Herders, farmers and land, and how a conflict model is built

Mobile herders and settled farmers share dry lands in many parts of the world, and their dealings run from trade and grazing agreements to quarrels over trampled crops and, at times, violence on a large scale. The question an agent-based model asks is narrow and useful: what individual needs and rules, acting together, are enough to turn a scattering of small disputes into collective violence, and what changes would stop that from happening?

An agent-based model (ABM) is a simulation in which many individual decision-makers, the agents, follow simple rules on shared ground, and the large-scale pattern is whatever emerges from their interactions. If the general idea is new, start with the explainer on agent-based models in anthropology. This note builds a herder and farmer model step by step, runs a small piece of it by hand, and then turns to a published model and the real setting it was built for.

Building the model in five decisions

Every model of this kind has to answer the same five questions. The numbers below belong to an imaginary region, chosen so the arithmetic stays small.

1. The land: a grid of cells, say 10 by 10. Each cell is pasture or field. Pasture grows grass at a rate that depends on rainfall; fields grow crops for the household that farms them.
2. The agents: farming households, each tied to its fields, and herding camps, each with a herd that must eat every month and can move.
3. The movement rule: each month a herd moves to the best pasture it can see. If the pasture it can reach falls short of what the herd needs, the herders take the shortfall from fields.
4. The response rule: when a herd enters a field, the farming household may let it pass, ask for compensation, or start a dispute. Disputes leave a grievance that the household remembers.
5. The collective rule: households with enough grievances may join a raid, but only if enough others are joining. This is the step where individual anger becomes group action, and it is the hardest one to justify from evidence.

A fair model should also let the two groups cooperate, for instance by letting herds graze harvested fields by agreement, so that conflict is something the model can produce, not the only thing it can do.

Here are the parameters for our toy version.

A run you can do by hand

Start with each herd ranging over 3 pasture cells. Then suppose farming expands and each herd loses one cell to fields, leaving 2. For one herd in one month:

Now add up a year for all 5 herds.

• With 3 cells: wet months give 0, dry months give 6 × 0.5 = 3 per herd. Total 5 × 3 = 15 expected disputes.
• With 2 cells: wet months give 6 × 0.25 = 1.5, dry months give 6 × 0.75 = 4.5, so 6 per herd. Total 5 × 6 = 30.

Losing a third of the pasture doubles the expected disputes, because a wet year on the smaller range now behaves like a dry year used to. That is the first lesson such models teach: the response to steady change can be sharply non-linear.

Now try a policy inside the model. Suppose local mediation lowers the dispute chance from 0.25 to 0.1. With 2 cells the total becomes 30 × (0.1 / 0.25) = 12, which is below the original 15. In the toy world, an institution that settles field entries offsets the lost land. A model is good at exactly this kind of "what if".

From disputes to a raid

The collective rule deserves its own small example. Mark Granovetter's threshold model (1978) gives each person a threshold: the number of others who must already be acting before that person joins. Take ten households with thresholds 0, 1, 2, and so on up to 9. The household with threshold 0 acts alone; that makes one, which is enough for the household with threshold 1; that makes two, and so on until all ten have joined.

Change one number. If the household with threshold 1 had threshold 2 instead, the first household acts, nobody else has their threshold met, and the cascade stops at one. Two nearly identical villages, two completely different outcomes. Thresholds like these are close to impossible to measure in real communities, which is why a modeler should report how results change across many plausible threshold sets, not one.

What a model like this can and cannot show

It can show:

• that a set of rules is sufficient to produce a pattern, such as raids clustering in dry years;
• where the tipping points are, and which parameters move them most;
• how interventions such as new water points, open livestock routes or mediation change the outcome in the model world.

It cannot show:

• that those rules are what actually drove events, since other mechanisms may produce the same pattern;
• values for parameters nobody has measured, such as dispute chances and thresholds;
• the influence of anything left out, which in real conflicts usually includes the state, armed groups, weapons, markets and the history of land rights.

A run that ends in violence shows that the rules can produce violence. It does not show that the real outcome was inevitable.

