The Impact of Kinship Networks on the Adoption of Risk-Mitigating Strategies in Ethiopia

World Development Vol. 43, pp. 100–110, 2013 Ó 2012 Elsevier Ltd. All rights reserved. 0305-750X/$ - see front matter www.elsevier.com/locate/worlddev

http://dx.doi.org/10.1016/j.worlddev.2012.10.011

The Impact of Kinship Networks on the Adoption of Risk-Mitigating Strategies in Ethiopia SALVATORE DI FALCO University of Geneva, Switzerland

and ERWIN BULTE * Wageningen University, Netherlands Summary. — The adoption of certain farm management practices, such as tree planting and soil and water conservation, can reduce exposure to weather shocks. However, in many countries the adoption of such risk mitigating measures is far from complete. We explore how risk-sharing networks in the form of kinship, characterized by the moral imperative of within-group sharing, affects the adoption of risk mitigating activities in rural Ethiopia. We find suggestive evidence that compulsory sharing invites free riding and attenuates incentives for self-protection against weather shocks. We also find evidence of the existence of a possible substitution effect between credit and social networks. Ó 2012 Elsevier Ltd. All rights reserved. Key words — kinship, compulsory sharing, drought, flood, Ethiopia, Africa

1. INTRODUCTION

insurance under the imperfect commitment model, and consider the implications of altruism entering in sharing relations. Altruism tends to ameliorate commitment problems, and increases the potential gains from income pooling and mutual insurance. Information advances or altruistic preferences probably imply gradual differences between kin and friendship networks. In addition, and more importantly for our purposes, kin networks may be different because they define sharing obligations for kin members (see below). While compulsory sharing within a kin network reduces idiosyncratic risk for members, this service comes at a cost because of several adverse incentive effects. Specifically, compulsory sharing of kin members invites free riding behavior––reducing incentives for selfprotection as members can fall back on others. Of course this is not unlike the classical moral hazard problem in other insurance contexts. Moreover, compulsory sharing may attenuate incentives for hard work as excelling members are prone to be approached for assistance by their kin. 4 In this paper we focus on the potentially adverse effects of kinship linkages in the context of self-insurance against weather shocks in Ethiopia, in the horn of Africa. 5 Redistribution implies shifting resources toward households that are worst affected by droughts and floods. Importantly, droughts and floods are systemic risks, affecting many members of the network simultaneously, limiting the scope to pool risks via local, informal mechanisms. While the impact of weather shocks can be, to some extent, buffered by the adoption of risk-mitigating technologies, free riding introduces a social dilemma in the

Farmers in the Horn of Africa are exposed to regular weather shocks (Intergovernmental Panel on Climate Change (IPCC), 2007). Various farm management innovations, such as tree planting and soil and water conservation, are available to reduce exposure to such shocks, but their uptake is far from complete. This is perhaps puzzling, in light of limited opportunities for smoothing consumption via “formal” financial and insurance markets. Self-protection and risk-sharing via informal community and family structures are the most prominent approaches to reducing exposure to risk. 1 Informal sharing arrangements have been analyzed by economists in detail. Such sharing usually focuses on self-enforcing arrangements as subgameperfect equilibria of repeated games, where binding participation and incentive constraints typically imply limited mutual insurance possibilities. 2 Alternatively, income pooling and redistribution can be organized in extended family or kinship networks. These relations are determined via bloodlines or marriage and therefore “are not the result of individual choice” (La Ferrara, 2007). This is an important difference between kinship groups and other types of groups, where individuals can choose to participate, or not. 3 The extended family is one of the key components of social capital throughout Sub-Saharan Africa. Kinship represents a primary principle of social organization, regulating access to resources and services, and governing social relationships and marital customs. Redistribution of assets (sharing) within the network is a prominent means to provide economic and social security to kinship members. Kinship may matter because “the ties of common experience, altruism, and heritage among family members enable families to transcend some of the information problems barring the development of impersonal markets” (Rosenzweig, 1988, p. 1167). Moreover, blood relations promote altruism (e.g., Hamilton, 1964). Foster and Rosenzweig (2001) extend the basic mutual

* Funding for this research was provided by the Swedish Research Council Formas through the program Human Cooperation to Manage Natural Resources (COMMONS). We want to thank three anonymous referees and seminar participants at the universities of Gothenburg, Gottingen, Oxford, and Tilburg for comments and suggestions. Remaining errors are our own. Final revision accepted: October 15, 2012. 100

THE IMPACT OF KINSHIP NETWORKS ON THE ADOPTION OF RISK-MITIGATING STRATEGIES IN ETHIOPIA

network. Low levels of mitigating effort may emerge as an equilibrium outcome. However, cutting back on self-protection is arguably not a socially optimal equilibrium strategy, as this would imply there are no unaffected community members to provide net transfers when disaster strikes. The main objective of this paper is to explore whether sharing norms within kinship networks imply adverse incentive effects for the adoption of specific technologies. Specifically, we seek to understand whether kinship obligations are correlated with the adoption of technologies that reduce exposure to weather risk (drought and floods)—ex ante risk mitigation. To do this, we test whether larger kinship networks are associated with reduced investments in self-protection against weather shocks. We establish a statistical relationship between investments in tree planting and soil and water conservation on the one hand, and the size of one’s kinship network on the other. The paper is organized as follows. In Section 2 we briefly discuss why kinship may affect productive and protective investments, and highlight the key tradeoffs for kin members. In Section 3 we hone in on the impacts of climate change for farmers in the horn of Africa—Ethiopia in particular—and discuss the various innovations that are available for farmers to partially self-protect against some of the relevant weather shocks. In Section 4 we present our data and outline our empirical strategy. Section 5 presents the main results, focusing on the drivers of adoption in Ethiopia and, in particular, on the role of kinship. Section 6 concludes. 2. THE MORAL ECONOMY OF KINSHIP Within kinship networks, individual members may claim assistance from others when necessary. In this respect, Scott (1976) refers to the “moral economy” of societies characterized by kinship relations, Baland, Guirkinger, and Mali (2011) refer to “forced solidarity,” and Hoff and Sen (2006) mention “social contracts” (see also Bloch, 1973). The moral element of kinship ties is underlined by Gulliver (1971, p. 217) who remarks that the statement “you must help a man because he is your kinsman” has the same constraining quality as the statement “you must cultivate because you need food to live.” Platteau (2000) discusses the role of witchcraft, ostracism, and other social sanctions to support them. “To fail in kinship obligation is to be a witch. . ., in other words to be the opposite of a moral being: a murderer, a bestialist, a lover of death, etc.” (Bloch, 1973, p. 78). Social stigma as well as retaliation “can thus fall on the defectors as well as on other members of their clan, increasing the cost of breaching the contract” (La Ferrara, 2003, p. 1733). This anthropological perspective—sharing without reckoning—suggests a safety net for the unlucky that is immune to selfish calculation. However, this perspective may be romantic or naı¨ve, as it discounts all economic reasoning and associated free riding issues. Some empirical evidence, albeit very limited, is available to support this concern. 6 Theorizing about free riding in the context of kinship networks is complex, and beyond the scope of this paper. However, some important insights can be gleaned from two papers by Alger and Weibull (2010, 2012). Their base model may be interpreted as dealing with the case of two family members who have to decide about the optimal level of self-protection against (weather) shocks. Individuals can undertake “effort” to reduce the risk of earning a low income in case a weather shock occurs, but this effort comes at a private cost. In case a shock affects the income of one of the family members, a sharing rule dictates that the other individual should provide assistance in the form of a

