Cents and sociability : household income and social capital in rural Tanzania

POLICY RESEARCH WORKING PAPER 1796 Cents and Sociability Matching a measure of social capital with data on household income in certain Household Income and Social Capital rural villages inTanzania in Rural Tanzania shows that 'social capital" is indeed both capital (in that it raises incomes) and social (in that household incomes Lant Pritchett depend on village, not just household, social capital) The World Bank Social Development and Development Research Group Poverty and Human Resources July 1997 H  l POLICY RESEARCH WORKING PAPER 1796 Summary findings Narayan and Pritchett construct a measure of "social The magnitude of social capital's effect on incomes is capital" in rural Tanzania, using data from the Tanzania impressive: a one standard deviation increase in village Social Capital and Poverty Survey (SCPS), a large-scale social capital increases a household proxy for income by survey that asked individuals about the extent and at least 20 to 30 percent. characteristics of their associational activity and their This is as great an impact as an equivalent increase in trust in various institutions and individuals. nonfarming assets, or a tripling of the level of education. They match this measure of social capital with data on Data from the two surveys make it possible to identify household income in the same villages (both from the some of the proximate channels through which social SCPS and from an earlier household survey, the Human capital affects incomes: better publicly provided services, Resources Development Survey). In doing so, they show more community activity, greater use of modern that "social capital" is indeed both capital (in that it agricultural inputs, and greater use of credit in raises incomes) and social (in that household incomes agriculture. depend on village, not just household, social capital). This paper - a joint product of Social Development, and Poverty and Human Resources, Development Research Group- is part of a larger effort in the Bank to understand the social determinants of sustainable development. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Sheila Fallon, room N8-030, telephone 202-473-8009, fax 202-522-1153, Internet address sfallon(ciworldbank.org. July 1997. (41 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should he cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Produced by the Policy Research Dissemination Center  Cents and Sociability: Household Income and Social Capital in Rural Tanzania Deepa Narayan Lant Pritchett   I Cents and Sociability: Household Income and Social Capital in Rural Tanzania' Beyond apparently now old fashioned 'physical" capital, human capital, natural capital, institutional capital and social capital all clamor for attention. With capitalism all the rage, perhaps the proliferating rechristenings as 'capital" of otherwise perfectly serviceable concepts is understandable. But fashion aside, the popularity of 'capital'isms is due in part to the robust usefulness of the underlying metaphor: stuff that augments incomes but is not totally consumed in use. The attraction of investigating the incomes of households by examining their ownership of the various "capitals" is obvious. However, while obvious, an exclusive focus on households is seriously incomplete. Factors both at the national level of policies and institutions (Olson, 1996) and at the community level affect the fortunes of households and are potentially as important as the household's own capitals. In this paper we show that associational relationships and social norms of villages in rural Tanzania are both capital and social. After outlining the various concepts of social capital in the introduction we tell how (and why) we created data on social capital using a large The Social Capital and Poverty Survey was conducted as part of a Participatory Poverty Assessment led by Deepa Narayan as a joint activity of the government of Tanzania, the University of Dar Es Salaam, and the World Bank, funded by the British Overseas Development Agency. The Human Resource Development Survey data used in this paper come from a nationally representative survey of 5,000 households in Tanzania which was a joint effort undertaken by the Department of Economics of the University of Dar es Salaam, the Government of Tanzania, and the World Bank, and was funded by the World Bank, the government of Japan, and the British Overseas Development Agency. We would like to thank Jonathan Isham, and Sushenjit Bandopadhyay for collaboration in the early stages of the research, Christiaan Grootaert, Dean Joliffe, Michael Kremer, Peter Lanjouw, and Jonathan Morduch for useful comments, and Deon Filmer for help and insights on the econometrics.  2 scale household survey in rural Tanzania designed to query households about their social connections and attitudes. Second, using this and data on incomes we show that a village's social capital has an effect on incomes of the households in that village, an effect that is empirically large, definitely social, and plausibly causal. Finally, we use the two data sets to examine a number of proximate channels through which social capital appears to operate. Introduction Social capital, while not all things to all people, is many things to many people. A dramatic restriction of what one might mean must precede any attempt to estimate either "social capital' or its impact. What do we mean (and what do we not mean) by social capital and why do we think it might affect incomes? By "social capital" we mean the quantity and quality of associational life and the related social norms. The basic survey instrument, the Social Capital and Poverty Survey (SCPS), asked individuals a variety of questions about three dimensions of social capital. First, individuals were queried about their membership in various voluntary associations or groups to investigate the raw magnitude. For each group in which an individual reported membership, questions were asked about that group's characteristics in several dimensions relevant to that group's contribution to social capital. For instance, if the group's membership is 'inclusive' we assumed any given individual's membership in that group contributed more to social capital than membership in a group in which membership is "exclusive' to a particular clan or ethnic group. With this data on the frequency of membership and the characteristics of groups we created an index of the village associational life, which we argue is a proxy for social capital.  