Successful prevention produces no invoice, no inquiry and no headline. Its value lies in a world that did not occur — which makes it easy to ignore, and just as easy to exaggerate.
A century of nothing
Great Britain has been free of rabies in terrestrial animals since 1922. For most of the century that followed, that freedom was maintained by a quarantine regime that was expensive, unpopular with pet owners and, in any given year, apparently uneventful. The Pet Travel Scheme replaced much of it from 2000 with vaccination, identification and documentation requirements — cheaper for travellers, but still a cost, still a control, and still a system whose success is recorded as a series of years in which nothing happened.
What was that century of nothing worth? The honest answer is that nobody can observe it. There is no ledger of dog bites that did not occur, post-exposure treatments that were never needed, or deaths that did not happen. Any figure would have to be built from a counterfactual: an estimate of what Britain’s rabies burden would have been without the controls. That estimate would depend on assumptions about wildlife reservoirs, pet movements, vaccination rates and public-health response that cannot be tested against the world as it is.
This is the problem this article addresses. The flagship article in this issue, The Economics of Prevention, identified the invisible return as the first of five structural reasons why biological security underinvests before the crisis. This piece examines it in detail: why successful prevention is so hard to value, how avoided-cost claims are built, why the most quoted of them deserve caution, and what an honest alternative looks like.
The accounting problem
Public accounts are built around transactions. They record what was spent, on what, and by whom. They are very good at showing the cost of a surveillance programme, a vaccine stockpile or a laboratory. They have no line for the outbreak that did not happen, because an outbreak that did not happen generates no transaction.
That asymmetry has consequences beyond the accounts themselves. Performance frameworks inherit it: prevention programmes are judged on outputs that can be counted — samples tested, animals vaccinated, inspections completed — because the outcomes they exist to produce cannot be. Budget negotiations inherit it: a line of spending with no visible return is the natural candidate for savings. And political attention inherits it: the minister who funds prevention pays for it now and, if it works, receives nothing to point to.
The result is what might be called the £0 outbreak. The more successful a preventive programme is, the more closely its record resembles a programme that achieved nothing. In the accounts, effective prevention and wasted money look the same.
The software industry’s experience with the year-2000 date problem is the best-known illustration outside biology. Substantial sums were spent to prevent computer failures at the turn of the millennium; very little failed; and the lack of failure was widely taken, afterwards, as evidence that the problem had been overstated. Whether it had been or not, the absence of damage could not settle the question — and that is exactly the point.
Four ways to build a counterfactual
Every claim about the value of prevention rests on a counterfactual: a statement about what would have happened without it. There are four common ways of constructing one. Each is legitimate. Each has a characteristic way of going wrong.
1. Before and after
The simplest approach compares the same place before and after an intervention. Foot-and-mouth disease in Britain is the classic case: the 2001 epidemic cost billions, the 2007 outbreak far less, and the difference is attributed to the preparedness improvements made in between.
The weakness is that other things change too. As the flagship showed, the 2007 outbreak began very differently — from a known laboratory site, with a laboratory strain, in a geographically compact area — so some of the difference between the two outbreaks has nothing to do with preparedness. Before-and-after comparisons are also vulnerable to regression to the mean: interventions are often introduced after an unusually bad year, and the next year is likely to be better regardless.
2. With and without
A second approach compares places that adopted an intervention with places that did not. If vaccinated districts have fewer cases than unvaccinated ones, the difference is attributed to vaccination.
The weakness is selection. Districts that adopt interventions early may differ from those that do not — in wealth, in veterinary capacity, in baseline risk — and those differences may explain the gap. Careful studies control for this; many published claims do not.
3. Modelled
A third approach builds an epidemiological model of how disease would spread without the intervention, and compares its output with what was observed or with a second run of the model including the intervention. Modelling is the only way to estimate the counterfactual for events that are rare or have never happened, and it is indispensable for biological security.
The weakness is model dependence. The avoided value is only as good as the assumptions about transmission, detection, contact patterns and response. Models are also easy to present with more precision than they possess: a single central estimate can conceal a range spanning an order of magnitude.
