Présentatioon

Based on my work with youth, women, mining communities, and funding in Kenya/Africa, here are the *3 real-world problems where agent-based modelling (ABM) would give you the most leverage*:

*Top candidate: "How funding delays + trust breakdowns cause youth mining cooperatives to collapse"*

*Why ABM fits perfectly*:  

You have _individual agents_ (youth miners, women traders, funders, middlemen, chiefs) with different goals, trust levels, and cash flow. Their interactions create system outcomes like cooperative survival or collapse that are hard to predict with spreadsheets.

*The problem*: Funding arrives late/unpredictably → youth can’t buy shared equipment → miss bulk buyer deadlines → lose trust → take exploitative loans → cooperative fragments → funders see "failure" and pull out. Classic feedback loop.

*How ABM helps*: Simulate "what if funding was monthly vs quarterly?" or "what if 30% of youth leave after 2 late payments?" Test interventions before risking real money.

---

*Other strong candidates for your context:*

Real-world problem Why ABM helps Key agents to model

**2. GBV + economic empowerment program uptake** Women decide to join/stay/leave programs based on safety, peer influence, transport cost, husband reaction. ABM captures social network effects Women, husbands, chiefs, WEL staff, perpetrators, peers

**3. Mercury use vs. safe mining adoption** Miners copy neighbors. If 3 friends use retorts safely + earn more, you switch. If funders stop retort subsidies, everyone reverts. Tipping points matter Miners, mercury traders, trainers, buyers, children

---

*To make an ABM useful + credible for Problem #1, you’d need:*

*1. Data - minimum viable set*

- *Individual level*: 50-100 anonymized profiles of youth miners: age, gender, weekly income, debt level, who they trust, distance to market

- *Network data*: "Who does Mary trade with?" "Who borrows from who?" Even 3-5 key links per person helps

- *Financial flows*: Timing + amount of last 12 months of funding, equipment costs, buyer payment cycles

- *Outcome data*: Which groups survived/collapsed in your county past 3 years, and why

_Don’t need perfect data_. Start with focus groups in Kisumu/Migori to map 1 cooperative. Calibrate later.

*2. Assumptions you must state openly*

1. *Decision rules*: “Youth leaves coop if 2 consecutive payments missed AND debt >KSh 5,000” - test if 2 or 3 changes results

2. *Trust decay*: “Trust drops 20% per month without funder contact” - validate with community

3. *Copying behavior*: “60% chance to adopt safe practice if 2+ trusted peers use it” 

4. *External shocks*: Rainy season cuts income 40%, police raids happen 1x/year

_Rule_: Every assumption gets a sensitivity test. If result changes wildly when you tweak it, go collect real data on that point.

*3. Stakeholder inputs - who must co-design it*

1. *Youth miners + women*: Game it out with them. “If funding was late, what would you _actually_ do?” Their answers become your agent rules

2. *Funders/Uwezo Fund officers*: What triggers them to release/stop funds? Budget cycles, reporting needs

3. *Local chiefs/traders*: They control access, markets. Model their veto power

4. *WEL or similar NGOs*: What behavior change have they seen work in GBV/economic programs?

5. *County mining officers*: Legal/shutdown risks that break the system

_Credibility test_: Run a 1-hour “participatory simulation” where stakeholders move tokens on a board following the model rules. If they say “no, I’d never do that,” your model is wrong.

*Fast way to start in your context*

1. *Pick 1 county, 1 cooperative* - e.g. gold miners in Migori, 30 members

2. *Map it on paper first*: Draw agents, flows of cash/trust/ore in a 2-hour workshop

3. *Build v1 in NetLogo or GAMA* - free, used widely in African research. Takes 2 weeks if you have 1 tech person

4. *Validate*: Show animation to the cooperative. Do they laugh and say “that’s exactly us”? Or “that would never happen”? Iterate

*Biggest pitfall in Africa/community work*: Building a complex model with no community input. It becomes extractive research. If stakeholders co-build it, they’ll trust the results and use them to negotiate with funders/government of Kenya