← The Library

Research

Correlation, Causation, and Relative Risk: A Practical Guide

Two things can move together without one causing the other. And a large relative change can describe a very small absolute difference.

By The Mindful Matrix Editorial Team · ~9 minutes · · Evidence reviewed August 2026 ·

The Library provides educational information and does not diagnose medical conditions or replace individualized medical care.

Links below go to Zinzino's own site. I'm an independent Zinzino Partner and earn a commission on purchases made through them.

Two data pathways diverging through confounders into relative and absolute risk frames

Key takeaways

  • Correlation means variables are associated; it does not by itself identify cause.
  • Confounding, selection, measurement error, reverse causation and chance can create or distort associations.
  • Relative risk compares groups; absolute risk shows the size of the difference in context.
  • Causal confidence grows from converging evidence, not one impressive number.

How to read the evidence

Evidence map

Established

Supported by established physiology, nutrition, or research-method evidence.

  • Risk ratios compare event probability between groups.
  • Absolute risk difference describes the difference in event rates.
  • Observational associations can be distorted by confounding and bias.

Supported

Evidence supports the idea, but effect size and application depend on context.

  • Temporality, consistency, dose-response patterns, plausible mechanisms and intervention evidence can strengthen causal inference when considered together.

Debated

Evidence is incomplete, inconsistent, or not ready for universal interpretation.

  • Any single checklist that proves causation.
  • Treating a statistically significant association as necessarily important or causal.

Action

A reasonable next step that follows from the evidence without turning uncertainty into a promise.

  • Ask what else differs between groups.
  • Translate relative numbers into absolute event rates.
  • Look for replication and designs that address alternative explanations.

Correlation is a pattern, not a mechanism

An association tells you that an exposure and an outcome differ together in the observed data. It does not tell you why. The exposure could affect the outcome, the outcome could affect the exposure, a third factor could influence both, or the pattern could reflect bias or chance.

Associations are still valuable. They can identify signals, generate hypotheses and reveal patterns that cannot be tested experimentally for ethical or practical reasons. The mistake is not observing correlation; it is skipping the work required to interpret it.

The alternative explanations

  • Confounding: a third factor is related to both exposure and outcome.
  • Reverse causation: the early outcome changes the exposure rather than the exposure causing the outcome.
  • Selection bias: the people included differ in a way that affects the relationship.
  • Measurement error: exposure or outcome is classified inaccurately.
  • Multiple testing: many comparisons increase the chance of an apparently positive result.
  • Residual confounding: statistical adjustment cannot fully measure or remove every important difference.

Relative risk needs an absolute baseline

Suppose an event occurs in 2 of 1,000 people in one group and 1 of 1,000 in another. The relative risk is doubled, but the absolute difference is one additional event per 1,000. Both statements are mathematically valid; only together show the scale.

Now imagine rates of 200 and 100 per 1,000. The same two-fold relative risk represents an absolute difference of 100 per 1,000. Baseline risk changes the practical meaning.

  • Risk ratio: event risk in the exposed group divided by event risk in the comparison group.
  • Absolute risk difference: exposed event rate minus comparison event rate.
  • Odds ratio: compares odds, not risks; it can look more dramatic when outcomes are common.
  • Confidence interval: a range of values compatible with the model and data under its assumptions.

How causal confidence grows

Randomization can balance known and unknown confounders on average, making a well-conducted trial a strong design for many intervention questions. But trials can be short, small, unblinded, poorly adhered to or focused on surrogate outcomes.

For exposures that cannot be randomized, causal reasoning draws on temporality, replication, magnitude, dose-response, negative controls, natural experiments, mechanistic evidence and methods designed to reduce confounding. No one feature automatically proves the case.

A five-question headline check

  • Was the study observational or experimental?
  • What were the absolute event rates?
  • What confounders were measured—and which important ones may remain?
  • Could reverse causation or selection explain the pattern?
  • Does the language say associated with or caused, and is that wording justified?

Evidence in context

What the evidence supports

Established

  • Association and causation are different claims.
  • Absolute and relative measures answer different questions.
  • Confounding and bias can distort observational results.

Supported

  • Converging evidence from stronger designs, replication and plausible mechanisms can increase causal confidence.

Debated

  • A universal threshold or single criterion that proves causality in every health question.

Good interpretation keeps the denominator, the alternative explanations and the study design visible at the same time.

Evidence notes

  • The numerical examples are hypothetical and illustrate interpretation; they are not estimates of a real health outcome.
  • Causal inference methods continue to evolve and must be judged in the context of the specific question and data.

Limitations

  • This guide does not cover hazard ratios, competing risks, Bayesian inference or advanced causal models in depth.
  • Odds ratios and time-to-event measures require additional care beyond the simplified examples.
  • A full causal assessment often requires specialist domain knowledge.

Related product records

Products in this learning context

These are navigation connections, not evidence that a product produces an outcome.

Sources

  1. Centers for Disease Control and PreventionEpidemiology training module
    Principles of Epidemiology: Measures of Association ↗

    Used for definitions of risk ratio and measures of association.

  2. National Institutes of HealthResearch-literacy guide
    Understanding Clinical Studies ↗

    Used for the causal strengths of randomized trials and the associational role of observational designs.

  3. National Heart, Lung, and Blood InstituteNIH critical-appraisal tools
    Study Quality Assessment Tools ↗

    Used for confounding, temporality, selection and measurement-bias questions.

  4. Centers for Disease Control and Prevention / NIOSHEpidemiology methods paper
    Estimation of Risk and Inferring Causality in Epidemiology ↗

    Used for risk measures and the complexity of causal inference in chronic-disease epidemiology.

  5. National Center for Complementary and Integrative HealthKnow the Science guidance
    The Facts About Health News Stories ↗

    Used for practical questions about study type, magnitude, uncertainty and reporting.

Where to go next

Choose the next useful step.

Keep learning, return to the process, or consider an optional tool only when it answers a clear question.

Keep learning

Use the framework

Return to the study-reading guide and apply the risk and causation questions to a complete paper.

Understand your next step

Return to the Matrix

Use Information → Education → Action to decide what—if anything—comes next.

Review the process →