4. Scaled from the worst case
The fourth approach is the most common in advocacy and the least reliable. It takes the cost of a past catastrophe — a pandemic, an epidemic, a major outbreak — and sets it against the cost of a prevention programme: “COVID-19 cost trillions; this programme costs millions.”
The comparison is not meaningless; it establishes the order of magnitude at stake. But on its own it omits three things. It omits probability: the catastrophe may be rare, and prevention reduces its likelihood rather than eliminating it. It omits attribution: no single programme would have prevented all of the loss. And it omits the counterfactual path: some of the cost of a past catastrophe resulted from how it was handled, not from the fact that it occurred.
The anatomy of “£1 saves £X”
Ratios of the form “every £1 spent on prevention saves £X” are the currency of prevention advocacy. They are memorable, portable and persuasive. They are also, very often, detached from the conditions under which they were estimated.
Three well-known examples show how.
Public health interventions. A 2017 systematic review by Masters and colleagues in the Journal of Epidemiology and Community Health reported a median return on investment of 14.3 for public health interventions in high-income countries, rising to a median of 34.2 for health protection interventions such as vaccination. Those figures are widely quoted. Less often quoted are the range — from −21.3 to 221 — and the authors’ own caveats: the 52 included studies used a variety of calculation methods, could not be pooled, varied in quality, and were subject to publication bias towards positive results. The median is a fair summary of the published studies. It is not an estimate of what any particular intervention will return.
Disaster preparedness. Healy and Malhotra’s 2009 study of US disaster policy, cited in the flagship, estimated that a dollar of preparedness spending was worth about fifteen dollars in future damage mitigated. It is an influential finding, and its central argument about voter incentives is well supported. But it was estimated from US federal spending and natural disasters, and a figure like that is easily carried into contexts — pandemics, animal disease — that it was not designed to describe.
Pandemic preparedness. The return-on-investment case prepared by WHO and the World Bank for the G20 underpins the international call for around US$31 billion a year in pandemic prevention, preparedness and response. In August 2026, Brown and colleagues argued in Health Economics, Policy and Law that the case rested on improbable assumptions, including full mitigation of a pandemic’s economic impact, and on crude baselines. The flagship discussed that critique and the funding behind it. Whatever view one takes of the authors’ wider position, the specific point — that a ratio which assumes complete mitigation is not a credible estimate of what preparedness would buy — is correct.
The common problem is not that these ratios are false. It is that they are single numbers standing in for distributions, and that they travel without the assumptions that produced them. A ratio quoted without its counterfactual, its range and its evidence base is not wrong so much as unreadable.
Six questions to ask of any prevention ratio
Before relying on a “£1 saves £X” claim, it is worth asking:
What is the counterfactual? What is assumed to happen without the intervention, and on what evidence?
Whose costs are counted? Government only, or farmers, supply chains, health systems and households? A ratio for society and a ratio for the payer can differ by a factor of several, as The Prevention Paradox in this issue shows.
What probability is assumed? Is the ratio conditional on the event happening, or weighted by how likely it is?
How much of the avoided loss is attributed to this intervention? All of it, or the share it plausibly changes?
Over what period, and discounted how? Costs today and benefits in twenty years are not comparable without discounting.
What kind of evidence supports each step? Observed, empirically estimated, modelled, transferred from another setting, or assumed?
A ratio that can answer all six is worth taking seriously, whatever its size. One that cannot answer them is not yet an estimate.
The problem of the tail
There is a deeper reason why valuing prevention is hard, and it concerns the shape of biological risk.
Most years, most biological threats do little damage. Occasionally one does a very great deal. The distribution of losses is not a bell curve but a long tail, in which a small number of extreme events account for most of the total. Marani and colleagues, analysing four centuries of novel epidemics in a 2021 paper in PNAS, estimated that the annual probability of a pandemic on the scale of COVID-19 is around 2% — rare in any year, but likely within a lifetime. Cirillo and Taleb, writing in Nature Physics in 2020, argued that pandemic fatalities follow a fat-tailed distribution in which the average is dominated by the most extreme events and is therefore poorly estimated from limited data.
Two consequences follow for the economics of prevention.