101

specific transfer. Family members are altruistic and care about each other’s well-being, so they are willing to provide some assistance. What happens when the sharing norm prescribes a transfer that exceeds the one that would be voluntarily provided? Alger and Weibull demonstrate the level of effort varies with the level of altruism. 7 They distinguish between the so-called empathy effect of sharing within the kin network (driven by altruism), capturing the incentive to support one’s kin (and reduce the probability of having to draw on one’s kin’s resources) and the so-called free-rider effect. This free rider effect is associated with forced sharing, and captures both the ability to live off the efforts of kin, as well as the disincentive to put in effort because there is always a risk that the returns of such investments will have to be shared with kin members with low payoffs. When the sharing norm exceeds voluntary transfers, the norm adversely affects incentives to engage in self-protection. In what follows, we seek to test the prediction that kinship ties adversely affect self-protection. We analyze the case of tree planting and soil conservation to self-protect against weather shocks in Ethiopia, and consider the implications of variation in the number of kinship links. This is a non-trivial step from the Alger–Weibull model, because it is not immediately obvious how the case of enlarging the size of the network compares to the case of different levels of altruism. Exploring this issue, Alger and Weibull (2012) develop a model that explicitly captures the impact of enlarging the size of the network. In this model, individuals in a population are pairwise and randomly matched to other individuals—kin members and non-kin members. The share of matches involving kin members increases as the number of kin members increases. Individuals may also be matched multiple times in each period, as “the machinery applies to each combination of game and kinship relation separately.” Hence, the strategic incentive effect becomes more pronounced as the share of kin members in the population increases, ceteris paribus. Applied to our context, if compulsory sharing dictates more generous transfers among kin members than those forthcoming because of voluntary sharing (altruism), then networks with more kin members will discourage self-protection against weather risk. However, the assumption of random matching is obviously a simplification of the true interaction within networks. In reality, matches are “chosen” by the unlucky, who approach others from whom they may expect to receive assistance. Note that the comparative statistics of such a model are much more complex. Being a member of a dense network implies that many kin members may approach you for assistance, but simultaneously it is also true that “the unlucky” may approach many other kin members for contributions (or perhaps the transfer can be jointly paid by some subsamples of kin members). In other words, it is unclear how the sharing rule is affected. In that sense, our empirical approach is exploratory: we analyze the relation between kinship links and self-protection, and explore whether the incentive effects are sufficiently large to dominate opposite effects due to, say, dilution of the norm or altruism (but also because of additional effects, such a social learning within the network).

3. WEATHER SHOCKS AND MITIGATING RESPONSES IN ETHIOPIA Ethiopia is one of the least developed countries in the world, with a per capita income of approximately USD 1000 (PPP). Agriculture is mainly traditional and employs more than

102

WORLD DEVELOPMENT

80% of the labor force, accounting for 45% of GDP and 85% of export revenues (MoFED, 2006). Ethiopian agriculture also heavily depends on natural rainfall. For the existing cultivated area only 4–5% is irrigated (Awulachew et al., 2007). Rainfall and temperature are important determinants of crop harvests, and unfavorable realizations of either the amount or the temporal distribution of rainfall triggers food shortages and famine. Ethiopia repeatedly suffered from weather shocks in the form of droughts and, to a lesser extent, floods. In light of this evidence, it is no surprise that Stige et al. (2006) identify Ethiopia as one of the most vulnerable countries to climate change. Improving the agricultural sector’s capacity to adequately respond to weather shocks is widely perceived as a top priority for decision-makers and developmental agencies. A number of studies, in the literature on climate change, have considered the factors governing farmers’ decision to adopt adaptive measures and their impact on decision on harvests (Molua, 2009; Seo & Mendelsohn, 2008; Seo, Mendelsohn, Dinar, & Kurukulasuriya, 2009). In the context of the Nile basin some previously published papers using the same database as we do have highlighted the set of strategies adopted by farmers to reduce the household’s exposure to the vagaries of changing climate (see Deressa, Hassan, Ringler, Alemu, & Yesuf, 2009; Di Falco, Veronesi, & Yesuf, 2011). Based on these studies we identify two well-known practices that are widely considered as risk-mitigating strategies: tree planting, and the implementation of soil and water conservation strategies. Tree planting helps to maintain soil moisture, conserve soil organic matter, reduce soil loss due to erosion and flooding, and provides shades for other crops (Kassie, Zikhali, Pender, & Kohlin, 2010; Melillo et al., 2011; van Kooten, Shaikh, & Sucha´nek, 2002). Soil and water conservation measures include soil bunding, cultivation of hedges, contour ploughing, irrigation, and water harvesting activities (so as to bridge dry spells). 4. DATA AND THE ECONOMETRIC MODEL (a) Data We use data from the 2004–2005 farming season, of a rural household survey of 1000 households located in Ethiopia’s Nile Basin. The sampling frame considered the traditional typology of agro-ecological zones in the country (i.e., Dega, Woina Dega, Kolla, and Berha), the percentage of cultivated and irrigated land, average rainfall as well as rainfall variability, and vulnerability of the population (as proxied by the number of food aid dependent people), such that a representative sample of the Nile basin eventuated. The procedure resulted in the inclusion of 20 villages, and 50 households were randomly sampled from each village. Data were collected on production and input use at the plot level for two cropping seasons—the Meher (long rainy season) and the Belg (short rainy season). Few plots, however, have a bi-annual cropping pattern (i.e., crop growth during both seasons). 8 Descriptive statistics are presented in Table 1, which provides information on a number of standard household characteristics—age, literacy, and size of the household—and agricultural variables, such as the quantity of fertilizer and manure applied per unit of land, and a (subjective, self-reported) measure of soil quality. For the latter we inserted a categorical variable, based on the farmers’ response to the request to evaluate plot quality as either low, average, or high. Data on labor allocated to each plot were also collected. While

the survey distinguished between male, female, and child labor, without much loss we lump these classes together into one aggregate labor measure (using adult equivalents). 9 Given the dependence on climatic conditions for farming success, monthly rainfall and temperature data were collected from meteorological stations in the country. The Thin Plate Spline method of spatial interpolation was used to impute household specific rainfall and temperature values using latitude, longitude, and elevation information of each household. 10 In order to control for the role of the current weather on the farmers’ decision to adopt, we used these interpolated farm-specific data to construct the annual average of rainfall and temperature. 11 Table 1 also summarizes links of the household with the outside world, including whether the household had access to formal extension or farmer-to-farmer extension activities. Given the objectives of this paper we are especially interested in the size of the kinship network. Our kin variable is simply defined as the (self-reported) number of relatives of the household living in the same village, but not in the same household. 12 In this study we consider family members up to cousins, nephews, and nieces as kin (more distant relatives are not included in the survey). We can therefore map, per each single farm household, the number of links existing in the village. We also know the distance between the household and the village. Below, we will combine these two pieces of information to construct instrumental variables. Unfortunately we do not have information on outside-village kinship size, so our measure of total kinship (pressure) is incomplete and we cannot explore whether there is a distinction between local and non-local kin in terms of risk sharing. The sample mean for the entire sample equals 12.3 relatives. 13 When restricting the sample to those who planted trees, the sample mean is 16.8. Similarly, the mean number of kinship links for households engaged in soil conservation equals 13.6. These differences in sample means are statistically not significant. Recently, the literature has highlighted the important role of networks in determining adoption decisions (e.g., Bandiera & Rasul, 2006; Conley & Udry, 2010). Farmers may adopt technologies because of social learning or imitation of their relatives, neighbors, or friends. Unfortunately, we do not have detailed information on how this process may happen. But we do know how many farmers in the village are engaged in soil conservation or tree planting. We use this variable to capture potential network effects that are different from the ones we speculate are at play (for a discussion on the estimation of how one’s decision is affected by others, see Glaeser & Scheinkman, 2001). Kinships network are not the only risk sharing network available in this setting. Two non-market institutions for risk sharing are widely used by farm households in our sample: Iddirs and Iqqub. No less than 66% of the households have access to these. Both Iddirs and Iqqub are indigenous organizations involved in providing self-help and shelter against risk in Ethiopia (e.g., Besley 1995; Dercon, De Weerdt, Bold, & Pankhurst 2006; Pankhurst 2003). Iddirs are historically established to provide mutual aid in burial matters, but also address other concerns. Households join the association and regularly pay a contribution. In case a member dies, the association handles the burial and related ceremonies (Pankhurst, 2003). Iqqubs are traditional savings and credit institutions, allowing members to borrow and lend. Members pay periodically a fixed amount of money. This is collected in a common pot which members receive on a rotational basis. The rotating scheme may be adjusted to reflect shocks, hence iqqubs also

THE IMPACT OF KINSHIP NETWORKS ON THE ADOPTION OF RISK-MITIGATING STRATEGIES IN ETHIOPIA

103

Table 1. Variable list and descriptive statistics Variables

Mean

Std dev.

