To translate probability, odds and risk accurately, you must preserve which event is being measured, which population forms the denominator and which comparison group acts as the reference. A probability of 20% corresponds to odds of 1 to 4 in favour of the event, not odds of 1 to 5. A risk ratio of 2 does not mean risk increased by two percentage points, and an odds ratio of 2 does not automatically mean probability doubled. These are related statistical ideas, but they are not interchangeable translation terms.
This guide explains how to translate probability, odds, absolute risk, risk difference, risk ratio, relative risk and odds ratio without strengthening or weakening a source claim. It is intended for educational, research, public-information and general technical writing. All health-style examples are fictional teaching cases, not medical advice, treatment guidance or claims about real interventions. The translation task is to preserve statistical relationships, not to tell a reader what decision to make.
The working method is to reconstruct every source claim as a small table before rewriting it. Record the event count, non-event count, total population, probability or risk in each group, and which group appears in the numerator of any ratio. Then translate the prose. This prevents fluent target sentences from reversing reference groups, turning odds into probability, or replacing an absolute difference with a more dramatic relative comparison.
A fifty-second orientation
Probability or risk expresses events relative to all relevant opportunities under the stated design. Odds compare events with non-events. A risk ratio compares the risk in one group with the risk in another. An odds ratio compares two sets of odds. The CDC’s Principles of Epidemiology section on measures of association distinguishes risk ratios, rate ratios and odds ratios and shows why the reference group matters.
Start every translation with four questions: What event counts as the outcome? What is the denominator? Which group is the comparison group? Is the source reporting an absolute quantity or a relative quantity? If those answers survive, the target wording can change naturally without changing the statistical meaning.
Keep mathematical verification separate from editorial interpretation. You can calculate a risk difference or odds privately to check the source, but do not add a new measure to the public text unless the assignment permits explanation or adaptation. Translation fidelity includes preserving which statistical summary the author chose to report.
1. Probability describes an event relative to possible cases
Suppose a fictional dataset contains 20 events among 100 observations. The event probability or risk under this simple setup is 20/100 = 0.20, or 20%. The denominator contains both the 20 events and the 80 non-events.
A translation that says “20 events per 80 non-events” has changed the measure into odds. The numbers are drawn from the same table, but the denominator is now different. Probability answers “what fraction of the total experienced the event?” Odds answer “how many events occurred relative to non-events?”
This distinction is especially important when target-language terminology uses one everyday word for several kinds of chance. In technical writing, use the source’s defined concept rather than the most familiar colloquial synonym. When necessary, add a short explanatory phrase that makes the denominator visible.
2. Odds compare events with non-events
With 20 events and 80 non-events, the odds in favour of the event are 20:80, which simplify to 1:4. The odds against the event are 80:20, or 4:1. The order of the words determines the order of the numbers.
Do not translate “odds of one to four in favour” as “one in four probability.” Odds of 1:4 correspond to one event among five total cases, giving a probability of 20%. A one-in-four probability is 25% and corresponds to odds of 1:3.
A simple QA rule is to rebuild the total from odds. For odds a:b in favour of an event, the corresponding probability under the simple two-outcome interpretation is a/(a+b). If the target prose gives a different probability without explanation, the translation needs review.
3. Probability and odds use different scales
A probability of 50% corresponds to odds of 1:1. A probability of 75% corresponds to odds of 3:1. A probability of 90% corresponds to odds of 9:1. As probability approaches 100%, odds can grow without a fixed upper bound.
This nonlinear relationship means that a fixed change in odds does not correspond to a fixed change in probability. Translators should therefore avoid substituting one measure for another merely because the target audience finds one easier to understand.
If an explanatory adaptation converts odds into probability, calculate the conversion explicitly and label it. Preserve the original measure somewhere when traceability matters. The source’s statistical scale should not disappear inside a friendly rewrite.
