Most economic and demographic surveys are held at the household level. Hence, we do not observe the individual behaviour of members residing in the household, nor do we observe the members of the family who do not reside in the household. In a recent interesting paper by Amar Hamoudi and Ducan Thomas – Endogenous Co-residence and Program Incidence: South Africa’s Old Age Pension(1) – the authors empirically study the behaviour of the extended family using household data. They make a case for extending the definition of the household beyond a spatially determined unit, in particular, by including non-co-residing family members.
To understand why this might be important let us first look at the specific program the authors study. The South African Old Age Pension (OAP) is a public transfer to support the income of women over 60 and men over 65 years. The eligibility is based on an age and income/asset criterion that is very generous, making the program available to the majority of South Africa’s black population. The transfer is substantial at around the median income of a black 20-50 year old’s income. So we would expect this exogenous and substantial shock in income to have some impact on those targeted.
The impact of the OAP has been studied in numerous papers, which show that the transfer has impacts not only on the pension recipient but also the co-residents. The papers also show differential impacts depending on the recipient being a man or woman, thereby rejecting the unitary household model(2). However, these papers assume that the household composition/living arrangements do not change in response to the OAP transfer. To study if this assumption is potentially misleading, the authors proceed to analyse the re-sorting of family members into households. In particular, they study how adult members are sorted to co-reside with the OAP beneficiaries as a result of the transfer.
How do the authors analyse this question? For their outcomes they use two indicators of adult human capital – education and height. The idea is that these indicators are pretty much fixed for an adult. Hence any variation in them between pension and non-pension households is due to the household composition and not as an impact of the transfer.
Though the transfer is a very generous one, the take-up is incomplete. The authors suspect that those who take-up the transfer differ from the ones who do not in terms of the outcome variables. Hence, they want to account for the household receiving the transfer relating to any unobserved characteristics of the household. The authors do this by using an instrumental variable approach. What that entails is – in the first stage they model the take-up choice by a dummy for the household having at least one person eligible for the pension. The authors argue that this is a valid instrument as they show that age-eligibility is highly correlated with take-up of the transfer, and they think that no changes would be caused to the outcome variables at this age-eligibility cut-off. In the second stage they use the predicted value of take-up to estimate the impact on the outcomes of interest.
What do they find? Adults who live with pensioners are negatively selected. Men aged 20-55 living with a pension recipient have nearly 12% (ie. 1.02 years) less education than the total average education for men in this age group. Similarly for women aged 20-55 the difference in education is of around 6%. For height they find that women living with a pensioner are shorter than those who do not co-reside with pensioners.
Why this negative selection on human capital? The authors hint towards a couple of possibilities. In a collective household model setup, the pension transfer will increase the income of the recipient giving them greater power in household decision-making. This could shift the household demand towards services preferred by the pensioners like personal care at home. Such services are intensive in domestic labour, and the most efficient way to meet them would be by adults with lower human capital to live with pensioners as their market wages are lower. The other channel could be a shift in family time-allocation. The pension transfer could finance out-migration of adults with higher human capital as they would expect higher market wages.
The authors push us to rethink the assumptions we make while evaluating the effects of a program or resource at household or individual level. Not taking into account some family level responses can be misleading and underline the importance of the role of family and networks in household decision-making.
(1) Hamoudi, Amar, and Duncan Thomas. “Endogenous Coresidence and Program Incidence: South Africa’s Old Age Pension.” Journal of Development Economics (2014).
(2) The Unitary model of the household considers the household as a single decision-making unit. On the other hand, the more general collective household models view the household as a collection of individuals making joint decisions. Collective models are covered in detail in an earlier blog – https://poresp.wordpress.com/2014/01/31/conference-poverty-and-family-some-insights-on-intrahousehold-allocations/