The first is that expected values are highly sensitive to assumptions about the tail. A modest change in the assumed probability or size of the worst event can change the value of prevention several times over. A ratio built on a central estimate of the tail conceals most of the uncertainty that matters.
The second is that short-run evaluation will systematically undervalue prevention. If most of prevention’s value comes from rare large events, then most years will show no return at all. A programme evaluated over three or five quiet years will look like cost without consequence, even if its expected value is high. This is the same logic as insurance: a household that judged its buildings insurance on the basis of the years in which the house did not burn down would cancel it. The next article in this issue, Insurance Before Emergency, takes that analogy seriously as a financing mechanism.
The tail cuts both ways. It is also why scaled-from-the-worst-case comparisons are so seductive: choosing the most extreme past event as the benchmark makes almost any prevention look cheap. Honest analysis has to show both the central case and the tail, and say how much of the result depends on each.
An honest alternative
None of this is an argument for giving up on estimating the value of prevention. Decisions about prevention are made either way; the question is whether they are informed by transparent estimates or by slogans and silence. Five practices would make avoided-cost claims more honest and more useful.
State the counterfactual in words
Before any number, say what is assumed to happen without the intervention and why. “Without the programme, we expect an outbreak in roughly three of every ten five-year periods, based on incursion frequency in comparable countries over the past twenty years” is a claim that can be examined and challenged. A counterfactual embedded in a formula cannot.
Show ranges, not points
Every avoided-cost estimate should come with at least a low, central and high case, built from consistently pessimistic, central and optimistic assumptions. If the low case is below break-even, say so. A central ratio of 3 with a low of 0.4 is a very different finding from a central of 3 with a low of 2.
Ask what would have to be true
Often the most useful question is not “what is prevention worth?” but “what would have to be true for it to be worth doing?” This is break-even or threshold analysis, and it turns an unanswerable question into one that experts can discuss.
The fictional illustration in the OHS Prevention Ledger methodology note shows how. A five-year surveillance programme costs £1.9 million in present-value terms. An outbreak, if it occurred, would cost around £21 million. With a modest adjustment for timing and discounting, the programme breaks even if it reduces the probability of an outbreak over five years by about ten percentage points. The analyst’s central estimate is fifteen points; the low estimate is five. The debate then becomes a focused one — is a ten-point reduction plausible? — rather than an exchange of incompatible ratios.
Measure capability, not only absences
If outcomes cannot be observed, the capability that produces them can. Detection times, vaccination coverage, laboratory throughput, veterinary vacancy rates, tracing completeness and the confidence of farmers in reporting are all measurable before any outbreak. They are leading indicators: they change before outcomes do. The flagship argued that the decay of successful capability is one of the five mechanisms of underinvestment precisely because it is invisible in outcome data. Measuring capability directly makes the preventive system visible even in years when nothing happens. The OHS-TPI instrument published with Issue 001 applies the same logic to trust.
Keep the evidence type attached to the number
Every estimate is some mixture of observation, empirical research, modelling, transfer from elsewhere and assumption. Recording which is which — and how much of the avoided value depends on each — lets readers judge a ratio on its foundations rather than its size.
From principles to practice: the Prevention Ledger
These practices are built into the OHS Prevention Ledger v1.0, published with this issue. The Ledger does not calculate the value of an outbreak that never happened. It records the chain of evidence and assumptions required to estimate the economic value of changing the probability or consequences of that outbreak: investment, capability, risk change, outcomes and economic consequences, each with its evidence type, across low, central and high scenarios, with the counterfactual stated and the distribution of costs and benefits shown.
The Ledger is not a solution to the counterfactual problem, which has none. It is a discipline for being honest about it. The applied research article in this issue, What Does Prevention Actually Buy?, runs the first real case through it: mass dog vaccination for rabies control in Ethiopia.
What would prove this wrong?
If ex-post evaluations of prevention programmes consistently confirmed the headline ratios claimed for them in advance, the caution urged here would be unnecessary.
If biological losses were not dominated by rare extreme events — if they were distributed more evenly across years — short-run evaluation would be more reliable than this article suggests, and the case for capability indicators would be weaker.