4. Absolute risk keeps the event probability in a particular group
Imagine a fictional group A with 10 events among 100 people and group B with 20 events among 100 people. The absolute risks are 10% and 20%. These values are directly interpretable as event proportions under the simplified example.
A translation should preserve which risk belongs to which group. Moving the group labels while leaving the numbers in place reverses the evidence. This can happen when target syntax naturally introduces the comparison group first.
Keep a temporary note: A = 10%; B = 20%. After rewriting the sentence, confirm the same mapping. This small discipline protects against one of the most consequential errors in comparative statistical translation: the correct numbers attached to the wrong populations.
5. Risk difference is an absolute difference
Using the fictional 10% versus 20% example, the risk difference B minus A is 10 percentage points. If the subtraction order is A minus B, the result is −10 percentage points. The sign depends on which group is defined as the first term.
Do not translate a ten-percentage-point difference as a 10% relative increase. From 10% to 20%, the relative increase is 100%. The absolute and relative descriptions are both mathematically related to the same endpoints, but they answer different questions.
The existing Translate guide to percentages and percentage points develops that distinction in detail. In risk translation, preserve whether the source reports the absolute gap or the relative comparison.
6. Risk ratio compares risks by division
The CDC defines risk ratio, also called relative risk, by dividing risk in one group by risk in a comparison group. In the fictional example, 20% divided by 10% gives a risk ratio of 2 when group B is compared with group A.
A risk ratio of 2 means the numerator group has twice the risk of the denominator group under the stated comparison. It does not mean the absolute risk is 2%, two percentage points higher, or that every individual’s personal risk doubled.
Translate the group order explicitly at first mention: “risk in B divided by risk in A.” This is especially useful when a source uses labels such as exposed and unexposed, treatment and control, or current and reference. The ratio only becomes interpretable when the reference group is visible.
7. Reversing the reference group inverts the risk ratio
If B-to-A risk ratio is 2, then A-to-B risk ratio is 0.5. These values describe the same pair of risks from opposite reference directions. A translation that changes which group is foregrounded must also preserve the mathematical direction.
This is a common source of accidental contradiction. A target sentence may say “A had half the risk of B” and later repeat the original ratio of 2 without changing its direction. The prose and number then disagree.
Use a reciprocal check. If the target reverses the group order, multiply the original ratio by the target ratio. If both are exact reciprocals, the product should be one. This is a quick way to detect direction errors in comparative claims.
8. A risk ratio of one has a special interpretation
A risk ratio of 1 means the compared risks are equal. Values above 1 indicate greater risk in the numerator group under the specified direction; values below 1 indicate lower risk in that group. The ratio itself does not establish causation.
Translate words such as associated with, linked to, observed among and caused by according to the source. Do not strengthen an observational comparison into a causal claim merely because the ratio differs from one.
Likewise, do not replace “no difference detected” with “identical risk” unless the statistical evidence and source wording support that stronger statement. Translation must preserve evidence status as well as numerical direction.
9. Relative risk reduction and absolute risk reduction answer different questions
Suppose a fictional event risk falls from 20% to 10%. The absolute reduction is 10 percentage points. The relative reduction is 50% because the decrease of ten points is half of the original 20% risk.
A target sentence saying “risk fell by 50 percentage points” would be impossible here because the final risk would be negative. A sentence saying “risk fell by 10%” would also be misleading if the source means an absolute ten-point decrease.
Record the baseline, endpoint, absolute difference and relative change separately. Then publish only the measure the source actually reports. Private arithmetic can prevent public exaggeration.
10. Odds ratios compare odds, not probabilities
An odds ratio compares the odds in one group with the odds in another. The CDC shows the standard two-by-two-table structure and the cross-product formula. Because the measure uses odds, its numerical value should not be translated as though it were directly a risk ratio in every setting.
Take a fictional group A with probability 20%. Its odds are 0.20/0.80 = 0.25. Group B has probability 40%, giving odds 0.40/0.60 ≈ 0.6667. The odds ratio B versus A is about 2.667, while the risk ratio is 2.
