported in the GHS. Child poverty is therefore likely to be over-estimated.
Commentary
One way of identifying how many children are living without enough resources to meet their needs is to use a poverty line and measure how many children live under it. As money is needed to access a range of services, income poverty is often closely related to poor health, reduced access to education, and physical environments that compromise personal safety. A lack of sufficient income can therefore compromise children’s rights to nutrition, education, and health-care services, for example.
 
International law and the South African Constitution recognise the link between income and the realisation of basic human rights, and acknowledge that children have the right to social assistance (social grants) when families cannot meet children’s basic needs. Income poverty measures are therefore important for determining how many people are in need of social assistance, and evaluating the State’s progress in realising the right to social assistance.
 
No poverty line is perfect. Using a single income measure tells us nothing about how resources are distributed between family members, or how money is spent. But this measure does give some indication of how many children are living with severely constrained resources.
 
South Africa has very high rates of child poverty. In 2010, three quarters of children lived below the upper poverty line (R1060 per person per month), 60% were below the lower poverty line (R575 per month) and 35% were below the ultra-poverty line of R290 per month. lived in households below this poverty line. Income poverty rates have fallen consistently since 2003. Significant decreases in child poverty are evident irrespective of the poverty line used, and occur across all provinces except the Northern Cape.This poverty reduction is likely to be partly the result of a massive expansion in the reach of the Child Support Grant over the same period.

There are substantial differences in poverty rates across the provinces: Using the lower poverty line, over 70% of children in Limpopo and the Eastern Cape are poor. Gauteng and the Western Cape have the lowest child poverty rates – calculated at 38% and 32% respectively.

There are glaring racial disparities in income poverty: while two thirds (67%) of African children lived in poor households in 2010 (using the 'lower' poverty line), only 4% of White children lived below this poverty line, and poverty rates for Coloured and Indian children were 31% and 14% respectively.
 
While other indicators span the period from 2002 onwards, the poverty analysis uses 2003 as its baseline. This is because the General Household Survey did not capture information on social grants in its first year, and so income from grants could not be included in household income for 2002.
 
Strengths and limitations of the data
The data are derived from the General Household Survey2, a multi-purpose annual survey conducted by the national statistical agency, Statistics South Africa, to collect information on a range of topics from households in the country’s nine provinces. The survey uses a sample of 30,000 households. These are drawn from Census enumeration areas using multi-stage stratified sampling and probability proportional to size principles. The resulting estimates should be representative of all households in South Africa.
 
The GHS sample consists of households and does not cover other collective institutionalised living-quarters such as boarding schools, orphanages, students’ hostels, old age homes, hospitals, prisons, military barracks and workers’ hostels. These exclusions should not have a noticeable impact on the findings in respect of children.
 
Changes in sample frame and stratification
The current master sample was used for the first time in 2004, meaning that, for longitudinal analysis, 2002 and 2003 may not be easily comparable with later years as they are based on a different sampling frame. From 2006, the sample was stratified first by province and then by district council. Prior to 2006, the sample was stratified by province and then by urban and rural area. The change in stratification could affect the interpretation of results generated by these surveys when they are compared over time.
 
Provincial boundary changes
Provincial boundary changes occurred between 2002 and 2007, and slightly affect the provincial populations. Comparisons on provincial level should therefore be treated with some caution. The sample and reporting are based on the old provincial boundaries as defined in 2001 and do not represent the new boundaries as defined in December 2005.
 
Weights
Person and household weights are provided by Statistics South Africa and are applied in Children Count – Abantwana Babalulekile analyses to give estimates at the provincial and national levels. Survey data are prone to sampling and reporting error. Some of the errors are difficult to estimate, while others can be identified. One way of checking for errors is by comparing the survey results with trusted estimates from elsewhere. Such a comparison can give an estimate of the robustness of the survey estimates. For this project, GHS data were compared with estimates from the Statistics South Africa’s mid-year estimates, as well as the Actuarial Society of South Africa’s ASSA2003 AIDS and Demographic model.
 
