Ties That Bind: The Genetic and Social Correlates Of Potential Financial Assistance∗ Simon D. Halliday Economics Master’s Dissertation December 10, 2006

Abstract This paper considers the role that relatedness, either social or genetic, plays in the possibility that individuals are identified as sources of financial assistance for households during times of emergency. It uses a sample of non-residents from the KwaZulu Natal Income Dynamics Survey and the household respondent’s reports of their willingness to help during crises. In addition to controlling for relatedness, the paper investigates the roles of other possible determinants of transfer behaviour such as gender, household pension eligibility, whether the household has experienced a negative shock, the household’s endowment of social capital, the household’s acccess to remittances and the household’s ties through the marriage practice of iLobolo. I find that genetic relatedness is not the most important factor in determining financial assistance. Conversely, there is robust evidence suggesting that non-relatives are the most likely to be identified as potential providers of financial assistance for a household during an emergency.

∗ I wish to thank Justine Burns for her dedicated supervision particulary for her time spent with me individually and for time spent reading the drafts of this paper. Additionally, I would like to thank Malcolm Keswell for his input, both theoretically and practically. I would also like to thank the members of SALDRU who gave valuable feedback on my seminar during the writing of this paper, the input of which aided dramatically in improving it. Further thanks go to Amy Miller and Seraj Haroun.

1

1

Introduction

The household has traditionally been viewed as an institution for sharing risk and for pooling income. In Economics we have used the household as a unit for basic preference aggregation. Hence, that members in families should have preferences that include benefits to their relatives seems not only a demonstrable fact but an intuitively obvious one at that. This is not necessarily the case. Although there is substantial evidence in Economics and Genetics to propose that families and family members help one another based on genetical, or familial, relations, there is evidence for alternative theories. Social network theory proposes an alternative. During times of need family members, or members of households, may be unable to help one another. This is not to say that they would not if they could, but simply that they are either incapable of doing so because of lack of resources, or they are constrained by other factors such as information asymmetries. In this case, families and individuals in households generally turn to more distant individuals - those outside of the household who could help the household in an emergency, the unrelated and possibly quite distant social relatives or friends. In this paper, I examine how relatedness, be it genetic or social, affects the likelihood of non-resident individuals with links to a household to be identified as potential financial assistants for a household in an emergency. I interrogate further whether this likelihood of being identified as a source of financial help is affected by the respondent household characteristics, the foremost of which is the eligibility of the respondent household for a state pension. I show that a genetic relatedness index does not significantly increase the likelihood of househo identifying individuals as sources of potential private transfers. Rather, non-relatives are significantly more likely to be identified as potential financial assistants than other relatives (genetical or legal). These findings are robust even when controlling for remittances, marriage assets (lobolo), social capital and the household’s experience of negative shocks. The structure of the paper is as follows. Section two discusses the literature on genetics, social networks and on the relevant household characteristics. Section three proposes the model that I use, based initially on work on altruism and adapted to my purposes. In section four I discuss the dataset and provide some summary statistics. Section five assesses the regression results from the probit regressions. I comment and critique the results in section six, weighing their relevance and the insights that can be gained from them offering some final conclusions.

2

Which Ties Bind? The Household, Genes and Social Relatedness

As the question above asks, which ties bind? The main thrust of this paper is to assess how strong the role of genetic relatedness is in the reported possibility of an individual to provide financial assistance to a household. It is furthermore 1

important not only to assess the role of genetic relatedness, but also the role of the networks in which individuals operate - their social networks and the resources to which this allows them access. Moreover, it is crucial to account for the contexts in which individuals operate - the characteristics of individuals and the characteristics of households that could drive potential non-market transfer behaviour during crises. Gender is a particularly important factor - empirically women seem to respond to intra-household dynamics differently to men. Women spend differently to men, allocating more money to food and education [Duflo and Udry, 2003, Thomas, 1991]. The saving patterns of men and women also differ women’s behaviour is stable and generally relates to benefiting the household, whereas men’s behaviour is affected by information availability and communication [Ashraf, 2004]. This is mirrored by Informal Savings Group behaviour in South Africa, which seems to be more common among females than it is among males [Irving, 2004].There are furthermore differential responses to increases in household income where women evidently increase their labour supply, men do not seem to be as affected [Bertrand et al., 2003, Posel et al., forthcoming]. There are additional results showing that women are more altruistic than men are, spending more of their own income on food and care for their children than men [Himmelweit, 1998, Thomas, 1991].1 Age is another demographic characteristic that plays a role in private transfer behaviour. Age can predict inter-generational support, or norms by which inter-generational support is accrued [La Ferrara, 2004, 2002]. In many cases a ‘Grandmother Effect’ is recorded in which grandmothers provide resources to the younger generations [Buss, 2004].2 Duflo [2001] observed this empirically for South Africa. In households where grandmothers were present the level of welfare of their granddaughters was higher. There was no similar effect present for grandfathers. In the discipline of Genetics, researchers have conversely been assessing the role of selection at the level of the gene, and reproduction of the gene being the level at which decision-making and behaviour should be analysed in animals [Hamilton, 1964a,b, Dawkins, 1989, 1982, Trivers, 1972, 1985]. Understandably, this is problematic with humans who have the cognitive capacity to overrule their genetic coding [Dawkins, 1989, Ridley, 2004]. However, some have argued that genetics and genetic relatedness play a role regardless of these limitations, precisely that genetic relatedness is instrumental in the structure of decisions that affect survivability of genes [Trivers, 1985, Buss, 2004, Bowles and Posel, 2005]. In this paper I use Hamilton’s Rule in which individuals make decisions 1 Several other models also look at intrahousehold allocations, the power dynamics thereof and how these are located in the household literature see [Browning et al., 1993, Bourguignon and Chiappori, 1992, Browning et al., 1994] 2 Grandmothers are guaranteed that their grandchildren are their genetic offspring - they know that their daughters carry their genes and hence they are guaranteed to have at least some of their genes carried by their grandchildren. The story is entirely different for grandfathers who are unable to state with certainty whether their wife’s daughter is their own offspring, taking this down one generation further exacerbated the effects. Hence, there is no real ’grandfather effect’ of which to speak.

2

based upon their genetic relatedness to other individuals to predict the likelihood of an individual being identified as a source of financial help - the intricacies of this are explored in Section 3. Reconciling the work in Genetics with that in Economics can be difficult. Economics has historically aggregated preferences to the level of the household, whereas Genetics is interested in a level of disaggregation (the gene) that is entirely outside of the economic realm. Is it therefore the gene that predetermines behaviour in economic decision-making, or is behavior separate to genetic programming? Most agree that there is an intersection of genetics with behaviour and that most of the evidence points towards an understanding of behaviour that is moderated by both genetics and by behavioural cues that impact on phenotypic development [Buss, 2004, Ridley, 2004, Pinker, 1998, 2002].3 Therefore I argue that there is some interaction of the two drivers, behavioural and genetic. Notwithstanding this interaction, researchers in Economics have observed how genetic structures affect economic decision-making and intra-household dynamics. There is evidence that step-parents are less likely than genetic parents to provide inputs into education, either in the form of monetary resources or in terms of allocation of time to step-children [Anderson et al., 1999]. Step-parents have also been shown to be more likely to be violent or homicidal towards nongenetic relatives [Daly and Wilson, 1982, 1983, Wilson and Daly, 1992]. Household spending is also different in households where a child has step, foster or adoptive parents - the evidence suggests that less is spent on food. Specifically in South Africa, when the household head is the mother of the children, or when the mother of the children in the household is the spouse of the household head, more is spent on milk, vegetables and fruit than on tobacco or alcohol [Case et al., 1999]. More recent evidence suggests that genetic relatedness can positively predict the incidence of remittances - households with members that are related by some higher value in the index receive remittances more regularly.Alternatively, individuals who are related to the household head are more likely to send remittances to the household. Moreover, genetic relatedness positively predicts the size of these remittances - the closer the genetic tie, the larger the remittance [Posel, 2001, Bowles and Posel, 2005]. The choice to make a private transfer is not necessarily only a genetically driven one. In fact it is quite possible that there may be numerous social factors that will impact on the likelihood to be identified by a household as a potential financial assistant.4 I assess the impact that social networks may have on potential financial assistance. A social network is a network that is constituted of a set of people, a group of contacts or interactions between these [Scott, 2000, 3 The vociferousness with which different authors pursue this point though is where much of the debate occurs see for example the debate following Wilson [1975]’s Sociobiology and Pinker [2002]’s The Blank Slate. 4 In fact it could be a strategic decision as La Ferrara [2002, 2004] argues in her papers on inter-vivos transfers and on kin-based models of reciprocity. Equally there could be an impact of preference aggregation and strategic behaviour therein asBecker [1974], argues in his ‘Rotten Kid Theorem’. Alternatively see Becker [1993].

