At a Glance
What Does a Theory of Change Contain?
A complete theory of change has four components. The written narrative is the primary artifact; the diagram visualizes it; the assumptions and evidence base make it testable.
| Component | What it captures | Typical length |
|---|---|---|
| Long-term outcome statement | The specific population-level change the program is designed to contribute to | 1-2 sentences |
| Causal pathway | The sequence of intermediate outcomes, outputs, and activities that produce the long-term outcome | 2-3 pages narrative + 1-page diagram |
| Assumptions | The conditions outside program control that must hold for each causal step | 1 page, structured by pathway step |
| Evidence base | Prior research, evaluation evidence, or context analysis supporting each causal claim | Short citations inline, or annex reference list |
A theory of change is not just a diagram. The diagram helps communicate the theory but does not replace the narrative. Programs that ship only a diagram (often called a "theory of change" when really only the boxes-and-arrows picture exists) are missing the reasoning behind each arrow. See theory of change for the concept definition and logframe vs theory of change for the critical comparison.
What Order Do You Write a Theory of Change In?
Write the theory of change in this order. Each step builds on the previous.
| # | Step | Output |
|---|---|---|
| 1 | Name the long-term outcome | 1-2 sentences at the top of the pathway |
| 2 | Backwards-map the intermediate outcomes | 2-4 intermediate outcomes, each a necessary precondition for the long-term outcome |
| 3 | Identify outputs and activities that produce each intermediate outcome | Specific, program-controlled deliverables and actions |
| 4 | Surface the assumptions at each step | 1-2 testable assumptions per causal step |
| 5 | Link each causal claim to evidence | Short citations or explicit acknowledgment of untested claims |
| 6 | Draw the diagram | One-page visual showing pathway, outcomes, and key assumptions |
| 7 | Write the narrative | 2-5 pages explaining the pathway in prose |
Most failed theories of change skip to the diagram (step 6) before the logic of steps 1-5 is worked out. Resist this. The diagram should be the last step, not the first.
Backwards Mapping
Step 1-3: From Long-Term Outcome Backward
The long-term outcome is where you want to end up. It is the population-level change your program is designed to contribute to, with a measurement horizon justified by the pathway, which may fall during or after implementation. Good long-term outcomes are concrete enough to test but ambitious enough to motivate: "Increased school completion rates for girls aged 11-16 in target districts," not "Improved education outcomes."
Once the long-term outcome is named, work backward.
Intermediate outcomes (step 2): ask "what changes in participants or systems must occur for the long-term outcome to happen?" These are typically 2-4 intermediate outcomes, each representing a specific change in behavior, capability, or condition. For a girls' education program: "Families support daughters continuing to secondary school," "Girls stay enrolled through transitions," "Schools provide quality girl-friendly education."
Outputs and activities (step 3): for each intermediate outcome, ask "what deliverables and actions does the program produce to drive this change?" Outputs are tangible products or services (trainings delivered, materials distributed, systems built). Activities are the actions that produce outputs (designing curricula, recruiting teachers, conducting outreach). Each output and activity should contribute to at least one intermediate outcome.
Backwards mapping is harder than forward planning. Forward planning starts with what you want to do; backwards mapping starts with what you want to change. The discipline is to resist starting with activities, because forward planning produces theories that rationalize pre-existing program designs rather than revealing where the logic breaks.
Step 4-5: Assumptions and Evidence
Step 4 (assumptions): at each step in the causal pathway, identify 1-2 assumptions that must hold for that step to work. Assumptions are conditions outside your control that you are counting on.
Examples of testable assumptions:
- Activity-to-output: "Local partners maintain operational capacity through implementation period."
- Output-to-intermediate-outcome: "Trained participants remain in their roles long enough to apply new skills (minimum 6 months)."
- Intermediate-to-long-term: "National education policies supporting girls' secondary school completion remain stable through the measurement horizon."
Assumption quality: specific enough to be checked. "External conditions remain favorable" is a placeholder. "The national water policy reform stays on its current schedule through project year 2" is a testable statement.
Step 5 (evidence): for each major causal claim, link to an evidence base. This is not a full literature review; it is short citations to prior evaluation evidence, behavior-change research, or context studies that support (or acknowledge as untested) each arrow in your pathway.
Three categories of evidence:
- Direct Program Evidence: prior evaluations of similar programs, published impact evaluations, sector meta-analyses
- Behavioral or Policy Research: theoretical or empirical work on the behavior/system change being produced
- Context Analysis: local studies, situational analyses, or stakeholder engagements that ground the theory in the specific program context
Where evidence is thin or nonexistent, the theory of change should say so explicitly and frame the program as testing the claim rather than assuming it holds.