A translation that says “B was 2.667 times as likely” may overstate or ambiguously restate the odds ratio as a probability ratio. Preserve the term odds ratio when that is the measure. If an explanatory sentence is added, phrase it in terms of odds unless a justified conversion has been performed.
11. Odds ratios can approximate risk ratios when outcomes are uncommon, but that is not a universal identity
The CDC notes that when an outcome is uncommon, an odds ratio can provide a reasonable approximation to a risk ratio in appropriate contexts. That does not mean the two measures are always equal or that the translator may freely substitute the names.
As event probabilities grow, odds and risk diverge more strongly. The previous 20% versus 40% example already shows an odds ratio of about 2.667 against a risk ratio of 2. A target that collapses the distinction can change the apparent strength of association.
Preserve the original measure name and any source explanation of study design. If the source later interprets the odds ratio cautiously, retain that caution. A translation should not upgrade an approximation into an equality.
12. Two-by-two tables protect the structure of odds-ratio claims
A simple two-by-two table separates group status and event status. The four cells show group A with event, group A without event, group B with event and group B without event. This structure makes both risk and odds calculations traceable.
When translating, preserve row and column labels exactly enough that the cells keep their roles. A table can contain the correct four numbers and still produce the wrong interpretation if event and non-event headings are swapped.
A useful QA method computes one known measure before and after translation. If the source and target table yield different risks or odds despite identical cell values, the labels have likely changed the structure. Tables should be checked as systems, not as independent text boxes.
13. “Twice the odds” is not automatically “twice the probability”
Suppose baseline probability is 20%, with odds 0.25. Doubling the odds gives 0.5. Converting those odds back to probability gives 0.5/(1+0.5) = 1/3, or about 33.3%. Probability has not doubled to 40%.
This matters in headlines and summaries because “twice the odds” can sound like “twice as likely” to general readers. Preserve the statistical measure the source actually uses, and avoid substituting an everyday likelihood phrase if it changes the numerical implication.
Where a plain-language explanation is required, provide the underlying group probabilities when available rather than inventing a direct probability multiplier from the odds ratio. Clear explanation can reduce confusion without distorting the measure.
14. Baseline risk controls the practical meaning of a relative measure
A risk ratio of 2 can describe a change from 1% to 2% or from 20% to 40%. The relative comparison is the same, but the absolute difference is one percentage point in the first case and twenty points in the second.
Therefore a translation that reports only “risk doubled” can be faithful if that is all the source says, but it should not imply a particular absolute increase. If the source supplies baseline risk, keep it visible where the explanation depends on practical magnitude.
This distinction is important for educational clarity. Relative measures compare scales; absolute measures tell readers how far apart the event probabilities are. Translators should preserve which one the author uses rather than replacing it with whichever sounds more impressive.
15. “More likely” and “higher risk” need numerical discipline
If a source says group B had a risk ratio of 1.5 relative to A, “50% higher risk” can be an appropriate relative interpretation under the stated comparison. “Risk was 50 percentage points higher” is not equivalent. The latter requires absolute endpoints separated by fifty points.
Similarly, “1.5 times the risk” and “150% higher risk” are not the same. A ratio of 1.5 means the risk is 150% of the reference risk, which is 50% higher than the reference. The phrase “150% higher” would correspond to 250% of the reference.
Write an endpoint example in your notes. If reference risk is 10%, a 1.5 ratio gives 15%. The increase is five percentage points or 50% relative. This one small example can prevent several common language errors.
16. Protective-direction language depends on the reference group
A risk ratio below one can indicate lower risk in the numerator group compared with the denominator group. But the wording “protective” may carry causal meaning that the source does not justify. Preserve the author’s level of inference.
