A game theory model to explore the role of cooperation and diversity in community food security: the case of Southern Malawi

1 A game theory model to explore the role of cooperation and diversity 1 in community food security: the case of Southern Malawi. 2 Stefano Balbi1,2, *, Unai Alvarez-Rodriguez1,3, Vito Latora3,4,5, 3 Alberto Antonioni6, Ferdinando Villa1,2 4 5 1 Basque Centre for Climate Change (BC3), Sede Building, Campus EHU/UPV, Leioa, Bizkaia, 6 Spain. [email protected], [email protected], 7 ferdinando.vi[email protected]rg 8 2 IKERBASQUE, Basque Foundation for Science, Bilbao, Bizkaia, Spain. 9 3 School of Mathematical Sciences, Queen Mary University of London, London E1 4NS, UK. 10 [email protected] 11 4 Dipartimento di Fisica ed Astronomia, Università di Catania and INFN, I-95123 Catania, Italy. 12 5 The Alan Turing Institute, The British Library, London, NW1 2DB, UK. 13 6 Grupo Interdisciplinar de Sistemas Complejos (GISC), Departamento de Matemáticas, 14 Universidad Carlos III de Madrid, E-28911, Leganés, Madrid, Spain. 15 [email protected] 16 *Corresponding author 17 18 Abstract 19 The Sustainable Development Goals aim at ending food insecurity by 2030. Therefore, 20 civil society needs to understand the inherent complexities of both socio-economic and 21 ecological dynamics and their interdependencies. In particular, the behavioural 22 dynamics that underpin human agents are crucial in driving the final outcomes in terms 23 of community food security and require further attention. Using household behaviour 24 within a rural village of Southern Malawi as an example, we describe a game theory 25 model representing cropping strategies: (1) cooperation, as driven by other-regarding 26 preferences, and (2) conformation, the tendency to converge to similar crop planting 27 choices as opposed to differentiation (and thus crop diversity). We find that the latter 28 plays a crucial role in driving the system towards successful strategies: how individuals 29 relate to social norms has greater effect. Cooperation is only necessary for community 30 success when the community converges on crop planting choices. On the contrary, 31 differentiation, the affirmation of the individual unique identity, can succeed with or 32 without cooperation. We further elaborate on the idea that community level sustainability 33 can be reached through different pathways, which might require food exchange 34 mechanisms within and beyond the system boundaries. 35 36 Keywords 37 human behaviour; game theory, social networks, diversity, social-ecological systems, 38 food security. 39 Length of Manuscript: 6100 words and 5 figures and 1 table. 40 This document is the Accepted Manuscript version of a Published Work that appeared in final form in: Balbi S., Alvarez-Rodriguez U., Latora V., Antonioni A., Villa F. 2020. A game theory model to explore the role of cooperation and diversity in community food security: the case of Southern Malawi. Regional Environmental Change. 20. (2) DOI (10.1007/s10113-020-01642-5).). © 2020, Springer-Verlag GmbH Germany, part of Springer Nature.. 2 1. Introduction 1 2 Food provisioning is a key challenge of coupled human-natural systems. The 3 Sustainable Development Goals (SDGs) adopted by the United Nations have set a clear 4 target to end food insecurity by 2030 (UN 2015a). In order to achieve this target 5 understanding the inherent complexities of both socio-economic and ecological 6 dynamics, and their mutual feedbacks across scales is crucial (Ostrom 2008). While 7 agro-ecological dynamics are well explored in the literature (e.g. Chappell and LaValle 8 2011), the behavioural dynamics that underpin human agents require more research to 9 understand how they drive community food security, especially in rural regions of the 10 Global South. 11 12 Using household crop planting choices within a rural village as an example, we 13 introduce a behavioural compass based on two dimensions: (1) cooperation, the degree 14 to which individual success depends on neighbours’ success due to other-regarding 15 preferences, which in turn captures a community scale objective function, and (2) 16 conformation, the tendency to converge towards similar crop planting choices as 17 opposed to differentiation, and as a consequence, crop diversity. We argue that by 18 acknowledging the interplay among these two dimensions in shaping household 19 cropping strategy, a rural agricultural system can better reorganize, spontaneously or by 20 means of policy interventions, to improve food security at the community level. 21 22 In the following paragraphs we explain the compass axes, used as a reference system 23 to characterize household behaviour and show how the compass applies to the selected 24 case study: a rural village of Southern Malawi. In Methods, we describe the context and 25 the data used for the application and the mathematical formalization of the model. We 26 also describe its behaviour as we change the main household parameters (other27 regarding preferences and homophily) within their interval of reference. In Results, we 28 assess the model performance when the game parameters (number of households, 29 number of crops and network topology) are set to match the case study characteristics, 30 deriving the household behaviour that delivers the best outcomes. In Discussion, we 31 elaborate on the implications of the model findings for the community food security. 