Rural institutions, social networks, and self-organized adaptation to climate change

The data was collected from the Takoli Panchayat in the lower Kullu Valley of Himachal Pradesh, India in 2011. Our analysis includes 273 households, which comprises all households that engage in agriculture in Takoli. The data identified 46 crops. We divided these crops into five types: vegetables, fruit, seeds, food grains, and ‘others’; each represents a distinct combination of labor requirements, inputs, market risk, and climate risks. We then calculated the proportion of land that each farmer has devoted to each of the five main crop types across one agricultural year. This was calculated as the total land of each crop type under cultivation divided by the gross cropped area for each farmer. Based on these variables, we undertook hierarchical cluster analysis using wards linkage, from which we derived 5 “agricultural portfolios”. This included three portfolios that specialize in fruit, vegetables, and food grains and two that are diversified across crop types with an emphasis on either food grains or vegetables. Households were also asked to identify the other households in Takoli they interact with on any issue regarding agricultural production. We used this data to create a bi-directional social network (using igraph package in R). We then calculated several measures of network centrality to capture different aspects of a household’s integration within the broader network: degree, betweenness, and eigenvector centrality. Within the full sample of 273 households, 4 households have no linkages within the network (they have a degree of 0). As such, they have no possible value for betweenness or eigenvector centrality (n=269). We also asked if households interacted with the six institutions of interest in our study: the Agricultural Department, Horticulture Department, the Bajaura Field Research Station (linked to state agricultural universities), Regional Banks, the Block Development Office (the lowest level of the development bureaucracy), and the Panchayat (local governmental unit). To calculate the network distance between households and each institution included in our study, we created a second social network that includes both households and institutions as ‘nodes’. We then calculated the additive inverse of the network distance to each of the institutions: a distance of ‘-1’ indicates that a household interacts directly with a given institution, a distance of ‘-2’ indicates that a household interacts with a household that interacts with the institution, and so on. Households with no network relationships can have no network distance (n=269). However, one household reports a direct interaction with the agricultural department and thus has a network position with regard to this institution (n=270). Please see detailed description in the publication SI.

本数据集采集自印度喜马偕尔邦（Himachal Pradesh）库鲁谷下游的塔科利村务委员会（Takoli Panchayat），采集时间为2011年。本次分析涵盖273户农户，覆盖了塔科利地区所有从事农业生产的家庭。本研究共识别出46种农作物，并将其划分为五大类别：蔬菜、水果、籽类、粮食作物及"其他类"；每一类作物均对应独特的劳动力需求、投入成本、市场风险与气候风险组合。随后，我们计算了单个农业年度内，每位农户投入五类主要作物类型的土地占比：该占比以各作物类型的总种植面积除以农户的总播种面积（gross cropped area）得出。基于上述变量，我们采用沃德联接法（Wards linkage）进行分层聚类分析（hierarchical cluster analysis），最终得到5类"农业种植组合"：其中三类分别以水果、蔬菜、粮食作物为专业化种植方向，另外两类为跨作物类型的多元化种植组合，分别侧重粮食作物与蔬菜种植。此外，我们向受访农户询问了其在农业生产相关事务中存在互动的塔科利地区其他农户信息，基于该数据构建了双向社交网络（social network，使用R语言的igraph工具包（igraph package）实现）。随后，我们计算了多项网络中心性指标以刻画农户在整体网络中的融入程度，包括度中心性（degree centrality）、介数中心性（betweenness centrality）与特征向量中心性（eigenvector centrality）。在全部273户样本中，共有4户农户未与网络中其他节点产生连接（度中心性为0），因此无法计算其介数中心性与特征向量中心性，有效样本量为269户。我们还向农户询问了其是否与研究关注的六类机构存在互动：农业部门、园艺部门、巴焦拉野外研究站（Bajaura Field Research Station，隶属于邦立农业大学）、地区银行、区块发展办公室（最低层级的行政发展官僚机构）以及村务委员会（Panchayat，地方基层行政单位）。为计算农户与各研究机构间的网络距离，我们构建了包含农户与机构两类节点的第二重社交网络，并据此计算了各机构对应的网络距离的加法逆元：若网络距离为"-1"，代表农户与该机构存在直接互动；若为"-2"，则代表农户通过与其有互动的农户间接与该机构建立联系，以此类推。无网络关联的农户无法计算网络距离，有效样本量为269户；但有1户农户报告与农业部门存在直接互动，因此该机构维度下有效样本量为270户。详细说明请参见论文补充材料（SI）。