Suppose a fictional ratio is 0.6. The numerator group has 60% of the reference group’s risk, corresponding to a 40% relative reduction. If the target reverses the comparison, the reciprocal ratio is approximately 1.667. The language must change with the direction.
A reliable translation makes the reference explicit whenever a below-one ratio could be confusing. “Compared with group A, group B had a risk ratio of 0.6” is clearer than presenting 0.6 without a visible denominator group.
17. Rates and risks should not be merged
A risk is typically a proportion over a defined population and period under the source design. An incidence rate can involve person-time in the denominator. The CDC distinguishes risk ratios from rate ratios for this reason.
A target that translates both as “risk” may erase the time component of a rate. Ten events per 1,000 person-years is not the same measure as ten events among 1,000 people observed for a fixed identical period.
Keep denominator units visible. If person-time appears in the source, preserve it in the target. Do not convert a rate into a probability without an appropriate model and explicit instruction; that is statistical analysis, not translation.
18. Counts, risks and ratios can all rise by different amounts
Imagine a fictional first period with 10 events among 100 observations and a second with 30 events among 300 observations. Event count triples, but risk remains 10% in both periods. A target saying “risk tripled” would confuse count growth with probability.
The opposite can also happen: event count stays the same while the denominator changes, causing risk to rise or fall. Translation should therefore keep count nouns, population nouns and rate nouns distinct.
A private two-column table is often enough: period, events, total, risk. If the target sentence describes a different column from the source, revise it before publication. This is a practical example of why numerical relationships should be made explicit during translation.
19. Rounding can change a ratio’s apparent relationship
Suppose risks are 9.6% and 19.4%. Rounded to whole percentages they appear as 10% and 19%, giving a visible ratio of 1.9. Using the unrounded values gives a ratio of about 2.02. The source may report a ratio calculated from full precision even when displayed risks are rounded.
Do not force a reported ratio to equal a ratio reconstructed from rounded display values. Check whether underlying data are available. A small apparent discrepancy may be a presentation effect rather than an error.
Preserve the source’s rounding policy and avoid adding unnecessary decimal places. More digits do not make a translation more truthful when the source measurement or estimate is not more precise.
20. Uncertainty language should travel with risk estimates
Words such as estimated, observed, modelled, approximately and projected are not optional tone markers. They describe evidence status. A source saying “estimated risk was 12%” should not become “risk was 12%” if the distinction matters.
Likewise, an association reported with an interval estimate should not be reduced to a single ratio in a summary unless the editorial brief allows that compression. The point estimate alone can suggest more certainty than the source intended.
This article does not replace a dedicated guide to confidence intervals. Its translation rule is narrower: preserve the source’s uncertainty markers and do not strengthen tentative statistical language into certainty.
21. Worked case: absolute and relative risk in the same paragraph
Consider this fictional source: “The event occurred in 4% of group A and 6% of group B. The absolute difference was two percentage points. The risk ratio for B relative to A was 1.5.” All three statements are consistent.
A flawed target might say: “Group B had 1.5 percentage points more risk, a 2% increase.” That confuses the ratio, absolute difference and relative increase. The correct relative increase from 4% to 6% is 50%, while the absolute difference is two percentage points.
The repair is to preserve each measure in its own grammatical slot: endpoint risks, absolute difference, then risk ratio. The private check recomputes 6/4 = 1.5 and 6 − 4 = 2 percentage points.
22. Worked case: odds ratio versus risk ratio
Use a fictional dataset where group A has 10 events and 90 non-events, while group B has 20 events and 80 non-events. Risks are 10% and 20%, giving a risk ratio of 2.
Odds in A are 10/90 ≈ 0.1111. Odds in B are 20/80 = 0.25. The odds ratio B versus A is 0.25/0.1111 ≈ 2.25. The odds ratio is larger than the risk ratio here.
A target sentence should not replace “odds ratio 2.25” with “risk was 2.25 times as high.” Preserve odds terminology unless the source explicitly provides a risk interpretation. The arithmetic example shows why measure names matter.
