The Gold Standard of Innovation Economics

Chapter index

A workshop discovers a way to reduce material waste. The method works in a demonstration, but someone must pay for testing, adapt the production line, train operators and bear the risk that ordinary orders behave differently from the demonstration. Other workshops might eventually learn the method without paying for its discovery. These are different economic questions, even though they concern the same technical idea.

Innovation economics asks who chooses the research effort, what alternatives that effort displaces, who can capture the benefits, and what happens to people outside the transaction. It also asks why a method spreads quickly in one setting and slowly in another. A technically impressive result is only one input to those explanations.

Keep the stages separate

The OECD/Eurostat Oslo Manual uses implementation to distinguish innovation from an unimplemented idea or invention. Its reference point matters: a significant improvement can be new to an organisation without being new to the world. This statistical distinction should be kept separate from a judgement that the change has produced net benefits.

In the workshop case, write four dates: the first workable method, the first production trial, the first sustained use and the first reliable evaluation. Each date answers a different question. A patent application, if one exists, supplies another record; it does not fill the gaps between the four dates.

Locate the neighbouring subjects

Innovation examines the practical journey into useful change. Innovation Studies develops the wider implementation and evaluation picture. Innovation Policy focuses on public choices. Here the central task is explaining resource allocation and returns: why particular research, financing and adoption decisions are made, and whether private incentives align with wider gains.

For a first exercise, replace “this workshop needs to innovate” with a decision: should it spend 100 units developing the waste-reduction method, buy an existing method, or continue its current process? Specify the same output quality and time period for all three. Without those common boundaries, the comparison can reward whichever proposal leaves the most costs outside its description.

Contents · Next chapter

One cutting machine cannot perform two different physical jobs in different workshops at the same instant. A written method can, in principle, inform both workshops simultaneously. Economists describe this difference as rivalry: using a rival resource leaves less of it available for another use, while use of a non-rival idea need not do so.

Paul Romer's model of endogenous technological change places deliberately produced, non-rival and partly excludable knowledge at the centre of growth. “Endogenous” means that technological change is explained through decisions inside the model rather than simply arriving as an unexplained outside improvement. That is a modelling contribution, not a promise that every research expenditure raises growth.

Separate copying from implementing

Suppose producing a reliable instruction manual costs 600 units. Each workshop then needs 40 units of training and calibration to use it. For one workshop, the total is 640. For ten workshops sharing the same manual, it is 600 + 10 × 40 = 1,000, or 100 per workshop. The discovery cost is spread across more users; the implementation cost remains.

It would be wrong to say that the tenth workshop costs nothing because knowledge is non-rival. It still needs equipment time, staff attention and adaptation. It would also be wrong to charge the full 600 discovery cost to every workshop when measuring the resources used by the group. Non-rivalry helps identify which cost is shared and which cost repeats.

Access and understanding are separate constraints

Excludability concerns whether access can be restricted. A method may be protected by confidentiality, contractual arrangements or intellectual property rights. Even openly available information can be difficult to use without experience. Reading a process description does not automatically supply calibration skill or local knowledge of materials.

The diagnostic question is therefore precise: is a workshop unable to obtain the method, unable to understand it, unable to finance adaptation, or unable to coordinate the change? Those barriers call for different responses. Making another copy solves only the first kind of problem when information access is the actual constraint. This is why counting downloads is a weak measure of productive diffusion.

Previous chapter · Contents · Next chapter

A firm usually needs a way to recover its expenditure. Society also cares about benefits that reach customers, other producers and future researchers. The firm's cash flow and the wider resource account answer different questions. Neither should be quietly substituted for the other.

Consider a fictional process-development project with an immediate cost of 100. Its expected additional operating surplus for the developer is 80, after all subsequent implementation costs. Other workshops obtain expected real resource savings of 50 through knowledge they can use without paying the developer. Assume those savings are genuinely additional, occur over the same period, and do not overlap with the developer's 80.

The developer's expected net return is 80 − 100 = −20. The simplified wider net benefit is 80 + 50 − 100 = 30. A socially promising project can therefore look unattractive to the organisation asked to fund it. This is a numerical illustration of a possible incentive gap, not evidence that any unfunded project deserves public money.

Do not count the same benefit twice

Now suppose the developer charges the other workshops a total licence fee of 30. If that fee changes no real activity, it raises the developer's receipts by 30 and reduces the other workshops' net gain by 30. The private net return becomes 10. The wider total remains 30: 80 + 30 + 50 − 30 − 100.

