HEW-NODE-0134 · How Education Works · cost-effectiveness analysis, benefit-cost analysis, economic evaluation, value for money, education interventions, implementation cost, opportunity cost, scale, learning outcomes, equity, discounting, sensitivity analysis, uncertainty and policy choice
Two education programmes can both work and still be very different investments.
One may improve learning modestly at very low cost. Another may produce larger gains but require specialist staff, expensive technology and recurring licences. A third may barely change test scores while substantially reducing violence, dropout or inequality.
Effectiveness asks whether something works. Economic evaluation asks a second question: what did the system have to give up to get that result?
Education cost-effectiveness is not the search for the cheapest programme. It is the discipline of comparing what different choices produce for the resources they consume.
This node sits beside the How Education Works hub, Education Costing, Economics of Education, Education Public Expenditure Reviews & Spending Diagnostics, Results-Based Financing & Disbursement-Linked Indicators, Education Sector Analysis & System Diagnosis and Education Sector Planning & Annual Operationalisation.
Those pages keep their jobs. Education Costing owns what a policy scenario requires in resources. Economics of Education owns the wider relationship among education, incentives, labour markets and returns. Expenditure Reviews own spending patterns and public-finance diagnosis. Results-Based Financing owns conditional payment. This node owns comparative economic evaluation: how alternative education interventions are compared by combining credible outcome evidence with full resource cost, opportunity cost, uncertainty, scale and distribution.
The 60-Second Read
- Effectiveness asks whether an intervention changes an outcome.
- Cost-effectiveness asks how much outcome is produced per unit of resource.
- Benefit-cost analysis converts benefits and costs into a common monetary frame where defensible.
- Costing a programme is not the same as evaluating whether it is cost-effective.
- Budget cost is not the same as economic cost.
- Teacher time, classroom space, volunteer time and donated equipment have opportunity cost even when no invoice is paid.
- Startup cost and recurring cost should be separated.
- Average cost can fall or rise when an intervention scales.
- Pilot cost can be misleading because research teams often provide support the scaled system will not.
- Outcome units need to be comparable across alternatives.
- A standard deviation, month of progress, completion gained or dropout prevented answers different questions.
- One headline outcome can miss safety, wellbeing or equity benefits.
- Cost-effectiveness ratios should carry uncertainty from both cost and effect estimates.
- A programme with uncertain effect can have an unstable cost-effectiveness estimate even when its cost is known precisely.
- Discounting matters when costs occur now and benefits arrive years later.
- Benefit-cost analysis should avoid double-counting overlapping benefits.
- Distribution matters: an intervention can have lower average efficiency and higher value for a disadvantaged group.
- Decision-makers should compare interventions under the same perspective and time horizon.
- Sensitivity analysis should show which assumptions reverse the ranking.
- The objective is better allocation under scarcity, not reducing education to one monetary metric.
One-Sentence Definition
Education cost-effectiveness analysis compares the resources required by alternative interventions with a common educational outcome, while benefit-cost analysis compares monetised benefits with monetised costs over a defined time horizon.
The First Distinction: Costing Is Not Cost-Effectiveness
If a reading programme costs $20 million, costing has answered an important implementation question. It has not told us whether the programme is good value.
To answer that, we need an outcome and a comparison: how much learning did the programme produce relative to a credible alternative, and what would that alternative have cost?
The Second Distinction: Cheap Is Not Cost-Effective
A programme costing $5 per learner that produces no measurable benefit is not automatically better value than one costing $50 that produces a large benefit.
The ratio needs both numerator and denominator.
The Third Distinction: Cost-Effective Is Not Automatically Affordable
An intervention can produce excellent outcomes per dollar and still require a total budget the system does not have because the eligible population is enormous.
Affordability asks whether the budget can carry the scale. Cost-effectiveness asks whether the resources generate relatively strong outcomes.
World Bank “Smart Buys” Makes the Comparison Explicit
The World Bank-hosted Global Education Evidence Advisory Panel groups education interventions by both effectiveness and cost-effectiveness, distinguishing “great buys,” “good buys,” promising approaches, effective but relatively expensive approaches and “bad buys.” Its 2023 Smart Buys report synthesised hundreds of rigorous evaluations after a systematic search of more than 13,000 studies.