The Diagram
Step 6: Draw the Diagram
The theory of change diagram is a one-page visual showing the pathway from activities at the bottom to the long-term outcome at the top, with intermediate outcomes in between and key assumptions noted alongside. Common diagram styles include box-and-arrow flowcharts, Sankey-style pathway diagrams, and nested outcome frameworks.
A good diagram:
- Fits on one page (not one PowerPoint slide per level)
- Shows clear causal direction (activities → outputs → intermediate outcomes → long-term outcome)
- Notes the most critical assumptions alongside the arrows they support
- Is legible without the narrative (but does not replace it)
Avoid:
- Diagrams that show boxes without arrows
- Diagrams that show every activity instead of aggregating to output level
- Diagrams that embed every assumption in tiny side boxes (pick the critical ones)
The diagram communicates the theory; the narrative explains it.
Step 7: Write the Narrative
The narrative is the prose version of the theory of change. Typically 2-5 pages, structured by the pathway.
Standard narrative structure:
- Problem statement and context (1 paragraph)
- Long-term outcome (1 paragraph)
- Pathway description, walking from long-term outcome backward through intermediate outcomes and outputs (2-3 pages)
- Key assumptions organized by pathway step (1 page)
- Evidence base supporting the pathway (1 page or annex reference)
The narrative is the document that reviewers and evaluators read. The diagram is what they remember. Both are needed.
For the broader MEL plan context, see how to write a MEL plan. For the logframe operationalization, see how to write a logframe.
Sector Examples
Worked Example: A Half-Day Theory of Change Workshop
Healthy Futures is a fictional 36-month child nutrition program in Kasenda District, funded by the fictional Global Health Partnership Fund (GHPF). The funder liked the program's results chain but asked a harder question: why will this program work? The M&E officer proposed a half-day workshop instead of writing the theory alone. The room held the program manager, a nutrition specialist by phone, and the three zone leads: James Ochieng (Zone A), Amina Hassan (Zone B) and David (Zone C).
The first question was "Why do we believe that training community health workers (CHWs) and running nutrition sessions will reduce stunting here?" The first answer, "other programs have done it," drew a fair challenge from the nutrition specialist: which programs, and in what context? Over three hours the team mapped four pathways:
- Screening: CHW training and supervision, correct screening, early detection, referral, recovery
- Education: nutrition sessions, caregiver knowledge, changed feeding practices, better child nutrition
- Gardens: seeds and training, diverse crops, more food at home, better dietary diversity
- Learning loop: monitoring data, trends spotted, approach adapted (a feedback loop, not a causal pathway)
For each arrow the facilitator asked "What must be true for this to work?" The weakest assumption was that knowledge changes behavior. The team agreed to test it: if caregiver knowledge scores rise while dietary diversity does not, the curriculum needs redesign. By the Month 9 learning review the data showed exactly that, and the curriculum changed in Month 10.
Only then did the team draft the logframe: one Goal row (the long-term outcome), three Outcome rows (the intermediate outcomes) and six Output rows, with 18 indicators in all.
Sector Examples
The scenarios below are illustrative. Their figures, costs, and review outcomes are teaching assumptions, not reported studies or evidence of donor approval.
Health: Safe Delivery Program, East Africa
A 4-year program aims to reduce maternal mortality in 3 rural districts. Long-term outcome: "Reduced maternal mortality in target districts." Intermediate outcomes: (a) community health workers support timely access to qualified maternal care within their authorized scope, (b) women with danger signs receive urgent assessment and referral under the applicable clinical protocol, (c) health facilities provide quality emergency obstetric care when reached. Outputs distributed across the three: training curricula, supervision visits, transport vouchers, facility upgrades, community mobilization. Critical assumption (intermediate to long-term): "District-level health system capacity does not collapse due to broader sector disruptions." Evidence base cited: prior Lancet series on community-based maternal health, regional DHS data on referral pathways, local needs assessment commissioned in year 0. Two cautions: measure mortality with a district-level source for a comparable period and population, because a national baseline cannot stand in for a district baseline; and set referral timing from the applicable clinical protocol, never a generic 24-hour delay.