The fee is a transfer within this deliberately narrow accounting boundary. Administering the licence, enforcing it or changing how many workshops adopt would involve additional consequences. Those must be modelled separately. Calling every payment a social cost would understate the benefit; calling every receipt a new social gain would overstate it.

Ask whether the boundary is complete

The answer changes if the method creates more waste elsewhere, worsens quality or displaces valuable research. Add those effects explicitly rather than hiding them in a vague adjustment. If an excluded harm is 45, the wider net benefit becomes −15. The attractive spillover story alone is then insufficient.

For any calculation, label the payer, recipient, real resource change, timing and uncertainty. That small discipline prevents a large class of errors in reports claiming that sales, cost savings and economic impact are interchangeable.

Previous chapter · Contents · Next chapter

One company's research can help another company learn. It can also take customers away from another company. Both effects can appear as changes in firm performance, but their interpretation differs. A decline in a competitor's profits is not automatically proof that research has destroyed knowledge or reduced total wellbeing.

Bloom, Schankerman and Van Reenen explicitly distinguish technology spillovers from product-market rivalry. Their research uses firms' positions in technological and product-market space to examine the two channels. The conceptual lesson is to identify the channel before interpreting a financial outcome; the balance estimated in one sample is not a universal constant for all industries.

Follow an original two-firm example

Firm A develops a scheduling method. It gains 60 units of operating surplus. Firm B loses 25 units because some customers move to A, but gains 15 units of resource savings after learning part of the method. Customers receive an additional benefit of 20 after paying prices. Development costs 30. Assume the quantities are incremental, measured consistently and exclude transfers already counted elsewhere.

The total change in this simplified surplus account is 60 − 25 + 15 + 20 − 30 = 40. Looking only at A gives 30 after development costs. Looking only at B gives −10. Neither firm's result alone describes the whole example.

Separate a prediction from a measurement

The 15-unit knowledge benefit must be supported by evidence in a real study. Shared suppliers, similar inventions or patent citations might suggest a route for knowledge, but do not by themselves establish its causal value. B may have improved because demand changed, because its workers gained experience, or because it invested independently.

Likewise, the customer benefit cannot be inferred directly from A's revenue. Customers pay money and receive a service; the value above what they pay is a different quantity. Quality changes, waiting times and previously excluded users complicate measurement further.

A useful research note draws two maps: who competes for the same customers, and who can learn from similar technical knowledge. They may overlap without being identical. An instrument supplier can share technology with a distant industry while competing only weakly with it. This distinction makes research comparisons more informative than a simple list of companies labelled “innovative”.

Previous chapter · Contents · Next chapter

Research spending buys a chance of learning or success, not a guaranteed asset. A proposal should say what can be learned before the next commitment and which decisions remain reversible. Otherwise a single expected-return figure conceals the most important part of the process.

Consider a fictional project with a 40 percent chance of eventual technical and commercial success. Development costs 60 now. Success produces 180 at the end of year two, net of later production costs; failure produces zero. At a purely illustrative discount rate of 10 percent, the expected present value is 0.4 × 180 ÷ 1.1² − 60, approximately −0.50.

Calculate the threshold, then challenge its inputs

Break-even requires a success probability of 60 × 1.1² ÷ 180, approximately 40.33 percent. A difference of a few tenths of a percentage point changes the sign. That does not make the probability estimate reliable to that precision. It shows how fragile the decision is under the stated assumptions.

Now imagine a perfect feasibility test costing 8 immediately. With probability 0.4 it identifies success, after which the firm pays 60; otherwise the firm stops. Assume the test adds no delay and predicts the final outcome perfectly. The value is −8 + 0.4 × (180 ÷ 1.1² − 60), approximately 27.50. Testing is valuable here because it avoids development expenditure in cases that would fail.

Restore the complications before applying the idea

A real test is rarely perfect. False positives cause wasteful continuation; false negatives abandon worthwhile projects. A delay can reduce the payoff, and learning may itself change the probability of success. The perfect-test calculation is an upper-bound teaching comparison under these assumptions, not a forecast for a pilot.

Write the decisions in order: pay for the test, observe a result, decide whether to develop, then observe the outcome. Do not compare a strategy allowed to stop with another strategy forced to continue without mentioning the difference. And do not justify continuing merely because earlier costs were high: money already spent is different from the resources still required. Preserve useful findings from stopped work so that the next project need not repeat an avoidable mistake.