32 Finally, we conclude with the key messages of this study. 33 34 Inspired by the political compass of Maddox and Lilie (1984) and Lester (1994), we refer 35 to a behavioural compass (Fig. 1), a four by four grid where the horizontal axis 36 represents economic behaviour and displays competition at its left side and full 37 cooperation at its right side, while the vertical axis represents personal behaviour and 38 displays differentiation at the bottom end and conformation at its top end. 39 3 Economic behaviour describes how the individual relates to the others for the purpose 1 of achieving resources, including consumption and utilisation of natural resources. 2 Cooperation can be understood as the process of acting in coordination with others to 3 maximise individual or mutual benefit (Perc et al. 2017). Competition arises when at 4 least two parties strive for a goal that cannot be shared, or is desired individually but not 5 in sharing. 6 Personal behaviour describes how individual beliefs and values relate to social norms. 7 Conformation is the process of matching individual behaviour with the dominant social 8 norms, while differentiation—as in psychological differentiation (Witkin et al. 1974)—is 9 the process of affirming an individual unique identity against social norms. Thus, the 10 higher differentiation, the higher the diversity of behaviours within the system. 11 12 In an early example of a similar approach, Deutsch (1949) identifies three types of “goal 13 structures on achievement”: cooperative, competitive and individualistic. Cooperation 14 and competition have been extensively researched in ecology and economics (game 15 theory, in particular: see Fehr and Fischbacher 2002) as individual operating modes 16 (from the seminal work of Axelrod and Hamilton 1981 to more recent reviews, such as 17 Perc et al. 2017). Differentiation and conformation belong to the domain of behavioural 18 and experimental methods in psychology and economics and, in the context of this 19 study, are particularly close to the research on how social norms and the framing of 20 options influence individual behaviour, with significant repercussions for human 21 cooperation and natural resource exploitation (Kahneman 2011). 22 23 24 Fig. 1 The behavioural compass 25 4 2. Methods 1 2 In this article, we refer to community food security as an emergent property of 3 household food security, which is a consequence of how households interact, acquire 4 and utilise assets, including natural resources (Chambers and Conway 1991). In most 5 cases, the livelihoods of rural households remain largely dependent on agriculture. This 6 article explores data of a rural region located in the southern part of Malawi, a small, 7 landlocked African country home to approximately 15 million people. More than 90% of 8 the rural population are smallholder farmers, responsible for cultivating plots with an 9 average size of just 0.8 hectares (NSO 2012). Maize is the socially preferred crop 10 among smallholders and the main staple diet of the population (NSO 2012). According 11 to Chinsinga and Chasukwa (2012), although over 97% of smallholder farmers grow 12 maize, only 10% are net sellers and up to 60% are net buyers. 13 Typically, most of the households within a rural village are at least acquainted with each 14 other and it is common practice to exchange on-farm labour and food donations in 15 periods of need (Dobbie et al. 2018). Similar dynamics have been witnessed in rural 16 settings around the globe (Patel 2009) including: sharing of unused ingredients that 17 would otherwise be trashed, sharing of knowledge about farming practices, good 18 nutrition and recipes, invitations to join meals, exchange of seeds and food commodities 19 to increase diversity of diet, exchange of portions of meals in different periods in time to 20 cushion temporary scarcity (i.e. food banking). In Malawi, this system of informal social 21 exchanges coexists with a market-oriented structure that heavily relies on local and 22 regional food markets for the integration and redistribution of food commodities, as a 23 consequence of the liberalisation reforms promoted by the International Monetary Fund 24 and the World Bank (Dorward and Kydd 2004). 25 26 2.1 The case of food security in Southern Malawi 27 In Malawi over 50% of the population live on less than one US dollar a day and the 28 proportion of ultra-poor people (defined as the proportion of population below the 29 minimum level of dietary energy requirement) is highest within Southern Malawi at 30 approximately 34.2% (Gondwe 2014). Food security is particularly problematic in rural 31 areas where agriculture is primarily rain-fed, leaving smallholders vulnerable to climatic 32 shocks (Sahley et al. 2005). A single rainy (growing) season, between the months of 33 November and March, is followed by a dry season from April to October. Only a limited 34 number of farming households with access to dimba fields of the valleys located at the 35 source of streams, creeks, or rivers may take advantage of residual moisture and 36 extend cultivation beyond the end of the rains (Orr et al. 2009). High population 37 densities, small average plot size and poor soil quality further increase food insecurity. 