The NOMAD model

Lawrence Kuznar and Robert Sedlmeyer (2005) built a simulation of this general type, called NOMAD, for interaction between pastoral nomads and sedentary farmers, with settings that can be adjusted to fit a particular environment and society. The model gives environmental and material conditions the leading role in individual decisions, and treats large-scale raiding as something that can arise from those decisions rather than something written in from the start. The authors acknowledged ethnic and political factors, national and international, and took their runs as support for scholars who trace the crisis to land use failing under population growth, poor habitat and severe ecological stress.

The real setting, briefly

Darfur lies in western Sudan. The war there began in 2003 and had displaced more than two million people by the time of Kevane and Gray's 2008 study. One explanation that drew attention tied it to drought and shrinking land. Michael Kevane and Leslie Gray (2008) checked rainfall records and found that the main drop in rainfall came around 1971, three decades before the war, with no clear trend in the thirty years just before it. Rainfall data, they concluded, give only weak support to the climate explanation. Research on land-use conflicts elsewhere in the Sahel points the same way. Tor Benjaminsen and colleagues (2012), working with court records from Mali's inner Niger delta, found little link to climate variability and pointed instead to farms spreading across herders' routes, a growing political vacuum, and corruption among officials.

None of this makes the model wrong. It moves the argument to the inputs. In a herder and farmer ABM, scarcity is usually something the modeler sets, and the debate over whether scarcity drove a real conflict is a debate about that setting. A careful study runs the model with scarcity alone, with institutional and political factors alone, and with both, and reports all three.

Where this kind of modeling went next

Two developments matter most for anyone reading such models now. The first is transparency. The ODD protocol (Overview, Design concepts, Details), proposed by Volker Grimm and colleagues in 2006 and revised since, gives a standard way to describe an ABM so that someone else can rebuild it and check it. The second is a widening set of models of pastoral life and conflict, from Joshua Epstein's 2002 models of civil violence, one of which simulates communal violence between two groups under a central authority, to agent-based models of pastoral nomad households such as the MASON HouseholdsWorld model of Claudio Cioffi-Revilla, J. Daniel Rogers and Maciek Latek (2010). More on the family of approaches is collected under models in anthropology and in our overview of research areas.

The toy model above fits on one page, and that is the point of it: change the regrowth from 2 to 3 in the dry season, run the table again, and watch how much of the conflict was weather and how much was the rule.

Sources and further reading

• Kuznar, Lawrence A., and Robert Sedlmeyer. 2005. "Collective violence in Darfur: an agent-based model of pastoral nomad/sedentary peasant interaction." Mathematical Anthropology and Cultural Theory 1(4). https://escholarship.org/uc/item/67x4t8ts
• Epstein, Joshua M. 2002. "Modeling civil violence: an agent-based computational approach." Proceedings of the National Academy of Sciences 99 (suppl. 3): 7243-7250. https://doi.org/10.1073/pnas.092080199
• Granovetter, Mark. 1978. "Threshold models of collective behavior." American Journal of Sociology 83(6): 1420-1443. https://doi.org/10.1086/226707
• Grimm, Volker, et al. 2006. "A standard protocol for describing individual-based and agent-based models." Ecological Modelling 198(1-2): 115-126. https://doi.org/10.1016/j.ecolmodel.2006.04.023
• Cioffi-Revilla, Claudio, J. Daniel Rogers, and Maciek Latek. 2010. "The MASON HouseholdsWorld model of pastoral nomad societies." In Simulating Interacting Agents and Social Phenomena, 193-204. Springer. https://doi.org/10.1007/978-4-431-99781-8_14
• Kevane, Michael, and Leslie Gray. 2008. "Darfur: rainfall and conflict." Environmental Research Letters 3(3): 034006. https://doi.org/10.1088/1748-9326/3/3/034006
• Benjaminsen, Tor A., Koffi Alinon, Halvard Buhaug, and Jill Tove Buseth. 2012. "Does climate change drive land-use conflicts in the Sahel?" Journal of Peace Research 49(1): 97-111. https://doi.org/10.1177/0022343311427343