Min

Max

Relative (number of relatives in a village) Access to formal extension (1 = yes; 0 = otherwise) Farmer extension (1 = yes; 2 = otherwise) Information received climate change through extension (1 = yes; 0 = otherwise) Household Size (number of members) Age of household head (in years) Education (in years) Land (in hectares) Labor per hectare (in adult days) Manure (kg per hectare) Fertilizer (kg per hectare) Belg: rainfall in short rainy season (mm per year) Meher: rainfall in long rainy season (mm per year) Temperature dispersion Soil fertility (1 = highly fertile, 2 = moderately fertile, 3 = infertile)

12.28 60% 50% 42% 6.57 45.66 1.64 0.38 101 206 60 322 1107 0.66 1:28% 2:56% 3:15% 16% 10% 1.7 9%

12.98 0.49 0.5 0.48 2.23 12.6 2.68 0.50 123 888 175 160 293 0.14 0.54

0 0 0 0 1 14 0 0.01 0 0 0 20 434 0.18 1

80 1 1 1 15 86 13 21 2128 1740 410 807 2075 1.45 3

0.044 0.99 2.9 0.071

0 0 0 0

1 1 50 1

Tree planting (1 = yes; 0 = otherwise) Soil and water conservation (1 = yes; 0 = otherwise) Distance (in km) Death in kin network (1 = yes; 0 = otherwise)

serve a risk sharing purpose (Calomiris & Rajaraman, 1998). In some of the models below we control for membership of such alternative risk-sharing institutions. 14 Finally we introduce our dependent variables. The survey aimed to analyze farmers’ adoption strategies in response to long run changes in climatic variables (Deressa et al., 2009), and included questions to investigate whether farmers noticed changes in mean temperature and rainfall over the last two decades (and probed the perceived cause of these changes). About 90% of the sample perceived such changes in mean temperature or/and rainfall. Farmers who observed climatic changes were also asked whether they had responded through the adaptive measures. The data suggest these measures were taken on 26% of the operated plots: 16% involved planting trees and 10% involved measures aimed at reducing soil and water conservation. 15 We assume these reflect a purposeful adaptation strategy to reduce exposure to climate change, and measure them via two dummy variables. While the survey instrument was designed to identify “adaptive strategies,” these investments may also raise productivity. Hence, they may serve additional purposes, and this should be noted as a caveat. (b) Econometric strategy We aim to estimate how the probability of adopting a specific risk mitigating strategy is affected by the extent of the family network and a set of controls. Let Aih represents the i-th strategy for household h, and xh ; xlh ; xch are (vectors of) household characteristics, land and climatic variables, respectively. The extent of the network is represented by Nh while, b is a vector of parameters and ehi is a household specific random error term. The econometric model is Aih ¼ Aðxh ; xlh ; xch ; N h ; bÞ þ ehi :

ð1Þ

As mentioned Nh represents the number of kin members of household h, or a proxy of the size of the safety network upon which household h can fall back in times of hardship. Note that this is a simplification, as the capacity of the safety

network to accommodate shocks also depends on the (protective) actions chosen by kin members. In other words, a game theoretical approach may be followed, where one solves for the equilibrium level of protection of households in the kin network (as in Alger and Weibull, 2008, 2010). However, we do not have data on technology choices of all kin members. For our simplest identification strategy to be valid we assume that the size of one’s kinship network is exogenous (as in Isham, 2002; Kinnan & Townsend, 2010). The nature of kinship (determined by bloodlines, marriage, or adoption) suggests such an assumption may be plausible (La Ferrara, 2007). However, it is perhaps not evident that this assumption holds in all contexts. For example, kinship may be correlated with wealth or status, and the number of local kinship members may vary with asset holdings. For instance, respondents from families with excellent soil quality may have more kin members in the family (because fewer family members migrated out, or because of differential mortality or fertility). If soil quality also affects technology adoption decisions (which is likely), then the kin and adoption variables are correlated but this correlation does represent a causal effect of kinship on adoption. This potential problem would be eliminated if we had access to a complete set of control variables—in fact, as mentioned, we have a proxy of soil quality—but this might not always be true; omitted variable problems might remain. Our response is multi-pronged. First, we control for land and manure (key assets capturing “wealth” or endowments) and other important controls to attenuate the risk of omitted variables. Second, we implement an instrumental variables probit approach to deal with possible endogeneity of the kinship variable. The choice of instruments is complex as we need a variable that is correlated with the kinship metric, but not with the error term of the adoption models. We considered variables that can encapsulate access or reachability from the kin group. As mentioned previously, we were able to map both the size of the network in the village and the geographical distance (in km) between the household and the village. We stress that this distance variable displays considerable variation (from 0 to 50 km). We also considered idiosyncratic

104

WORLD DEVELOPMENT

shocks affecting the size of the extended family, and use a dummy capturing whether the household reported the death of a kin member in the last 10 years. Such deaths may be correlated with the size of the current network, and are unlikely to be directly correlated with the propensity to adopt new technologies. 16 We are quick to point out that neither instrumental variable is perfect, but we will demonstrate that the test statistics support the idea that they help to bolster our case. 17 Third, we include fixed effects. We add village fixed effects to address unobservable heterogeneity at the village level. Further, to address issues such as social status or cultural differences (or other plot-invariant household unobservables), we estimate some models at the plot level and include household pseudo-fixed effects to control for unobservable heterogeneity (Wooldridge 2002). That is, we run a random effects model while controlling for unobserved heterogeneity using Mundlak’s approach (Mundlak, 1978; Wooldridge, 2002). This is a very general and convenient way of inserting fixed effects in a probit model. It is also a useful specification in the presence of some farm households invariant variables. Besides our metric for kinship, the size of the resident household, age of the household head, race, and access to information do not change within the household (unlike land size, fertility, and other inputs which vary across plots). This approach exploits the existence of plot-varying exogenous variables, and models heterogeneity as a linear function of all exogenous variables (Mundlak 1978; Wooldridge, 2002). The right hand-side of our pseudo-fixed effect regression equation includes the mean value of the plot-varying explanatory variables, and assumes that unobserved effects are linearly correlated with explanatory variables, as specified by: wh ¼ xa þ gh ;

gh  iidð0; r2g Þ;

ð2Þ

where x is the mean of the plot-varying explanatory variables within each household (cluster mean), a is the corresponding vector coefficient, and g is a random error term that is unrelated to x. 18 The vector a will be equal to zero if the observed explanatory variables are uncorrelated with the random effects. This allows us to test for the relevance of the fixed effects via an F test–testing whether the estimated coefficients in vector a are jointly equal to zero. We reject this null hypothesis, so controlling for the pseudo fixed effect helps to obtain consistent estimates. Besides the possible correlation with unobservables, our kinship metric can be correlated with other observables. If so, the interpretation of the coefficients may be less neat. We therefore estimate an auxiliary regression regressing our kinship metric on all other control variables––including potentially collinear ones like education and household size, and proxies for endowments such as manure or farm size. The results of this auxiliary regression are used to net out the impact of the kinship variable from other influences (the so-called regression anatomy approach, see Angrist & Pischke, 2009). Hence, after regressing kinship against all explanatory variables, we use the residuals from this regression in place of the original kinship variable to partial out the effect of other covariates. 19 The key variable of interest is the number of kinship linkages, and this variable enters in all models. To analyze how kinship impacts on adoption of soil conservation measures and tree planting we first estimate two separate probit models. To check the consistency of our results across different specifications, we present the results of different model specifications where we gradually include various determinants of adoption. In the Appendix we demonstrate that the results are robust to estimating a bivariate probit model, considering the two adoption decisions jointly. Since we observe multiple

plots operated by the farm household, we relax the assumption that the error term is iid and cluster standard errors at the household level, allowing the errors to be correlated in some “unknown” fashion.