3

Wasserman and Faust, 1994, Newman, 2003, Newman and Park, 2004]. Within social networks, patterns of (inter)marriage, friendship and business have been studied [Newman, 2003]. A seminal contribution to this literature on social networks is that of Granovetter [1973, 1983] the main theory of which he called ‘The Strength of Weak Ties’. His argument was that individuals in the same social cluster would have access to similar resources. Consequently if an individual in the social network required a resource to which individuals in that cluster did not have immediate access then this individual would have to look elsewhere. Particularly, individuals who are in a social network and have strong ties are separated by a short geodesic distance, individuals who are more distant in the social network and who may have access to other resources are generally separated by a greater geodesic distance.5 Granovetter argues that it is these individuals, from whom individuals are separated by a greater geodesic distance, that can provide access to resources. The applications of these patterns of access to resources has been applied to labour markets and job-seeking behaviour [Granovetter, 1983, Gans, 1974]. It was found that more distant social ties are those that are able to supply individuals with job opportunities, rather than immediate or ‘strong’ ties [Ericksen and Yancey, 1980, Granovetter, 1973, 1983].6 Similar conditions could arguably apply to access to financial assistance in emergencies. When a specific household is affected by an emergency, then it follows that it would not help individuals in that household to try to use one another for insurance. They may instead look towards more distant social relatives or friends in order to obtain help. In the framework that I present, individuals are either genetically related (parents, children), legally related (spouses, in-laws) or unrelated (friends, co-workers). I consider the first two groups as strong ties and the latter group as a weak tie. Following Granovetter’s reasoning it is this group of individuals that could be important in understanding social insurance dynamics in emergencies. Granovetter [1973, 1983] was furthermore concerned with the possibility of there being differential strength of ties over the income distribution, particularly different responses depending on the class of the households and individuals involved. Although I cannot strictly control for class, I proxy for this by expenditure per capita in the household. There are several other factors that affect non-market transfer behaviour. Households that have access to social security or government grants often end up having altruistic behaviour towards them, or private transfers that may have been made to them, displaced by these grants - the presence of grants or old age pensions crowded out private transfers [Cox, 1987, Cox and Jimenez, 1992, 1995, Goldstein et al., 2002, Jensen, 2003]. Cox and Jimenez [1992, 1995] furthermore found that education (at the personal level) and negative household shocks (at the household level) had impacts - both of these characteristics positively and 5 Geodesic distance is a measurement of the distance across a network, generally the number of nodes that separates one node from another, or more practically the number of separations between individuals. Practically, we could measure it by physical distance, genetic relatedness or by reported strength of ties if that is recorded in a survey. 6 I provide an exact definition of the strength of ties in Section 3.

4

significantly drove transfer behaviour. Work on the phenomena of remittances and the old age pension in South Africa has been prolific. The recent work by Bowles and Posel [2005] furthered earlier work by Posel [2001] on remittances that showed links between genetic relatedness and remitting behaviour. The recent work on this phenomenon in South Africa by Jensen [2003] was precisely on the crowding out impacts of the pension. This advanced similar research by Cox and Jimenez [1995] that indicated a negative relationship between the incentive to transfer and household pension eligibility. The negative labour market supply [Bertrand et al., 2003] and the out-migration for work-search purposes [Posel et al., forthcoming] as responses to pension availability have also been explored. Generally, both household remittances and household pension eligibility have particularly strong effects.7 In terms of other social effects, the role of social capital has been explored in depth by many researchers. Putnam [1995, 1998] as a general exploration of social capital showed the necessity and power of social capital, which has been mirrored by other work in the social sciences. Narayan and Pritchett [1999] observed a positive correlation between household income and its social capital endowment. Grootaert [1999] noticed that better household accumulations of social capital reduced the prospect of poverty. In South Africa, Maluccio et al. [2002] found that social capital increased welfare. Although I cannot construct an index such as that used by Maluccio et al. [2002], I can facilitate the basic participation of household members in groups such as Informal Savings Groups (stokvels) and Burial Societies, both of which have been labeled as strong indicators of social capital [Maluccio et al., 2002, Irving, 2004]. The social network effects of marriage is also an important factor. Himmelweit [1998] highlights the role that the bride price has to play in Indian households, facilitating close ties and acting as a social contract. This was reinforced by Townsend [1994] who commented on the risk-pooling evidence of bride prices. Similar cultural practices appear in South Africa, namely the practice of Lobolo. It has been noted that the use of the nomenclature ’Bride Price’ is generally insufficient to describe the cultural saliency of the practice, so I refer to Lobolo as a collective term for any practice which results in the transfer of goods from one family to another as a result of marriage [Comaroff, 1980, Dlamini, 1983, Bennett, 2004]. Marriage as a determinant of social networks has also been explored and been shown to have positive impacts on the strength of social networks [Newman, 2003, Padgett and Ansell, 1993].8 In the current investigation, the emphasis is on the possibility of transferral rather than the amounts or the incidence thereof. Hence, my analysis is different to that of Bowles and Posel [2005] and Jensen [2003]. The dependent variable is explained in section 4 - it is a yes/no answer to the question: “Would [X] help the household in a financial emergency?” The dynamics of this question are explored later. 7 A comprehensive review of the role of the Old Age pensions can be found in the study by Case and Deaton [1998]. 8 Where strength in this paper is a measure of the strength of ties under the framework from Granovetter [1973].

5

There are several reasons to instigae this investigation as a departure from research into actual transfer behaviour. Firstly, the question under interrogation is centred around the incidence of an emergency. Remittances however are regular transfers made from individuals employed outside of the home, they do not tell us how those who send remittances would act when the household was in an emergency, or even if they would be at liberty to respond.9 Furthermore, individuals in a household may feel constrained by social norms not to return constantly to one source of remittance income and hence when a crisis occurs they may feel it appropriate to seek other sources of income. This is where the approach that I take is substantively different to that taken by researchers looking at the incidence and size of actual private transfers rather than the possibility of transfers during crisis situations.

3

Modelling Private Transfer Behaviour

Models of altruism assume that there is interdependence of utility amongst individuals such that one individual’s level of utility is contingent both on their own utility and on the utility of others. Instrumentally, this comes down to the income level of the individual and the income level of the relative or friend [Becker, 1974]. The model that is used follows closely to that of Jensen [2003].10 It should be noted further that it is not assumed that individuals in the model must be parents and children as is conventionally the case. Instead it is assumed that they are related, socially or genetically, and hence individuals are indexed i=1,2, which is generalisable to the i th case. Ui (Ci , C− i) = Ui [Vi (Ci ), V−i (C−i )] U−i (Ci , C− i) = U−i [Vi (Ci ), V−i (C−i )]

(1)

In the above Ui and U−i are utilities, Ci and C−i are the consumption of the individuals within their personal utility sets Vi and V−i . A weighting factor α is included in the structure of the model. For αi this weighting factor represents the importance of the utility of individual -i to individual i. Similarly for α−i . The weighting factors and corresponding utilities are depicted below for two individuals: 1 and 2. U1 (C1 , C2 ) = (1 − α1 )V1 (C1 ) + (α1 )V2 (C2 )) U2 (C1 , C2 ) = (1 − α2 )V2 (C2 ) + (α2 )V2 (C1 ))

(2) (3)

Where α has the constraint: α, 0 ≤ α ≤ 1. It is furthermore necessary to consider the budget constraint of individuals, in addition to including income 9 They need to know whether a household is in an emergency, without this knowledge they mayn’t even know to respond - this could be more a problem of information asymmetry because of lack of communication than because of inability to provide help. 10 There is a plethora of other literature on the subject of modelling altruism see for example, Stark and Lucas [1988], Stark [1989], Stark [1991], Stark [1993], Stark [1995], Burnstein et al. [1994].

6

transfers of one individual to the other. C1 = Y1 − T1 , C2 = Y2 + T1

(4)

Here Yi is income for individual i and Ti is the transfer by individual i. Hence in the above, Individual 1 transfers to Individual 2. From (2) - (4) above one obtains a first order condition maximum.

V10 (C1 ) =



α 1−α



V20 (C2 )

(5)

Equilibrium occurs where the marginal utilities of the individuals are equal. If there are disparities between these, then private transfers are the means by which individuals can redistribute wealth. Genetic relatedness is measured using an index. This follows Hamilton’s method of an index that decreases by a proportion of one half for each degree of separation. See Table 1 in the index for a summary of this. Hamilton used this index to assess altruistic behaviour and established Hamilton’s Rule. The rule states the conditions under which an agent will behave altruistically towards another agent. This behaviour is inherently dependent on the costs and benefits involved in the action. Definition 1 (Hamilton’s Rule) bih ∗ rih > cih

(6)

Individual i in household h will observe their benefits, b, from acting altruistically towards another individual to whom they are related by the genetic relatedness measure, r, the enaction of which would cost them c. 11 Although I am not strictly investigating altruistic behaviour,12 I consider that it is plausible for private transfers by one individual to another to be altruistic. The only complication in this instance is that the information with which we are provided by the dataset only gives the relationship of the nonresident with links to the household to the household head. I do not have access to information about other individuals in the household who may be relatives to the potential assistant and thus may be the individuals to whom a financial assistant may wish to transfer. The weighting factor is reconsidered below. It is now constructed as a function of r the genetic relatedness index. This function should be positive and increasing in r. In which case, α(r) will take higher values as r increases. Hence, the weighting factor will be higher for those to whom i is related relative to those to whom i is less related or unrelated entirely. Hamilton’s Rule can be reconsidered in terms of the benefit alone: bih = rcih ih considered in the literature review, making such a transfer could be a strategic act, rather than an altruistic one. Furthermore, it could be reciprocally altruistic act rather than a purely altruistic one. 11 Equivalently, 12 As

7

V10 (C1 )

