Skip to main content
← Back to Explore
RESEARCH & PERSPECTIVES / Research methods

Why an association is not an explanation

Confounding, reverse direction and design choices change what can reasonably be concluded.

Editorial sample: This reading has not yet undergone Mind Assembly's editorial review. Verify all sources, interpretation and permissions before public release.

What a correlation does tell you

If two variables are associated in a study, their values vary together in a particular sample. A positive association does not establish that one variable caused the other.

Competing explanations

Three possibilities deserve attention: an effect from A to B, an effect from B to A, or a third influence related to both. Selection bias, measurement problems and chance can also complicate interpretation.

Why timing and design help

Longitudinal designs can establish temporal ordering for repeated observations, but timing alone cannot eliminate confounding. Randomised experiments are powerful for causal questions when ethically and practically feasible; their findings also depend on what was manipulated and how outcomes were measured.

An example

Suppose a questionnaire study finds an association between late-night screen use and sleep problems. It might reflect effects of screen use on sleep, poor sleep leading to screen use, stress affecting both, or some combination. A one-time survey cannot disentangle these possibilities.

Reporting is not quality certification

Guidelines such as PRISMA improve the transparency of reporting research syntheses. They should not be confused with proof that the underlying studies are bias-free or that a causal claim is justified.

A better conclusion

Say what the study observed, explain what the design permits, and describe plausible alternative explanations.

Sources & further reading

← Return to reading catalogue