Environmental crime
Illegal fishing, pollution, trafficking. Which enforcement actually deters, and which only moves the problem elsewhere.
Causal inference for people and nature
Better worlds have a but-for.
We measure what interventions cause, and calculate which ones to make next, for organizations that protect people and nature.
Illegal fishing effort in a patrolled zone
The difference is the counterfactual.
Before-and-after comparisons mislead. Illegal fishing drops after a patrol programme starts, but it might have dropped anyway. A credible answer needs an estimate of the world without the programme, and a measure of how sure we can be.
- 01 · Frame
Turn the question into a causal model
What you did, what you wanted to change, and what else could explain the outcome. Every assumption is written down where you can challenge it.
- 02 · Estimate
Build the world without it
From your data and open data, with methods that fit the setting: synthetic control, difference-in-differences, Bayesian models, causal discovery.
- 03 · Improve
Report the effect, then act on it
The size of the effect, its uncertainty, and what would have to be true for the conclusion to change. Then what to keep, fix or stop.
Our work
First case study
West Africa: when a port state joins the port agreement, where do foreign vessels go?
220,254 port arrivals in 22 coastal states, 2015 to 2026. A staggered event study and a hierarchical Bayesian model, with counterfactuals, uncertainty and the power to detect each effect.
Open the study (access by invitation) →Where we work
For organizations that act to make a better world.
Biodiversity conservation
Protected areas, restoration, species programmes, measured against what the land or sea would look like without them.
Public health
Prevention programmes and health policies, judged on the outcomes they cause rather than the activity they report.
Justice
Policies, tools and evidence in the justice system, examined for what they cause, and for whom.
How we work
- Assumptions in the open
- Every estimate comes with the causal model behind it, so anyone can see what it rests on.
- Uncertainty, always
- A number without its range is a guess dressed up. We report both.
- Findings, not accusations
- Probabilities stay probabilities. Naming anyone takes evidence, review and a right of reply.
- Reproducible
- Code and methods can be rerun and checked, within what partners allow us to share.
Bring us a question.
Tell us about an intervention you want to evaluate or improve. A rough question is fine. Framing it well is part of the work.
Or write to [email protected]