Measures of Association
Measures of association are the quantities epidemiology uses to express how strongly an exposure or intervention is linked to an outcome. They translate counts of disease and exposure into a single comparable number — a ratio or a difference — that captures the size and direction of an effect and allows studies to be summarised, compared, and pooled.
Definition
A measure of association is a statistic that quantifies the strength of the relationship between an exposure (or treatment) and an outcome, expressed either as a ratio of two risks, odds, or rates, or as their arithmetic difference.
Scope
This area orients the reader to the family of effect measures used across epidemiologic and clinical research: relative (ratio) measures such as the relative risk, odds ratio, and hazard ratio; absolute (difference) measures such as the risk difference; and the patient-facing reciprocal, the number needed to treat. It explains how these measures relate to one another and when each is appropriate, treating them as methodological tools rather than clinical instructions.
Sub-topics
Core questions
- How strongly is an exposure or treatment associated with an outcome?
- When should a relative (ratio) measure be used rather than an absolute (difference) measure?
- How do the risk ratio, odds ratio, hazard ratio, and risk difference relate to one another?
- How can the same effect look large or small depending on which measure is reported?
Key concepts
- Relative (ratio) measures
- Absolute (difference) measures
- Risk, odds, and rate
- Reference group and direction of effect
- Time-to-event and the hazard
- Baseline risk and clinical interpretation
Mechanisms
Effect measures fall into two broad families. Ratio measures (relative risk, odds ratio, hazard ratio) divide the outcome frequency in an exposed or treated group by that in a comparison group, so that a value of 1 means no association; they capture the strength of an association and travel well across populations with different baseline risks. Difference measures (the risk difference) subtract one outcome frequency from another, so that 0 means no association; they capture the absolute public-health or clinical impact and depend on baseline risk. The number needed to treat is the reciprocal of the risk difference and re-expresses an absolute effect in patient terms. Choosing among these measures, and reporting both relative and absolute forms, is central to honest evidence appraisal, because a constant relative effect can correspond to very different absolute effects depending on how common the outcome is.
Clinical relevance
Measures of association are the language in which trials, cohort studies, and systematic reviews report their results, so understanding them underpins critical appraisal of the evidence. They describe how strongly evidence links exposures and outcomes; they are tools for interpreting research, not prescriptions for individual diagnosis or treatment.
Epidemiology
These measures are used throughout observational and experimental epidemiology and form the output of meta-analysis. The choice of measure is partly dictated by study design — cohort and trial data support risk ratios and risk differences, case-control data naturally yield odds ratios, and survival data yield hazard ratios — and partly by the communication goal, since relative measures convey strength while absolute measures convey impact.
History
The formal vocabulary of effect measures grew up alongside twentieth-century epidemiology and biostatistics: the odds ratio and its link to relative risk were clarified in mid-century cancer epidemiology, the proportional-hazards framework introduced the hazard ratio for time-to-event data in 1972, and the number needed to treat was proposed in 1988 to make absolute effects easier to communicate. Modern textbooks consolidated these into a coherent system of relative and absolute measures.
Debates
- Relative versus absolute measures
- Relative measures convey the strength of an association but can exaggerate perceived importance when the baseline risk is low; reporting absolute measures alongside them is widely recommended so that the real-world impact of an effect is not misjudged.
- When the odds ratio misleads
- The odds ratio approximates the risk ratio only when the outcome is rare; for common outcomes the two diverge, and treating an odds ratio as if it were a risk ratio overstates the effect.
Key figures
- Kenneth Rothman
- Sander Greenland
- David Sackett
- Jerome Cornfield
Related topics
Seminal works
- rothman-2008
- cook-sackett-1995
- davies-1998
Frequently asked questions
- What is the difference between a relative and an absolute measure of association?
- A relative measure (such as a risk ratio) divides one group's outcome frequency by another's and expresses how many times more likely the outcome is; an absolute measure (such as a risk difference) subtracts them and expresses the extra outcomes per population, which depends on the baseline risk.
- Which measure of association should a study report?
- Good practice is to report both a relative and an absolute measure, because the relative measure conveys the strength of an association while the absolute measure conveys its public-health or clinical impact; the appropriate ratio measure also depends on the study design.