5. REGRESSION RESULTS Our first set of econometric results is reported in Tables 2 and 3. We focus on the effect of the number of kinship links on two adoption strategies—soil conservation and tree planting. We first present the results of a parsimonious specification that includes only the kinship variable. We then control for household characteristics and extension services (model 2), farm inputs and soil fertility (model 3), weather variables (model 4), and spillover and risk-sharing network effects (model 5). Models 6 and 7 are the IV probit models where we instrument for kinship. Finally, models 8 and 9 are the elaborate plot level models controlling for pseudo fixed effects. Since all farmers have multiple plots (and this analysis is at the plot level), the sample size of the analysis increases. As mentioned, we partial out the influences of other covariates (regression anatomy approach) on the kinship variable. All specifications include village fixed effects. The first result is that the number of kinship links is significantly correlated with a reduced probability to invest in soil conservation (Table 2). While this stands in contrast to earlier studies, which tend to document a positive effect of measures of networks on the rate of adoption of new technologies (e.g., Boahene, Snijders, & Folmer, 1999; Bandiera and Rasul, 2002; Isham, 2002), these earlier studies are based on network proxies that include kin as well as non-kin. By focusing exclusively on relatives—a subset of social capital—our analysis allows honing in on potentially adverse incentives induced by forced solidarity. We document robust evidence of such adverse incentive effects associated with kinship ties, at least for the case of soil conservation. This is also true for the IV models, which satisfy the relevant test statistics. 20 We estimated the average marginal effect for the kinship variable, which we found to be 0.005. Roughly, adding one member to the network (or a 10% increase in the kinship variable) reduces the probability of investing in soil conservation measures by 0.5%. The average marginal effects of the other two network variables (number of adopters in the village and availability of other risk mitigating networks), are respectively 0.02 and 0.29. Hence, an increase in 10% in the number of adopters in the village will increase the probability of undertaking soil conservation by 2%, and having access to other risk mitigating networks increases that probability by 29%. The negative effect of the kinship network is therefore modest when compared to the impact of other variables. Interestingly, and this is our second result, the estimates for tree planting summarized in Table 3 are qualitatively different. The most striking difference is that there exists a positive correlation between kinship and adoption. This finding is consistent with the perspective that trees serve a double function: reducing exposure to weather shocks and enhancing tenure security 21 by signaling ongoing use and investment (Deininger & Jin, 2006; Mekonnen, 2009). 22 To attenuate the risk of losing one’s property to a kinship member—perhaps most likely to be successful when challenging tenure rights because of the common heritage and bloodlines—farmers may resort to planting trees. 23 To further probe this interpretation, we identified if the household officially owns the plots it is cultivating, as guaranteed by official certification. Only 20% of the plots are titled. When we split the sample and distinguish between tenure

THE IMPACT OF KINSHIP NETWORKS ON THE ADOPTION OF RISK-MITIGATING STRATEGIES IN ETHIOPIA

105

Table 2. Kinship and adaptation to climate change—The case of soil conservation Probit

IV Probit

(1)

(2) *

(3) *

(4) *

(5) *

0.0086 (0.00487)

0.010 (0.006) 0.0259 (0.0522) 0.0121 (0.0085) 0.0444 (0.0361) 0.464* (0.270) 0.0303 (0.262) 0.316 (0.204)

0.0105 (0.006) 0.0386 (0.0522) 0.0156* (0.0089) 0.0442 (0.0359) 0.458* (0.270) 0.0335 (0.266) 0.242 (0.208) 1.088** (0.519) 0.0010 (0.0009) 0.00001 (0.00001) 0.00001 (0.00006) 0.325** (0.165)

0.0102 (0.0062) 0.0404 (0.0565) 0.0163* (0.0096) 0.0589 (0.0378) 0.437 (0.283) 0.0053 (0.273) 0.162 (0.213) 1.242** (0.532) 0.0017 (0.0014) 0.00001 (0.00001) 0.000002 (0.0001) 0.347** (0.175) 0.944** (0.393) 0.0411*** (0.0150) 0.00213** (0.0009)

Village fixed effects Farm fixed effects

Yes No

Yes No

Yes No

N

461

461

0.1

0.13

Relative HH size Age Education Extension Farmer extension Information on CC Land Labor Manure Fertilizer Fertility

**

Yes No

Yes Yes

0.037* (0.0210) 0.181 (0.158) 0.0359* (0.0203) 0.0058 (0.123) 0.0217 (0.550) 1.511** (0.658) 0.358 (0.361) 0.705*** (0.205) 0.0039** (0.0015) 0.00001 (0.00001) 0.0003** (0.0001) 0.159 (0.190) 0.681 (0.685) 0.0390** (0.0186) 0.0020* (0.0011) 0.244*** (0.0523) 1.777* (0.921) Yes Yes

461

461

1381

1381

461

461

461

0.16

0.25

0.4

Number of adopters in village Other risk sharing networks

**

(9)

0.0243 (0.0111) 0.0486 (0.0860) 0.0173 (0.0147) 0.00155 (0.105) 0.676* (0.372) 0.142 (0.304) 0.0999 (0.354) 0.228 (0.139) 0.0003 (0.0010) 0.000001 (0.00001) 0.00001 (0.0001) 0.0202 (0.176) 0.828 (0.593) 0.0411 (0.0273) 0.0021 (0.0014)

Yes No

Rain  temperature

(8)

0.0255 (0.0119) 0.0669 (0.0880) 0.0238 (0.0165) 0.0887 (0.0717) 0.112 (0.327) 1.025 (0.789) 0.536 (0.452) 1.453 (1.028) 0.0009 (0.0017) 0.00001 (0.00001) 0.00004 (0.0001) 0.105 (0.298) 0.424 (0.311) 0.0205 (0.0168) 0.00105 (0.0009) 0.718* (0.373) 0.715 (0.784) Yes No

0.0191 (0.0091) 0.0205 (0.0471) 0.0109 (0.0204) 0.0129 (0.131) 0.316 (0.312) 0.189 (0.309) 0.245 (0.272) 0.883 (1.613) 0.0011 (0.0025) 0.00001 (0.00002) 0.00003 (0.00004) 0.127 (0.420) 0.622 (0.806) 0.0264 (0.0415) 0.0014 (0.0021)

Rain

Pseudo R

*

IV Probit pseudo FE (7)

0.0149 (0.0084) 0.0519 (0.0836) 0.0197* (0.0112) 0.0148 (0.0569) 0.0279 (0.392) 1.534*** (0.498) 0.533 (0.332) 2.872*** (0.994) 0.0001 (0.0015) 0.000001 (0.00001) 0.0001 (0.00004) 0.297 (0.254) 0.403 (0.255) 0.0220** (0.0107) 0.0013** (0.0006) 1.016*** (0.163) 0.717 (0.437) Yes No

Temperature

2

(6)