The important idea is not the labels themselves. It is that evidence of positive impact is not enough when governments must choose among many positive options under a finite budget.
EEF Makes Cost Visible Alongside Impact
The Education Endowment Foundation’s Teaching and Learning Toolkit reports estimated impact alongside implementation-cost categories and evidence strength. This makes a useful decision principle visible: impact, cost and confidence in the evidence should be read together rather than treated as three separate conversations.
Define the Decision Before the Metric
“Which literacy intervention should we fund for Grade 2?” permits a direct cost-effectiveness comparison if alternatives target the same outcome. “Should we fund literacy tutoring or school meals?” is harder because the programmes produce different bundles of outcomes.
The evaluation method should fit the decision, not force unlike interventions into a misleading common metric.
Choose the Analysis Perspective
- ministry budget;
- whole government;
- school;
- household;
- employer;
- society.
A free course to the ministry may impose travel and lost-wage costs on families. A programme funded by another ministry may look cheap to education while remaining expensive to government overall.
State the perspective before counting costs.
Budget Cost Is Only One Cost
Economic evaluation values resources even when no new cash changes hands. Existing teachers spending two hours per week on a programme have opportunity cost because those hours cannot simultaneously be used elsewhere.
Ignoring non-cash resources makes programmes that consume existing staff time look artificially cheap.
Use the Ingredients Method
One disciplined approach identifies every resource required to reproduce the intervention:
- staff time;
- training;
- materials;
- technology;
- facilities;
- transport;
- administration;
- data systems;
- monitoring;
- maintenance;
- licensing;
- replacement cycles;
- household time where relevant.
Then value quantities using transparent unit costs.
Separate Development From Delivery
A curriculum package may cost $2 million to develop and $5 per learner to deliver. At 10,000 learners, development dominates. At 10 million learners, the fixed cost becomes small per learner.
Scale changes the cost structure.
Scale Can Also Increase Marginal Cost
The cheapest schools to reach may enter first. Expansion into remote regions can require transport, smaller training cohorts, additional languages and stronger support.
Do not assume average pilot cost remains constant at national scale.
Pilot Cost Contains Research Infrastructure
Trials may employ extra coaches, researchers, enumerators and project managers. Some are necessary for delivery; others exist only to evaluate the programme.
Scale costing should distinguish intervention cost from research cost while preserving any support that was actually necessary for effectiveness.
Do Not Remove the Ingredient That Made the Programme Work
A structured pedagogy intervention may work because teachers receive both materials and ongoing coaching. Scaling only the printed lesson plans can reduce cost and also remove the mechanism.
Cost reduction is not value improvement if effectiveness collapses.
Choose a Common Outcome Unit for Cost-Effectiveness
- cost per 0.1 standard deviation learning gain;
- cost per additional learner reaching proficiency;
- cost per additional year of schooling completed;
- cost per dropout prevented;
- cost per additional transition to secondary school;
- cost per attendance day gained;
- cost per credential completed.
The outcome should match the policy objective and be measured comparably across alternatives.
Standardised Effect Sizes Help and Hide
Standard deviations allow comparison across tests with different raw scales. But the same 0.2 SD can have different practical meaning depending on age, assessment reliability and score distribution.
Translate statistical effects back into educational meaning where possible.
“Months of Progress” Is an Interpretation, Not a Natural Unit
Some evidence syntheses express effect sizes as additional months of learning. This can be intuitive for practitioners, but the conversion rests on assumptions about typical progress rates.
Use the metric consistently and avoid implying calendar precision that the evidence does not contain.
Incremental Cost-Effectiveness Compares the Next Best Option
If Programme A costs $100 and produces 0.10 SD, while Programme B costs $160 and produces 0.14 SD, the relevant question may be what the additional $60 buys: 0.04 SD.
Incremental analysis prevents decision-makers from comparing every option only with “do nothing.”
Dominated Options Should Be Visible
If one intervention costs more and produces less of the same outcome than another, it is economically dominated for that decision unless it has important benefits omitted from the outcome metric.
Cost-Effectiveness Thresholds Are Policy Choices
There is no universal educational price for one standard deviation of learning or one prevented dropout. The threshold depends on available budget, competing opportunities, policy priorities and context.