Education: Girls' Education Program, South Asia
A 5-year program aims to increase girls' secondary school completion in 60 communities. Long-term outcome: "Increased rates of girls completing secondary education in program communities by year 5." Intermediate outcomes: families supporting girls' continued education; girls staying enrolled through transitions; schools providing girl-friendly environments. Activities and outputs distributed: mentor program for 2,400 girls, community dialogue events, teacher training, stipend delivery, menstrual health support. Critical assumption: "Ongoing government policy commitments to girls' education do not reverse during program period." Evidence base: prior program evaluations in neighboring country, cross-cultural research on girls' education barriers, country-specific adolescent health and education studies.
WASH: Community Water Sustainability, West Africa
A 4-year program aims to improve durable water access in 80 rural communities. Long-term outcome: "80 communities maintain JMP-basic or better water service at year 9 (5 years after this 4-year program)." Intermediate outcomes: (a) communities operate and maintain water systems after handover, (b) households use safely managed water sources routinely, (c) local capacity exists to repair and upgrade systems. Outputs: water points installed, water committees trained, community contribution framework established, local technician network formed. Critical assumption: "Local government capacity to support handover remains adequate over the 5-year sustainability measurement window." Evidence base: prior WASH sustainability studies, JMP longitudinal service data, rural sanitation literature.
Food Security: Pastoralist Livelihoods, Sahel
A 3-year program aims to improve food security among pastoralist households through drought cycles. Long-term outcome: "Reduced prevalence of moderate and severe food insecurity in target communities, sustained through at least one drought cycle." Intermediate outcomes: diversified income sources; functional savings groups; adaptive herd management. Outputs: agricultural training, savings group formation, small-business grants, livestock vaccination campaigns. Critical assumption: "No catastrophic drought during program period (if one occurs, the theory shifts and adaptive response replaces steady-state strategy)." Evidence base: prior pastoralist resilience studies, climate trend data, savings group meta-analyses.
Mistakes & Checklist
Common Mistakes
Mistake 1: Starting with activities and aggregating forward. Forward planning produces theories that rationalize what you already wanted to do. Backwards mapping from the long-term outcome forces you to ask whether your planned activities actually lead there.
Mistake 2: Skipping the assumptions column. A theory of change without explicit assumptions is a declaration of intent, not a theory. Each causal step should have at least one named assumption that could be tested during implementation.
Mistake 3: Presenting claims without evidence. Novel causal logic needs grounding in prior research, evaluation evidence, or explicit acknowledgment that the program is testing an untested claim. Unmoored theories are weaker in review.
Mistake 4: Confusing the diagram for the theory. The diagram is a communication tool, not the theory itself. Programs that ship only a diagram without narrative are missing the reasoning behind each arrow.
Mistake 5: Long-term outcomes stated as program deliverables. "5,000 women trained" is not a long-term outcome; it is an output. Long-term outcomes are population- or sector-level changes often measured at or after program close.
Mistake 6: Too many intermediate outcomes. 2-4 intermediate outcomes is typical. 7-10 is a sign that you have confused outputs for outcomes, or that the program is trying to produce change at too many levels simultaneously.
Mistake 7: Assumptions that are not testable. "Supportive external environment" is not an assumption; it is a wish. An assumption should be specific enough to check during implementation. If it cannot be tested, it cannot inform adaptive management.
Mistake 8: Theory of change disconnected from logframe. The two documents should tell the same story in different formats. When the theory of change says one thing and the logframe shows another, reviewers notice. Update both together.
Theory of Change Completion Checklist
Run through this before treating the theory of change as final.
Pathway structure:
- Long-term outcome stated clearly (population- or sector-level change)
- 2-4 intermediate outcomes mapped backward from long-term outcome
- Outputs and activities identified for each intermediate outcome
- Each level plausibly produces the level above it
Assumptions:
- 1-2 assumptions named at each causal step
- Assumptions specific enough to be tested during implementation
- Critical assumptions flagged (the ones the theory most depends on)
Evidence:
- Major causal claims linked to evidence base (direct program evidence, behavioral/policy research, or context analysis)
- Untested causal claims flagged as such rather than assumed
Diagram and narrative:
- One-page diagram fits on one page
- Narrative is 2-5 pages
- Diagram and narrative tell the same story
Integration:
- Theory of change matches the logframe (if both exist)
- Theory of change matches the proposal narrative
- Key assumptions mirrored in the risk register
For the logframe operationalization, see how to write a logframe and logframe vs theory of change. For the broader MEL plan, see how to write a MEL plan. For the proposal-section integration, see how to write the M&E proposal section. For current document workspaces, see the Playbooks catalog.