 =

α(r) 1 − α(r)



V20 (C2 )

(7)

Now, the ratio of the weighting factors shows the relevance of individual 2’s utility in its contribution to the utility of individual 1. It is important to note however that, at its base, this formulation is different to that presented by Jensen [2003]. In the formulation considered in this paper, the weighting factor cannot differ for two related individuals - they will be related by the factor r and hence their values of α(r) should be equivalent. Demonstrably, one individual cannot be related to another individual by a different value of the index than that by which the other individual is related to them.13 To confirm the above, it is necessary to show what would happen to the above if r increased. Hence, one takes the partial derivative of V1 (C1 ) with respect to r to show that the change in r results in a strictly positive response. δV10 (C1 ) = δr



α0 (r) (1 − α(r))2



V20 (C2 )

(8)

By assumption α(r) is positive and increasing in r, in which case α0 (r) should be positive. Hence, both the denominator and the numerator of the above are positive. In which case, an increase in genetic relatedness would increase V1 (C1 ) through the α(r) function. This is backed by Hamilton’s assessment that individuals weight the survivability of closer genetic relatives more than they do those of less proximal or unrelated agents. On account of the fact that I use Granovetter’s theory of the strength of weak ties, it is necessary to have a precise definition of what this means. It is this definition that I use in asserting that there will be differential responses between different types of ties. This proposition is set up as an alternative to the above in which genetics drives the decision to make private transfers [Granovetter, 1973]. I utilise his argument on the extrapolation of dyadic relationships to larger social structures.14 Definition 2 (Strength of Ties) There are two individuals A and B and a set of individuals [C, D, E...,] ∈ S with ties to either or both of A and B. For the extrapolation of these dyadic ties to larger structures: 1. The stronger the tie between A and B the greater the number of individuals in the set S will know both A and B. 2. Conversely, the weaker the tie between A and B the fewer the number of individuals in the set S will know both A and B. 13 I cannot test the reversal of this - the data that I have is one-way and hence cannot verify whether the values of α(r) would be the same in reality. 14 Where a dyad is a relationship between two individuals or two nodes in a sociogram or map of a social network.

8

3. If ties are absent then individuals in the set S should know only one of A or B. It is this definition that I use when discussing the strength or weakness of ties later in the paper.15

4

Empirics

I use the 1998 Kwa-Zulu Natal Income Dynamics Survey (KIDS) data set. This is a cross-sectional dataset based on surveys carried out in 1998 for which household in the Kwa-Zulua Natal province of South Africa were interviewed. The KIDS dataset has the conventional household roster providing data with which to assess the impact of genetic relatedness to the household head. Moreover, additional information on pensions, social capital, household shocks (negative or positive) and marriage assets is present in the dataset allowing for an in depth analysis of the possible determinants of private transfers if a household were in an emergency. The sample used in this analysis is of those individuals that the respondent reported as having a link to the household. These individuals are non-resident individuals with links to the household that send or receive remittances, lend or borrow money, lend or borrow housing, land, or livestock, or individuals that would provide the household with financial help. To be exact, the sample is comprised of those individuals for whom the respondent reported any of the following roles. Are there any individuals who? 1. Send remittances to this household or receive remittances from this household? 2. Lend money to this household or borrow money from this household? 3. Lend a house, land or livestock to this household or borrow any of those things from the household? 4. If you had some financial trouble, such as funeral expenses, who would you turn to for assistance besides those mentioned? A link is defined as existing if there are extant relationships as defined by categories (1) through (4) above. It should also be clarified that categories (1) through (3) define pre-existing links that the household has, whereas category (4) asks the respondent to hypothesise about individuals who may help the household in emergencies, rather than those who may already have done so. Hence. I am working with two sets of information. The first comprises data 15 A more convivial explanation is given by Granovetter in which the strength of ties is a linear combination of: time, emotional intensity, intimacy and reciprocal services [Granovetter, 1973].

9

of households that identify these links (Table 2) and the second is of those individuals who are identified as having links to households (Tables 3 and 4). Table 2 provides the means of the respondent household characteristics of the households who identified individuals with links. Column 1 presents these means for all households in the sample defined above. I then differentiate between those households with more than the mean number of links (column 2) and those with fewer than the mean number of links (column 3). 16 Those households with more than the mean number of links have the means of their household characteristics tested to indicate whether they are significantly different to the means of these statistics relative to the households with fewer than the mean number of links.17 Of those identified as having links to the household, households identify that genetic relatives comprise 80%, legal relatives 14% and non-relatives 6% of those with links. A large proportion, 73%, of the households are pension eligible. Households with more than the mean number of links identify a greater percentage of genetic relatives (84% against 78%) and fewer non-relatives (3% against 7%) as having links to the household. Households with more than the mean number of links are more likely to be pension eligible (84% relative to 68%), less likely to have experienced a negative shock18 (59% relative to 74%),19 and less likely to have a Lobolo agreement (42% relative to 57%). Each of these indicate that there are differences in the structures of the households with fewer than and more than the mean number of links. The dependent variable that I use in the probit regressions is from the question [Would [X] help the household in a financial emergency? (Yes=1, No=0)], where [X] is the non-resident individual with links to the household. I note that the question above asks the respondent to hypothesise about whether the individuals identified as having links with the household would indeed aid or not during times of crisis.20 16 Households could list several individuals linked to the household, hence it is important to control for this to infer whether there are systematic differences between households that have more than and fewer than the mean number of linked individuals and whether this affects the outcomes of the question later. 17 The mean number of links was 4.4 hence the sample is split into those with fewer than 4.4 (4 or fewer) and those with more than 4.4 links (5 or greater). 18 Where a negative shock includes: losing/reduced remittance income, a household member losing a job, the household losing access to a grant (such as the Old Age Pension), a death/serious injury to household member, or family which prevents normal activities, abandonment or divorce, crop failure/diseased or death of livestock, theft, fire, or destruction of household property. Moreover, because these are objective events we are not concerned that there are factors endogenous to these shocks. 19 It may be argued that there could be reporting bias to this, that those household with more than the mean number of links are less likely to be affected by a negatve shock. However the question in the survey did not ask about effects, it simply asked whether any negative shocks (as in the categories defined) had occurred. 20 It is feasible that those who provide the household with aid (remittance income, loans, etc) are those who the respondent lists as ’Yes’ responses, whereas those for whom the household provides aid (remittances, loans, etc) are those who the respondent lists as ’No’ responses. Nevertheless, I believe that the fourth category of the links should add sufficient variation for this question to be interesting and for it to provide us with additional and illuminating information. Moreover, we control for remittances and the results remain robust.

10

In Table 3, I provide the mean characteristics for those individuals identified as having links to the household. The first category is defined for all of those individuals with links to the household.21 I separate those with links into two groups, those with links whom the respondent said would provide financial assistance (column 2) and those for whom the respondent said would not provide financial assistance (column 3). The means of these groups are tested to confirm whether they are equal or not using Mann-Whitney tests. The last column (4) provides the means for these individual characteristics for those individuals who are not in the sample as defined by the links above. This is meant to serve as a basis for comparison of these linked individuals relative to the rest of the KIDS 98 dataset. All of the statistics are for individuals above working age.22 The sample of those with links is comprised of more females (56%) than it is males (44%). Approximately 38% are in regular employment with 25% being unemployed. 23 It is important to note however that contrasting this with the rest of the dataset (column 4) we see that approximately 22% of the rest of KIDS is in regular employment, with 26% unemployed. The rest of KIDS is involved more in formal education (20%) than those with links (13%). To dig deeper into those individuals who have links, I separate them into those who are listed as potential financial assistants (column 2) and those who would probably not provide assistance (column 3). There are substantial differences here. Those who are identified as potential assistants are older on average (41 years of age rather than 32). Similarly there are differences in the amount of education of the two groups, we observe that those identified as potential assistants have 7.17 years of education on average, relative to the 5.95 years of potential non-assistants. There is a larger proportion of males in the financial assistants than in those who were listed as probably not being financial assistants (the former having 47% males, the latter 42%). Those listed as financial assistants are more likely to have regular employment (59%) relative to those who aren’t financial assistants (23%). This impacts on the unemployment rates of the different groups where we see that those who are listed as financial assistants have a 10% rate of unemployment relative to the 36% who were not listed as assistants. Only 3% of the financial assistants are involved in formal education, relative to 20% of those who aren’t listed as assisting.24 There are significant differences in terms of the structure of the relatedness of the groups in addition to the differences identified above. Those identified as potential assistants are more likely to be non-relatives (11% vs. 2%) and less likely to be genetic relatives (75% vs. 83%). 21 i.e.

for whom the respondent said fitted into categories (1) - (4) in the question above. conventionally as 16 years and greater. 23 Unemployment is defined in KIDS98 as not having regular, casual or self- employment, but having looked for work in the past week. 24 This gives credence to the argument proposed earlier which said that those who were not listed as financial assistants are those for whom the household provides aid - students often require loans from their family in order to complete their studies. When studying they are also less able to help during financial emergencies. This could be the reason for the disparity in education levels - if so many non-assistants are still obtaining an education it would imply that on average they should have less now but may end up having a similar level later. 22 Defined