Benefit-Cost Analysis Changes the Unit
Benefit-cost analysis monetises both benefits and costs so very different programmes can be compared through measures such as net present value or benefit-cost ratio.
Benefit-cost ratio = present value of monetised benefits ÷ present value of costs.
A ratio above one means monetised benefits exceed monetised costs under the model. It does not mean every important benefit has been measured or that the distribution is fair.
Education Benefits Often Arrive Late
Higher earnings, productivity, lower unemployment risk, improved health or reduced crime can emerge years after an intervention. Costs are often paid now.
Discounting places future costs and benefits onto a present-value basis, making the discount rate a consequential assumption.
Discount Rate Should Be Tested, Not Hidden
A higher discount rate reduces the present value of distant benefits. Early-childhood programmes with benefits extending over decades can therefore look very different under different rates.
Report sensitivity rather than presenting one rate as a law of nature.
Do Not Double-Count Benefits
If higher earnings already reflect some productivity gain, adding both full earnings gains and a separate productivity estimate may count the same mechanism twice.
Build a benefit map before summing values.
Some Benefits Should Stay Non-Monetary
Dignity, rights, inclusion, cultural continuity and democratic participation can matter without a credible monetary conversion. Multi-criteria decision analysis can display those dimensions alongside economic evaluation.
Equity Can Change the Decision
An intervention costing $500 per additional completer may be less “efficient” than one costing $300, but the $500 programme may serve remote learners who otherwise have no route to completion.
Distributional weights, subgroup cost-effectiveness or explicit equity criteria can stop average efficiency from becoming the only value.
Average Cost-Effectiveness Can Hide Who Benefits
A national programme may be highly cost-effective because it works especially well in already strong schools. Disaggregate results before scaling claims to harder contexts.
Context Transfer Is Not Free
A programme evaluated in a country with lower wages, shorter travel distances or stronger instructional infrastructure may have different cost and effect elsewhere.
Transport evidence through mechanisms, not ratios alone.
Exchange Rates Can Mislead International Comparisons
Converting local costs at market exchange rates can distort the relative cost of labour-intensive services. Purchasing-power adjustments may help for some comparisons, but local budget decisions still occur in local currency and wage structures.
Inflation and Price Year Must Be Explicit
A 2018 programme cost and a 2026 programme cost cannot be compared cleanly without bringing them to a common price year.
Uncertainty Lives on Both Sides of the Ratio
Costs may be estimated from samples. Effects have confidence intervals. Attrition, implementation variation and missing data add uncertainty.
A single ratio can look precise while both its components are uncertain.
Use Probabilistic or Scenario Sensitivity Where Possible
Instead of changing one assumption at a time, analysts can examine combinations: lower effect, higher wage cost, weaker scale economies and shorter benefit duration.
The useful question is often: under how many plausible futures does this intervention remain good value?
Implementation Fidelity Is an Economic Variable
A cheaper version of an intervention may be cheaper because coaching frequency, material quality or supervision fell. If those ingredients drive effectiveness, the low-cost version is a different intervention.
Capacity Constraints Create Shadow Costs
A programme may require thousands of qualified coaches who do not exist. Training them has cost; diverting experienced teachers into coaching has opportunity cost.
Economic evaluation should not assume scarce capabilities can be expanded instantly at current prices.
System Interactions Can Change Value
A tutoring programme can be highly effective while masking weak classroom instruction. A strong curriculum reform can raise the productivity of teacher coaching. Two interventions may be complements or substitutes rather than independent menu items.
Portfolio Choice Is Different From Ranking
The “best” intervention cannot consume the entire education budget. Systems need portfolios: foundational services, equity obligations, infrastructure, teacher salaries, safety and innovation.
Cost-effectiveness informs the margin where choices are real; it does not abolish the architecture of a functioning education system.
Case Study: The Cheap Tablet Programme
Invented example: a tablet programme appears inexpensive because devices were donated. Economic costing adds teacher training, device management, connectivity, replacement and IT support. Learning impact remains near zero because instructional use is weak.
The programme is no longer “free,” and the low impact makes its cost-effectiveness unattractive.
Case Study: The Expensive Tutoring Programme
Invented example: intensive tutoring costs far more per learner than a messaging intervention but produces much larger gains for students two years behind.