3 In addition to the information on associational life we sought to explore the existence and role of social and civic norms and individual's attitudes towards others, focusing in particular on the degree of trust individuals felt towards social groups, such as family, village or tribe, and towards government authorities, at the local, district, and national level. While social capital thus defined as the quantity and quality of local associational life is clearly social, is it 'capital"? Does it fit the 'capital' metaphor of something accumulated which contributes to higher income (or, more broadly, better outcomes). Five mechanisms have been proposed for how local social capital affects outcomes. From an economist's viewpoint, all of these share the characteristic that pure non-cooperative action would lead to inferior outcomes and hence that greater social capital potentially leads to better outcomes by facilitating greater cooperation. First, Putnam's (1993) fascinating analysis of the variations in public sector efficacy of the newly created regional governments in Italy suggests that regions of Italy in which people had greater degrees of horizontal connections had more efficacious governments. He documents a close connection between the numbers of voluntary associations and the efficacy of the regional government. Putnam finds that the more likely a region's citizens are to join football clubs and choral societies the faster the regional governnent is in reimbursing health care claims. One way of understanding this result is that monitoring the performance of the government is facilitated by greater social capital, either directly, because the government agents themselves are more embedded in the social network or perhaps indirectly because the monitoring of the public provision of services is a public good (and this is true even if the  4 publicly provided service is itself a private good as long as quality cannot be individually differentiated)2. Second, independent of the efficacy of governmental activity the role of group or community cooperative action in solving problems with a local 'common property' elements is potentially imnportant. Ostrom's (1990) work suggests that the ability of local groups to cooperate plays a large role in avoiding the negative consequences of the excessive exploitation or under maintenance of assets that would result from purely individualistic behavior under open access. She points out that the infamous "tragedy of the commons" based on purely individualistic behavior is only one possible outcome and that cooperative action can be a stable outcome. Ajuha (1996) shows that in Cote d'Ivoire the degree of land degradation is worse in more ethnically heterogenous villages, suggesting difference in the effectiveness of community controls and cooperation depends on social factors. Wade (1988) documents wide differences in the extent of cooperation within villages in Southern India, which he attributes to a significant degree to differences in the benefits from cooperation due to differences in the physical characteristics of the irrigation network serving the villages. Social capital may facilitate greater cooperation in the provision of services which benefit all members of the community. 2 Alesina, Baqir, and Easterly (1997) show that greater ethnic fragmentation in US cities leads to lower spending on productive public goods (e.g. education, roads, sewers) and is negatively related to the share of local spending on welfare. While not able to measure efficacy directly, the results also suggest higher public employment with greater fragmentation, possibly the results of higher patronage.  5 Third, diffusion of innovations might be facilitated by greater linkages among individuals. In his review of empirical work on the diffusion of innovations Rogers (1983) reports studies which suggest that "social participation," "interconnectedness with the social system,' 'exposure to interpersonal communication channels" and "belonging to highly interconnected systems' are each positively associated with the early adoption of innovations. Recent research on the adoption of Green Revolution innovations suggest that village level spillovers played a role in individuals' adoption decisions, but do not examine the role that social capital may have played in mediating the village level effects (Besley and Case 1994, Foster and Rosenzweig 1995). Fourth, greater associational activity may lead to less imperfect information and hence lower transactions costs and a greater range of market transactions in outputs, credit, land and labor leading to higher incomes. Social links among parties to economic transactions may increase their ability to participate in economic transactions which involve some uncertainty about compliance, like credit. There are two possible mechanisms at work. Social capital could lead to a better flow of information between creditors and borrowers and hence less adverse selection and moral