11

In Table 4, the sample of individuals with links (i.e. the same sample in column 1 of table 3) is split up into two general categories: all of those with links (columns 1 to 3) and all of those with links listed as potential financial assistants (columns 4 to 6). These individuals are then further split into their relationship categories: genetic relatives, legal relatives or non-relatives. This is their relationship relative to the household head as listed in KIDS. I present the mean individual characteristics for each of these linked groups. Dealing with all the linked individuals first, the statistics display that there are substantial differences between the relatedness groups for the linked individuals, specifically with respect to the age, education and gender compositions of the groups. Genetic relatives are the youngest on average at a mean of 34 years, they have 6.5 years of education and are comprised of 47% males. Contrast this with the legal relatives who are 41.7 years of age on average, have 5.7 years of education and comprise only 30% males. Lastly, non-relatives on average are the oldest at 42.4 years, have the highest average education at 7.67 years and have a fairly low proportion of males at 36%. In terms of employment, there are further important differences. Non-relatives are far more likely to be employed, over half (53%) have regular employment, with an additional 12% either casually employed or self-employed. Commensurate with this is the lowest proportion of unemployed individuals at 11%. Genetic relatives and legal relatives are generally quite similar on this front, with the only significant and pertinent difference being the proportion of genetic relatives involved in formal education(16%) relative to the legal relatives (1%). There is also a relatively greater ratio of housewives in the legal relative group relative to the genetic relative group.25 Contrasting columns 1 to 3 with columns 4 to 6, we can see that there are substantial differences between the group in terms of those who would potentially provide help relative to the entire group of linked individuals. On average, for each of the relatedness categories those who are listed as potential financial assistants are older. For genetic relatives there is a 5 year difference, for legal relatives a 4 year difference and for non-relatives a difference of approximately 1.5 years. Moreover, there are differences in the levels of education, with genetically and legally related potential financial assistants having approximately a year more of education on average. There are furthermore differences in the percentage of the individuals who are employed, with far more of the potential financial assistants being employed than the linked individuals alone. (61%, 51% and 58% against 37%, 37% and 53%). There is also therefore a concomitant drop in the percentage of unemployed individuals in the potential financial assistant groups relative to the group of individuals with links, with a drop in over 10% for each of the genetically and legally related groups. 25 This

is unsurprising as it is highly unlikely that any of the households have a genetic relative as a spouse.

12

5

Regression Results

The results that follow are all based on a probit model run using the question [Would [X] help the household in a financial emergency? (Yes=1, No=0)] as the dependent variable. Hence, of the sample of individuals with links, those who the respondent reported as financial assistants would have a ’Yes’ [1] response, whereas those whom the respondent reports as not being financial assistants would have a ’No’ [0] response. The results are presented in tables 5 through 7 in the appendix. All regressions control for clustering at the level of the enumeration area. All individuals are of working age and above.

5.1

Result 1: The Index of Genetic Relatedness does not positively predict potential private transfers from individuals identified as having links to the household

Genetic relatedness is constructed as an index as described earlier and attributable to Hamilton [1964a,b]. A priori I expect the index to have a positive and significant sign as argued in Section 3. Referring to Table 5, when included as the sole determinant of financial assistant behaviour the genetic relatedness index is negative, but not significant (column 1). Upon inclusion of further controls the coefficient on the index becomes significant and remains negative (column 2). Additionally, there is a non-linear effect when genetic relatedness is interacted with gender - the coefficient on the interaction term of gender and genetic relatedness is negative and significant which overrides the genetic relatedness index which has a positive and significant coefficient (column 3). These results mean two things, firstly that closer genetic relatives are less likely to be identified as providers of financial assistance to the household (columns 1 and 2) and, secondly, that male, genetic relatives are less likely than female genetic relatives to be identified as potential financial assistants. This is in accordance with Posel [2001] in which women were more likely than men to provide remittances to the household. Assessing the gender aspects in both the genetically related model and the relationship dummy model (result 2), I show that genetically related male relatives are less likely to be financial assistants (regresions 2 and 3). Placing this in the context of the second model with relatedness dummies, we see that male legal relatives26 and male non-relatives are more likely than the base female genetic relative group to provide financial assistance. These results are indicated by the interaction terms included, initially in Table 5 for the genetic index and gender (column 3) and again for the legal relative and non-relative dummies with gender (column 5). The importance of gender plays into the inclusion of Lobolo as investigated later. Nevertheless, the result that the genetic relatedness index was negative is in direct contrast with the result from Bowles and Posel [2005] in which ge26 Most

likely husbands and fathers-in-law.

13

netic relatedness positively and significantly determined the incidence and size of remittances. This result is crucially important as it is the basis for investigating alternative hypotheses in order to understand how households search for resources outside of the family in times of need.27

5.2

Result 2: Non-relatives are more likely to be listed as potential financial assistants

Considering Granvetter’s theory of the strength of weak ties, dummies for the different relatedness groups were included in the regressions in columns 4 and 5. Non-relatives are significantly more likely than genetic relatives and legal relatives to make a private transfer in a financial emergency. This result is presented in the regressions from Table 5 through to Table 7. The coefficient on the non-relative dummy is consistently positive and significant, whereas that for the legal relatives is consistently negative and significant.28 Moreover, performing Wald Tests to assess whether the coefficients of the legal relative is equal to the coefficient of the non-relative variable displays that the coefficients are consistently and significantly unequal. Hence, to rank the groups from most likely to least likely to be identified as potential sources of financial help by relatedness elicits the following ranking: 1. Non-relatives 2. Genetic relatives 3. Legal relatives In column 5 we observe that there is a positive effect from the interactions between the male gender dummy and the legal and non-relative dummies. Legal relatives are less likely to be identified as assistants, moreover this is particularly the case for female legal relatives who are less likely than males to be identified as potential financial assistants. In terms of the non-relatives, we see that male non-relatives are more likely than female non-relatives to be identified as potential sources of financial assistance. To tackle Granovetter’s concerns that individuals higher in the income distribution would find it less necessary to use weak ties for resources, I show that this is a needless concern by the use of regressions by expenditure quintile.29 If richer households, correlated with those which spend more per capita, find 27 It is worth noting that I ran regressions with the remittance dummy as the dependent variable and the genetic relatedness index as an explanatory variable. In those regressions, controlling for demographic and household characteristics, the genetic relatedness index was significant and positive. However, this was simply in order to see whether the Bowles and Posel [2005] result was consistent in this datatset, which it seemingly was. However, the question we address is substantially different to theirs so there is no specific reason to believe that the results should be consistent. 28 The base group is formed of those relatives who are genetically related to the household head. The legal relatives dummy incorporates the spouse and spousal relatives (in-laws). 29 I use Deaton [1997] as the basis for the use of expenditures over income.

14

it less opportune to use weak ties then we would predict that the non-relative dummy would not be significant for these groups. However, for those at the top of the income distribution we find that weak ties are robustly positive - the dummy for non-relatives is significant and positive in a series of regressions run to test this. These results are presented in Table 6. Table 6 also offers insight into what happens at the bottom of the distribution. It seems as though genetic relatives are important, or at least as important as non-relatives because we cannot assert that the coefficient on the non-relative dummy is different to that of the base genetically related individuals. However, this does not dismiss the claim in the pooled regressions that non-relatives are identified as significant potential providers of aid. The coefficients on the legal relative dummies were still significantly negative at the bottom of the distribution, indicating that the legal relatives still do not seem to be as likely as either genetic or non-relatives to provide assistance.30 It is important to note that the Wald χ2 statistics all indicate that there remain significant differences between the non-relative and the legal relative responses from the bottom to the top of the expenditure distribution.

5.3

Result 3: Pension eligibility crowds out the likelihood that individuals with links will be identified as a source of financial assistance during crises

Social grants provided by government have been found to have several impacts on household dynamics, the nuances of which were explored in Section 2. Households that obtain a pension have been shown to be less likeley to receive private transfers [Jensen, 2003]. Consequently, I include a dummy variable for household pension eligibility to control for this phenomenon and to investigate whether the effect is present in the dataset. Household pension eligibility is signalled by the presence of an individual of pensionable age in the household. 31 These results are reported in column 11 of Table 7. The crowding out of potential private transfers operates in an intuitive manner. Individuals respond to the utility levels of others, thus if their household income increases through access to a pension then their consumption should increase thus increasing their utility. Additionally, the converse could be valid, if a household has a positive income shock (in the form of access to pension income) then they could now have incentives to make transfers out of their household. The results from the data display a significant and negative relationship32 30 One problem here could be that those with jobs and thus those who spend more, have better resources in terms of their social networks and thus the non-relatives to which they have access. Hence the social networks into which they can tap are different to those that individuals without jobs and who are unable to spend as much have access to. 31 The dummy takes a value of 1 if the household has a male of age 65 or greater, or a female of age 60 or greater, or both present in the household. 32 Here I make the assumption that individuals are not irrationally prejudiced against old individuals - which could hypothetically be what we measure here by the definition of the pension eligibility. However, this seems farfetched and hence is not considered as a driving factor.

15

between household pension eligibility and potential financial assistance. This would indicate that there is indeed a possible crowding out effect that occurs as a result of access to social security, or public transfers. The intricacies of interpreting this are discussed in Section 6. Furthermore, testing for the pension eligibility of the individual with links to the household indicated a negative and significant effect. This contradicts the possibility proposed above in which individuals or households with increased income should have a greater incentive to provide financial assistance to other individuals or households. Notice that the results on the positive nature of non-relative linked individuals and the negative results from legal relatives remain robust even with the inclusion of this variable for pension eligibility. Thus individuals with links are less likely to be identified as sources of financial assistance if respondent households receive pensions.

