Who Are The Hungry Children? And How Far Can Increases In Social Grants Help Them? Evidence From The 2019 General Household Survey.

Introduction

Hunger, and particularly child hunger, have been discussed recently as a justification for extending the social grant system[1]. This brief probes the relationship between child hunger and social grants using data from the 2019 General Household Survey (GHS)[2].

As a preliminary, the pervasiveness of the social grant system should be noted. The GHS estimates that 45.5% of all households in South Africa are in receipt of at least one social grant, that the average number of grants in grant receiving households is 2.27, and that 77.8% of children[3] live in household receiving at least one social grant.

Outline of the analysis

The analysis proceeds in four stages:

1. The imputation of caregivers to children, and a division of children into those attracting a child support grant and those not doing so.
2. An analysis of income and expenditure in households and among caregivers and their spouses[4], if any.
3. Relation of hunger status among children to household resources
4. Discussion of policy options for the reduction of child hunger.

Notes on the details of method are provided in the Annexure.

All financial amounts are in 2019 prices.

1. Children, caregivers and child support grants. A household roster, with age, gender, marital status, relationship to the household head, and whether the mother and/or father is in the household forms part of the GHS data set. Caregivers of children are not identified and have to be imputed. Six possibilities are considered in descending order: the mother, the father, the grandmother, the grandfather, the oldest woman in the house, and the oldest man[5].

These are gendered assumptions, implying that nearly all caregivers are women. Thus would not only be a reflection of custom, but would also be rational from the point of view of eligibility for a child support grant, since women in a given household are likely to have lower incomes than men. The imputations are only approximate. For reasons which will become clear below, the inexactness does not matter too much.

Table 1 sets out the distribution which results.

Table 1

2. Income and expenditure of households and income of caregivers. The GHS missed a great deal of data on personal income. Only 49% of those indicated as having a job and/or a business have a determinate income or income band reported. However, 91% of households reported their monthly income and 96% reported their monthly expenditure. The correlation coefficient between household income and expenditure for households reporting both is 0.65.

One wants to know two things. The first is whether a child is eligible to attract a CSG for its caregiver and the second is whether a household is in poverty or not. Give the gaps in earnings data, the best one can assume is that all children who receive them are eligible for them. Otherwise, only when the caregiver and spouse if any, have a recorded income of below R 4 200 per month for a single or R 8 400 for a couple is there eligibility even if a grant is not claimed.

Households are classified as poor or non-poor[6] using Statistics South Africa’s upper-bound poverty line. This refers to the food poverty line plus the average amount derived from non-food items of households whose food expenditure is equal to the food poverty line. In short, households receiving this income (R 1 227 per person per month in 2019) are able to meet their basic needs and feed their children adequately[7]. Hunger in such households testifies to internal maldistribution of consumption rather than lack of the necessary resources.

Table 2 sets out the results of the classification of children according to whether they are in a poor household or not.

Table 2

The Don’t Knows are a result of the incompleteness and incoherence of the data in the GHS.

3. Hungry children and their relationship to economic and child support grant status. The GHS distinguishes five levels of hunger among children: children who are never, seldom, sometimes, often and always hungry. Hunger is reported at the household level[8], so the assumption is that, if any child goes hungry, all children in the household go hungry. This is generally plausible, though there will be exceptions[9]. Children can be divided into five categories:

1. Children in households known to be not poor
2. Children eligible to attract a child support grant but the grant is not claimed.
3. Children age 7 to 17 in households which are known to be poor or which may be poor.
4. Children age 3 to 6 in households which are known to be poor or which may be poor.
5. Children under the age of 3 in households which are known to be poor or which may be poor.

As will be argued below, the appropriate policy response for each category will be different.

Table 3 presents estimates of the number of children in each category

Table 3

Thousands of children

Comparison with the National Income Dynamics Study’s Coronavirus National Income Dynamics Study (CRAM) in October 2020

The reports on the third wave of CRAM, based on information collected in October 2020, contain a study of hunger[10]. There, it was found that the prevalence of child hunger in the week before the survey was 16%. 24% of children who were hungry in that week had been hungry every day or almost every day.

A comparison is difficult to make because the reference period in the GHS (the past year) is different from that in the CRAM survey. An approximate comparison, based on a simple model[11] suggests that the prevalence of hunger found by the CRAM survey was more than double that found by the GHS, even though conditions in the fourth quarter of 2020 were more similar to those in 2019 than in the second or third quarters of 2020. This is not the only respect in which fourth quarter CRAM findings differ from those of Statistics South Africa. The CRAM survey also found a considerably larger bounce back of employment in the third quarter than the Quarterly Labour Force Survey[12].

4. Policy. The appropriate policy responses would differ by category:

Category A (Households not in poverty): Even though these households have generally sufficient resources to feed their children adequately, some may run occasionally into transient cash flow problems. There is little that policy can be do about that. The appropriate policy response when children in this category are often or always hungry is a good scolding of the household head.

Category B (Households where children should attract CSG grants, but where these grants are not claimed.) The first response should be the claim of the entitlement. Policy cannot help when entitlements are not claimed.