hazard in the market for credit. Social capital also potentially expands the range of enforcement mechanisms for default on obligations in environments in which recourse to the legal system is costly or impossible. Fifth, greater sharing of household risk and informal insurance may allow households to pursue higher return but more risky activities and production techniques. If this is so then a social safety net that mitigated the consequences of adverse outcomes would lead farmers to  6 undertake higher return but also higher risk activities (Morduch, 1995). Increased social capital could lead to greater risk sharing among villagers and act as an informal safety net. What do we not mean by social capital? There are many other equally plausible and perhaps empirically important definitions of 'social capital" which we do not explore. In order to distinguish our work from the previous literature it helps to begin with a more general definition. In the abstract a 'society" can be thought of as a series of nodes (e.g. individuals, households) and a set of connections between those nodes. The connections between the nodes can be any kind of relationship whether social relationship (e.g. familial, ethnic), shared beliefs (e.g. religious), group identification (e.g. national, local) or a voluntary association, whether economic (e.g. employee, creditor) or non-economic (e.g. social club). Different notions of social capital can be distinguished by two features. First, whether the focus is on the nodes themselves, and hence on individual's social ties, or on the connections between the nodes, the intrinsically social. The second distinguishing feature of the existing empirical studies on social capital is the specification of what 'connections" between individuals are counted and how much weight each different type of link should receive. While we examine the social by examining the effect of the density of associational life on village outcomes, there is a considerable body of work on "social capital' that examines individual's ownership of social capital by examining the worth of each individual's social connection to other nodes. The emphasis is on the effects on the individual of having social links to valuable nodes, like having a rich uncle, or growing up in a good neighborhood (Case and Katz, 1991), or being a member of a successful ethnic group (Borjas, 1994).  7 We examine the links between individuals created by memberships in voluntary associations and social norns, which potentially excludes other dimensions of social capital. First, we do not examine the impact of any sense of affiliation with a nation or nation-state or any measure of distributional or ethnic conflict within the polity. Second, we do not examine as 'capital' the institutional "capacity' either of specific government or non-government organizations nor in the broad sense of society possessing a well known and legally sanctioned set of "rules of the game." Third, we do not analyze any 'cultural" values or attitudes, such as degrees of compassion, altruism, respect, tolerance4. Fourth we do not examine the issues explored recently by Knack and Keefer (1996) on the relationship between trust, norms of civic cooperation, associational activity and aggregate economic growth and investment rates. I. Data on Social Capital? We cannot examine the effects of what we mean by social capital, only what we measure. The sections below describe the survey and the procedure we used to construct our measure of social capital. The Social Capital and Poverty Survey (SCPS) was carried out in rural Tanzania in April and May of 1995 as part of a larger participatory poverty assessment exercise (Narayan, 1997). While the households were chosen randomly within clusters, the 3 Easterly and Levine (1996) have shown that a measure of "ethnic heterogeneity" is empirically associated with the adoption of bad economic policies, which they attribute to the importance of distributional conflict among groups. ' Explaining economic performance by such "cultural" characteristics such as the "Protestant work ethnic" or "Confucianism" has a history which is long and checkered, as it easily veers to self-congratulation or condescension. Recent entries include Harrison (1992) on culture and Fukuyama's examination of trust (1995).  8 sampling clusters themselves, which correspond roughly to villages in rural areas, were the same as those randomly selected for use in the 1993 Human Resource Development Survey (HRDS), hence the SCPS and HRDS data can be matched village by village5. The total usable SCPS sample is 1376 households located in 87 clusters6. Social capital module of SCPS. The survey's social capital component queried a household respondent about three dimensions of social capital: first, their membership in groups, second, the characteristics of those groups in which the households were members, third, the individuals values and attitudes, particularly their definition, and expressed level, of trust in various groups, and their perception of social cohesion. In this work we describe the groups only briefly, with a fuller description of the groups, their activities, and the results of qualitative information from interviews and participatory data collection methods in a companion paper (Narayan, 1997). The first set of questions was simply the number of groups in which an individual was a member. The average number of groups per person was 1.5 and table 1 lists the most prevalent groups, individuals' responses as to their 'most important' group and the groups they would join if they could join only one group. Most groups are Christian churches, Mosques, the village burial society, women's groups, and the political party. The more purely economic ' Also, rather than carry out new