5.4

Result 4: The Probability that an individual is identified as a financial assistant is positively and significantly increased if the household has experienced a negative shock

If a household has experienced a negative shock and we do not measure for this, then this experience could be what would drive the provision of private transfers to these households. Negatives shocks include a household member losing a job, the household losing remittance income, the household losing access to some resource to which their members had access previously33 and several other effects all given earlier. A priori a household that has experienced this kind of shock has greater need, with this greater need comes a concomitant imperative by others connected to the household to insure the household during this period of need. Table 7 presents regressions that include a dummy variable for negative household shocks in column 12. The coefficient on the dummy is consistently and significantly positive. This therefore means that those households that have experienced a negative shock are more likely to identify individuals with links as sources of potential private transfers, most likely as a result of one of the following reasons. Either they can list individuals who would help them as a consequence of historic bias - they have been helped historically (when they experienced the shock) and hence they are able to list individuals who would help them.34 Or alternatively, the households are more likely to be helped as a direct result of having experienced a negative shock - a negative shock causes decreased income and therefore requires an increased incidence of private transfers. Notwithstanding the above result, the inclusion of this variable does not change the result that non-relatives are more likely tbe identified as potential 33 Such

as the Child Support Grant or the statutory Old Age Pension. hypothesis is not supported by the summary statistics presented in Table 2 - households with fewer links had more negative household shocks. 34 This

16

financial assistants. The result for non-relatives was thus not masking an underlying negative shock effect.

5.5

Result 5: Social Capital Endowments decrease the probability that a household will identify a linked individual as a source of financial help

As mentioned previously, the pooling of risk in informal savings groups (stokvels) or through membership in burial societies can provide a household with social capital into which it can tap during situations of emergency. These results are reported in column 14. Household memberships in these groups was recorded in KIDS 98. I consequently include these memberships as a characteristic of the household. They are both dummy variables that take values of 1 if any household member has a membership in one of these groups. Thus it is assumed that the household at large garners the benefit of an individual in that household being a member in a group. This is consistent with the literature on the subject [Maluccio et al., 2002, Grootaert, 1999]. The coefficient on the stokvel membership dummy is consistently negative and significant. The burial society memberships on the other hand are significant and negative in one formulation of the model, but are later insignificant one additional controls are included. Consequent to this, it would seem as though household social capital would have a negative impact on the likelihood of that household identifying a person with links as a source of potential non-market transfers. The most likely reason for this is that the household believes that it can self-insure because of the membership in the stokvel - for example the individuals in the stokvel may be able to negotiate their rotation of money earlier. This does not require the household to receive an actual private transfer from others, but in reality there is pooling of risk occurring.

5.6

Result 6: Remittances significantly and positively predict the likelihood that linked individuals are identified as a source of financial assistance

Remittances are the mechanism by which individuals make non-market transfers to households. The main result from Bowles and Posel [2005] was that genetic relatedness positively predicted the likelihood of remittances, in addition to positively predicting the size of remittances. Hence it is relevant to my considerations here. To control for the presence of remittances in the household, I include a dummy variable for the household having received a remittance in monthly or yearly installments, for either cash or in-kind transfers. This dummy is included in the regressions presented in Table 7 to observe whether it is the driving force behind our positive non-relative result.

17

The coefficient on the remittance dummy variable is persistently significant and positive. This indicates therefore that households that receive remittances are more likely to identify individuals with links as potential sources of financial assistance in an emergency. The inclusion of this remittance variable does not affect the robust and positive effects that we observe occurring with non-relatives, nor the robust negative effects of the legal relative dummy. Additional regressions were run to test for non-linear interactions between remittances, gender and the relationship dummies. They were not significant and are unreported.

5.7

Result 7: Lobolo significantly and positively predicts a link being identified as a potential source of financial assistance

The phenomenon of Lobolo is an example of a bride price used as a risk pooling device for future reciprocal benefit. The bride price was explored as a social practice in India by Himmelweit [1998] and as a correlate of village insurance by Townsend [1994]. I consider its role as a similar practice in South Africa in my analysis of private transfers. The inclusion of Lobolo, separate to the inclusion of the legal relative dummy is important as it may explain effects that were not captured by the inclusions of the legal relative dummy. Lobolo can be thought of as a social contract between two families, and thus as a risk-pooling phenomenon. Lobolo is a household level characteristic and is included as a dummy variable indicating whether a household has a Lobolo agreement with another household. In the regressions, household lobolo agreements are significantly and positively related to the likelihood to individuals with links being identified as potential sources of inter-household transfers. The intuition would be that households that have Lobolo agreements have individuals, or other households, to whom they can turn in times of financial emergency. This result is important in the light of the legal relative dummy being consistently negative and significant, from which we concluded that legal relatives were less likely than the other relationship groups to help the household in an emergency. The Lobolo variable thus indicates that the social contract that results from marriage could decrease this negative legal relative effect. Additional regressions were run to test for possible interactions between Lobolo, gender and the relationship dummies. They were not significant and are unreported. The inclusion of the Lobolo variable did not reduce the probability of nonrelatives being the most likely to be listed as sources of financial assistance.

6

Binding Ties: A Discussion

Genetic relatedness did not prove to be a positive predictor of potential private transfer behaviour. This is contrary to what Bowles and Posel [2005] found, as

18

well as contradicting much of the literature on altruistic behaviour.35 However, I reiterate that my inquisition is different to those of actual remittances and moreover the context of the emergency may alter the outcomes relative to actual remittances. Nevertheless, that genetics was a negative predictor for households to identify potential private transfers indicates various possibilities. If there is a negative correlation between genetic proximity and households identifying individuals for possible private transfer behaviour, then genetics is not the sole driver, nor the most important factor in the decision by households to identify individuals as potential financial assistant. What makes the initial negative result all the more interesting is the replication (although in a simplistic manner) of the Bowles and Posel result with this dataset - the genetic relatedness index positively predicted remittance sending behaviour.36 Notwithstanding the above considerations, it is worthwhile noting that my construction of the genetic relatedness index does not capture any relationship other than that to the household head. Moreover, we cannot tell whether those individuals who are listed as potential financial assistants in the dataset are not in any way related genetically to other inhabitants in the household. I do not believe that it is possible to capture all the nuances of Hamilton’s index with this, or most other South African datasets. I conclude that this genetic relatedness index, of genetic relatedness to the household head, in this dataset did not prove to be a robustly positive predictor of potential private transfer behaviour.37 These considerations are even more important in the context of the extant literature on intra- and inter-household dynamics. If we could address the interests of those Individuals who are identified as having links to the household, these individuals have incentives to make transfers to particular individuals in a household. The respondent may believe that other individuals would transfer to the household but they do not on list them as assistants because of their knowledge of that individuals beliefs about the intra-household dynamics, such as whether the money would go to its intended recipient. All of these intuitions could mean that the index of genetic relatedness may not give an accurate representation of underlying genetic forces. Consequent to the contradictory results for the genetic relatedness index, it became important to test another hypothesis to understand how households might insure during periods of emergency. Doing so, I tested Granovetter’s theory of the strength of weak ties [Granovetter, 1973, 1983, 1985]. The evidence presented suggests that the theory of the strength of weak ties is a robust tool 35 For

example Jensen [2003], Posel [2001], Lucas and Stark [1985], Stark [1989], Stark and Lucas [1988]. 36 It should be noted that the non-relative dummy, used as an explanatory variable in this simplified version of the Bowles and Posel model was positive and significant in several construction of that model, but not to as great an extent as it is in this model. While a genetic relatedness index proves to be robustly positive in the prediction of remittances, it is not so for potential financial assistance behaviour in emergencies. 37 Were it possible to identify individuals and to whom those individuals are related then we may be able to obtain a better understanding of the intricate workings of private transfer behaviour and the interactions that genetics may have with such behaviour.

19

for interpreting individual transfer behaviour in social contexts. Specifically contexts in which the behaviour under investigation applies to theory on pooled risk, insurance-seeking behaviour and access to out of household resources. The problem with this is that I could not adhere to Granovetter [1973]’s strict definitions in which individuals are asked whether someone is close or not, nor am I able to proxy for this by distance or frequency of interaction. In this case, I know only if the individual is family or not - genetic and legal relatives were strong ties whereas non-relatives were weak. I do not know in actuality whether the respondent would have considered the non-relatives weak ties and the families strong - these need to be reported by the individuals in order for such phenomena to be depicted accurately. Although I could not strictly control for class, I performed regressions by expenditure as the closest proxy. The evidence lends credence to the concern that households lower in the distribution may identify potential donors differently to those higher in the expenditure distribution. However, the reverse reasoning - that those in higher classes may not need weak ties for help because of having high incomes and having enough close ties with resources was unwarranted. Those at the top end of the expenditure distribution identified non-relatives as sources of aid during crises.What we do know is that the non-relative group had a positive and significant coefficeint across most of the expenditure quintiles and in the pooled regressions. There were furthermore salient interactions between relatedness and gender. More proximal genetically related males were less likely to be identified as sources of financial assistance and male legal and non-relatives were more likely to be identified as potential providers of assistance. There are several possible reasons for this. For males related to, but not resident in a household it is plausible that they could be related down the genetic line and hence would not wish to transfer backwards along that line to the household head.38 In terms of the legal and non-relative males who are more likely to assist, my interpretation of this is as follows: for the male legal relatives the result would be driven by nonresident spouses and fathers-in-law who wish to insure that their wives/children remain safe and secure. For non-relatives this most likely has to do with either the head-ship of other households or the power dynamics of those households.39 The evidence also suggests that households that are eligibile for pensions, or are involved in burial societies or stokvels are less likely to identify individuals with links to the household as sources of financial assistance during emergencies. Social security, in the form of pensions, proved to follow trends in the international literature. The evidence suggests that informal security measures, such as potential financial assistance, are crowded out by eligibility for a state 38 Genetically

the reasoning for this would be as follows: those older than you cannot continue to procreate and thus will not pass on your (selfish) genes. It is instead better for your genes for you to be altruistic down the gene line [Dawkins, 1989, Buss, 2004, Ridley, 2004]. 39 The other possibility is that there are other factors endogenous to this: males may be more likely to be employed, or they may have better access to resources, or something equivalent to this which will make them capable of helping where females that the household knows may not be as capable.