38 The rural population is anticipated to grow from approximately 8.4 million in 1990 to 39 almost 20 million by 2030 (UN 2015b) and this growth will have wide ranging impacts 40 5 upon land and labour availability as well as market prices and productivity. Additional 1 exogenous trends such as soil degradation and climate change further compound food 2 insecurity (Schmidhuber and Tubiello 2007). 3 4 Indeed, food security is multidimensional issue (Connolly-Boutin and Smit 2016). A total 5 of four dimensions are recognized under the “four pillars” framework—created by FAO 6 and operationalised for modelling purposes in Dobbie and Balbi (2017)—including food 7 availability, access, utilisation and stability. The production of food is related primarily to 8 food availability (Headey and Ecker 2013). Access refers to the amount of food a 9 household can produce, purchase from the market and/or derive from other means 10 (Burchi and De Muro 2016). Households might draw upon social safety nets such as 11 food for work programmes or adopt coping strategies like selling livestock or borrowing 12 food (Devereux 2016). A third dimension, utilisation, refers to the ability of households 13 to process accessible food. This is dependent upon the household ability to obtain 14 sufficient quantities of fuel and clean water. Finally, stability dictates how robust 15 availability, access and utilisation dimensions are to shocks and stresses over time 16 (Burchi and De Muro 2016), such as those related to climatic and demographic change. 17 This article focuses primarily on the first two pillars: availability and access. 18 19 In the following paragraph we present the alternative principles of the behavioural 20 compass applied to the cropping strategies of farming households, and illustrate how 21 the combination of these principles is relevant for understanding community food 22 security (Fig. 1). The four storylines shown in Table 1, which describe observed 23 household cropping behaviours in rural regions of the Global South, are synthesized 24 from the Participatory Rural Appraisals (PRA: Schreckenberg et al. 2016) that took 25 place in multiple villages of the Zomba Region, in Southern Malawi. 26 27 According to the competition and conformation storyline (Table 1), a household would 28 maximise the yield of maize using all the available inputs subsidized by the government 29 (e.g. hybrid maize seeds and chemical fertilizer), and sell the produce at market. This is 30 currently the most common strategy within the villages of Southern Malawi, where 31 liberalisation reforms have failed to the produce the intended improvements and the 32 social norms regarding what to grow are widely and voluntarily accepted. Food 33 preferences are key in this regard. Maize has become the main food asset in the local 34 markets and is considered a strategic commodity. This social norm is not only 35 detrimental in terms of diet variety and healthy nutrition, but also makes the region more 36 at risk in the face of climate change by imposing a climate-vulnerable crop to the 37 majority. With cooperation and conformation a household would plant maize and share 38 the produce with the neighbours that have provided the labour to farm the land. This 39 used to happen more often in the past when the farm clubs were popular, before the 40 6 reforms, although mainly for cash crops. With cooperation and differentiation a 1 household would maximise crop diversity choosing a mixed cropping pattern with 2 different proportions of maize, pigeon pea, sweet potato, cassava, fruit trees, and 3 vegetables. With differentiation and competition a household would maximise profit by 4 planting cash crops, such as tobacco, peppers, or cotton. The emphasis on profit rather 5 than on nutrients makes it a risky paradigm from a food security perspective, especially 6 if we assume that self-sufficiency is relevant at the village level, as opportunity cost may 7 drive agricultural entrepreneurs to invest on non-food crops beyond the social optimal 8 level (e.g. Anderman et al. 2014). 9 10 Table 1. Behavioural compass: village food security storylines. 11 2nd Quadrant: Competition & Conformation The household produces the socially preferred crop, maximizing possible yield in order to have more commodity to be exchanged at the market. Seed and fertilizer subsidies influence this behaviour. 1st Quadrant: Cooperation & Conformation The household produces the socially preferred crop, after which the produce is aggregated to that of other producers belonging to a farm club, and shared equitably. 3rd Quadrant: Competition & Differentiation The household chooses a cropping pattern that maximizes income, with the objective of obtaining a more valuable commodity (diversified quality) to be exchanged at the market (e.g. cash crop). 4th Quadrant: Cooperation & Differentiation The household plants crops in order to maximize crop diversity, differentiating cropping pattern vs. its neighbours. Later food exchanges increase diet diversity. 