And if decision-makers given ranges, counterfactuals and threshold analysis made no different decisions from those given single headline ratios, the practical value of greater honesty would be limited — although it would still be more honest.
Conclusion
The £0 outbreak is the defining difficulty of prevention economics. Success leaves no trace, so it is undervalued; and because it leaves no trace, its value can be asserted without being tested, so it is also overclaimed. Both errors come from the same place: treating the counterfactual as something to be assumed rather than argued.
The way through is not a bigger number or a smaller one. It is to show the working: what would have happened otherwise, how likely it was, how much of it the intervention could plausibly change, who would have borne the cost, and how sure we are of each step. That will not make the value of prevention visible in the accounts. It will make it visible to anyone willing to read the reasoning.
Questions & Answers
Why is a prevented outbreak so hard to value?
Because the outcome of successful prevention is an absence, and accounts record transactions rather than absences. The most successful preventive programme can therefore look, in the ledger, like money spent for nothing.
What is wrong with “£1 spent on prevention saves £X”?
Those ratios travel well precisely because they leave their assumptions behind. Every avoided-cost claim rests on a counterfactual — a statement about what would have happened otherwise — and the ratio hides which counterfactual was used and how wide the uncertainty is.
Does that mean we should stop estimating?
No. The honest alternative is to state the counterfactual, show ranges rather than a point estimate, and ask what would have to be true for prevention to pay. That is harder to put in a headline and considerably more defensible.
References
- Brown, G.W., von Agris, J., Tacheva, B. and Bell, D. (2026) ‘An investment too good to be true?: Reassessing the World Health Organization and World Bank return-on-investment estimates for pandemic preparedness’, Health Economics, Policy and Law, published online 6 August 2026. doi.org/10.1017/S174413312610067X
- Cirillo, P. and Taleb, N.N. (2020) ‘Tail risk of contagious diseases’, Nature Physics, 16, pp. 606–613. doi.org/10.1038/s41567-020-0921-x
- Healy, A. and Malhotra, N. (2009) ‘Myopic voters and natural disaster policy’, American Political Science Review, 103(3), pp. 387–406. doi.org/10.1017/S0003055409990104
- HM Treasury (2022) The Green Book: Central Government Guidance on Appraisal and Evaluation. www.gov.uk/government/publications/the-green-book-appraisal-and-evaluation-in-central-government
- Marani, M., Katul, G.G., Pan, W.K. and Parolari, A.J. (2021) ‘Intensity and frequency of extreme novel epidemics’, Proceedings of the National Academy of Sciences, 118(35), e2105482118. doi.org/10.1073/pnas.2105482118
- Masters, R., Anwar, E., Collins, B., Cookson, R. and Capewell, S. (2017) ‘Return on investment of public health interventions: a systematic review’, Journal of Epidemiology and Community Health, 71(8), pp. 827–834. doi.org/10.1136/jech-2016-208141
- Rose, G. (1981) ‘Strategy of prevention: lessons from cardiovascular disease’, British Medical Journal, 282, pp. 1847–1851.
- World Bank and WHO (2022) Analysis of Pandemic Preparedness and Response (PPR) Architecture, Financing Needs, Gaps and Mechanisms. Paper prepared for the G20 Joint Finance & Health Task Force.
Key Takeaways
- The outcome of successful prevention is an absence. Accounts record transactions, not absences, so the most successful preventive programme can look, on paper, like a budget that bought nothing.
- Every avoided-cost claim rests on a counterfactual: a statement about what would have happened otherwise. There are four common ways of constructing one, and each has a characteristic way of going wrong.
- Headline ratios such as "£1 spent on prevention saves £X" travel well precisely because they leave their assumptions behind. The best-known examples come with ranges and caveats that rarely travel with them.
- Biological risk is dominated by rare, very large events. That makes average returns highly sensitive to assumptions about the tail — and means that judging prevention on a few quiet years will systematically undervalue it.
- The honest alternative is not to stop estimating. It is to state the counterfactual, show ranges, ask what would have to be true for prevention to pay, and measure the capability that prevention buys rather than waiting for the outbreak that shows it was missing.