20

pension. This result is intriguing because it is the consequence of self-reporting by individual respondents in households that are pension eligible. 40 What this means is that individuals in households that are pension eligible, when listing those individuals with links to their household are less likely to identify linked individuals as a source of financial help. Those households with pension eligible individuals therefore either believe that they do not require the help,41 or in terms of the crowding out hypothesis, they still require the help but their access to pensions really does drive down the likelihood that other individuals will help them and the respondent is simply reporting this recorded fact. Each of these constitute important interpretations of the result. The first is important because it means that households with pensions could encourage the crowding out behaviour of others - they believe that they are more capable and hence may act in such a way that means that they would discourage financial assistance from other individuals. The second means that there is most likely underreporting - regardless of whether they believe that they are capable of dealing with the an emergency or not, the household should still be able to list individuals that would aid it.42 Households that receive regular remittances, have experienced a negative shock or have a lobolo agreement with another household are more likely to identify individuals with links as sources of potential financial assistance. The positive significance of the remittance and the lobolo agreement variables suggests that there are long term relationships that exist between the household and either other individuals or other households. This also suggests that households bonded by social contracts such as marriage, and the traditions inherent in that bond such as lobolo, may facilitate pooled insurance behaviour such as is evidenced by Himmelweit [1998] and Townsend [1994]. Notwithstanding the robustness of the results, there is a possibility that the cognitive limitations of respondents in the survey could bias the sample - the sample could be a non-random selection of individuals linked to the household rather than a random sample of individuals linked to the household. If the respondent is asked to provide details on individuals with links to the household then it is possible that there may be a cognitive association with those who would only help the household - they reference these individuals as the most recent connections. The respondents are specifically primed to discuss individuals with existent links and those who would help the household - the sample therefore may not be a random sample of individuals linked to the household, in fact it is distinctly non-random. It could be that the sample is separated into two disparate groups: those who the household helps and those who have helped or would help the household. Feasibly we could have Yes responses for those who 40 Remember we have the information from the respondent not from the actual individual with links. 41 On account of receiving pensions - although this should not impact on whether someone would help them if they needed it. 42 There was a secondary result in which individual pension eligibility of the linked individual to the household had a negative and significant effect on the likelihood to transfer - this could be as a result of similar reason to that produced above. The household respondent does not believe that households with pensions should or need to receive private transfers.

21

have links with the household and who have provided help for the household and No responses fo those who have links with the household and for whom the household has provided help. This means that we would have a high collection of Yes and No answers without variation in between of those who may have done either depending on the context. Depending on the weighting of the on group against the other - the sample could bias the result upwards or downwards. The questions that we use may not be specific enough for this kind of in depth inquisition into the role of relationships. We need far more specific questions on individuals, their characteristics, the specific individuals to whom they would (do) transfer money and their relationships to these people specifically. Most household surveys do not and cannot go into this much depth on one topic. Therefore, there are several significant openings for original research to gather data in this area - moreover such specific information would also provide data on the social nature of such interactions and hence also help validate or invalidate theories of social interactions and networking such as the stength of weak ties. Ultimately, the evidence presented displays that there are several important factors that may affect the probability of an individual with links to a household providing financial assistance during crises. That non-relatives were shown to be the most likely to provide such potential assistance is indicative of social dynamics that warrant further investigation with nuanced and detailed surveys for future research into the question of household insurance during crises.

22

Relationship Parent/Child/ Full Sibling Grandchildren/ Grandparents Niece/Nephew/ Aunt/Uncle Great-grandchild/ Great-grandparent/ Children of (Niece/Nephew/Aunt/Uncle)

Coefficient of Relatedness 0.5 0.25 0.25 0.125

Table 1: Genetic Relatedness of Diploid Animals

23

Household Characteristic

All Households

% Gene Relatives % Legal Relatives % NonRelatives % Households Pension Eligible % Households Experienced a Negative Shock % Households with Stokvel Membership % Households with Burial Society Membership % Households with Lobolo Agreement % Households Receive Remittances Number of Household Links n

(1) 0.80 (0.40) 0.14 (0.35) 0.06 (0.23) 0.73 (0.44) 0.69 (0.46)

Households with more than the mean number of links (2) 0.84 (0.37)** 0.13 (0.34) 0.03 (0.18)*** 0.84 (0.37)*** 0.59 (0.49)***

Households with fewer than the mean number of links (3) 0.78 (0.41) 0.15 (0.36) 0.07 (0.25) 0.68 (0.47) 0.74 (0.44)

0.21 (0.41) 0.69 (0.46)

0.21 (0.40) 0.69 (0.46)

0.21 (0.41) 0.70 (0.46)

0.52 (0.50) 0.47 (0.50) 4.38 (3.84) 1479

0.42 (0.49)*** 0.46 (0.50) 8.30 (4.27)*** 497

0.57 (0.49) 0.48 (0.50) 2.39 (1.11) 982

*, **, ***, Mann-Whitney Test that the mean for Households with greater than the mean number of links = mean for Households with fewer than the mean number of links.

Table 2: Household Characteristics

24

Individual Characteristic

% Male Age Education % In Regular Employment % In Casual Employment % SelfEmployed % Housewife % Unemployed % In Formal Education % Retired % Engaged in Other Activities % Genetically Related to Household Head % Legal Relatives of Household Head % Unrelated to Household Head n

All those with Links

(1) 0.44 (0.50) 35.59 (14.31)††† 6.45 (4.63)††† 0.38 (0.49)††† 0.02 (0.15) 0.03 (0.18) 0.07 (0.25) 0.25 (0.43) 0.13 (0.34)††† 0.08 (0.27) 0.01 (0.10) 0.80 (0.40)†† 0.14 (0.35)†† 0.06 (0.23)††† 1479

Those with Links who were listed by respondent as potentially giving financial assistance (2) 0.47 (0.50)** 40.84 (14.46)*** 7.17 (5.29)*** 0.59 (0.49)*** 0.02 (0.15) 0.06 (0.24)*** 0.07 (0.26) 0.10 (0.30)*** 0.03 (0.17)*** 0.12 (0.32)*** 0.00 (0.07) 0.75 (0.43)*** 0.14 (0.35) 0.11 (0.31)*** 606

Those with Links who were listed by respondent as not potentially giving financial assistance (3) 0.42 (0.49) 31.94 (13.01) 5.95 (4.05) 0.23 (0.42) 0.03 (0.16) 0.01 (0.12) 0.06 (0.24) 0.36 (0.48) 0.20 (0.40) 0.06 (0.23) 0.01 (0.11) 0.83 (0.37) 0.14 (0.35) 0.02 (0.15) 783

†, ††, †††Mann-Whitney Test that mean for those with links=mean for those without links 10%, 5% and 1% levels of significance respectively. *, **, *** Mann-Whitney Test that mean for those with links listed as financial assistants=mean for those without links not listed as financial assistants 10%, 5% and 1% levels of significance respectively.