12 13 2.2 Data and model formalization 14 Household-level data for a village in Southern Malawi was collected over a period of 15 four days in July 2015 as part of the larger research project (Schreckenberg et al. 16 2016). Four trained Malawian enumerators used a household questionnaire to collect 17 information on farming practices, crops planted, harvested and sold, other income 18 generating activities, perceived food security, and socio-demographic characteristics of 19 the households. After a village mapping exercise, in which three village representatives 20 listed the household heads and mapped their locations, all households (N = 46) were 21 selected (census) to participate. This same dataset was also used to calibrate an agent22 based model to simulate the community food security into the future (Dobbie et al. 23 2018). In this study we utilize the information on the crops planted, harvested and sold, 24 which include: maize, groundnuts, tubers, pigeon peas, peas, sorghum, beans, soy, 25 vegetables, and cotton. 26 We propose a game theory model for describing the household cropping strategies. We 27 assume that each household is able to change its main crop type, at each time step, 28 7 and that there is only one planting season per year. The crop selection depends on two 1 main factors, which describe personal and economic behaviour, as per behavioural 2 compass. The first one is homophily, which determines the household tendency to 3 imitate its neighbours and is therefore the parameter corresponding to the 4 differentiation-conformation axis (i.e. personal behaviour). The second one describes 5 other-regarding preferences, the influence of community level satisfaction in the 6 household decision, and corresponds to the competition-cooperation axis (i.e. economic 7 behaviour). The main rules of the game are given in the following bullet points: 8 9 ● The topology of interactions is given by an initial network of N nodes, each node 10 representing one household, whose connections, represented in a matrix A with 11 elements 𝑎𝑖𝑗, remain constant in time. 12 ● Each node has a state, σi, which represents the household chosen main crop, 13 and is updated asynchronously at each time step. 14 ● The state, σi ∈ {1, S}, with S being the total number of crops, changes at each 15 time step according to the best response mechanism. This, implies that each 16 household tests all the possible crops and selects the one that maximizes its 17 payoff. Crops are initialized according to a uniform distribution. 18 ● The other-regarding preferences parameter α ∈ [0, 1], weights the importance of 19 the community in the strategy decision for each household. Its value is fixed in 20 time. 21 ● The homophily parameter, h ∈ [0, 1] represents the desired proportion of 22 neighbours matching the same crop planting choices. The homophily parameter 23 is also fixed in time. 24 ● At each time step all households update their state σi to maximize their payoff 25 (last bullet point). The total number of time steps or iterations is given by T. 26 ● The satisfaction function si ∈ [0, ki], where ki is the degree of node i, depends on 27 the distance ∆i, which measures how close are the links of a node from their 28 optimal configuration. In other words, it retrieves the distance with respect to the 29 number of connected households with same crop, according to h. 30 31 (1) 𝛥𝑖 = |ℎ − 1 𝑘𝑖 ∑ 𝑗𝛿𝜎𝑖𝜎𝑗 𝑎𝑖𝑗| 32 33 (2) 𝑠𝑖= 𝑘𝑖 𝑒 −∆𝑖 34 35 ● The payoff function Πi ∈ [0, max ki] depends on the individual and the community 36 satisfactions, weighted by the individual other-regarding preferences parameter. 37 8 Therefore the payoff is not a purely economic measurement, but it depends on 1 both economic and personal behaviour. 2 3 (3) 𝛱𝑖= (1 − 𝛼)𝑠𝑖 + 𝛼 𝑁 𝛴 𝑠𝑖 4 5 The homophily parameter can capture both coordination (h ≈ 1) and anti-coordination 6 dynamics (h ≈ 0) and the situations in between. The last two points indicate that 7 individual satisfaction is subjective to each household and driven by own level of 8 homophily: own satisfaction is assessed in comparative terms. Other households are 9 considered in terms of their state (i.e. the planted crop) and we assume full knowledge 10 of nodes state within the network. 11 Payoff combines individual and community satisfaction; here the other-regarding 12 preferences parameter is key: the higher the more important will be the satisfaction of 13 others. This formalization allows the consideration of two scales of concern (individual 14 and communal) in the objective function driving household behaviour. 15 We consider cooperation as driven by the importance assigned to the community in 16 terms of neighbours’ satisfaction. In this framework, high other-regarding preferences (α 17 ≈ 1) means to adapt crop selection so that neighbours’ satisfaction can maximize 18 individual payoff in addition to individual satisfaction. Low other-regarding preferences 19 (α ≈ 0) means to not consider neighbours’ satisfaction. Regardless of this parameter all 20 households maximize their own payoff and thus are modelled as self-interested 21 individuals. 22 23 In the following section, we describe the model behaviour by means of numerical 24 simulations on artificial networks of different kinds. Then, in Results, we apply the model 25 to the data of a village in Southern Malawi and summarize the main findings. 