“The risk doubled” tells you a ratio. It does not tell you the starting risk. A change from one event per 10,000 people to two per 10,000 and a change from one per ten to two per ten are both doublings, but their absolute differences are very different.
Absolute risk describes the probability of an event in a defined population over a specified period. Relative risk compares that probability between groups. Both can be useful. A clear health statement keeps them together rather than allowing a striking ratio to stand in for the whole comparison.
Every number below is invented for arithmetic illustration. The examples do not estimate a real disease, recommend a treatment, or predict an individual's future.
Start with two groups and the same observation period
Imagine two groups of 1,000 people followed for one year. In Group A, 40 experience a defined event. In Group B, 20 experience the same event. Assume complete follow-up and the same counting method for this simplified example.
Group A's observed risk is 40 divided by 1,000, or 4 percent. Group B's is 20 divided by 1,000, or 2 percent. Those are absolute risks over the stated year.
The risk ratio for B compared with A is 2 percent divided by 4 percent, or 0.5. B's observed risk is half A's. That can also be expressed as a relative reduction of 50 percent.
The absolute difference is 4 percent minus 2 percent: two percentage points. In the common denominator of 1,000 people, that is 20 fewer observed events. “Fifty percent lower” and “20 fewer per 1,000” are compatible descriptions of the same comparison.
| Measure | Calculation | Meaning in the example |
|---|---|---|
| Risk in A | 40 ÷ 1,000 | 4% over one year |
| Risk in B | 20 ÷ 1,000 | 2% over one year |
| Risk ratio, B relative to A | 2% ÷ 4% | 0.5 |
| Relative reduction | 1 − 0.5 | 50% lower relative to A |
| Absolute difference | 4% − 2% | 2 percentage points |
The comparison alone does not establish why the groups differ. If a study is observational, other differences between groups may matter. Even in a trial, uncertainty and the study's particular conditions belong with the estimate.
Change the baseline and preserve the same ratio
Now imagine another pair of groups, also followed for one year, in which the risks are 0.4 percent and 0.2 percent. The risk ratio is still 0.5, and the relative reduction is still 50 percent.
The absolute difference is now 0.2 percentage points, or two fewer events per 1,000 people. That is one tenth of the absolute difference in the first example, despite the identical relative reduction.
| Fictional comparison | Starting risk | Comparison risk | Relative reduction | Absolute difference |
|---|---|---|---|---|
| Higher baseline | 4% | 2% | 50% | 20 per 1,000 |
| Lower baseline | 0.4% | 0.2% | 50% | 2 per 1,000 |
This is why a relative percentage cannot identify the number of people affected without a baseline. It also explains why a result estimated in one population may have different absolute implications in another population with a different underlying risk.
That observation is not permission to insert any convenient personal baseline into a published ratio. The ratio itself may not apply equally across populations, and a valid risk estimate may require information the reader does not have. The example explains the arithmetic, not an individualized calculator.
Percentage points are not percent change
A fall from 4 percent to 2 percent is a decrease of two percentage points. Calling it a “2 percent reduction” is ambiguous and, if meant as a relative change, incorrect.
To calculate the relative decrease, divide the difference by the starting value: two percentage points divided by four percent equals 50 percent. To calculate the absolute difference, subtract the two percentages without that additional division.
The guide to percentage points and percentage change explains the same arithmetic outside health. In a health context, the error matters because it can make an effect sound much larger or smaller than the source supports.
When writing a comparison, add the unit explicitly. “Two percentage points” and “50 percent relative reduction” leave less room for misunderstanding than a bare “two percent” or “half the risk.”
The reference group determines the direction
In the first example, B has half the risk of A. Viewed in the reverse direction, A has twice the risk of B. The ratio changes from 0.5 to 2 because the reference group changes.
Relative percentage changes are therefore not symmetric. Moving from 4 percent to 2 percent is a 50 percent decrease. Moving from 2 percent to 4 percent is a 100 percent increase. Both describe the same pair of values with a different starting point.
This is not a trick; it follows from the denominator. A reader can resolve it by writing “compared with which group?” beside the ratio. The words exposed, unexposed, treated, untreated, or reference group need to match the study's actual definitions.
If a chart reverses the reference category while retaining the old verbal description, the interpretation can become wrong even though the numerical calculation was performed correctly. Labels deserve the same scrutiny as arithmetic.
The outcome cannot change halfway through the sentence
A risk of developing a condition is different from a risk of being hospitalized with it. A risk of any symptom is different from a risk of a severe outcome. The ratio belongs to the particular outcome used in the study.
Imagine a fictional analysis reporting a lower risk of hospitalization. Rewriting that as a lower risk of every infection changes the claim. The study may not have measured every infection, especially those that never led someone to seek care.