Table 3: Individual Characteristics by Link Category

25

Those without Links

(4) 0.45 (0.50) 35.51 (16.45) 6.08 (4.38) 0.22 (0.42) 0.05 (0.22) 0.03 (0.17) 0.07 (0.25) 0.26 (0.44) 0.20 (0.40) 0.09 (0.29) 0.01 (0.11) 0.83 (0.38) 0.16 (0.37) 0.01 (0.08) 5406

Individual Characteristic % Male Age (in years) Education (in years) % Regularly Employed % Casually Employed % SelfEmployed % Housewife % Unemployed % In Formal Education % Retired % Other Activity n

All Linked Individuals Genetic Legal NonRelatives Relatives Relatives (1) (2) (3) 0.47 0.30 0.36 (0.50)††† (0.46) (0.48)‡ 34.01 41.74 42.43 (14.04)††† (13.74) (13.35)‡‡‡ 6.50 5.70 7.67 (4.55)††† (4.44)** (5.84) 0.37 0.37 0.53 (0.48) (0.48)** (0.50)‡‡‡ 0.03 0.02 0.02 (0.16) (0.14) (0.15) 0.03 0.04 0.10 (0.16) (0.19)** (0.30)‡‡‡ 0.05 0.14 0.07 (0.23)††† (0.34) (0.26) 0.27 0.23 0.11 (0.44) (0.42)** (0.31)‡‡‡ 0.16 0.01 0.04 (0.36)††† (0.12) (0.19)‡‡‡ 0.07 0.13 0.10 (0.26)††† (0.34) (0.30) 0.01 0.01 0.01 (0.10) (0.07) (0.11) 1184 212 83

Potential Financial Assistants Only Genetic Legal NonRelatives Relatives Relatives (4) (5) (6) 0.48 0.47 0.41 (0.50) (0.50) (0.50) 39.40 46.06 44.06 (14.06)††† (16.24) (12.63)‡‡‡ 7.22 6.35 7.88 (5.17)† (5.25) (6.08) 0.61 0.51 0.58 (0.49) (0.50) (0.50) 0.02 0.02 0.02 (0.15) (0.15) (0.13) 0.05 0.05 0.13 (0.22) (0.21)* (0.33)‡‡ 0.06 0.11 0.9 (0.24)† (0.32) (0.29) 0.11 0.08 0.06 (0.31) (0.28) (0.24) 0.04 0.00 0.00 (0.19)† (0.00 (0.00) 0.10 0.22 0.11 (0.30)††† (0.42)* (0.32) 0.01 0.00 0.02 (0.07) (0.00) (0.13) 456 86 64

†, ††, †††Mann-Whitney Test that Genetic Relative Mean=Legal Relative Mean at 10%, 5% and 1% levels of significance respectively. ‡, ‡‡, ‡‡‡Mann-Whitney Test that Genetic Relative Mean=Non-Relative Mean at 10%, 5% and 1% levels of significance respectively. *, **, *** Mann-Whitney Test that Legal Relative Mean=Non-Relative Mean at 10%, 5% and 1% levels of significance respectively.

Table 4: Individual Characteristics by Relatedness Category

26

Genetic Relatedness Index Legal Relative Dummy Non-Relative Dummy Male Dummy Education (in years) Age (in years) AgeSquared Male* Genetic Relatedness Male*In-Law Dummy Male*NonRelative Constant Observations Pseudo-R2 Log PseudoLikelihood Wald χ2

[Potential Financial Assistance Dummy] (1) -0.064 (1.19)

Dependent Variable [Potential [Potential Potential Financial Financial Financial Assistance Assistance Assistance Dummy] Dummy] Dummy] (2) (3) (4) -0.129 0.213 (2.23)** (2.33)** -0.196 (5.02)*** 0.725 (5.98)*** 0.107 0.343 0.101 (7.30)*** (11.84)*** (6.10)*** 0.071 0.069 0.069 (31.32)*** (29.81)*** (25.06)*** 0.099 0.102 0.097 (18.22)*** (19.16)*** (19.23)*** -0.001 -0.001 -0.001 (10.06)*** (10.82)*** (10.49)*** -0.726 (7.77)***

[Potential Financial Assistance Dummy] (5)

-0.418 (8.34)*** 0.555 (5.45)*** -0.006 (0.31) 0.067 (24.62)*** 0.099 (19.43)*** -0.001 (10.84)***

-0.209 (11.04)*** 1521 0.0001 -1029.1559

-3.201 (35.73)*** 1479 0.1311 -869.75156

-3.338 (37.37)*** 1479 0.1344 -866.44552

-3.199 (34.06)*** 1479 0.1440 -856.7946

0.80 (13.94)*** 0.513 (5.16)*** -3.162 (33.03)*** 1479 0.1511 -849.67334

-

-

-

76.27 0.0000

140.50 0.0000

*, **, *** Indicate significance at the 10%, 5% and 1% levels of significance respectively. Heteroscedasticity robust standard errors are reported in brackets. Regressions clustered at the enumeration area level.

Table 5: Baseline Regressions for the Genetic Relatedness Index and Relatedness Dummies

27

[Potential Financial Assistance Dummy]

Legal Relative Dummy Non Relative Dummy Male Dummy Education (in years) Age (in years) AgeSquared Pension Eligibilitya Constant Observations Pseudo-R2 Log PseudoLikelihood Wald χ2

020% (6) -0.708 (6.49)*** 0.934 (1.44) 0.161 (4.51)*** 0.050 (4.34)*** 0.114 (12.84)*** -0.001 (8.82)*** -0.354 (5.57)*** -3.645 (25.05)*** 377 0.1896 -164.06672 6.02 0.0141

Dependent Variable [Potential [Potential Potential Financial Financial Financial Assistance Assistance Assistance Dummy] Dummy] Dummy] Expenditure Quintiles 21 41 61 40% 60% 80% (7) (8) (9) -0.414 -0.348 -0.069 (4.98)*** (4.68)*** (3.91)*** N/A 0.257 1.074 (5.04)*** (74.12)*** 0.221 0.233 -0.153 (6.40)*** (5.36)*** (5.53)*** 0.061 0.016 0.035 (22.37)*** (4.64)*** (27.97)*** 0.099 0.059 0.130 (11.69)*** (5.75)*** (43.18)*** -0.001 -0.000 -0.001 (11.56)*** (2.50)** (33.72)*** 0.033 -0.623 -0.077 (0.65) (11.73)*** (3.82)*** -3.167 -1.327 -3.320 (17.07)*** (8.20)*** (40.64)*** 264 247 256 0.0924 0.1006 0.1365 -152.5131 -153.9024 -152.78807 N/A -

46.98 0.0000

39726.79 0.0000

[Potential Financial Assistance Dummy] 81 100% (10) 0.154 (0.95) 0.569 (11.85)*** 0.111 (1.50) 0.086 (22.93)*** 0.114 (8.56)*** -0.001 (5.71)*** 0.018 (0.20) -3.648 (13.03)*** 329 0.1740 -187.30246 4.73 0.0296

*, **, *** Indicate significance at the 10%, 5% and 1% levels of significance respectively. Heteroscedasticity robust standard errors are reported in brackets. Regressions clustered at the enumeration area level. a Indicates a household level dummy variable.

Table 6: Relatedness Dummies and Pension Eligibility Regressions by Expenditure Quintile

28

Legal Relative Dummy Non-Relative Dummy Male Dummy Education (in years) Age (in years) AgeSquared Pension Eligibilitya Male*Legal Relative Male*NonRelative Negative Shocka Lobolo Agreementa Stokvel Membershipa Burial Society Membershipa Receives Remittancea Number of Links Constant Observations Pseudo-R2 Log PseudoLikelihood Wald χ2

[Potential Financial Assistance Dummy] (11) -0.455 (8.71)*** 0.537 (5.44)*** -0.013 (0.64) 0.065 (23.74)*** 0.099 (20.06)*** -0.001 (10.93)*** -0.253 (8.72)*** 0.781 (13.27)*** 0.454 (4.68)*** -2.964 (29.71)*** 1479 0.1559 -844.91657

[Potential Financial Assistance Dummy] (12) -0.446 (8.99)*** 0.493 (4.91)*** -0.008 (0.40) 0.065 (23.99)*** 0.097 (21.65)*** -0.001 (11.54)*** -0.294 (9.27)*** 0.791 (13.96)*** 0.471 (4.77)*** 0.298 (14.15)*** -3.136 (33.22)*** 1479 0.1631 -837.69371

169.36 0.0000

142.60 0.0000

Dependent Variable [Potential Potential Financial Financial Assistance Assistance Dummy] Dummy] (13) (14) -0.454 -0.453 (9.34)*** (9.38)*** 0.487 0.505 (4.74)*** (4.64)*** -0.010 -0.009 (0.51) (0.50) 0.065 0.065 (23.09)*** (21.83)*** 0.098 0.098 (21.63)*** (22.29)*** -0.001 -0.001 (11.65)*** (12.04)*** -0.261 -0.266 (7.40)*** (7.30)*** 0.762 0.763 (12.62)*** (12.68)*** 0.437 0.416 (4.51)*** (4.22)*** 0.296 0.297 (14.17)*** (14.50)*** 0.133 0.139 (5.65)*** (6.02)*** -0.093 (3.63)*** -0.054 (2.57)** -3.240 -3.178 (31.65)*** (29.71)*** 1479 1479 0.1647 0.1653 -836.06654 -835.46815 134.06 0.0000

123.06 0.0000

[Potential Financial Assistance Dummy] (15) -0.410 (9.02)*** 0.439 (3.72)*** -0.038 (2.15)** 0.069 (22.15)*** 0.091 (20.32)*** -0.001 (10.40)*** -0.333 (8.13)*** 0.717 (13.37)*** 0.394 (3.87)*** 0.247 (12.61)*** 0.150 (6.08)*** -0.070 (2.92)*** 0.006 (0.29) 0.583 (31.06)*** -3.339 (30.90)*** 1479 0.1966 -804.15232

[Potential Financial Assistance Dummy] (16) -0.475 (10.90)*** 0.413 (3.74)*** -0.042 (2.42)** 0.060 (21.02)*** 0.093 (22.19)*** -0.001 (11.97)*** -0.179 (4.97)*** 0.811 (14.25)*** 0.391 (3.85)*** 0.103 (5.73)*** 0.077 (3.49)*** -0.050 (2.14)** 0.002 (0.08) 0.561 (32.02)*** -0.131 (17.24)*** -2.723 (22.83)*** 1479 0.2399 -760.83308

76.80 0.0000

97.26 0.0000

*, **, *** Indicate significance at the 10%, 5% and 1% levels of significance respectively. Heteroscedasticity robust standard errors are reported in brackets. Regressions clustered at the enumeration area level. a Indicates a household level dummy variable.