randomization, the households in the sample were the same as those sampled in the 1994/95 Agricultural Survey, with the addition of up to five non- agricultural households randomly selected within the cluster. ' The survey was implemented in two parts, a social capital module and a household module devoted primarily to measuring household expenditures, but unfortunately the second part was only administered in every other cluster so only 53 clusters have SCPS expenditure data.  9 associations (cooperatives, rotating credit groups) are much less important. In the construction of our measure of social capital we deliberately do not differentiate by type of group, as the main purpose is to examine whether groups with non-economic functions have village level spillover effects on economic outcomes'. Table 1: Groups in rural Tanzania, by membership and characteristics Group as a Number of 'Most important 'If you could join only percent of households group in your one group, which one all with life at present?" would it be?" membership members Church 21 230 29 24 Political party 17 195 10 3 (CCM ) _ _ _ _ _ _ _ _ _ _ _ _ Burial society 15 167 19 14 Women's group 9 104 5 8 Muslim group 9 109 11 8 Farmer's group 8 87 8 16 Other 21 252 _ _ Notes: In this table 'Other' includes (with percent reporting): Youth group (7), Primary society (4), Cooperative (2), Rotating Credit Societies (2), Dairy/cattle (1) and Other (5). In addition to questions about membership a second set of questions were asked about the characteristics of the each group in which the individuals reported membership. These were grouped into five categories: 1) kin heterogeneity of membership, 2) income heterogeneity of membership, 3) group functioning, 4) group decision making, and 5) 7 Early results suggested that excluding economic groups altogether had very little influence on the findings, as would be expected given their small share.  10 voluntary membership. The five questions in the three categories listed in detail in table 2 were those that proved useful in defining social capital for the empirical analysis below8. ' Earlier work on this data set used principal components to create an index but this was abandoned for three reasons. First, using a multiplicative rather than an additive index to combine membership and characteristics was appropriate. Second, the principal components methodology was not appropriate as the inter-correlations amongst these dimensions are not particularly high (the first principal component only "explained" 35 percent of the total variation). Third, the results on the first principal component alone were not robust when extended to other data and variables.  I1 Table 2: Questions for sub-components of social capital Question Responses Freq. Number of How many [from a prompted 0 32.3 groups enumeration of groups in the 1 40.4 village] are you a member of? 2 17.4 3 or greater 9.9 Answers to questions below recorded for up to three groups for each individual: Kin Who are (the group's] members? I-Close relatives 1.09 Heterogeneity Are they the same kin or the same 2-Same clan 2.95 clan? 3-Different tribes 25.7 4-Anyone in the village 70.2 Income Are all members from the same 1- All same livelihood 5.7 Heterogeneity economic group, do they all make 2-Most are the same 11.9 a living in the same way? 3-Mixed 82.3 Do the leaders or group officials 1 -Different livelihood 33.2 earn their living in the same way 2-Same livelihood 66.7 as other members or in different ways Group Overall, how would you rate the 1-Very poorly 2.1 functioning group functioning? 2-Poorly/weakly 8.5 3-Average 23.7 4-Good 47.4 5-Excellent 18.3 If there is a fee, what happens if a 1-Asked to leave the group 30.1 member does not pay the fee? 2-Delay in payment accepted 17.4 3-Nothing happens 52.5 Notes: Based on the non-missing observations for each category. To combine these questions into a single numerical index various strong (and arbitrary) assumptions were necessary about the weights and the aggregation. Since the variables are on different discrete scales, all the variables are first rescaled under the assumption that the observed indicator divided up evenly some underlying uniformly distributed continuous  12 variable ranging from 0 to 100 into N categories9. Second, we assume that the contribution to social capital of being a member of each group was greater if the group was more heterogenous across kinship groups, more inclusive and horizontal, and better functioning. Hence the contribution of each group to social capital is an equally weighted sub-index of these three characteristics. The village level social capital index is the product of the average number of groups with the average characteristics of those groups. Since the absolute scale for this index is arbitrary we re-normalize the index to have mean zero and standard deviation 10 one Before examining the specific hypotheses about social capital, some information on the situation and activities of the people surveyed will set the context. Rural Tanzania is a clear case of arrested economic development. Tanzanians are very poor, the average per person consumption expenditures reported in the 1993/94 HRDS in rural areas is 50 cents a day ($180 per person per year)". Most of the population is employed in traditional agriculture, with a substantial subsistence component as the imputed value of production for own consumption accounts for half of consumption expenditures. Nearly all agriculture is rain fed and uses 9 The numerical value to being the k' of the N, categories for the lI indicator is: V, = (IO0/N)*k - IOOI(N *2) . As long as the value assigned to each category is the mean of the observations in that category this procedure will not induce inconsistency in the resulting estimates. '