National average cost-effectiveness favours messaging; subgroup analysis shows tutoring may still be the stronger choice for the highest-need group.
Case Study: The Pilot That Could Not Scale at Pilot Prices
Invented example: a programme uses excellent university graduates as coaches at a low research-project stipend. National scale would require market wages, supervision and travel to remote schools.
Scale costing triples delivery cost. The programme still works, but the policy ranking changes.
Case Study: The Early-Childhood Programme With Long Benefits
Invented example: a high-quality preschool programme has substantial upfront cost. Short-term test-score effects are moderate. Long-run evidence suggests higher completion and earnings.
Cost-effectiveness on immediate test scores and benefit-cost analysis over decades answer different policy questions. Both should be shown rather than collapsed.
Failure Modes and Repairs
- Budget cost only: repair by valuing all resources and opportunity costs.
- Pilot price treated as scale price: repair with fixed, variable and expansion cost models.
- Effect without uncertainty: repair with confidence intervals and sensitivity analysis.
- One outcome for everything: repair by matching the metric to the decision and preserving other material outcomes.
- Cheapness mistaken for value: repair by comparing cost with effect.
- Benefit double-counting: repair with an explicit benefit pathway map.
- Discount rate hidden: repair by reporting alternative plausible rates.
- Average efficiency erases equity: repair with subgroup and distributional analysis.
- Context copied: repair by re-costing and re-testing mechanisms locally.
- Ranking mistaken for portfolio design: repair by combining economic evidence with system obligations and complementarities.
The Economic Evaluation Operating Chain
- Define the decision.
- Define the comparison alternatives.
- Choose the analysis perspective.
- Choose the time horizon.
- Map intervention ingredients.
- Measure resource quantities.
- Assign unit costs.
- Separate startup and recurring cost.
- Separate research cost and delivery cost.
- Model scale effects.
- Identify opportunity costs.
- Select comparable outcomes.
- Estimate intervention effects.
- Represent effect uncertainty.
- Calculate cost per outcome.
- Compare incremental alternatives.
- Identify dominated options.
- Map additional benefits.
- Monetise benefits only where defensible.
- Choose and report discount rate.
- Calculate net present value or benefit-cost ratio where appropriate.
- Test double-counting.
- Analyse equity and subgroups.
- Test context transfer.
- Run sensitivity and scenario analysis.
- Identify capacity constraints.
- Assess complementarity with other reforms.
- Test affordability at scale.
- Present economic evidence alongside non-monetary values.
- Update after real implementation costs and effects arrive.
A Cost-Effectiveness Dashboard
- intervention;
- comparator;
- population;
- analysis perspective;
- price year;
- startup cost;
- annual delivery cost;
- cost per learner;
- teacher time;
- household cost;
- effect estimate;
- effect uncertainty;
- cost per outcome unit;
- incremental cost-effectiveness;
- scale scenario;
- remote-area cost;
- subgroup effect;
- equity weighting if used;
- discount rate;
- benefit-cost ratio where relevant;
- assumptions that reverse the ranking.
Canonical Owner Boundaries
- Education Costing owns the resource requirements of a policy or programme.
- Economics of Education owns the broader economics of education, incentives and returns.
- Education Public Expenditure Reviews & Spending Diagnostics owns system-level spending patterns and expenditure diagnosis.
- Results-Based Financing & Disbursement-Linked Indicators owns conditional payment tied to verified results.
- Education Sector Planning & Annual Operationalisation owns conversion of chosen strategies into implementable plans.
This node owns comparative economic evaluation: identifying full resource cost, connecting cost to credible outcomes, comparing incremental alternatives, monetising longer-term benefits where appropriate, testing uncertainty and showing how efficiency interacts with scale and equity.
The Return Path
Return to a ministry table with five programmes that all have advocates and only enough money for three.
Evidence that each programme works does not remove the choice. Scarcity remains. The system must know what resources each programme truly consumes, what outcomes those resources buy, how uncertain the estimates are, who receives the benefits and what disappears from the budget when one option is chosen.
Economic evaluation does not make the value judgement disappear. It stops hidden resource trade-offs from pretending not to exist.
The best-value education programme is not the one with the smallest invoice. It is the one whose full cost buys an educational result the system actually needs at a scale it can sustain.
Return to the How Education Works hub.