Previous chapter · Contents · Next chapter

A project can have an attractive expected return and still struggle to raise funds. The developer may know more about its prospects than an outside financier. Results may take years to observe, and much of the expenditure may create knowledge that cannot readily be sold if the project fails. These features make assessing and contracting over research difficult.

Hall and Lerner's review of R&D financing examines financial-market explanations for underinvestment distinct from spillovers. It reports particular difficulties for small and new innovative firms and cautions against treating venture capital as a complete solution. The review is a research synthesis, not a current financing offer or a universal finding about every firm.

Keep project value and financing terms separate

Suppose a project needs 50 today and will return either 140 or zero one year later, each with probability one half. Ignoring discounting, the expected gross return is 70 and the expected net project value is 20. An investor who supplies all 50 and receives only 60 in the successful state expects 30, so those terms do not cover the investment.

An investor receiving 100 in the successful state expects 50 and just breaks even before administration, risk preferences and other costs. The developer retains 40 in success. Whether that leaves enough incentive for effort, or compensates the developer for other contributions, is an additional question. Positive total value alone does not specify a workable contract.

Diagnose the missing agreement

If the developer believes success probability is one half while investors believe it is one quarter, the two sides disagree about the project itself. If they agree on the probability but cannot verify future revenue, the obstacle concerns enforceability. If a prototype can resolve uncertainty but nobody finances it, staging may help. These are not equivalent problems.

The repair is to identify what credible evidence or arrangement changes the disagreement. It might be independently checked test results, smaller commitments, clearer rights or access to complementary expertise. It might also be rejection after better evidence shows weak prospects. A disciplined account permits both outcomes rather than defining every refusal as market failure. Organizational Economics extends this question to how decisions and information are arranged inside firms.

Previous chapter · Contents · Next chapter

“Competition encourages innovation” and “profits finance innovation” can both describe real mechanisms. A useful analysis identifies what changes in the return to undertaking research. The level of profit before innovation is not the same as the extra profit caused by innovation.

Aghion and colleagues' competition-and-innovation study develops competing mechanisms and examines an inverted-U relationship using UK firm patenting data. Competition can increase incentives to escape close rivals while reducing incentives for firms further behind. This is a reason to examine industry conditions, not a rule that every industry has the same optimal concentration.

Compare incremental rewards

Imagine an incumbent earning 80 without a new product and 110 with it. The incremental operating reward is 30. A challenger earns zero without entering and 70 after successful entry with an improved product. Its incremental reward is 70. With development costs of 40 and certainty for this exercise, the incumbent declines while the challenger proceeds.

Now alter the incumbent's alternative. If it expects a rival's innovation to reduce its no-innovation profit to 20, while its own successful innovation still delivers 110, its incremental reward is 90. The same incumbent can now have a stronger incentive to invest. The key change is the counterfactual profit, not a claim that large firms are intrinsically reluctant to innovate.

Add feasibility and timing

The challenger may lack manufacturing capacity or distribution. The incumbent may already possess technical knowledge that lowers its development cost. Both may face uncertainty about what the other will do. These conditions can overturn the simple ranking, which is why market share alone cannot explain research effort.

In a real case, record the product market, the technical frontier, entry conditions and relevant complements. Ask whether the innovation replaces existing sales, reaches new users or reduces costs across an established product line. Each creates a different pattern of rewards.

Finally, distinguish innovation incentives from customer outcomes after innovation. A strong incentive to win a market does not establish that every resulting business practice improves access, choice or quality. Those outcomes require their own evidence and the applicable competition framework; a classroom profit calculation cannot settle a regulatory question.

Previous chapter · Contents · Next chapter

An inventor faces a tension: sharing technical information can make collaboration and use possible, while unrestricted imitation may reduce the return to discovery. Intellectual property arrangements affect that tension, alongside lead time, complementary capabilities, reputation and the practical difficulty of copying.

The World Intellectual Property Organization's patent guidance explains that patents involve exclusive rights and disclosure, and that patent rights are territorial and limited in duration. The precise protection and permitted activities depend on the applicable jurisdiction and claims. Public disclosure of technical information is not equivalent to permission to commercially use a protected invention.