The distinction also applies to a composite endpoint that combines several events. A result for the combined measure does not necessarily describe the same effect on each component. The source needs to show the component results before that claim can be made.
Our article on vaccine efficacy and effectiveness uses this principle in a specific evidence setting. The percentage should remain attached to the outcome, population, and observation period rather than becoming a universal product score.
One year is not a lifetime
Risk accumulates over a stated interval. A one-year estimate and a lifetime estimate answer different questions. Comparing them directly without aligning the time horizon can create a dramatic but meaningless contrast.
Nor can a one-year probability always be multiplied by a number of years to obtain a valid long-term probability. The chance can change with age or circumstances, and people who have already experienced the event may no longer belong to the same at-risk group. Other events can also affect what happens during follow-up.
For a simple illustration unrelated to disease, repeated annual probabilities cannot exceed a total probability of one. Blind multiplication can eventually produce a value above 100 percent, revealing that the method is not a general rule for cumulative risk.
The article on incidence and prevalence also distinguishes risks from person-time rates. A figure per 1,000 person-years has time built into its denominator. It should not be silently rewritten as the percentage of people who will experience the event during any chosen interval.
Odds and risk use different denominators
Risk compares events with all people in the relevant group. Odds compare events with non-events. When an event is uncommon, the numerical values can be close; as it becomes more common, the difference can be substantial.
In a fictional group where 20 of 100 people experience an event, risk is 20 divided by 100, or 0.2. The odds are 20 divided by the 80 without the event, or 0.25. Those are not two alternative ways of writing exactly the same number.
Suppose a second group has 40 events among 100 people. Its risk is 0.4 and its odds are 40 divided by 60, about 0.667. The risk ratio is 0.4 divided by 0.2, or 2. The odds ratio is about 0.667 divided by 0.25, or 2.67.
Calling that odds ratio “2.67 times the risk” would overstate the risk ratio in this example. A paper's stated measure should be preserved. Case-control studies and adjusted models may report odds ratios for reasons related to their design; the terminology is not optional decoration.
Zero observed events is not proof of zero possible risk
If a small study observes no events, the observed count is zero. That does not establish that the event can never occur in the wider population. The amount of observation and statistical uncertainty still matter.
A ratio can also become difficult to calculate when the reference group's observed risk is zero. Dividing by zero does not produce a meaningful ordinary risk ratio. Researchers may use methods suited to sparse data, but a reader should not replace the undefined calculation with a confident statement of infinite or perfect effect.
For a fictional example, observing zero events among 20 people provides a different amount of information from observing zero among 20,000 people over the same period. Both counts are zero, but the observation base is very different.
Keep the sample size and follow-up visible when a headline emphasizes “no cases.” That phrase describes what was observed, not a guarantee about what is possible.
Precision and causation are separate issues
A confidence interval describes statistical uncertainty under the analysis. It does not automatically account for every bias, data-quality problem, or difference between the studied population and a reader's circumstances.
Likewise, a precisely estimated association is not necessarily causal. If two groups differ in other relevant ways, those differences may help explain the result. A large dataset can estimate a biased comparison very precisely.
The question “How uncertain is the number?” is therefore different from “Does this design support a causal conclusion?” Both matter. A credible summary discusses the design, the estimate, and its limits rather than relying on the number of decimal places.
This also prevents an unhelpful all-or-nothing response to uncertainty. A study can contribute useful information while leaving important questions unresolved. The right conclusion should match the evidence rather than swinging between certainty and dismissal.
Repair a risk claim by restoring its missing fields
Take the sentence “This lowers risk by 30 percent.” To make it interpretable, look for the event, reference group, baseline risk, population, period, and type of measure. Is 30 percent a relative reduction, an absolute percentage-point difference, or a different statistic being summarized loosely?
If the original source supplies those fields, a clearer summary can state both group risks and the difference over the same interval. If it does not, leave the missing information visible rather than inventing a baseline to make the claim feel complete.
Finally, separate the evidence explanation from a personal decision. Benefits, harms, alternatives, preferences, and individual circumstances can all matter in care. Arithmetic helps clarify an estimate, but it does not by itself decide which option a person should choose.
Absolute risk tells how often an event occurs in the defined group and period. Relative risk tells how that frequency compares with another group. Their value comes from using them together, with the outcome and denominator intact.
Sources
- NCI: Absolute risk
Absolute risk describes the probability of a defined event over a specified period.
- NCI: Relative risk
Relative risk compares the risk of the same event in two groups and is also called a risk ratio.
- NCI: Odds ratio
An odds ratio compares odds rather than risks and is often used in case-control research.
- NCI: Cancer screening overview, risk explanations
Risk communication needs a population, time period, and a distinction between absolute and relative measures.