26 27 2.3 Model behaviour 28 The game theory model has three control parameters, namely the other-regarding 29 preferences parameter α, the homophily parameter h, and the number of possible node 30 states S, which represent the main crop types. The model can be implemented on 31 different network topologies (Latora, Nicosia and Russo 2018). Here we compare three 32 networks, namely Regular Lattices (RL), Random Networks (RN) and Scale Free 33 Networks (SF), with the same number N = 100 of nodes and the same average degree 34 <k> = 4. Regular Lattices of <k> = 4 correspond to a uniform distribution of the nodes in 35 the plane, where each node is connected to its two nearest neighbours in each 36 dimension. In order to have <k> = 4 for all the nodes, we considered periodical 37 boundaries. Random Networks (Erdos and Rényi 1960) of N nodes and <k> = K are 38 achieved by fixing the link probability to K/(N-1). Their degree distribution can be well 39 approximated with a Poisson probability distribution. Scale Free Networks (Barabási 40 9 and Albert 1999) are characterized by a power law decay in the degree distribution, 1 allowing the existence of hubs, or nodes with a high degree. 2 Simulations have been performed for α ∈ {0, 0.5, 1}, h ∈ {0, 0.5, 1}, and S ∈ {2, 3, 4, 5}, 3 and the results have been averaged over ensembles of 100 different network 4 realizations. In each simulation we have run the game for T = 20 iterations under a best 5 response update rule, which assumes that each of the nodes tests all the possible 6 states and selects the one with the highest payoff. The value of T = 20 is enough to 7 guarantee the convergence of the system to the equilibrium state. The combination of 8 the best response update mechanism and the small size of network explains why most 9 of the information of the equilibrium solution is provided for small values of T (e.g. T = 10 5). 11 12 The asymptotic values of payoff and dominating state fraction are shown in Fig. 2 and 13 Fig. 3. The first is the value of payoff when the system approaches equilibrium, the latter 14 represents the proportion of households within the village matching the same crop 15 planting choices. Both figures show the average values of output as a function of other16 regarding preferences and homophily for different topologies and values of S. The 17 columns correspond to the topology, RL, RN and SF, while the rows correspond to the 18 total number of states. In each subplot the colour in the white-blue scale represents the 19 value of the payoff for the specific selection of α and h. 20 21 From the simulations we can infer the most relevant features of the model: 22 ● The relation between S (number of states/crops) and <k> (average degree of the 23 network) determines the strategy with the highest payoff. For low S/<k>, 24 conformation is more favourable than differentiation. The opposite is true for 25 higher values of S/<k>. 26 ● The system is polarized between consensus (high dominating state fraction (df)), 27 for h = 1, and dissension (low df), for h = 0. The intermediate situations imply 28 lower payoffs. This indicator suggests the existence of a phase transition 29 between these two regimes. 30 ● Cooperation is especially relevant when the community converges on crop 31 planting choices: it is a necessary ingredient for the success of households that 32 conform. The opposite is not true. Differentiation can survive without cooperation. 33 ● The dynamics are stable except for the combination of h = 0.5 and α ∈ {0.5, 1}. 34 We expand on this finding when studying the empirical case study. The higher S, 35 the lower the fluctuations. The tendency towards stability depends on the 36 topology. RL achieve it faster than SF, and these are faster than RN (See Online 37 Resource 1: Fig. 6 and Fig. 7. Here we plot the absolute value of the discrete 38 derivative at T = 20, as a mechanism for evaluating if the result corresponds to 39 an equilibrium situation). 40 16 It is reasonable to think that that agri-food systems can be mapped within a continuum 1 between these two paradigms and dynamically move in one direction or the other 2 according to social and ecological conditions. For example, a country of the Global 3 South like Malawi has been witnessing a shift towards IFA under the liberalisation 4 reforms that happened since the eighties (Dorward and Kydd 2004). At the same time 5 NFA is gaining momentum in the cities of the Global North, where the masses have 6 been relying on the IFA paradigm, almost since the industrial revolution (Ilieva 2016). 7 8 In this study we adopted a village-centric perspective and a system approach, whereby 9 the system performance is stemming from the behaviour of and interactions among the 10 individual components, to explore how household behaviour can affect food security at 11 the community level. We implicitly considered that self-sufficiency in food production is a 12 positive feature for a rural agri-food system of the Global South, but here we discuss its 13 consequences. 