THE HILL REPORT A Number at Risk Is a Count of ExposureConnor Hill · InsightfulWord · August 30, 2026  Figures of the form "X million people are at risk" appear constantly, across health, finance, safety and policy, and they are among the most misread statistics in circulation. The misreading is not about the arithmetic, which is usually correct. It is about what the number counts. In nearly every case, such a figure is a count of people who share an exposure — who take a medication, live in a floodplain, hold a type of account, work in an industry, carry a genetic variant. It is a description of a population, not of an outcome. The number of people who will actually be harmed is a different quantity, obtained by multiplying the exposed population by the probability of harm given exposure. That probability is frequently very small, and it is the number that almost never appears alongside the headline. The two get conflated because the language does not distinguish them. "At risk" is true of anyone whose probability of harm exceeds zero, which includes essentially everyone for essentially everything. Stated without a probability, the phrase has no information content beyond the size of the group. There is a second and subtler problem. Even where a probability is supplied, the relevant question is usually not the absolute risk but the excess risk attributable to the exposure — how many of those harms would have occurred anyway. A population with a background rate of an outcome will produce that outcome whether or not the exposure exists, and attributing all of it to the exposure is the most common error in this genre. None of this implies that large risk figures are always exaggerated. Some exposures carry substantial probabilities, some populations are genuinely large, and some warnings are correct and underplayed. The point is narrower: the headline figure alone cannot distinguish those cases from the others, and the distinguishing numbers are usually published. What follows is how these figures are constructed, the difference between absolute and attributable risk, why the denominator determines everything, what makes a risk figure checkable, and the questions that resolve one in a few minutes. How the Figures Are ConstructedThree ingredients go into any population risk estimate, and the headline usually reports only the first. The exposed population is a count — how many people have the characteristic. This is generally the most reliable component, drawn from prescription data, census data, survey data or administrative records, and it is usually accurate to within a reasonable margin. The risk given exposure comes from studies, and its quality varies enormously. It may derive from a randomized trial, from a cohort study, from a registry, or from a model. Its confidence interval is frequently wide and is frequently dropped when the figure travels. The baseline risk — the rate in a comparable unexposed population — is the third and is omitted most often. Without it, no statement about excess harm is possible. Multiplying the first by the second produces an expected count of events among the exposed. Subtracting what the baseline would have produced gives the excess. Only the last of those three quantities describes harm attributable to the exposure, and only the first routinely reaches a headline. Absolute, Relative and AttributableThree ways of expressing the same finding produce very different impressions, and all three are correct. 📈 Number of the Day Exposure is not outcome A figure describing how many people are at risk counts a population sharing an exposure. The number harmed is that population multiplied by the probability of harm, and the number harmed because of the exposure is that figure minus what the background rate would have produced anyway. The first quantity is the one that travels; the third is the one that answers the question. Source: Centers for Disease Control and Prevention, principles of epidemiology in public health practice. |
Support or oppose: should risk statistics be required to state absolute alongside relative figures? Supporters argue that relative figures without absolutes are the single most reliable way to mislead without lying, that both are computed from the same data, and that several editorial bodies already require it. Opponents answer that absolute risks vary by subgroup so a single figure misleads differently, that mandating format invites compliance without comprehension, and that the discipline belongs in editing rather than regulation. Which would improve things more? Hit reply — one line is enough. |
A relative figure states the ratio between exposed and unexposed — a doubling, a fifty percent increase. It is what studies usually report because it is comparatively stable across populations. An absolute figure states the rate itself — so many events per thousand people per year. A doubling of a rate of one in a hundred thousand produces two in a hundred thousand, which is a doubling and is also a change of one event per hundred thousand people. An attributable figure states how much of the total is explained by the exposure, given how common the exposure is. A small relative increase in a very common exposure can produce more total harm than a large relative increase in a rare one. Each answers a different question. A relative figure is what a person with the exposure asks about themselves. An absolute figure translates it into consequences. An attributable figure is what a policymaker needs. Quoting one and implying another is the standard mechanism by which technically true statements mislead. The direction of the distortion is predictable. Where a baseline rate is low, relative figures sound alarming and absolute figures sound negligible, so anyone wishing to alarm reports the relative and anyone wishing to reassure reports the absolute. Both are accurate. Seeing only one is the reason the same finding can support two opposite headlines on the same day. Why the Denominator Determines EverythingReports of harm without a denominator are the weakest form of evidence and the most emotionally effective. Context — where the numbers themselves are published Population risk estimates in the United States are generally traceable. Adverse event reports for medicines appear in the FDA's public dashboard, disease and injury rates in CDC's WONDER and surveillance systems, exposure prevalence in the National Health and Nutrition Examination Survey, and hazard exposure in agency-specific registries. Each states its own limitations, and the limitation statement is usually the most useful part of the document. Anyone who wants to check a figure can generally find its source in under half an hour, and the source usually contains the qualifications the figure lost on the way to the headline. |
A count of adverse events establishes how many were reported. It does not establish how many people were exposed, so it cannot yield a rate. Reporting systems for medicines and devices work this way by design: they exist to detect signals, not to measure incidence, and the agencies running them say so explicitly. A rising count can therefore reflect rising harm, rising exposure, rising awareness, a change in reporting rules, or media attention. Distinguishing among these requires the denominator and the reporting-rate history, and both are published. The same applies to lists of cases. A collection of individual accounts, however numerous and however genuine each one is, cannot establish a rate, because nobody assembled a comparison group. The accounts may all be true and the inference from them may still be unsupported. This is worth stating carefully, because it is frequently heard as dismissal of the people in the accounts. It is not. Individual reports are how nearly every genuine safety signal in history was first noticed, and taking them seriously is exactly what a functioning system does. What a collection of them cannot do is measure how often something happens, and the step from noticing to measuring requires a denominator that the collection itself does not contain. What Makes a Risk Figure CheckableFive items make any such claim evaluable, and their presence or absence is usually apparent at a glance. The source of the exposure count, which determines whether the population figure is a measurement or an estimate. The study design behind the risk estimate, since a randomized trial, a cohort study and a model carry very different weight. The absolute risk, stated as a rate over a defined period. The baseline rate in a comparable unexposed group. And the confidence interval, which describes how much the estimate could move with more data. A figure supplying all five is doing serious work. A figure supplying only the population count is a description of how many people share a characteristic, which is a fact about demography rather than about danger. The Questions That Resolve One QuicklyFour questions, asked in order, settle most claims of this shape. How many people have this exposure, and how was that counted. This is usually easy to verify and is usually accurate. Out of every thousand exposed people, how many experience the outcome in a year. If this cannot be answered, the claim has no rate behind it. How many of every thousand unexposed people experience it. This is the comparison, and its absence is the single most common defect. And what happens to the number if the exposure is removed. That is the question the whole exercise is meant to answer, and it is answerable only from the previous three. The general observation worth carrying is that a large number of people is not the same as a large amount of harm, and that whichever of the two a claim is actually about is usually determinable from the published source in less time than it takes to argue about it. The bill, not the debate A figure describing millions at risk counts people who share an exposure, and it says nothing about how many will be harmed until it is multiplied by a probability and compared against what would have happened anyway. Both of those numbers are usually published in the same document the headline came from. When a population figure is put in front of you, do you know its denominator? Connor Hill reads every reply. |
Connor Hill · InsightfulWord |
| 