**

Village clustered standard errors in parentheses. Constant not reported. Farm level analysis: Wald test of exogeneity (/athrho = 0):chi2(1) = 1.04 Prob > chi2 = 0.3. Plot level analysis: Wald test of exogeneity (/athrho = 0):chi2(1) = 0.38, Prob > chi2 = 0.5. * p < 0.10. ** p < 0.05. *** p < 0.01.

secure and insecure households, the kinship variable only enters significantly for the subsample of tenure insecure households—see Table 4. This suggests that when tenure security is achieved via alternative means, planting trees becomes not necessary to signal property rights. The average marginal effect is such that a 10% increase in the size of the network increases the probability of planting trees by 2.5%. We do not report average marginal effects for other network measures as their coefficients are not statistically significant. We implemented a Wald test for exogeneity for both soil conservation and tree planting equations. While we have no evidence of possible endogeneity in the former, we can reject the null hypothesis in the latter, the standard probit estimation results in the tree planting analysis can therefore be plagued by endogeneity bias. Nevertheless, it is comforting to observe that

the results for the kinship variable are qualitatively unaffected across all specifications (in the sense that the directional changes in the quantitative results are the same as before). Finally, we are interested in the interplay between access to formal credit and the incentive effects implied by the kinship network. Access to credit is an important variable in explaining the propensity to invest, and its omission could result in biased parameter estimates. The survey contains the following question: “Did you borrow in 2004–2005?” In the survey, about 25% of the sample reported to have used credit. Obviously, this variable may be correlated with unobservable household characteristics, and should be treated as potentially endogenous in regression models. We use the household credit question to construct a community credit variable (which is arguably exogenous to household characteristics). Specifically,

106

WORLD DEVELOPMENT Table 3. Kinship and adaptation to climate change—The case of tree planting Probit

IV Probit

IV Probit Pseudo FE

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

0.0037* (0.0019)

0.0036* (0.0020) 0.0617 (0.0381) 0.0073 (0.0091) 0.0122 (0.0417) 0.112 (0.251) 0.0966 (0.265) 0.538*** (0.202)

0.0050* (0.0026) 0.0619 (0.0388) 0.0068 (0.0092) 0.0140 (0.0431) 0.101 (0.253) 0.100 (0.269) 0.518** (0.211) 0.0304 (0.567) 0.00109 (0.0007) 0.00001 (0.00002) 0.00001 (0.00004) 0.257 (0.192)

0.0049* (0.0026) 0.0650* (0.0381) 0.0074 (0.0093) 0.0143 (0.0439) 0.0942 (0.255) 0.112 (0.269) 0.535*** (0.207) 0.0559 (0.579) 0.0011 (0.0007) 0.00001 (0.00001) 0.00001 (0.00004) 0.255 (0.193) 0.136 (0.203) 0.0006 (0.007) 0.00001 (0.0005)

0.0171*** (0.0061) 0.0179 (0.0219) 0.0043 (0.0058) 0.0525 (0.0320) 0.102 (0.109) 0.0802 (0.152) 0.138 (0.100) 0.305 (0.380) 0.0005 (0.0003) 0.000004 (0.00001) 0.00002 (0.00001) 0.0631 (0.154) 0.114 (0.0898) 0.0014 (0.0021) 0.0001 (0.0001)

Yes No

yes No

Yes No

Yes No

Yes No

0.0133*** (0.0049) 0.0597 (0.0450) 0.0113 (0.0082) 0.0769* (0.0467) 0.404 (0.387) 0.107 (0.276) 0.149 (0.208) 0.458 (0.539) 0.0014* (0.0007) 0.00002 (0.00001) 0.00003 (0.00003) 0.208 (0.161) 0.172 (0.171) 0.0010 (0.00442) 0.0002 (0.0003) 0.0173 (0.0669) 0.119 (0.231) Yes No

0.0150*** (0.0051) 0.0122 (0.0290) 0.0050 (0.0070) 0.0598 (0.0370) 0.161 (0.137) 0.0073 (0.191) 0.0016 (0.254) 0.0037 (0.131) 0.0001 (0.001) 0.00001** (0.00000) 0.0002** (0.0001) 0.0953 (0.0738) 0.00712 (0.129) 0.0011 (0.004) 0.0001 (0.0003)

Village fixed effects Farm fixed effects

0.0071* (0.0042) 0.0547 (0.0469) 0.0027 (0.0086) 0.0089 (0.0403) 0.361 (0.267) 0.183 (0.246) 0.181 (0.252) 0.328 (0.403) 0.002*** (0.0007) 0.00002 (0.00002) 0.00001 (0.00003) 0.180 (0.177) 0.269** (0.134) 0.0026 (0.0054) 0.0002 (0.0003) 0.147*** (0.0298) 0.454** (0.216) Yes No

Yes Yes

0.0123*** (0.0041) 0.0273 (0.0309) 0.0099 (0.0067) 0.0641 (0.0401) 0.0810 (0.293) 0.350 (0.312) 0.0809 (0.204) 0.0684 (0.136) 0.0005 (0.0007) 0.00002*** (0.00001) 0.0005*** (0.0001) 0.210*** (0.0792) 0.0042 (0.150) 0.0022 (0.0041) 0.0001 (0.0002) 0.0153 (0.0149) 0.104 (0.203) Yes Yes

N

450

450

450

450

450

450

450

1234

1234

0.2

0.23

0.24

0.25

Relative HH size Age Education Extension Farmer Extension Information on CC Land Labor Manure Fertilizer Fertility Temperature Rain Rain  temp Number of adopters Other risk sharing networks

Pseudo R

2

Village clustered standard errors in parentheses. Constant not reported. Farm level analysis: Wald test of exogeneity (/athrho = 0)::chi2(1) = 8.98 Prob > chi2 = 0.002. Plot level analysis: Wald test of exogeneity (/athrho = 0):chi2(1) = 9.44 Prob > chi2 = 0.0021. * p < 0.10. ** p < 0.05. *** p < 0.01.

Table 4. Tenure security and tree planting Plot level

Farm level

Certified land

Not certified land

Certified land

Not certified land

Relative

0.00482* (0.00256)

0.00365 (0.00795)

0.00430* (0.00258)

0.00884 (0.00782)

N

1609

463

597

146

Village clustered standard errors in parentheses. Other controls not reported. Significance codes: * p < 0.10. ** p < 0.05. *** p < 0.01.

THE IMPACT OF KINSHIP NETWORKS ON THE ADOPTION OF RISK-MITIGATING STRATEGIES IN ETHIOPIA Table 5. Kinship and investments in the presence of access to credit Soil conservation

Tree planting

Access to Access to Access to Access to credit 6 15% credit > 15% credit 6 15% credit > 15% (1) (2) (3) (4) Relative 0.205*** (0.0752)

0.00507 (0.00668)

0.0251** (0.0101)

0.00426 (0.00582)

N

1582

674

1582

449

Village clustered Robust standard errors in parentheses. Other controls not reported. Significance codes: * p < 0.10. ** p < 0.05. *** p < 0.01.

