Grandmothers and Granddaughters: Old-Age Pensions and Intrahousehold Allocation in South Africa2003

Back in 2003, and maybe still today, cash transfers are rare in developing economies, but are proposed to redistributive resources. How to do such a thing optimally? Conditional? Unconditional? She seems to praise PROGRESA (the Mexican program) which was horrifying but then asks two interesting questions that she aims to resolve in the paper. Firstly, do unconditional cash transfers have positive effects? Are these sensitive to the gender of the recipient? She answers them using South African pensions which were indeed unconditional. These pensions were highly unequal because of the Apartheid, a system of white supremacy and racial segregation. A commitment was done in 1989 to reach parity in the pensions between their invented "racial groups", which was allegedly reached in 1993. Benefits for Africans rose while those for whites declines. In 1992 the means test that was used for the pension was unified across "races". Payments under this regime went to woman over 60 and men over 65. The means test was not particularly rigorous, a couple's resources were divided in two and other household members' income is not taken into account.

The survey was applied to nine thousand households during the last 5 months of 1993, with height and weight of children under 7 measured for all ages in months. The author constructs height-for-age and weight-for-height z-scores by subtracting the median and dividing by SE in corresponding ge and sex group with well nourished American kids as reference. She then runs the main regression \[ w_{ijk} = \pi_f E_f + \pi_m E_m + \sum_{l=1}^{4} \gamma_l \mathbb{1}_{(l=k)} + W_{ijk}\lambda + X_{ijk}\delta + \omega_{ijk} \] where \(w_ijk\) is the weight-for-height z-score, for child \(i\), in family \(j\), and birth cohort \(k\). \(E_f\) and \(E_m\) is a dummy for if there is an eligible man or woman in the household respectively, \(\mathbb{1}_{(l=k)}\) are birth-year dummies, \(W_ijk\) is a vector that controls for the presence of near-eligible elderly (ages 50/55/60 cutoffs), and \(X_ijk\) includes family background controls (parental education, age, residence, household composition). The big assumption here is that there is no difference in nutrition between eligible an non-eligible households with an elderly member.

In terms of results, being in a household with an eligible person is associated with \(\approx 0.35\) SD higher weight-for-height z-score for girls. For boys, it is not significant. Being in a household with an eligible woman increases the effect to 0.6 SD higher weight-for-height z-score for girls. Boys insignificant but positive. She then estimates a 2SLS variant \[ w_{ihk} = \alpha_f \, \text{PENS}_f + \alpha_m \, \text{PENS}_m + \sum_{l=1}^{4} \gamma_l \mathbb{1}_{(l=k)} + W_{ihk}\lambda + X_{ihk}\delta + \omega_{ihk}\] to check that the different coefficient that comes from women and men pension receivers is not due to the fact that men claim it less than women. \(\text{ PENS}_f, \text{PENS}_m\) are dummies indicating if a woman/man in the household actually received the pension. Here the implied effect is also bigger. We see a disaggregation and see a higher coefficient when it is the mother's mother as the eligible grandma. Then, she presents a second empirical strategy, a DiD comparing old vs. young children within eligible and non eligible households. \[ \begin{split} h_{ijk} = \pi_f (\text{YOUNG} \times E_f) + \pi_m (\text{YOUNG} \times E_m) + \beta_f E_f + \beta_m E_m + \\ \sum_{l=1}^{4} \gamma_l \mathbb{1}_{(l=k)} + X_{ijk}\delta + \sum_{l=1}^{4} \mathbb{1}_{(l=k)} \times X_{ijk}\lambda_l + \varepsilon_{ijk} \end{split}\] Here the outcome is height, and the effect is 1.16 SD for girls from a woman's pension, and no effect for boys, which is similar as before. She then runs robustness checks for different stories. Firstly, it could be that eligible households differ from non-eligible ones for reasons unrelated to the pension. Secondly, families might have rearranged themselves, moving a grandmother in because of the money.

She gives two interpretations of the results. One is that resources are spent differently when in a woman's hands. This implies that if a policy maker cares about children's nutrition they ought to prioritize women as cash transfer receivers. The second is that the pension in present-value terms for men is worth less than for women, since men die sooner. It is expected then that this transitory income be saved and not spent on kids. She regresses household savings on pension income by gender, and shows that since households do not save more out of a man's pension, this story is less likely to be true.

This is a very interesting paper, and the fact that there seems to be an all-woman effect through generations is also very interesting, as well as the other highlighted result that there is almost a no-men effect also. The paper ends by asking why do grandmothers prefer girls? The data is not very strong by 2026 standards. I like that it criticizes and shows the fragility of a unitary household model.