14 15 Because satisfaction, and thus payoff, is not measured in terms of nutritional 16 requirements we can only suggest how human behaviour is expected to influence food 17 security. Indeed, our model is primarily concerned with behavioural dynamics in crop 18 adoption and the level of dissension/consensus within the community. To draw 19 conclusions on the nutritional level of community food security, we would have to couple 20 this behavioural model with agricultural and environmental modules (see Balbi et al. 21 2015) and a module of exchanges, including market interactions and bartering 22 dynamics, possibly considering food imports and exports. In the work presented here 23 we assume that all the households have the same nutritional requirements, and that the 24 production costs and economic profit are equal for all crop types. We also don’t 25 consider any difference in the crops that are consumed by the producers and the ones 26 that are sold in market. Accessing food via production or via market doesn’t affect the 27 benefits and food price is exactly the production cost, thus the community is assumed to 28 operate through direct food exchanges. 29 30 Although past studies have emphasized the role of cooperation in self-organized 31 systems (see Perc et al. 2017 for a review), our game theory model suggests that 32 homophily—the continuum between conformation and differentiation—can play a 33 relevant role. Both conformation and differentiation can be successful strategies leading 34 to high payoffs at the individual level, but according to which prevails, they drive the 35 system to very different community outcomes. In the first case, we have a community 36 converging on crop planting choices, in the second case the community maintains crop 37 diversity. 38 39 17 Both cases can happen regardless of the importance given to others satisfaction, which 1 is what drives cooperation in our model. This can have significant consequences in 2 terms of food security at the village level. The first case is a specialization case that 3 could succeed only if the village is connected to the outside via food exchange 4 mechanisms (e.g. regional food markets) through which households can sell their own 5 produce and buy the lacking nutrients. The alternative is a community that can perhaps 6 self-sufficiently meet its own caloric needs, but not a healthy diet diversity. The second 7 case is a diversification case that can succeed under a self-sufficiency scenario, but 8 requires internal exchanges to happen (via local market or bartering) within the village 9 to meet the necessary diet diversity at the household level. 10 11 However, the case study we analysed in this article, using empirical data from a rural 12 village in Southern Malawi, has provided more insights about the role of cooperation. 13 Our analysis suggests that the village is located in a very particular region of the 14 parameters space where the influence of other-regarding preferences is relatively high 15 for any point in the curve presented in Figure 5. This curve acts as the boundary 16 between the regions representing the two cases mentioned above. Thus cooperation 17 has a key role to play: the higher other-regarding preferences the more the system is 18 driven to consensus, and the analysis of expected payoffs suggests that the village is 19 better off with consensus. Partial dissension can survive for other-regarding preferences 20 below 0.9. It thus seems that sacrificing some degree of cooperation in favour of more 21 crop diversity would reduce the individual expected payoff, although it would make 22 sense from a nutritional diversity perspective at the community level. Indeed, the current 23 village situation is confirming these insights: in this part of Malawi food exchanges are 24 largely market-driven with regional markets playing an important role and maize is by far 25 the socially preferred crop (Schreckenberg et al. 2016). Moreover, the results reinforce 26 the idea of close to perfect knowledge among households at the village scale, and the 27 convenience of considering others satisfaction in own payoff, thus being more 28 cooperative. 29 30 One additional limitation of our work is the assumption about the real topology of the 31 village network. Although we expect an error of approximation of the actual social 32 network, our assumption is based on previous studies and the sensitivity analysis in 33 Section 2.3 explains the similarity of model behaviour when the topology is altered. 34 Accordingly, we expect the relation between the other-regarding preferences and the 35 homophily parameters to be robust against changes in the topology. For what concerns 36 village-level results, we expect them to hold for small communities (e.g. 30 to 100 37 households corresponding to 100-500 individuals) with a relatively high degree of 38 mutual knowledge among community members and parcelled access to the resource— 39 in this case farming land. 40 18 Further analysis will explore the explanatory power of the behavioural compass at 1 multiple spatial scales. We argue that the behavioural compass could be extended on a 2 third axis, thus representing a three dimensional space, to capture multiple levels of 3 governance (Lebel et al. 2006) and their institutions (from the microto macro-scale), in 4 line with the thinking of Waring et al. (2015). 5 6 5. Conclusions 7 8 The main finding of our simulations suggests that individual and community success is 9 more strongly related to personal behaviour than to economic behaviour. In other 10 words, how individuals relate to social norms has greater effect than how much they 11 care about the success of their neighbours (i.e. our working definition of cooperation). 