An economic licensing exercise

Suppose a method generates 12 units of annual gross benefit for each adopting workshop. Adoption requires 4 units of real annual support and a licence fee of 5. The adopter retains 3. With 20 identical adopters, the fee yields 100 for the rights holder, while real support consumes 80 and adopters retain 60.

The combined annual benefit before discovery and licensing-administration costs is 20 × (12 − 4) = 160. It is also 100 + 60. Adding the licence income once more would double-count the transfer. If the fee rises above 8, these identical workshops no longer obtain a positive net benefit under the model; adoption can change discontinuously.

Ask what the simple model excludes

Real users differ in benefit, support cost and bargaining power. The licence can finance maintenance or enable access that would otherwise be impossible. Conversely, negotiation costs can make a collection of separately useful technologies hard to combine. The effect depends on the arrangement rather than the mere presence of a right.

Do not use patent totals as a substitute for this analysis. A patent count does not directly reveal commercial use, customer benefit, research quality or the cost of assembling permissions. Some valuable improvements are not patented, and patents vary greatly in economic importance.

For learning, draw the cash flows and the resource flows separately. For an actual invention, decisions about ownership, disclosure, filing and permissions require the relevant official guidance and qualified advice. The economic exercise explains incentives; it does not determine whether a particular design infringes a right or qualifies for protection.

Previous chapter · Contents · Next chapter

A technology can be available at an attractive price and still be expensive to use well. The missing investment may be staff training, a reliable electricity supply, compatible equipment, data cleaning or a change in the workflow. Adoption is therefore a bundle of decisions rather than a single purchase.

Imagine a scheduling system costing 30. It promises annual savings of 20 for three years. A buyer looking only at those figures sees 60 − 30 = 30 before discounting. But installation costs 8, training costs 7 and annual maintenance costs 4. The same undiscounted comparison is now 60 − 30 − 8 − 7 − 12 = 3.

Change one complementary condition

If poor data causes the first year's savings to fall from 20 to 8, the net becomes −9. A small-looking implementation omission reverses the conclusion. If correcting the data costs 5 and restores the lost 12, it improves the result by 7, bringing the net to −2. Even a worthwhile repair does not necessarily make the whole purchase worthwhile under the assumed three-year horizon.

These calculations deliberately omit discounting to make the missing cost visible. A later investment appraisal must align timing and include residual value, uncertainty and alternatives. Do not compare a three-year cost with a five-year benefit because the longer horizon makes the proposal attractive.

Explain uneven diffusion

One workshop may already have suitable data and trained staff; another may need to build both. Identical purchase prices then conceal different effective costs. A low adoption rate can reflect these differences rather than ignorance or resistance. Equally, high adoption can reflect a subsidy without demonstrating lasting productivity gains.

How Technological Maturity Works helps explore how stable designs, interfaces and accumulated experience alter the work required around a technology. The history of containerisation in Singapore's port provides a route into a larger system of equipment, skills and coordination. Use that historical guide to investigate complements, rather than borrowing its events as proof that any present technology will succeed.

For a practical diagnostic, ask a non-adopter to reconstruct the whole transition. Separate lack of information from an unattractive transition cost. A better explanation of benefits cannot fix an incompatible production line.

Previous chapter · Contents · Next chapter

Research expenditure, researcher numbers, patents, launches, adoption and wellbeing occupy different points in a causal chain. A rise in one measure does not imply an equal rise in the next. Even a carefully collected input statistic cannot tell the whole story of what the input produced.

Bloom, Jones, Van Reenen and Webb examine whether ideas are becoming harder to find. Their evidence across several settings links rising research effort with declining measured research productivity. The study invites careful attention to the output produced per unit of research effort; it does not establish that every research field follows an identical trend or that a particular new tool has solved the problem.

Work through a comparable-output example

A fictional team uses 10 researcher-years to produce a five-percent improvement in a defined technical performance measure. Later it uses 20 researcher-years to obtain another five-percent improvement on the same measurement convention. The simple ratio falls from 0.5 to 0.25 percentage points per researcher-year.

That ratio halves, but several interpretations remain possible. The technical frontier may be harder. The second improvement may be more reliable, serve more users or require additional safety work. The team may have shifted resources toward a different objective not captured by the chosen measure. The arithmetic describes a ratio; explaining it requires evidence about comparability.

Watch the denominator and the time lag

If research effort is measured in money, changes in wages and equipment prices can alter the input without changing hours. If outcomes arrive years after spending, dividing this year's patents by this year's budget mismatches cohorts. If only successful projects remain in the dataset, the ratio may omit resources used by failures.

