Suggest the Right Methodology
Choose an evaluation methodology from the actual questions, evidence and constraints, with practical implementation advice.
On this page
- What You Need
- Describe the evaluation question, intended use, population, program logic, available data, capacity, and deadlines.
- Choose This Prompt When
- Use this to choose an evaluation approach. Choose Walk Me Through Data Analysis when you need a sequenced analysis plan for a defined question and data type, or Help Me Understand Contribution Analysis for that specific causal approach.
Recommend an evaluation methodology for the supplied program, question and intended use. Ask for missing population, available data or major constraints before making a firm recommendation. Do not assume an impact evaluation or reject a sound single-method design.
Explain the recommended name/type and why it fits the goal, theory, context, question and constraints. Compare a second plausible option only when the tradeoff helps the decision. Describe planning, collection, analysis and reporting steps; required skills, staffing, budget considerations, technology and time; suitable collection tools; strengths, limitations, challenges and mitigations.
Explain relevant safe disaggregation and how the method serves actual stakeholder decisions. Give a conditional timeline with dependencies; staffing, cost and timing estimates remain provisional until local capacity is known. Keep supplied sources, preferences and plans distinct from confirmed evidence or agreed requirements. Return a concise, actionable recommendation covering the ten areas: method/type, fit, steps, resources, tools, strengths/limits, disaggregation, stakeholder use, challenges and timeline.
Worked Example
Illustration only. The setting and figures are invented to show the pattern.
Scenario: A team running a community water program wants to know if it should commission an impact evaluation. It has a baseline survey, a six-month window and a modest budget.
What You Give the AI: The evaluation question (did household water access improve), the program theory, the baseline and monitoring data, the budget level, the timeline, the reporting needs of the funder, and the districts where the program works.
What the AI Might Return: A recommendation of a before and after design with contribution analysis for causal claims, and a note that a randomized design does not fit the timeline. A list of steps from planning to reporting follows, with the skills needed, suitable tools such as a household survey and key informant interviews, and a conditional timeline. A second option, a matched comparison, appears with its tradeoff.
When to Choose This Prompt
- You need to choose an evaluation approach from your question, data and constraints.
- You want one recommended method, with a second option only where the tradeoff helps the decision.
- You want implementation advice with it: steps, skills, tools and a timeline that depends on stated conditions.
When Not to Choose It
- You know the question and the data and need analysis steps. Walk Me Through Data Analysis
- You want to understand contribution analysis as an approach. Help Me Understand Contribution Analysis
- You are deciding if the program is ready to be evaluated at all. Create an Evaluability Assessment
- A design is already drafted and needs testing. Review My Evaluation Design
Checking the Output
- Read the stated fit against your question. A descriptive question does not need a comparison group, so a method that adds one is oversized.
- Every data source the plan relies on should appear in the list of available data you gave.
- Treat staffing, cost and timeline figures as provisional and compare them with local rates and your own calendar.
- The second option, if any, should add a real tradeoff. Cut it if it only restates the first.
- Disaggregation advice should name groups you supplied and warn about cells too small to report safely.
- The method's limitations and mitigations should be listed. Their absence means the answer is unfinished.
Common Problems and Fixes
- The AI defaults to an impact evaluation. Fix: State the purpose and the decision, and say that a single sound method is acceptable.
- The recommended method needs data you do not have. Fix: List your real data sources and ask for a method that works with them, or for the gap to be named.
- The timeline is optimistic. Fix: Give your team's capacity and ask for dependencies to be shown, such as approvals and tool testing.
- Your preferences are treated as requirements. Fix: Mark which constraints are fixed and which are preferences.
Adapting It to Your Context
- Remote or insecure locations: Describe access limits and ask for collection modes that suit them, with the risks they carry.
- Funder requires a specific approach: Paste the wording of the requirement and ask how far the recommended method meets it.
- Small team: State headcount and skills so the steps are sized to what you can run.