12 While we expected the latter to be the main driving force in determining community 13 success, we found that cooperation is only necessary for community success when the 14 community converges on crop planting choices. On the contrary, differentiation, the 15 affirmation of the individual unique identity, can succeed with or without cooperation. 16 17 Under a food security perspective, when there is a diversity of options about crop 18 adoption, differentiation is likely to deliver more positive outcomes. Only when the 19 options are few, then conformation is a winning strategy. In a situation such as a rural 20 region of the Global South—in this article we used the example of a village in Southern 21 Malawi—this translates into clear policy implications about the role of food sovereignty, 22 and in particular food knowledge sharing. Food security, among other things, depends 23 on a variety of nutrients for a healthy and balanced diet, thus enabling biological 24 diversity in agriculture can be beneficial. 25 Instead, it seems that the policy reforms of the past have pushed countries like Malawi, 26 similarly to other rural African regions, to specialize their agricultural production towards 27 one or two socially preferred crops, via crop-specific subsidies and other incentives. 28 This made sense under a scenario of market liberalisation that was going to deliver 29 efficient redistribution of food commodities among different regions (for food security 30 purposes), thus allowing beyond-system food transactions. Evidence from the ground 31 suggests that this paradigm has only partially delivered its promises (see Dorward and 32 Kydd 2004) and that stimulating more diverse crop adoption at the household level is a 33 reasonable strategy to improve community food security. 34 We argue that this process is not likely to happen endogenously, because, in the 35 current setting, households regard conformation as a safer strategy, due to the 36 dominant social norms on crops. 37 38 19 Acknowledgments 1 This research is supported by the Basque Government through the BERC 2018-2021 program 2 and by the Ikertzaile Doktoreentzako Hobekuntzarako doktoretza-ondoko Programa, and by 3 Spanish Ministry of Economy and Competitiveness MINECO through BC3 María de Maeztu 4 excellence accreditation MDM-2017-0714 and by the Leverhulme TrustResearch Fellowship 5 “CREATE: the network components of creativity and success. This work was also inspired and 6 partially funded by the ASSETS (Attaining Sustainable Services from Ecosystems) project 7 (NERC/ESPA NE-J002267-1). The authors wish to thank the advice and support of Marije 8 Schaafsma (University of Southampton), who provided the dataset analysed, Kate 9 Schreckenberg (King's College London), who led the work on the participatory rural appraisals 10 within the ASSETS project, Samantha Dobbie for the extensive work on sustainable livelihoods 11 achieved during her doctoral research at the University of Southampton, and Ortzi Akizu12 Gardoki (University of the Basque Country) for the design of Figure 1. 13 14 References 15 Anderman TL, Remans R, Wood SA, DeRosa V, DeFries RS (2014) Synergies and tradeoffs 16 between cash crop production and food security: a case study in rural Ghana. Food 17 Security 6(4):541–554. doi: 10.1007/s12571-014-0360-6 18 Axelrod R., Hamilton WD (1981) The evolution of cooperation. Science 211(4489):1390–1396. 19 doi: 10.1126/science.7466396 20 Balbi S, del Prado A, Gallejones P, Geevan CP, Pardo G, Pérez-Miñana E, Manrique R, 21 Hernandez-Santiago C, Villa F (2015) Modeling trade-offs among ecosystem services in 22 agricultural production systems. Environ Modell Softw 72:314-326. doi: 23 10.1016/j.envsoft.2014.12.017 24 Barabási AL, Albert R (1999) Emergence of scaling in random networks. Science 286(5439): 25 509-512. doi: 10.1126/science.286.5439.509 26 Burchi F, De Muro P (2016) From food availability to nutritional capabilities: Advancing food 27 security analysis. Food Policy 60:10–19. Doi: 10.1016/j.foodpol.2015.03.008 28 Chambers R, Conway G (1991) Sustainable rural livelihoods: Practical concepts for the 21st 29 century. IDS Discussion Paper 296. 30 http://opendocs.ids.ac.uk/opendocs/handle/123456789/775 Accessed 09 March 2020. 31 Chappell MJ, LaValle LA (2011) Food security and biodiversity: can we have both? An 32 agroecological analysis. Agric Hum Values 28(1):3-26. doi: 10.1007/s10460-009-9251-4 33 Chinsinga B, Chasukwa M (2012) Youth, agriculture and land grabs in Malawi. IDS Bulletin 34 43(6): 67–77. doi: 10.1111/j.1759-5436.2012.00380.x 35 Connolly-Boutin L, Smit B (2016) Climate change, food security, and livelihoods in Sub-Saharan 36 Africa. Reg Environ Change 16(2):385–399. doi: 10.1007/s10113-015-0761-x 37 Deutsch M (1949) An experimental study of the effects of cooperation and competition upon 38 group process. Human Relations 2(3):199–231. doi:10.1177/001872674900200301 39 Devereux S (2016) Social protection for enhanced food security in Sub-Saharan Africa. Food 40 Policy 60: 52–62. doi: 10.1016/j.foodpol.2015.03.009 41 Dobbie S, Schreckenberg K, Dyke J, Schaafsma M, Balbi S (2018). Agent-based modelling to 42 assess community food security and sustainable livelihoods. Journal of Artificial 43 Societies and Social Simulation 21(1):1-23. doi: 10.18564/jasss.3639 44 20 Dobbie S, Balbi S (2017) Design of an empirical agent-based model to explore rural household 1 food security within a developing country context. In: Jager W, Verbrugge R, Flache A, 2 de Roo G, Hoogduin L, Hemelrijk (eds.), Advances in Social Simulation 2015, Cham: 3 Springer, pp 81–94. doi: 10.1007/978-3-319-47253-9_7 4 Dorward A, Kydd J (2004) The Malawi 2002 food crisis: the rural development challenge. The 5 Journal of Modern African Studies 42(03):343–361. doi:10.1017/S0022278X04000229 6 Erdos P, Rényi A (1960) On the evolution of random graphs. Publ Math Inst Hung Acad Sci 7 5(1):17-60. 