we seek to distinguish between “credit villages” (villages with relatively easy access to formal credit) and “non-credit villages.” Rather arbitrarily we have set a threshold at 15%: if more than 15% of the respondents in a particular village responded to have access to credit, we define the village as a “credit village” (and assume that credit via formal banks or a system of moneylenders is “relatively readily” available). If less than 15% of the villagers have used credit, we assume obtaining credit is difficult. We have varied the nature of this threshold—considering thresholds of 25% or 35%––and found this did not affect the results. We use the distinction between credit and non-credit villages to partition the sample and perform separate regressions on subsamples of our respondents. Results are reported in Table 5. Importantly, we find that the results do not hold in “credit villages,” which may indicate the existence of a possible substitution effect between credit and social networks. 6. DISCUSSION AND CONCLUSION We explored the role of risk sharing network on the uptake of weather shocks management strategies in Ethiopia. We found that a specific cultural phenomenon—traditional sharing norms—may attenuate incentives to adopt technologies that reduce exposure to weather shocks. Understanding how this can act as a barrier to adoption is crucial especially in the context of climate change in the horn of Africa. Higher temperatures and changing precipitation levels caused by climate change are expected to depress crop yields in many countries during the coming decades. Agricultural sectors in many African countries are particularly vulnerable to climate change as they are largely based on rain-fed farming styles (IPCC, 2007), 24 and because adaptive capacities are modest. Prolonged droughts in the horn of Africa and elsewhere, interspersed with periods of flooding, have underscored the importance of enhancing the agricultural sector’s capacity to adequately respond to weather shocks. A better understanding of farmers’ responses to weather shocks is therefore crucial to design strategies to mitigate the adverse effects of climate change. Our results suggest that traditional sharing norms in kinship networks may attenuate incentives to adopt protective measures. Admittedly, the evidence presented on these pages must be taken with caution. The kinship variable captures the salience of sharing norms, but possibly various other effects as well (altruism, social learning), so our “reduced form” approach only allows us to test whether the overall effect of kinship ties is positive or negative. Also, endogeneity concerns may remain, and our data are rather crude (e.g., our kinship variable does not include kinship members living outside the village).

107

Finally, to assess whether investment levels are sub-optimally low (or high) requires a careful comparison of all relevant marginal benefits and costs—data that are not available. 25 We hope future research could address these caveats. However, and notwithstanding these caveats, we conjecture the romanticized view of “sharing without reckoning” systems within extended family systems held by many anthropologists—celebrating traditional sharing norms as a safety net for the unlucky in the absence of formal insurance mechanisms—may be naı¨ve in the sense that it overlooks efficiency-equity tradeoffs. Our findings are consistent with the hypothesis that incentives for selfish behavior exist, even in the “moral economy” of kinship networks. The outcome may be a poverty trap where self-protection is under-supplied. While risk-share enables individuals to reduce defensive efforts in the presence of idiosyncratic shocks (such as adverse health shocks or wildlife damage to a standing crop), risk sharing is not effective in the context of systemic effects like droughts and floods—affecting many or all members of the kin network simultaneously. While the socially optimal response to such systemic risks may be the universal adoption of defensive measures by all kin members, individuals have incentives to defect. The result is a social equilibrium that is Pareto dominated. If so, the redistributive network approach toward insurance becomes a dysfunctional institution—inviting collective decisions that are ex ante inefficient and hindering adequate responses to collective risks. For theoretical reasoning along these lines, refer to Platteau (2000) and Hoff and Sen (2006). These results support an earlier perspective by Bauer and Yamey (1957, p. 64): “The extended family [. . .] is an example of an institution which has many advantages in one stage of development but which may later become a drag on economic development.” Our results highlight another type of risk in Ethiopian farming. Tenure rights in rural Ethiopia are often uncertain, and may be contested by others—notably kin members. It has been suggested in the literature (and contested by others) that to reduce exposure to losing tenure, farmers in Ethiopia plant trees to signal ownership. Our results provide some support for this proposition; it seems that trees serve a double function in terms of self-protection. From an agronomic view they protect against temperature increases associated with climate change, and from a socio-economic view they protect the household’s resource base against claims by others. Households with a greater kin network are more likely to plant trees. In other words, while being locked in a kin network reduces incentives for soil conservation, it increases incentives for tree planting. Finally, if traditional sharing norms may hinder development, how do modern insurance mechanisms (access to formal credit) affect adoption of risk mitigating strategies? Our tentative evidence is consistent with the view that modern and traditional insurance mechanisms can be substitutes. Households living in villages with access to formal credit may opt out of the network and shed its moral imperatives. Credit may “crowd out” some forms of network if it can supply insurance benefits at a lower cost. Indeed, we find that the negative effect of the network on adoption of risk-mitigating measures disappears once we focus on credit villages. If confirmed by other studies, an important recommendation for policy makers may be to promote expansion of the financial system into rural areas allowing individuals to evade the family tax, enable them to adopt efficient levels of self-protection, and hence reduce exposure to climatic shocks. Financial expansion may dissolve poverty trap outcomes—not only because capital is scarce and necessary for development, but also because it attenuates culturally-induced distortions.

108

WORLD DEVELOPMENT

NOTES 1. See, e.g., Coate and Ravallion (1993), Fafchamps (1992), Fafchamps and Gubert (2007), Fafchamps and Lund (2003), Kimball (1988), Rosenzweig (1988)Townsend (1994), and Udry (1994). 2. For example, Coate and Ravallion (1993) analyze stationary schemes, Ligon, Thomas, and Worrall (2003) consider non-stationary schemes with history-dependent transfers—blurring the distinction between insurance and credit—and Genicot and Ray (2003) analyze coalition stability in the presence of potential defection by subgroups of individuals. 3. Of course individuals can also “opt out” of family networks, for example by migrating to another area. In the empirical analysis below we will further probe the assumption of exogeneity of the kinship group. 4. These ideas are not new. For example, Bauer and Yamey (1957, , pp. 66–67) noticed the moral hazard problems of mutual insurance within kin networks and argue that “. . . the [extended family] system . . . minimizes the inducement for people to improve their position because they can count on being provided with the means of subsistence at a level not very different from that of the majority of their kinsmen, including the energetic, thrifty, and able.” In addition, the prospect of forced redistribution may directly impact on investment and spending decisions. Lewis (1955, p. 114) recognizes the adverse incentive effects of compulsory sharing when writing about successful kinship members who “may be besieged by . . . increased demand for support from a large number of distant relatives.” 5. For a broad perspective on sharing and kinship systems in Ethiopia, please refer to Matsumara (2006). 6. For example, Baland et al. (2011) analyze borrowing behavior in Cameroon, finding that some people “pretend to be poor.” Excess borrowing is costly but may be used to signal poverty, and hence suggests an inability to respond to demands for financial assistance from kin members. Using household data from Kwazulu Natal, Di Falco and Bulte (2011) document households evading compulsory sharing through their consumption behavior. As the number of kinship linkages increases, accumulation of sharable goods (e.g., jewels) is reduced, and household consumption of non-durables (food) increases as does accumulation of non-sharables (e.g., furniture). Jakiela and Ozier (2012) vary the observability of endowments in a behavioral game and demonstrate that some respondents are willing to incur a cost to hide their endowments from their kin. Our analysis is the first to consider the effect of kinship links on ex investments in ex ante risk mitigation. 7. Obviously, when the cost of effort is sufficiently high, both family members undertake no effort. 8. Droughts occurred in 1965, 1974, 1983, 1984, 1987, 1990, 1991, 1999, 2000, and 2002, and floods occurred in 1997 and 2006. 9. We employed the OECD/EU standard conversion factor in the literature in developing countries, where adult female and child labor are converted into the adult male labor equivalent with the conversion factors 0.8 and 0.3, respectively. 10. The Thin Plate Spline is a physically based two-dimensional interpolation scheme for arbitrarily spaced tabulated data. The Spline surface represents a thin metal sheet that is constrained not to move at the grid points, ensuring that the generated rainfall and temperature data at the weather stations are exactly the same as data at the weather stations