Categories C, D and E (Households which are, or may be in poverty and where there are CSG grants.) New policies should be considered to assist children in these households. There are, however, different degrees of urgency between the categories. Least urgent is Category C (children between the ages of 7 and 17) since older children are better able to withstand hunger than younger children and children in this category are more likely to be in school (the GHS found that 97% of them were) and more likely to be able to access food at school (the GHS found that 82% of them could). More urgent intervention is needed for Category D children (those between the ages of 3 and 6) and the most urgent need is for Category E children (those under the age of 3). The prevailing wisdom is that inadequate nutrition in the first 1 000 days of a child’s life permanently damages health throughout the rest of it.

In an era of fiscal constraints, policy should concentrate on the most urgent problems. In this case, this means focusing on Category D and E children. What are the policy options?

Option 1 would be to increase the child support grant for children up to the age of six. Assume an increase of the grant by R 150 per month (in 2019 prices). This would cost R 9.9 billion per year. The trouble with cash grants is that they go into the general household income pot and are not necessarily used for the purpose for which they are granted. Nonetheless, there would be an impact on child hunger[13].

Option 2 would be to bias grants more heavily to children under three, giving an increase of R 200 per month to children under three and R 100 per month to children between three and six. This would cost R 9.2 billion a year. Option 2A would be to offer a food parcel costing the same amount instead. This parcel would be designed to include as many of the micronutrients and other elements of healthy nutrition as possible. Option 2A would confer more benefits on targeted children than Option 2 because the package is likely to be better nutritionally than anything most caregivers can purchase for the same coast and because the scope for diverting the benefit away from the children would be smaller.

Option 3 is the same as Option 2A, except that the parcels would be handed out a health facilities rather than social grant collection points, and they would be handed out when a child fails to thrive and a simple questionnaire indicates that low household income is likely to be the cause. Assume that all children who are always are or often hungry and half the children who are sometimes hungry fail to thrive, that the system errs on the side of generosity, so that twice as many parcels are given out than are strictly necessary, and that two thirds of relevant children would be observed at health facilities. Assume also that the parcels are 50% bigger in terms of cost. The total cost per annum would be R 1.4 billion.

Table 4 presents the estimated impact of each option.

Table 4

Conclusion

Child hunger is appalling, especially when it is frequent and affects small children often, and a decent society should do what it can to minimize it. The cost of Option 3 is modest, as is appropriate in the context of constrained fiscal circumstances. It is not provided for in the 2021/22 Budget. Perhaps it should be initiated when the billions of looted rand, which the Minister of State Enterprises has vowed to recover,[14] start rolling in.

Charles Simkins
Head of Research
charles@hsf.org.za

Annexure – Methods

Note 1

The algorithm for the imputation of the caregiver is:

i. Choose the mother if she is in the household.

ii. If she is not, choose the father if he is in the household.

iii. If neither mother nor father is in the household, choose the grandmother if one is in the household.

iv. If neither mother nor father not grandmother is in the household, choose the grandfather if one is in the household.

v. If neither parents nor grandparents are in the household, choose the oldest woman if there is one.

vi. If step v, fails choose the oldest man.

Note 2

The algorithm for determining whether a child is in a poor household or not is:

i. Start with total monthly household income as given in the data.

ii. Add up the value of private pensions, remittances, and social grants and earnings, where identified, by household, and replace total monthly household income with this value if total monthly income as given is lower.

iii. Divide total monthly income by household size to get monthly household income per capita

iv. Set POV1 (poverty as measured from the income side) at Yes if household income per capita is lower than the poverty line, No if its if equal to or higher than the poverty line and Don’t Know if household income per capita is zero or missing.

v. Start with total monthly household expenditure by expenditure band and set the expenditure at the midpoint of the band.

vi. Set POV2 (poverty as measured from the expenditure side) at Yes if household expenditure per capita is lower than the poverty line, No if its if equal to or higher than the poverty line and Don’t Know if household expenditure per capita is zero or missing.

vii. Assess the poverty of each household according to the following table:

Note 3

The model assumes that children are divided into two categories: those for whom hunger is permanent or almost permanent condition, and those who experience hunger on a more transient basis. Denote the proportion of children who are permanently hungry by a, and the transient probability of hunger on any day by p₁. Assume the daily transient probabilities are independent. Then the probability of a child being hungry in a period pf t days is:

a + (1 - a)p_t

where p_t = 1 – (1-p₁)^t

The table below shows the relevant calculations.

An assumption that a is 6.35% and p₁ is 0.00015 fits the GHS data fairly well. It predicts that the probability that a child would be hungry in the last week would be 6.45%, whereas the CRAM estimate is 16%.

Note 4

An ordered probit regression reveals that there is a statistically significant relationship with the expected sign between child hunger status and household income. The relationship is applied to post-grant household income for Options 1 and 2. Some children move out of the D and E categories altogether. Others improve their hunger status.

In the case of Option 3, the calculation is more straightforward and is based directly on the data in Table 3 and the assumptions of the option.

[1] See, for instance, Ferial Haffajee, Jobs bounce-back returns millions to work but more people hungry as grants end, Daily Maverick, 17 February 2020 and Ina Conradie ,Katharine Hall and Stephen Devereux, Transforming social protection to strengthen child nutrition security, Chapter 8 of J May, C Witten and L Lake (eds) South African Child Gauge 2020, Children’s Institute, University of Cape Town.