Reviewed: 21 October 2026
“Twice the risk” is not the same as “two extra people in every hundred”. When a cannabinoid research headline reports a percentage, the missing denominator can change what readers think it means. This guide explains absolute risk, relative risk and odds, using deliberately fictional examples rather than estimates of cannabis-related harm or benefit.
Start with the event, group and time period
Absolute risk describes the chance of a specified event. It needs a time frame and a clearly defined population. The National Cancer Institute’s definition of absolute risk illustrates why an outcome over a lifetime is a different statement from an outcome over a year.
Before comparing figures, write down what was counted: a new diagnosis, a self-reported symptom, a hospital admission or something else. Then ask who was studied and how long they were followed. “Ten people experienced a symptom” is incomplete without knowing whether there were 100 participants or 10,000, and whether observation lasted a week or several years.
Relative risk is a comparison, not a head count
The NCI definition of relative risk describes a comparison between the risk of the same event in two groups. A risk ratio of 2 means one group’s risk is twice the other’s. It does not reveal either group’s absolute risk on its own.
Here is a fictional one-year example with two equally sized groups:
- Comparison group: 4 events among 1,000 people, or 0.4%.
- Other group: 8 events among 1,000 people, or 0.8%.
- Relative risk: 0.8 divided by 0.4 = 2, a 100% relative increase.
- Absolute difference: 0.8% minus 0.4% = 0.4 percentage points, or 4 more events per 1,000 people.
Now change the fictional figures to 100 and 200 events per 1,000 people. The relative risk is still 2, but the absolute difference is 100 per 1,000. These calculations show why the same relative comparison can describe very different event counts. Neither example is a cannabis risk estimate.
Percentages and percentage points answer different questions
If a fictional event frequency falls from 12% to 9%, the decrease is 3 percentage points. Relative to the original 12%, it is a 25% reduction. Saying “a 25-percentage-point reduction” would be wrong: it would describe a different calculation.
Try translating a headline into a sentence with a common denominator: “Over this period, the event occurred in X out of 1,000 people in one group and Y out of 1,000 in the other.” If those numbers are unavailable, record that gap rather than filling it with a guessed baseline. A large relative increase should not automatically be dismissed, and a small absolute difference is not automatically unimportant: the seriousness of the outcome also matters.
Odds ratios are not risk ratios
Risk is events divided by everyone in the group. Odds are events divided by non-events. In another fictional example, 20 events among 100 people give a risk of 20% and odds of 20 to 80. Forty events among 100 give a risk of 40% and odds of 40 to 60. The risk ratio is 2, but the odds ratio is approximately 2.67.
Do not turn an odds ratio of 2 into “twice as many people will experience the event”. The Cochrane Handbook’s guide to effect measures distinguishes these measures. Hazard ratios and incidence-rate ratios also have their own meanings; they should not simply be relabelled as risk ratios.
A precise-looking number can still be uncertain
An estimate such as 1.4 is not a complete result. Look for its confidence interval and the authors’ account of limitations. An interval that includes 1 for a ratio measure may be compatible with effects in different directions; it does not establish that the groups are equivalent. Statistical significance also does not tell you whether a difference is important. Cochrane’s guidance on interpreting results explains why estimates need context.
For pooled cannabinoid findings, our guide to certainty in systematic reviews explains why a summary estimate and confidence in the underlying evidence are separate questions.
Keep cannabinoid headlines within the study’s boundaries
A percentage is not a substitute for checking study design. A measured association does not by itself establish a cause, even when it is expressed as an adjusted ratio. Our explanation of correlation and causation in cannabis reporting covers that separate issue.
Likewise, a number from one formulation, population or outcome cannot automatically describe another. Use our guide to laboratory findings and human evidence to check whether a headline has moved beyond what was actually studied.
A useful reading note keeps six items together: the event, population, comparison, time period, effect measure and uncertainty. If one is missing, say so. This article is about interpreting research, not estimating anyone’s personal risk or recommending treatment. Personal health questions belong with a qualified healthcare professional.