Build a measurement sheet with the unit of observation, input definition, output definition, price basis, research cohort and observation window. Record what happens to abandoned projects and organisations that close. A comparison that cannot explain those choices is difficult to interpret, however precise its chart looks.

For a small learning project, compare two versions of the same process using a stable task and quality standard. Record both preparation effort and operating performance. Then deliberately change the quality requirement and explain why the earlier productivity ratio may no longer be comparable. This makes the measurement problem tangible before moving to national innovation indicators.

Previous chapter · Contents · Next chapter

An organisation receiving support may subsequently grow, but it might have grown anyway. Innovation economics asks what activity and benefit the support causes beyond the relevant alternative. That question connects the incentive models to empirical evaluation without assuming that a particular funding instrument must work.

Sabrina Howell's research account discusses evidence from innovation-support programmes, including how selection and programme design affect results. Its lesson for evaluation is to examine who applies, what constraint the intervention changes and which outcomes follow. Results from a particular programme and group cannot simply be assigned to another scheme with different conditions.

Construct a transparent comparison

Suppose supported firms raise research spending from 100 to 150. Similar comparison firms increase from 100 to 130 over the same period. The raw supported-group increase is 50; the difference between changes is 20. Under a credible parallel-trends assumption and other design conditions, 20 could estimate additional spending attributable to support.

But similar starting values do not prove parallel trends. Supported firms may have had stronger projects ready to begin, different demand exposure or better access to staff. Applicants may select into the programme. Comparison firms may benefit from the funded research themselves, contaminating the comparison. Check these possibilities before treating 20 as a causal result.

Additional spending is an intermediate outcome

Even a well-identified increase in spending is not the same as an increase in social benefit. The new spending could buy scarce researchers away from another valuable project, or it could enable work that would otherwise not happen. Later outcomes may include better products, learning, customer gains and harms, with considerable delay.

Record administrative costs and the resources required to apply, monitor and comply. Also state the geographic and population boundary: a programme can benefit one place partly by drawing activity from another. Whether that is an acceptable objective is a policy judgement that should be visible.

The existing Innovation Policy guide develops instrument choice and accountability in more detail. The economic discipline here is narrower and demanding: identify the changed decision, build a defensible counterfactual, and avoid turning a gross activity total into a claim of net value.

Previous chapter · Contents · Next chapter

Imagine a report announcing that a new production method increased output by 25 percent. Begin by asking what output means. Units per hour, saleable units per paid hour, revenue and customer value are different measures. Then ask what changed alongside the method and whose costs are missing.

A quality-adjusted production case

Before adoption, a workshop makes 100 units in 10 hours, with 5 rejected. Saleable output is 95, or 9.5 units per hour. Afterwards it makes 125 units in 10 hours, with 20 rejected. Saleable output is 105, or 10.5 per hour. Gross production rises 25 percent, but saleable output per hour rises approximately 10.53 percent.

Now include two extra hours of inspection required by the new method. Saleable output per total hour becomes 105 ÷ 12 = 8.75, below the previous 9.5. This does not prove the innovation is worthless: material savings, quality differences or future learning may matter. It proves that the original headline did not specify a complete productivity boundary.

Match the repair to the mechanism

If rejects arise from poor calibration, repair the calibration process and retest. If inspection is temporarily high during learning, track its decline rather than assuming it will disappear. If the new method shifts work to another team, include that team's time. If the output mix changes, compare equivalent tasks or explain the quality adjustment.

Next examine incentives. A manager rewarded for gross output may prefer the flattering measure even when the wider process deteriorates. A supplier paid at installation may have little reason to help with later adaptation. These are hypotheses to investigate through contracts, workflow and observation, not accusations inferred from a disappointing figure.

Keep distribution visible

An aggregate gain can coexist with concentrated losses. Operators may need new skills, customers may face a less accessible interface, or smaller suppliers may incur compatibility costs. Report who benefits, who must adjust and whether the proposed remedy reaches them.

End the diagnostic with a decision that can be revised: continue a bounded trial, repair a specific defect, collect missing evidence or stop. State what observation would change that decision. How Innovation Works provides the practical implementation sequence; economic analysis adds the resource, incentive and comparison questions that keep the sequence accountable.