used for the interpolation. So, rainfall and temperature data at the weather stations are reproduced by the interpolation for those stations (see Wahba, 1990 for details). This method is one of the most commonly used to create spatial climate data sets. Its strengths are that it is readily available, relatively easy to apply, and accounts for spatially varying elevation relationships. However, it only simulates elevation relationships, and it has difficulty handling sharp spatial gradients. This is typical of coastal areas. Given that our study area is characterized by significant terrain features, and no climatically important coastlines, the choice of this method is reasonable (for details on the properties of this method, refer to Daly (2006). 11. Admittedly, rainfall and temperature averages may not be well suited to capture the role of weather extremes. 12. Kinnan and Townsend (2010) used such information to build a kinship variable, constructing a dummy variable taking the value of one if the household has kin in the village, and zero otherwise. However, since the extent of network pressure may also be determined by the size of the network, we extend this measure and document the number of relatives in the village (a count variable). 13. There is considerable variation in our kinship variable, from 0 to 80 kin members within the village. We assume this diversity reflects actual divergence in the extent of kinship links, and that all households interpreted the survey question in the same manner (i.e., we assume that the definition of uncle or nephew is the same for all our respondents). 14. Since these variables can be endogenous in some of our models, we do not emphasize their coefficient estimates. The main purpose for including them is to demonstrate that the coefficient of our main variable of interest (kinship) is not affected. 15. We take soil conservation and water harvesting together and treat it as a measure of soil moisture and nutrient management. Tree planting and soil and water management are basically the only investments applied by the farmers in the sample. Other adaptive strategies were changing crop varieties (30% of the farmers) and changing planting date (12%), but these are not investments and are therefore not included in the remainder of this study. Future research should be allocated to understand if networks play a role in the adoption of these short term strategies. 16. We also considered the timing of the death and built dummies to distinguish if the shock was experienced recently or more than 5 years ago. Results were very consistent. 17. For example, consider a potential omitted variable—altruism toward kin members. Families that are attached to one another are likely to prefer living close to one another. The distance to family variable is likely to be also correlated with other aspects of social interactions, like labor sharing, and information sharing, both of which might affect the technology adoption decision. 18. Therefore Mundlak’s formulation assumes “equicorrelation” between the unobservable fixed effect and the plot varying variables. An alternative formulation that relaxes this assumption, at the cost of degrees of freedom, is given by Chamberlain (1982). 19. Of course we have also implemented the analysis without partialing out the effect of other variables on our kin metric (entering the kinship proxy directly). Results are very consistent and available upon request.

THE IMPACT OF KINSHIP NETWORKS ON THE ADOPTION OF RISK-MITIGATING STRATEGIES IN ETHIOPIA 20. It should be noted that with the ML estimator, there is not really a “first-stage regression”, since it is estimated simultaneously with the main equation. Nevertheless to probe if we are instruments are relevant we did run an auxiliary first stage regression where our kinship variable is regressed against the instruments and the other exogenous variables. We found that both excluded instruments significantly correlated with our kinship variable (coefficient = 0.604**, clustered standard error = 2.47 for the deaths variable, and coefficient = 1.15*, clustered standard errors 0.62, for the distance variable). Our instruments are correlated with the kinship variable at the 5% and 10% significance level. The estimated coefficient for the “death” variable is negative, reflecting this shock reduces the size of the extended family. The estimated coefficient for the variable distance is positive, which may reflect that larger families are dispersed over larger distances. 21. We note this view is not uncontested in the case of Ethiopia. Land in Ethiopia is state owned, and occasionally redistributed by the government. For example, Ali, Dercon, and Gautam (2007) demonstrate that perceived expropriation risk discourages investments in tree crops. 22. An alternative explanation exists based on the potential difference in “observability” of the two types of investments. While planting trees is clearly visible, the same need not be true (or need not be true to the same extent) as some of the soil and water conservation measures are included in this study. Assume people seek to neutralize the free riding effect within kin networks by making sharing conditional on adoption of specific risk-

109

reduction strategies. Perhaps people plant trees to signal to their kin that they have made efforts to reduce their exposure to weather shocks. If so, free riding combined with contracting on technology could explain the opposing effects for tree planting (observable) and soil conservation (not easily observable), rather than free riding combined with tenure insecurity. 23. As a further robustness analysis we also estimated a bivariate probit model. In Table 6 in the appendix reports the same specification as in columns (5) of Tables 2 and 3. The results are consistent. Moreover, the testing procedure on the correlation coefficient of the error terms, q0, indicates we cannot reject the null hypothesis of zero correlation (see test result at the bottom of Table 6 for the Wald test). It is, therefore, appropriate to look at the two adoption strategies as independent from one another, as in Tables 2 and 3. 24. According to the IPCC, agricultural productivity at lower latitude, in tropical dry areas is expected to decrease “for even small local temperature increases (1–2°C)” (IPCC, 2007, p. 11). In many African countries access to food will be severely affected, “yields from rain fed agriculture could be reduced by up to 50% by 2020” (IPCC 2007, p. 13). 25. For example, Kinnan and Townsend (2010) document that kin may act as an “implicit collateral” facilitating investment. This clearly is a benefit associated with kinship ties, but the exact magnitude of this benefit is difficult to estimate.

REFERENCES Alger, I., & Weibull, J. (2010). Kinship, incentives and evolution. American Economic Review, 100(4), 1725–1758. Alger, I., & Weibull, J. (2012). A generalization of Hamilton’s rule – Love others how much?. Journal of Theoretical Biology, 299(4), 42–54. Ali, D., Dercon, S., & Gautam, M. (2007). Property rights in a very poor country: Tenure insecurity and investment in Ethiopia. Policy Research Working Paper #4363. Washington, DC: The World Bank. Angrist, J. D., & Pischke, J.-S. (2009). Mostly harmless econometrics: An empiricists companion. Princeton: Princeton University Press. Awulachew, S., Yilma, A., Loulseged, M., Loiskandl, W., Ayana, M., & Alamirew, T. (2007). Water resources and irrigation development in Ethiopia. Working Paper #123. Colombo, Sri Lanka: International Water Management Institute. Baland, J.-M., Guirkinger, C., & Mali, C. (2011). Pretending to be poor: Borrowing to escape forced solidarity in Cameroon. Economic Development and Cultural Change, 60(1), 1–16. Bandiera, O., & Rasul, I. (2006). Social networks and technology adoption in northern Mozambique. Economic Journal, 116(514), 869–902. Bauer, P., & Yamey, B. (1957). The economics of under-developed countries. Cambridge: Cambridge University Press. Besley, T. (1995). Property rights and investment incentives: Theory and evidence from Ghana. Journal of Political Economy, 103(5), 903–937. Bloch, M. (1973). The long term and the short term: The economic and political significance of the morality of kinship. In J. Goody (Ed.), The Character of Kinship. Cambridge: Cambridge University Press. Boahene, K., Snijders, T., & Folmer, H. (1999). An integrated socioeconomic analysis of innovation adoption: The case of hybrid cocoa in Ghana. Journal of Policy Modeling, 21(2), 167–184. Calomiris, C., & Rajaraman, I. (1998). The role of RoSCAs: Lumpy durables or event insurance?. Journal of Development Economics, 56(1), 207–216. Chamberlain, G. (1982). Multivariate regression models for panel data. Journal of Econometrics, 18(1), 5–46. Coate, S., & Ravallion, M. (1993). Reciprocity without commitment: Characterization of informal insurance arrangements. Journal of Development Economics, 40(1), 1–24. Conley, T., & Udry, C. (2010). Learning about a new technology: Pineapple in Ghana. American Economic Review, 100(1), 35–69. Daly, C. (2006). Guidelines for assessing the suitability of spatial climate datasets. International Journal of Climatology, 26(6), 707–721.