23. Worked case: same ratio, different baseline risk
Scenario one: fictional risk rises from 1% to 2%, giving a risk ratio of 2 and an absolute difference of one percentage point. Scenario two: risk rises from 25% to 50%, also giving a ratio of 2 but an absolute difference of twenty-five points.
If a summary says only “risk doubled,” both scenarios fit. A translator should not add “a small increase” or “a large increase” based solely on the ratio. Those evaluations depend on baseline, context and consequences.
When baseline values are available, preserve them because they help readers understand the absolute scale. When they are absent, do not invent them. Translation should make the source clearer, not complete missing statistical context through guesswork.
24. Practice clinic with explained answers
Practice one. Ten events occur among fifty observations. Probability is 10/50 = 20%. Odds in favour are 10:40 = 1:4. Do not translate the odds as one in four probability.
Practice two. Probability is 25%. Odds in favour are 0.25/0.75 = 1:3. A ratio of one event to three non-events creates four total cases.
Practice three. Risk rises from 5% to 10%. Absolute difference is five percentage points. Risk ratio is 2. Relative increase is 100%.
Practice four. Risk falls from 20% to 15%. Absolute reduction is five percentage points. Relative reduction is 25% because five is one quarter of the original twenty.
Practice five. A risk ratio of 0.5 for B versus A becomes a risk ratio of 2 for A versus B. Reversing the group order requires taking the reciprocal.
Practice six. Group A risk is 20%, group B risk is 40%. Risk ratio is 2. Odds ratio is about 2.667. The two measures must not be named interchangeably.
Practice seven. Odds are 3:2 in favour. Probability is 3/(3+2) = 60%. It is not 150%, even though 3 divided by 2 equals 1.5.
Practice eight. A report says 15 events per 1,000 person-years. Preserve the person-time denominator. Do not translate it as 1.5% of people unless the study design and source explicitly justify that transformation.
Practice nine. Event count doubles from 10 to 20 while population doubles from 100 to 200. Risk remains 10%. Count growth is not automatically risk growth.
Practice ten. A source reports an odds ratio of 1.8 and calls it an association. Do not translate it as proof that the exposure caused the outcome. Preserve the evidential verb.
Practice eleven. A risk ratio of 1.5 means risk is 150% of the reference level, which is 50% higher. Do not write 150% higher unless the ratio is 2.5.
Practice twelve. Baseline risk is 2% and final risk is 3%. The absolute increase is one percentage point; the relative increase is 50%. Keep those two statements separate.
25. A release checklist for probability and risk language
For every probability, identify events and total cases. For every odds statement, identify events and non-events. For every ratio, label numerator and reference groups. For every absolute difference, check the subtraction order. For every relative change, identify the baseline.
Then check terminology: probability, risk, rate, odds, risk ratio and odds ratio should not collapse into one generic word if the source distinguishes them. Check evidence verbs separately: observed, estimated, associated, predicted and caused carry different claims.
Finally, reconstruct one representative claim numerically. If the target language implies a different denominator, reference group or multiplier, repair the statistical layer before refining style. Fluent prose should be the last pass, not the first guarantee of correctness.
26. Where this article sits in the eduKate translation system
This specialist guide extends Translate | Names, Numbers, Dates and Units, the new guide to Fractions, Ratios and Proportions, and the guide to Percentages, Percentage Points and Basis Points. The broader owner remains Master Art of Translation.
The Vocabulary Learning Hub supports exact distinctions between terms such as risk, odds, rate and probability, while How English Works supports the grammar of comparisons, reference groups and qualifiers. Those systems remain the broad language owners; this article handles one narrow statistical translation problem.
The governing principle is to preserve the denominator and the reference group before polishing the sentence. Once you know who experienced the event, out of whom, and compared with whom, the language can move. Without those anchors, a translation can keep every number and still tell a statistically different story.