Who benefits from the South African Child Support Grant?: The role of gender and birthweight

ABSTRACT

Several studies have suggested that the South African Child Support Grant (CSG) reduces stunting in benefiting children. However, all of these studies have estimated the impact of the CSG on the mean of the height-for-age distribution. This paper investigates how this benefit varies across the quantiles of the height-for-age distribution. The result suggests that the positive effect at the mean is driven by children in the high quantiles and this group of children are more likely to be girls and children that did not experience low birthweight at birth. I argue that the CSG has not been able to address the malnutrition inequality that disadvantage male children and children born with low birthweight.

KEYWORDS:

• Unconditional cash transfer
• child health
• causal inference and nutrition

JEL:

• I38
• H53
• C21
• D13

Acknowledgement

The author is grateful for the comments received from the anonymous referee at Economic Research Southern Africa (ERSA) and for their useful suggestions. This paper has a working paper version that was published in ERSA working paper series.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The primary caregiver is the person that takes care of the needs of the child without payment.

2 Note that one could argue that motivation is time invariant and make an argument for a fixed effect model. First, this will put considerable difference between the methodology adopted in this study and the existing results it wishes to shed light on. Second, this will mean that the current study cannot make sense of the gender and birth weight coefficients since they are also time invariant. Lastly this will shrink the sample size significantly in latter waves of the data because year on year the number of children that benefit from the grant increases, meaning that control units become more and more scares.

3 Note that there is no need to check for balance after running the entropy weight algorithm because it is designed to automatically achieve balance.

4 Calculated as delay minus mean of delay over standard deviation of delay.

5 Note that while unobserved factors are not directly controlled for, unobserved characteristics related to delay (in receiving the CSG) may be captured by the motivation variable.

6 At the time of the NIDS 2008 survey a caregiver (who does not have to be a family member of the child) must have a monthly income below R800 in urban areas or R1 100 in rural areas (see Coetzee 2011).

7 An exception here is the income and expenditure variables. These variables were imputed by NIDS.

8 A dummy variable is used for this. The variable is equal to 1 if the child never visited the hospital in the last 12 months and 0 otherwise.

9 Food expenditure per adult equivalent scale (AES) = (Food expenditure/(Adult + 0.5 * children)^0.9).

10 An indicator for low birthweight is added in subsequent analysis to isolate its relationship with height-for-age after accounting for the effect of the treatment and other covariates. This provides an initial idea of how this variable is correlated with height-for-age.

11 The algorithm did not converge when an attempt was made to balance the skewness. This is, however, not a significant source of bias because, even without balancing the skewness, only two variables show differences in skewness (food expenditure and motivation). Since this variable will be included in the subsequent weighted regressions, this will help to mitigate the effect of problematic imbalance in skewness (if any).

12 Sometimes height-for-age result is reported in terms of the percentage of a standard deviation it represents since height-for-age is measured by z-scores see (Coetzee, 2011) for example.

13 A child is born with low birthweight if the weight of the child at birth is below 2.5 kg according to the World Health Organization’s definition.

14 These papers report a range of estimates under different tweaks to their method, the one reported here is the maximum of the reported values.

15 Note that Aguero et al. (2006) use the KwaZulu-Natal Income Dynamics Study while Coetzee (2011, 2013) use the National Income Dynamic Study.

16 The size of the estimate in DSD, SASSA and UNICEF (2012) is 19% of a standard deviation (0.19). The difference in size of the effect compared with the 27% of a standard deviation reported for girls in Table 5 can be attributed to difference in the samples and methodology used. The effect reported for boys in in DSD, SASSA and UNICEF (2012) is 0.007 (i.e. similar to the result in the current study the disparity in the estimates is large).

17 i.e. the quantiles are defined before the regression so that the quantile effects do not depend on the covariates and represent the difference between the marginal distribution of outcomes across treatment arms.

18 The treatment effect is also estimated using conditional quantile estimator. The result is presented in Figure A1 of the Appendix. It shows that the effect is not significant anywhere on the outcome distribution, although the effect is stronger at the higher quantiles. This is attributable to the difference between the conditional and unconditional quantile regression.

19 Stunting is defined as height-for-age score less than −2 standard deviations of World Health Organization’s reference standard.

20 The author is grateful to one of the reviewers for raising this point.

Additional information

Funding

The author is grateful for the comments received from the anonymous referee at Economic Research Southern Africa (ERSA) and for their useful suggestions.