Phase 1
Define the question
Clarify the decision, the audience, the required output, and what a useful answer would change.
Approach
Clear enough to review. Useful enough to act on.
The purpose of analysis is not to make uncertainty disappear. It is to establish what the evidence supports, make assumptions visible, and give decision-makers a dependable basis for action.
Phase 2
Understand and reconcile the data
Examine sources, definitions, timing, populations, missing information, transformations, and differences between reported results.
Phase 3
Design the analysis or system
Select methods appropriate to the evidence. Separate inputs and assumptions from calculations. Design the work so it can be reviewed and repeated.
Phase 4
Test assumptions and results
Reproduce calculations, compare alternatives, test sensitivities, investigate discrepancies, and determine where the result is robust or uncertain.
Phase 5
Explain, document, and transfer
Present the result in language appropriate to the audience. Document the method and, where appropriate, leave behind a process the client can continue to use.
Working principles
Rigor includes being honest about limits.
The quality of an engagement is shaped by how assumptions, evidence, uncertainty, and decisions are handled throughout the work.
Start with the question, not the software
The method should fit the problem. A preferred tool should not define the question.
Reconcile the data before modeling
A sophisticated model cannot correct data that represents different definitions, populations, or periods without explanation.
Keep assumptions visible
Important inputs and judgments should be identifiable, reviewable, and easy to update.
Preserve reproducibility
Where possible, the same data and assumptions should reproduce the same result.
State uncertainty honestly
Forecasts, scenarios, and economic estimates depend on assumptions and changing conditions. Limits should be stated plainly.
Fit the solution to its users
A useful system should fit the people who will operate it. Familiar working practices should be preserved where they help adoption and control.
Protect confidential information
Access, handling, and sharing expectations should be established before sensitive information is exchanged.
Build for continued use
When appropriate, models, data structures, and processes should be documented so the client can continue using them.
A focused first conversation