Analyses of the seven surveys from 2002 to 2008 suggest that over- and under-estimation may have occurred in the weighting process:
  • When comparing the weighted 2002 data with the ASSA2003 AIDS and Demographic model estimates, it seems that the number of children aged 0 – 9 years was under-estimated in the GHS, while the number of children aged 10 – 19 was over-estimated. The pattern is consistent for both sexes. The number of very young males aged 0 – 4 years appears to be under-estimated by 15%. Girls in this age group have been under-estimated by 15.8%. Males in the 10 – 14-year age group appear to be over-estimated by 5.7%.
  • Similarly in 2003, there was considerable under-estimation of the youngest age group (0 – 9 years) and over-estimation of the older age group (10 – 19 years). The pattern is consistent for both sexes. The results also show that the over-estimation of males (9%) in the 10 – 19-year age group is more than double the over-estimation for females in this age range (3.8%).
  • In the 2004 results, it seems that the number of children aged 7 – 12 years was over-estimated by 6%, as well as the number of persons aged 13 – 22 years. The number of very young children appeared to have been under-estimated. The patterns of over- and under-estimation appear to differ across population groups. For example, the number of White children appears to be over-estimated by 14%, while the number of Coloured persons within the 13 – 22-year age group appears to be 9% too low.
  • In 2005, the GHS weights seem to have produced an over-estimate of the number of males within each five-year age group. The extent of the overestimation is particularly severe for the 10 – 14-year age group. In contrast, the weights produce an under-estimate of the number of girls – the error seems greatest in respect of the younger age groups. These patterns result in male-to-female ratios of 1.06, 1.13, 1.10 and 1.09 respectively for the four age groups covering children (ie 0 – 4, 5 – 9, 10 – 14 and 15 – 19 years).
  • The 2006 weighting process yielded the same results as in 2005. The one exception is that the under-estimation of females is greatest in the 5 – 9 and 15 – 19-year age groups. This results in male-to-female ratios of 1.03, 1.10, 1.11 and 1.12 respectively for the four age groups covering children.
  • The 2007 weighting process produced an over-estimation for boys and an under-estimation for girls. The under-estimation of females is in the range of 3 – 5% while the over-estimation is in the range of 1 – 7%. This results in male-to-female ratios of 1.07, 1.06, 1.08 and 1.08 respectively for the four age groups covering children.
  • Overall, assuming the ASSA2003 Aids and Demographic model to be the ‘gold standard’, it appears that the GHS2008 over-estimates both male and female populations under the age of 19 years, except for 0 – 4- year-old females. The extent of over-estimation for boys is in the range 0 – 7%. It is particularly severe for boys aged 10 – 14 years. Over-estimation is in the range of 2 – 5% for girls aged five years and above. For girls aged 0 – 4 years, the ASSA2003 model suggests that these may have been under-estimated by about 1%. The GHS2008 suggests a sex ratio of 1.03 for children aged 0 – 4 years, which is higher than that of the ASSA model and Statistics South Africa's mid-year estimates.
The apparent discrepancies in the seven years of data may slightly affect the accuracy of the Children Count – Abantwana Babalulekile estimates. Since 2005 the male and female patterns vary in respect of a particular characteristic, which means that the total estimate for this characteristic will be somewhat slanted toward the male pattern. A similar slanting will occur where the pattern for 10 – 14-year-olds, for example, differs from that of other age groups. Furthermore, there are likely to be different patterns across population groups.
 
Disaggregation
Statistics South Africa suggests caution when attempting to interpret data generated at low level disaggregation. The population estimates are benchmarked at the national level in terms of age, sex and population group while at provincial level, benchmarking is by population group only. This could mean that estimates derived from any further disaggregation of the provincial data below the population group may not be robust enough.
 
Reporting error
Error may be present due to the methodology used, ie the questionnaire is administered to only one respondent in the household who is expected to provide information about all other members of the household. Not all respondents will have accurate information about all children in the household. In instances where the respondent did not or could not provide an answer, this was recorded as “unspecified” (no response) or “don’t know” (the respondent stated that they didn’t know the answer).
Technical notes
The General Household Survey asks a set of questions to establish whether household members over 15 years are economically active. For those who are economically active and report their earnings, these amounts are standardised to monthly values.

For those who report earnings in income bands rather than discrete amounts, each income bracket is split into deciles for those who indicated an income in that bracket, and a uniform distribution of income is assigned within each income bracket decile, for those who indicated an income in that bracket.

For those who are economically active but did not provide a discrete income amount or indicate an income bracket (unspecified/refused), the median income for men and women in each population group is allocated. The medians are calculated separately for each year.

The method for assigning income is derived from that used by Daniele Bieber and adapted by Debbie Budlender.

Total household income from earnings iscalculated as the total earnings for all household members over 15 years. Total household income from social grants is calculated by allocating the grant amounts for that year for each type of grant reported to be received by household members. Total household income is derived by adding total income from earnings and grants.

Three poverty lines are set in 2000 Rand values. This are inflated using CPIX reported by Statistics South Africa at July each year. Per capita income is calculated by dividing total household income equally by the number of household members.

There are many limitations to working with poverty lines, and this method almost certainly results in an over-estimation of the poverty rate because both income and social grants are under-reported in the General Household Survey.

There are numerous poverty lines to choose from.1 These three poverty lines were selected because they can be linked to poverty lines commonly used by economists in South Africa2. The upper and lower lines are derived from the work of Hoogeveen & Ozler3. The $2-a-day poverty line is an international poverty line used by the World Bank, the OECD and other international groups.

References
1 Woolard I & Leibbrandt M (2006) Towards a poverty line for South Africa: Background note. Cape Town: Southern Africa Labour and Development Research Unit, UCT

2 see,for example, Leibbrandt M, Woolard I, Finn A & Argent J (2010). Trends in South African Income Distribution and Poverty since the Fall of Apartheid. OECD Social, Employment and Migration Working Papers, No.101. OECD Publishing.

3 Hoogeveen J & Ozler B (Eds.) (2006) Poverty and Inequality in post-Apartheid South Africa: 1995-2000. Cape Town: HSRC Press.  

4 Statistics South Africa (2003-2009). General Household Survey 2002-2008 Metadata. Cape Town, Pretoria: Statistics South Africa.

Barnes H (2009). Child poverty in South Africa: A Money Metric Approach using the Community Survey 2007. Pretoria: Department of Social Development.

Hall K & Wright G (2010) A profile of children living in South Africa in 2008. Studies in Economics and Econometrics, 34(3): 45-68.

Ravallion, M (2010). Poverty lines across the world, Policy Research Working Paper 5284. Washington: World Bank Development Research Group.

Streak J, Yu D, & van der Berg S (2009). Measuring child poverty in South Africa: Sensitivity to the choice of equivalence scale and an updated profile. Social Indicators Research, 94(2), 183-201.



 

Department of International Development UK Children's Institute