Previous chapter · Contents · Next chapter

Return to a fictional research project, but rebuild the answer rather than remembering a favourable result. The developer pays 120 now. With probability 0.5, success generates 240 units of operating surplus at the end of year two; failure generates zero. Discount at 10 percent. Separately, successful research lets other users save 80 at the end of year two. They need an immediate collective adaptation investment of 25, paid before the outcome is known. Assume all figures are real incremental resources or surplus, and that no component overlaps another.

Attempt the first calculation

The developer's expected net present value is 0.5 × 240 ÷ 1.1² − 120, approximately −20.83. The expected net external benefit is 0.5 × 80 ÷ 1.1² − 25, approximately 8.06. The combined value is therefore approximately −12.77.

Positive knowledge spillovers do not make this project worthwhile under the specified assumptions. A learner who automatically recommends support because others benefit has transferred a slogan rather than the reasoning.

Change when adaptation can occur

Now suppose users can wait until success is known at the end of year two, then pay 25 and immediately realise the 80 saving. The expected net external benefit becomes 0.5 × (80 − 25) ÷ 1.1², approximately 22.73. Combined with the developer's −20.83, the total becomes approximately 1.90.

The underlying saving is unchanged. The difference is that adaptation costs are delayed and avoided after failure. Those timing and contingency changes reverse the sign of the combined value. Neither calculation allows a claim that the result is robust: an omitted coordination cost of more than 1.90 in present value would reverse the second conclusion again.

Explain the remaining decision

A positive combined value does not establish how to finance the project. The developer still faces a negative private return, users may be hard to organise, and promises about future benefits may be difficult to enforce. A feasible agreement, funding arrangement or redesign must be assessed on its own costs and incentives.

To demonstrate understanding, explain the result without using the phrase “innovation is good for growth”. Identify the uncertain event, the two decision dates, who pays in failure and the non-overlapping benefits. Then change the success probability to 0.4 and recalculate. Under the delayed-adaptation scenario, combined value becomes 0.4 × 295 ÷ 1.1² − 120, approximately −22.48. The decision is conditional on the model's inputs.

Previous chapter · Contents · Next chapter

Begin with one ordinary process rather than a famous invention. A booking method, production instruction or scheduling tool provides enough complexity to practise. Describe its old and new versions, identify the resources required for transition and write down who decides whether to adopt. Use hypothetical data unless you have permission to collect and report real observations.

In a second session, draw separate private and wider accounts. Mark licence payments as transfers where appropriate, and keep resource costs visible. In a third, add uncertainty and a decision that can wait for evidence. In a fourth, design a comparison that could reveal whether the improvement actually occurred. Present the limits alongside the result.

Is innovation economics only about high-technology companies?

No. Its questions apply to services, production processes, organisational changes and the adoption of established methods. The relevant economic issue may be financing, information, coordination or the distribution of returns. A modest change used widely can warrant analysis even when it does not resemble a laboratory breakthrough.

Does a large social return justify any subsidy?

No. The estimate may be wrong, benefits may already be included elsewhere, or the intervention may fail to change the relevant decision. Administration, displacement and alternative uses of resources matter. Compare feasible instruments and the no-intervention case, rather than comparing a real programme with an idealised world in which its benefits arrive without costs.

Are patents a good innovation measure?

They can answer specific questions about recorded inventive activity and rights, with suitable qualifications. They cannot by themselves establish adoption or welfare. Match the measure to the claim, account for differences across fields and time, and avoid treating every patent as an equal unit of useful output.

Why can a good technology spread slowly?

Users may face different transition costs, uncertain benefits, missing complements or financing constraints. Waiting can sometimes be rational. Investigate the actual obstacle before prescribing more publicity or blaming users. The adoption calculation in Chapter 9 gives a starting point for that investigation.

Where should I read next?

Use Innovation Studies to develop implementation and evaluation questions, Innovation Policy for public instruments, and Organizational Economics for decisions inside firms. For primary research, the linked Romer, Hall–Lerner, Aghion and colleagues, Bloom and colleagues, and Howell sources each address a different mechanism. Read the original study's setting and assumptions before transferring its findings.

The final learning test is a clear explanation of a changed case. You should be able to say what is new, what resources it requires, why people choose it, what others gain or lose, and which evidence would make you revise the conclusion. That capability connects the subject back to Education and gives an innovation claim something firmer than enthusiasm to stand on.

Previous chapter · Contents · Gold Standard master