Imagine three health advisories about the same study. The first says a common medication causes a 100% increase in the risk of a serious side effect. The second says the risk rose from 1 in 10,000 to 2 in 10,000. The third says the increase is estimated at somewhere between 20% and 250%. All three are accurate, and all three come from the same data.
This is the central challenge of public health communication. Health authorities draw their findings from thousands of patients, imperfect surveillance and evidence that keeps changing. They then have to compress those findings into a few sentences that millions of people will act on. Every choice of framing changes what the numbers appear to mean.
None of these framings is a lie. Each one emphasises something different, and each can mislead if you read it on its own. Learning to translate between them is a practical skill. It helps you tell genuine warnings apart from statistical noise and recognise when experts are honestly uncertain. It also shows you what a population-level number means for you personally. Three tools do most of the work.
Absolute vs Relative Framing
In October 1995, the UK Committee on Safety of Medicines warned that certain newer contraceptive pills roughly doubled the risk of venous thrombosis compared with older formulations. The headline figure, a twofold increase, was accurate. The baseline received far less attention. The risk rose from around 1 in 7,000 women per year to around 2 in 7,000.
The consequences of that framing were measurable. Many women stopped taking the pill abruptly. Later analyses linked the scare to a rise in unplanned pregnancies and an estimated 13,000 additional abortions in England and Wales the following year. Pregnancy itself carries a higher thrombosis risk than either pill. A relative number, presented without its baseline, led people to decisions that increased the very harm they feared.
Relative risk tells you how much a factor multiplies the baseline. Absolute risk tells you how many people are actually affected. A third figure is also useful: the absolute risk difference, which here is one extra case per 7,000 women per year. Turned around, it becomes the number needed to harm, meaning roughly how many people must be exposed for one additional person to experience the outcome. In this case, that number was about 7,000.
Each framing answers a different question. Researchers often focus on relative risk because it signals an effect worth investigating. Individuals need absolute risk to make decisions. When an advisory gives only a percentage increase or decrease, ask first: a percentage of what? The same logic applies to benefits. A treatment that halves risk sounds impressive until you learn it moves the risk from 2 in 1,000 to 1 in 1,000. That is real, but modest.
TakeawayA relative risk without a baseline is a multiplier with nothing to multiply. Whenever you see a percentage change, translate it into people per thousand before deciding how much it matters.
Uncertainty Communication
Early in an outbreak, every number is provisional. In March 2020, the World Health Organization reported that about 3.4% of confirmed COVID-19 cases had died. The calculation was correct for the data available, but many people heard it as their chance of dying if infected. It was not. The figure left out mild and undetected infections, which would have enlarged the denominator. It was also skewed by which patients got tested.
Later studies estimated the overall infection fatality rate at a fraction of that figure, with enormous variation by age. The lesson is not simply that the early number was wrong. A single figure implied a precision the evidence could not yet support. A plausible range spanning several-fold would have been harder to quote, but far more honest about what was actually known.
A confidence interval is designed to express exactly this. A 95% interval comes from a procedure that would capture the true value about 95% of the time across many repeated samples. Wide intervals signal small samples, noisy measurement or high variability. Narrow ones signal more data and more consistency. The tradition of statistical inference that Ronald Fisher helped build rests on a simple idea: an estimate is incomplete without a measure of its error.
Ranges also change how you should respond to updates. When an authority revises an estimate, the interval is often just narrowing as evidence accumulates. That is not a sign that experts were misleading you. Language like preliminary, likely or between X and Y signals that experts are matching their confidence to the evidence. Be more wary of a confident single number about something new. At the start of an emerging risk, that kind of precision usually has not been earned yet.
TakeawayAn honest estimate comes with its error bars attached. When the range is wide, let your caution match the uncertainty instead of leaning towards certainty in either direction.
Personal Risk Calculation
Health statistics describe populations. A figure such as 'one in 500 people will be hospitalised' averages across the young and the old, the vaccinated and the unvaccinated, and the frequently exposed and the barely exposed. Almost nobody sits at that average. To turn a population number into a personal one, ask how your circumstances differ from the group the number describes.
A useful mental model splits personal risk into two stages. The first is the probability of exposure: how likely you are to encounter the hazard at all, given where you live, work and travel. The second is the probability of harm given exposure: how likely a bad outcome is if you are exposed, which depends on age, health conditions and protective measures. Multiplying the two gives a rough personal estimate. A severe outcome you are very unlikely to encounter can matter less than a moderate one you are almost certain to face.
Relative risks help here, as long as you apply them to your own baseline rather than the population average. If a factor doubles risk and your baseline is a tenth of the average, your doubled risk is still well below average. If your baseline is ten times higher, the same doubling carries far more weight. One caution: relative risks are not always constant across groups. A factor that doubles risk in older adults may do little in younger ones, so look for subgroup data where it exists.
Personal risk calculation also has limits. Public health advice often targets the population rather than the individual, for two reasons. Small personal risks add up to large collective burdens, and one person's exposure affects others. Your estimate may give you reason to feel calmer. It rarely justifies dismissing guidance designed around transmission and shared resources. The aim is to know where you stand, not to find reasons to opt out.
TakeawayYour risk is the population figure adjusted for your exposure and your vulnerability. Estimate each one separately, and remember that some advice exists for everyone's sake, not just yours.
Risk communication is a translation problem. Behind every advisory sits a dataset, a set of assumptions and a choice about which number to show you. Reading an advisory well means reconstructing what was left out.
Three questions do most of the work. What is the absolute risk in people per thousand, rather than just the percentage change? How wide is the uncertainty, and is the estimate likely to shift as evidence accumulates? How do your exposure and vulnerability compare with the population the number describes?
None of these questions requires advanced mathematics. They require the habit of treating numbers as measurements rather than verdicts: imperfect, dependent on context and always worth a second look.