8 Fehr E, Fischbacher U (2002) Why social preferences matter–the impact of non-selfish motives 9 on competition, cooperation and incentives. The economic journal 112(478):C1–C33. 10 doi: 10.1111/1468-0297.00027 11 Gliessman, S. (2009) Editorial: A Framework for the Conversion to Food System Sustainability. 12 Journal of Sustainable Agriculture 33(1):1-2. doi: 10.1080/10440040802587249 13 Gondwe GE (2014) 2014 Millennium Development Goal Report for Malawi. Ministry of Finance, 14 Economic Planning and Development, Lilongwe, Malawi. 15 http://www.undp.org/content/dam/malawi/docs/general/Malawi_MDG_Report_2014.pdf 16 Accessed 09 March 2020. 17 Headey, D, Ecker O (2013) Rethinking the measurement of food security: From first principles 18 to best practice. Food Security 5(3):327–343. doi: 10.1007/s12571-013-0253-0 19 Ilieva, RT (2016) Urban food planning: seeds of transition in the global north. Routledge. 20 Kahneman, D (2011) Thinking, fast and slow. Macmillan. 21 Kremen C, Miles A (2012) Ecosystem services in biologically diversified versus conventional 22 farming systems: benefits, externalities, and trade-offs. Ecology and Society 17(4):40. 23 doi: 10.5751/ES-05035-170440 24 Latora V, Nicosia V, Russo G (2018) Complex Networks: Principles, Methods and Applications. 25 Cambridge University Press. 26 Lebel L, Anderies JM, Campbell B, Folke C, Hatfield-Dodds V, Hughes TP, Wilson J (2006) 27 Governance and the capacity to manage resilience in regional social-ecological systems. 28 Ecology and Society 11(1):19. 29 Lester JC (1994) The evolution of the political compass (and why libertarianism is not right30 wing). Journal of Social and Evolutionary Systems 17(3):231–241. 31 Ligon E, Schechter L (2012) Motives for sharing in social networks. J Dev Econ 99(1):13–26. 32 doi: 10.1016/j.jdeveco.2011.12.002 33 Maddox WS, Lilie SA (1984). Beyond liberal and conservative: Reassessing the political 34 spectrum. Cato Institute. 35 NSO (2012) Malawi Integrated Household Survey 2010-11 (IHS3). National Statistical Office, 36 Lilongwe. http://microdata.worldbank.org/index.php/catalog/1003 Accessed 09 March 37 2020. 38 Orr A, Mwale B, Saiti-Chitsonga D (2009) Exploring seasonal poverty traps: The ‘Six-Week 39 Window’ in Southern Malawi. The J Dev Stud 45(2):227–255. doi: 40 10.1080/00220380802265330 41 Ostrom E (2009) A general framework for analyzing sustainability of social-ecological systems. 42 Science 325(5939):419–422. doi: 10.1126/science.1172133 43 21 Patel R (2009) Food sovereignty. The Journal of Peasant Studies 36(3):663–706. doi: 1 10.1080/03066150903143079 2 Perc M, Jordan JJ, Rand DG, Wang Z, Boccaletti S, Szolnoki, A. (2017). Statistical physics of 3 human cooperation. Physics Reports, 687, 1-51. doi: 10.1016/j.physrep.2017.05.004 4 Petrini C (2003) Slow food: The case for taste. Columbia University Press. 5 Rausser G, Zilberman D, Kahn G (2015) An alternative paradigm for food production, 6 distribution, and consumption: A noneconomist’s perspective. Annual Review of 7 Resource Economics 7(1):309–331. doi: 10.1146/annurev-resource-100913-012549 8 Sahley C, Groelsema B, Marchione T, Nelson D (2005) The governance dimensions of food 9 security in Malawi. USAID Bureau of Democracy, Conflict, and Humanitarian Assistance, 10 Washington, DC. 11 https://sarpn.org/documents/d0001649/P1998-USAID_Malawi_Sept2005.pdf Accessed 12 09 March 2020 13 Schmidhuber J, Tubiello FN (2007) Global food security under climate change. PNAS 14 104(50):19703–19708. doi: 10.1073/pnas.0701976104 15 Schreckenberg K, Torres-Vitolas CA, Willcock S, Shackleton C, Harvey CA, Kafumbata D 16 (2016) Participatory Data Collection for Ecosystem Services Research: A Practitioner’s 17 Manual. ESPA Working Paper Series No 003/June 2016. ISSN 2058-9875 18 http://www.espa.ac.uk/files/espa/PRA-Manual.pdf Accessed 09 March 2020 19 Spielman DJ, Davis K, Negash M, Ayele G (2011) Rural innovation systems and networks: 20 Findings from a study of Ethiopian smallholders. Agric Hum Values 28(2):195–212. doi: 21 10.1007/s10460-010-9273-y 22 United Nations (2015a) Transforming our world: The 2030 agenda for sustainable development. 23 Resolution adopted by the General Assembly. 24 United Nations (2015b) World urbanisation prospects: The 2014 revision. The Population 25 Division of the Department of Economic and Social affairs of the United Nations. 26 Waring TM, Kline MA, Brooks JS, Goff SH, Gowdy J, Janssen MA, Smaldino PE, Jacquet J 27 (2015) A multilevel evolutionary framework for sustainability analysis. Ecology and 28 Society 20(2):34. doi: 10.5751/ES-07634-200234 29 Watts DJ, Strogatz SH 1998. Collective dynamics of ‘small-world’ networks. Nature 393(6684): 30 440–442 31 Witkin HA, Price-williams D, Bertini M, Christiansen B, Oltman PK, Ramirez M, Meel JV (1974) 32 Social conformity and psychological differentiation. International Journal of Psychology 33 9(1):11–29. 34