Table 7: Regressions including Household Characteristics that may affect Potential Financial Assistance 29

References Kermyt G. Anderson, Hillard Kaplan, David Lam, and Jane Lancaster. Paternal care by genetic fathers and stepfathers II: Reports by Xhosa high school students. Evolution and Human Behaviour, 20:433–451, 1999. Nava Ashraf. Spousal control and intrahousehold decision making: An experimental study in the Phillippines. Harvard University, 2004. Gary S. Becker. A theory of social interactions. Journal of Political Economy, 82:1063–1094, 1974. Gary S. Becker. A Treatise on the Family. Harvard University Press, 2nd edition edition, 1993. Thomas Bennett. Customary Law in South Africa. Juta and Company Ltd., Cape Town, 2004. Marianne Bertrand, Sendhil Mullainathan, and Douglas Miller. Public policy and extended families: Evidence from pensions in South Africa. World Bank Research Observer, 17(1):27–50, 2003. Fran¸cois Bourguignon and Pierre-Andr´e Chiappori. Collective models of household behaviour: An introduction. European Economic Review, 36:355–364, 1992. Sam Bowles and Dorrit Posel. Genetic relatedness predicts South African migrant workers’ remittances to their families. Nature, 434, 2005. Marianne Browning, Fran¸cois Bourguignon, Pierre-Andr´e Chiappori, and Valerie Lech`ene. Intra Household Allocations of Consumption: A Model and ´ some Evidence from French data. Annale d’Economie et de Statistique, 26: 137–156, 1993. Marianne Browning, Fran¸cois Bourguignon, Pierre-Andr´e Chiappori, and Valerie Lech`ene. Income and outcomes: A structural model of intrahousehold allocation. Journal of Political Economy, 102(6):1067–1096, 1994. Eugene Burnstein, Christian Crandall, and Shinobu Kitayama. Some neodarwinian decision rules for altruism: weighing cures for inclusive fitness as a function of the biological importance of the decision. Journal of Personality and Social Psychology, 67:773–789, 1994. David Buss. Evolutionary Psychology: The New Science of the Mind. Pearson Education Inc., New York, 2nd edition edition, 2004. Anne Case and Angus Deaton. Large cash transfers to the elderly in South Africa. The Economic Journal, 108(450):1330–1361, 1998. Anne Case, I-Fen Lin, and Sara McLanahan. How hungry is the selfish gene? NBER Working Paper, 1999. W7401. 30

James L. Comaroff, editor. The Meaning of Marriage Payments. Academic Press, London, 1980. Donald Cox. Motives for private transfers. Journal of Political Economy, 95: 508–546, 1987. Donald Cox and Emmanuel Jimenez. Social Security and private transfers in developing countries: the case of Peru. World Bank Economic Review, 6(1): 155–169, 1992. Donald Cox and Emmanuel Jimenez. Public Spending and the Poor: Theory and Evidence, chapter Private transfers and the effectiveness of public income redistribution in the Phillipines. Johns Hopkins University Press, 1995. Martin Daly and Margo Wilson. Homicide. Aldine, Boston, 1982. Martin Daly and Margo Wilson. Sex, Evolution and Behaviour. Willard Grant, Boston, 2nd edition edition, 1983. Richard Dawkins. The Extended Phenotype. W.H Freeman, Oxford, 1982. Richard Dawkins. The Selfish Gene. Oxford University Press, 2nd edition, 1989. Angus Deaton. The Analysis of Household Surveys. A Microeconomic Approach to Development Policy. Johns Hopkins University Press, 1997. Charles R.M. Dlamini. A Juridical Analysis and Critical Evaluation of Ilobolo in a Changing Zulu society. Master’s thesis, University of Zululand, Empangeni, 1983. Esther Duflo. Grandmothers and granddaughters: Old age pensions and intrahousehold allcation in South Africa. World Bank Economic Review, 2001. Esther Duflo and Chris Udry. Intra-household resource allocations in Cˆote dIvoire: Social norms, separate accounts and consumption choices. Economic Growth Center, 2003. Eugene Ericksen and William Yancey. Class, sector and income determination. Temple University, 1980. Herbert J. Gans. Gans on Granovetter’s ”the strength of weak ties”. The American Journal of Sociology, 80(2):524–527, 1974. Markus Goldstein, Alain de Janvry, and Elisabeth Sadoulet. Is a friend in need a friend indeed? Inclusion and exclusion in mutual insurance networks in Southern Ghana. WIDER, 2002. Mark S. Granovetter. The strength of weak ties. American Journal of Sociology, 78(6):1360–1380, 1973. Mark S. Granovetter. The strength of weak ties revisited. Sociological Theory, 1:201–233, 1983. 31

Mark S. Granovetter. Economic action and social structure: The problem of embeddedness. American Journal of Sociology, 91(3):481–510, 1985. Christiaan Grootaert. Social capital, household welfare and poverty in Indonesia. Local Level Institutions Working Paper No. 6, 1999. William D. Hamilton. The genetical evolution of social behaviour I. Journal of Theoretical Biology, 7:1–52, 1964a. William D. Hamilton. The genetical evolution of social behaviour II. Journal of Theoretical Biology, 7:1–52, 1964b. Susan Himmelweit. Understanding Economic Behaviour: Households, Firms and Markets, chapter Decision Making in Households. Open University Press, Milton Keynes, 1998. Margaret Irving. Informal savings groups. Centre for Social Science Research Working Paper, 2004. Robert T. Jensen. Do private transfers ’displace’ the benefits of public transfers: Evidence from South Africa. Journal of Public Economics, 88:89–112, 2003. Eliana La Ferrara. Kin groups and reciprocity: A model of credit transactions in Ghana. Bocconi University and IGIER, 2002. Eliana La Ferrara. Descent rules and inter-vivos transfers. Bocconi University and IGIER, 2004. Robert E. Lucas and Oded Stark. Motivations to remit: Evidence from Botswana. Journal of Political Economy, 93(5):901–918, 1985. John Maluccio, Lawrence Haddad, and Julian May. Social capital and household welfare in South Africa, 1993-98. Journal of Development Studies, 34(4):54– 81, 2002. Deepa Narayan and Lant Pritchett. Cents and sociability: Household income and social capital in rural Tanzania. Economic Development and Cultural Change, 47(4):871–897, 1999. Mark E. J. Newman. The structure and function of complex networks. SIAM Review, 45:167–256, 2003. Mark E.J. Newman and Juyong Park. Why social networks are different to other types of networks. Santa F`e Institute, 2004. John F. Padgett and Christopher K. Ansell. Robust action and the rise of theMedici. American Journal of Sociology, 98:1259–1319, 1993. Steven Pinker. The Blank Slate. The Penguin Press, 2002. Steven Pinker. How The Mind Works. The Penguin Press, 1998.

32

Dorrit Posel. How do households work? Migration, the household and remittance behaviour in South Africa. Social Dynamics, 27(1):165–189, 2001. Dorrit Posel, James Fairburn, and Frances Lund. Labour migration and household: A reconsideration of the effects of the social pension of labour supply in South Africa. Economic Modelling, forthcoming. Robert Putnam. Bowling alone: America’s declining social capital. Journal of Democracy, 6(1):65–78, 1995. Robert Putnam. Bowling Alone: The Collapse and Revival of American Community. Simon and Schuster, New York, 1998. Matthew Ridley. Nature Via Nurture. Harper Perennial, London, 2004. John Scott. Social Network Analysis: A Handbook. Sage Publications, London, 2000. Oded Stark. Altruism and the quality of life. American Economic Review, 79 (2):86–90, 1989. Oded Stark. The Migration of Labour. Basil Blackwell, Massachusetts, 1991. Oded Stark. Non-market transfers and altruism. European Economic Review, 37(7):1413–1424, 1993. Oded Stark. Altruism and Beyond: An Economic Analysis of Transfers and Exchanges within Families and Groups. Cambridge University Press, Cambridge, 1995. Oded Stark and Robert E. Lucas. Migration, remittances and the family. Economic Development and Cultural Change, 36(3):465–481, 1988. Duncan Thomas. Intra-household resources: an inferential approach. Journal of Human Resources, 25(4):635–664, 1991. Robert M. Townsend. Risk and insurance in village India. Econometrica, 62(3): 539–591, 1994. Robert Trivers. Sexual Selection and the Descent of Man 1871-1971, chapter Parental Investment and Sexual Selection, pages 136–179. Aldine, Chicago, 1972. Robert Trivers. Social Evolution. Menlo Park/Cummings, 1985. Stanley Wasserman and Katherine Faust. Social Network Analysis: Methods and Analysis. Cambridge University Press, Cambridge, 1994. Edward O. Wilson. Sociobiology: The New Synthesis. Harvard University Press, Cambridge, Massachusetts, 1975. Margo Wilson and Martin Daly. The Adapted Mind: Evolutionary Psychology and the generation of culture, chapter The Man Who Mistook His Wife for a Chattel. Oxford University Press, 1992. 33