Research programme
CEIS: Computational Evolutionary Institutional Science
CEIS examines how individually understandable choices can combine to produce unexpected patterns in organisations. It considers behaviour, group dynamics, incentives and rules.
It is a line of research, not a finished or validated theory.
Illustration · fictional mechanism
Sensible alone. Dysfunctional together.
The idea in plain language
People are not blank slates optimising a spreadsheet. They carry evolved capacities and biases: sensitivity to status, loyalty to groups, a talent for reading intentions, a poor feel for slow, diffuse risks. They have limited time and attention, so they use rules of thumb that mostly work. They respond to incentives, including informal ones such as reputation. And they form coalitions around shared interests, which change who gets heard.
Put many such people inside institutions — formal rules, budgets, norms, tools — and organisational behaviour emerges that nobody designed and few intended. Evolutionary lens: which practices persist, spread or die, and why? Computational lens: simple rules for individuals can be written down and simulated, to see which collective patterns they can and cannot produce. Institutional lens: rules and constraints shape which variations survive.
The ingredients we look at
When competent people produce a result nobody wants, we check these, one at a time, before we reach for a story about character.
- Incentives: what is rewarded and punished, formally and informally.
- Status: who gains or loses standing from an act, and how visible it is.
- Coalitions: who sides with whom, and who is left outside.
- Reciprocal exchange: favours owed and remembered. Who would be breaking an unwritten deal?
- Affordances: what the tools make easy. If a button exists, it gets pressed.
- Bounded rationality: limited time and attention, so rules of thumb.
- Formal and informal rules: what the policy says, and what people actually do.
- Emergence: the pattern that appears only when many people act at once.
A locally sensible act with a collectively poor result
Take a fictional organisation in which every team can mark its requests “urgent”. Each team has learned that unmarked requests wait. Marking is sensible for each team; collectively the label stops meaning anything, and the genuinely urgent are lost in the noise. Nobody is lazy or malicious. The pattern comes from incentives plus a shared tool plus limited attention.
Rival hypotheses and discriminating observations
- H1 · IncentiveTeams over-mark because attention is scarce. Observation: marking falls when a quota makes it costly.
- H2 · StatusMarking signals importance to others. Observation: marking is highest in teams with the lowest standing, regardless of workload.
- H3 · InformationTeams do not know what others have waiting. Observation: marking drops when queues are visible to all.
A fourth explanation lives in the exchange of favours: teams mark requests urgent for those who owe them, and the label becomes a currency. If so, flagging follows relationships, not workload. The point of CEIS is that these hypotheses differ in what they predict, so a small observation can tell them apart before an expensive reorganisation.
Lean tests
None of this needs software. Pull two months of the tracking log and sort it by team and by requester: that tests status and exchange. Show everyone the queue for one week: that tests information. Cap each team at five urgent flags for a fortnight: that tests incentive. Each takes days. Each can fail. If none moves the pattern, the explanation was none of these, and we say so.
Culture, alternatives and evidence
Cultures differ, and so do workable solutions: a practice that fits one setting may fail in another because the incentives, the history or the coalitions differ. CEIS therefore starts from a repertoire of alternatives and asks which survive the local evidence. It treats status and coalition dynamics as ordinary forces to be observed, not as embarrassing exceptions.
Model risk
A model is a simplification, and a simulation shows what is possible, not what is true. Typical risks: a model that fits because it has too many free parts; results that depend on an assumption nobody examined; confusing a plausible mechanism with a demonstrated one. We keep models small, state their assumptions, and treat predictions that fail as information. Simulated results from models do not count as empirical evidence about a particular organisation.
How we use it
CEIS raises questions that can be useful across different assignments: what explanations are possible, what observations would help distinguish them, and what might we expect to see? More on the Approach page.