Deininger, K., & Jin, S. (2006). Tenure security and land-related investment: Evidence from Ethiopia. European Economic Review, 50(5), 1245–1277. Dercon, S., De Weerdt, J., Bold, T., & Pankhurst, A. (2006). Group-based funeral insurance in Ethiopia and Tanzania. World Development, 34(4), 685–703. Deressa, T., Hassan, R., Ringler, C., Alemu, T., & Yesuf, M. (2009). Determinants of farmers’ choice of adaptation methods to climate change in the Nile Basin of Ethiopia. Global Environmental Change, 19(2), 248–255. Di Falco, S., & Bulte, E. (2011). The dark side of social capital: Kinship, consumption and investment. Journal of Development Studies, 47(8), 1128–1151. Di Falco, S., Veronesi, M., & Yesuf, M. (2011). Does adaptation to climate change provide food security? A micro perspective from Ethiopia. American Journal of Agricultural Economics, 93(3), 829–846. Fafchamps, M. (1992). Solidarity networks in pre-industrial societies: Rational peasants in a moral economy. Economic Development and Cultural Change, 41(1), 147–173. Fafchamps, M., & Gubert, F. (2007). The formation of risk sharing networks. Journal of Development Economics, 83(2), 326–350. Fafchamps, M., & Lund, S. (2003). Risk sharing networks in rural Philippines. Journal of Development Studies, 71(2), 261–287. Foster, A., & Rosenzweig, M. (2001). Imperfect commitment, altruism and the family: Evidence from transfer behavior in low-income rural areas. Review of Economics and Statistics, 81(3), 389–407. Genicot, G., & Ray, D. (2003). Group formation in risk-sharing arrangements. Review of Economic Studies, 70(1), 87–113. Glaeser, E., & Scheinkman, J. (2001). Measuring social interactions in social dynamics. In S. Durlauf, & P. Young (Eds.), Social dynamics. Cambridge: MIT Press. Gulliver, P. (1971). Neighbours and networks: The idiom of kinship in social action among the Ndendeuli of Tanzania. Berkeley: University of California Press. Hamilton, W. (1964). The genetic evolution of social behavior. Journal of Theoretical Biology, 7(1), 1–52. Hoff, K., & Sen, A. (2006). The kin system as a poverty trap. In S. Bowles, S. Durlauf, & K. Hoff (Eds.), Poverty traps. Princeton: Princeton University Press. Intergovernmental Panel on Climate Change (IPCC) (2007). Summary for policymakers. Climate change 2007: The physical science basis. Working group I contribution to IPCC fourth assessment report. Geneva.

110

WORLD DEVELOPMENT

. Isham, J. (2002). The effect of social capital on fertiliser adoption: Evidence from rural Tanzania. Journal of African Economies, 11(1), 39–60. Jakiela, P., & Ozier, O. (2012). Does Africa need a rotten kin theorem? Experimental evidence from village economies. Policy Research Working Paper #6085. Washington, DC: World Bank. Kassie, M., Zikhali, P., Pender, J., & Kohlin, G. (2010). The economics of sustainable land management practices in the Ethiopian highlands. Journal of Agricultural Economics, 61(3), 605–627. Kimball, M. (1988). Farmers’ cooperatives as behavior towards risk. American Economic Review, 78(1), 224–232. Kinnan, C., & Townsend, R. (2010). Kinship and financial networks, formal financial access and risk reduction. Mimeo: Northwestern University. La Ferrara, E. (2003). Kin groups and reciprocity: A model of credit transactions in Ghana. American Economic Review, 93(5), 1730–1751. La Ferrara, E. (2007). Descent rules and strategic transfers: Evidence from matrilineal groups in Ghana. Journal of Development Economics, 83(2), 280–301. Lewis, W. (1955). The theory of economic growth. London: George Allen & Unwin. Ligon, E., Thomas, J., & Worrall, T. (2003). Informal insurance arrangements with limited commitment: Theory and evidence in village economies. Review of Economic Studies, 69(1), 115–139. Matsumara, K. (2006). Moral economy as emotional interaction: Food sharing and reciprocity in highland Ethiopia. African Studies Quarterly, 9(1), 9–22. Mekonnen, A. (2009). Tenure security, resource endowments, and tree growing: Evidence from the Amhara Region of Ethiopia. Land Economics, 85(2), 292–307. Melillo, J., Butler, S., Johnson, J., Mohana, J., Steudler, P., Luxa, H., et al. (2011). Soil warming, carbon–nitrogen interactions, and forest carbon budgets. Proceedings of the National Academy of Sciences of the United States of America, 108(23), 9508–9512. MoFED (2006). Survey of the Ethiopian economy. Addis Ababa, Ethiopia. Molua, E. (2009). An empirical assessment of the impact of climate change on smallholder agriculture in Cameroon. Global and Planetary Change, 67(3–4), 205–208. Mundlak, Y. (1978). On the pooling of time series and cross-section data. Econometrica, 46(1), 69–85. Pankhurst, A. (2003). The role and space for iddirs to participate in the development of Ethiopia. In A. Pankhurst (Ed.), Iddirs: Participation and development, Proceedings of the Ethiopian national conference 20– 21 December 2001 (pp. 2–42). Addis Ababa: Agency for Cooperation in Research and Development (ACORD). Platteau, J. P. (2000). Institutions, social norms and economic development. Amsterdam: Harwood. Rosenzweig, M. (1988). Risk, implicit contracts and the family in rural areas of low-income countries. Economic Journal, 98(393), 1148–1170. Scott, J. (1976). The moral economy of the peasant: Rebellion and subsistence in Southeast Asia. New Haven: Yale University Press. Seo, N., & Mendelsohn, R. (2008). Animal husbandry in Africa: Climate change impacts and adaptations. African Journal of Agricultural and Resource Economics, 2(1), 65–82. Seo, N., Mendelsohn, R., Dinar, A., & Kurukulasuriya, P. (2009). Adapting to climate change mosaically: An analysis of African livestock by agro-ecological zones. The B.E. Journal of Economic Analysis & Policy, 9, [Contributions, article 4]. Stige, L., Stave, J., Chan, K., Ciannelli, L., Pattorelli, N., Glantz, M., et al. (2006). The effect of climate variation on agro-pastoral production in Africa. Proceedings of the National Academy of Sciences, 103(9), 3049–3053. Townsend, R. (1994). Risk and insurance in village India. Econometrica, 62(3), 539–591. Udry, C. (1994). Risk and insurance in a rural credit market: An empirical investigation in northern Nigeria. Review of Economic Studies, 61(3), 495–526. van Kooten, G., Shaikh, S., & Sucha´nek, P. (2002). Mitigating climate change by planting trees: The transaction costs trap. Land Economics, 78(4), 559–572. Wahba, G. (1990). Spline models for observational data. Philadelphia: Society for Industrial and Applied Mathematics.

Wooldridge, J. (2002). Econometric analysis of cross section and panel data. Cambridge, MA: MIT Press

APPENDIX

Table 6. Bivariate probit estimation Soil conservation (1)

Tree planting (2)

Village fixed effects Farm fixed effects Athrho

0.0296** (0.015) 0.0989 (0.0639) 0.0373** (0.0153) 0.0874* (0.0474) 0.330 (0.227) 1.783*** (0.648) 0.610 (0.482) 0.381 (0.283) 0.0016 (0.0014) 0.0000 (0.0000) 0.0000 (0.0001) 0.232*** (0.0765) 1.983 (1.440) 0.0673 (0.0471) 0.0039 (0.0025) 0.241*** (0.0424) 0.450 (0.395) Yes Yes 0.611

0.0121*** (0.0024) 0.103*** (0.0313) 0.0054 (0.0081) 0.0257 (0.0225) 0.852*** (0.306) 0.385* (0.222) 0.140 (0.259) 0.132 (0.100) 0.0018*** (0.0002) 0.0000 (0.0000) 0.0000 (0.0000) 0.202* (0.115) 0.0527 (0.176) 0.0053 (0.0043) 0.0003 (0.0003) 0.0725** (0.0297) 0.681*** (0.250) Yes Yes (0.389)

N

1381

1381

Relative HH size Age Education Extension Farmer extension Information on CC Land Labor Manure Fertilizer Fertility Temperature Rain Rain  temperature Number of adopters in village Other risk sharing networks

Village clustered standard errors in parentheses. Constant not reported. Wald test for rho = 0:hi2(1) = 2.46 Prob > chi2 = 0.12. * p < 0.10. ** p < 0.05. *** p < 0.01.