01Overview
Illusory correlation is seeing a systematic link between two things when none exists — or when the link is far weaker than perceived. Co-occurrence feels like causation because the brain is a pattern engine with loose quality control.
Designers correlate a banner with a signup spike, a colour change with support drop, or a persona trait with churn. Sometimes they are right. Often the timeline aligned by chance, seasonality, or a hidden third variable. Roadmaps chase correlations that dashboards invented.
02Detailed explanation
Illusory correlation thrives in noisy, low-sample product environments:
- A redesign ships the same week as a PR hit — credit goes to UI, not press.
- Support tickets fall when a tooltip is added — but a major bug was fixed the same release.
- Researchers link a demographic tag to a behaviour observed in two participants.
- Teams see "mobile users hate feature X" when mobile sessions coincided with a slow API region.
Stereotypes partly persist through illusory correlation: rare co-occurrences of group label and memorable behaviour become "always true" when base rates are ignored.
03Why it exists
Confirming co-occurrence is cognitively cheap. Disconfirming instances are less memorable — classic availability and confirmation dynamics.
Dashboards encourage line-up comparisons without experimental design. Visual proximity becomes narrative causality in slide decks.
What else changed at the same time — and what would disprove the link you are about to ship against?
04Effects on users
Users develop superstitions about products: "It breaks when I do X" when X and breakage are unrelated. Support must untangle illusory user correlations too.
They suffer when teams redesign from illusory links — removing features users loved because they coincided with a metric dip driven elsewhere.
05Effects on designers & teams
Teams enshrine illusory correlation in rituals:
- Release post-mortems without controls. Every adjacent metric movement gets a causal story.
- Persona superstitions. Traits stick after two examples — "finance users never scroll."
- Heatmap mythology. Eye movement correlated with success without task outcome linkage.
- Qual sample overfitting. Two quotes establish "correlation" between segment and need.
06Practical takeaways
- Hold a changelog of confounds. List everything that changed in a release window.
- Run controlled tests. Isolate one variable when claiming causality.
- Check base rates. How often does the behaviour happen without the supposed cause?
- Segment by hidden variables. Device, region, cohort, new vs returning.
- Delay attribution. Wait through seasonality before crediting UI.
- Teach correlation language. "Associated with" until proven "caused by."
07Design examples
The banner that saved signup
Signups rise after a hero banner change. The same week, an app store feature lands. Retrospective credits design. Holdout analysis shows banner-only users flat — illusory correlation.
Tooltip = fewer tickets
Tickets drop after a tooltip on billing. Engineers also fixed a tax calculation bug. Support credits content; finance knows otherwise.
Finance users don't scroll
Two finance personas never scrolled in lab sessions. Analytics on 40k sessions show identical scroll depth to other roles — two memorable cases built a rule.
Mobile hates video
Video engagement is low on mobile in one market. API latency was high there only. A "mobile doesn't watch" narrative almost killed a feature — correlation was infrastructure.
08Ethical risks
Illusory correlation fuels discriminatory design when spurious links between group labels and behaviours become requirements.
Users lose access to valuable features when teams misattribute outcomes and "fix" the wrong thing confidently.
Self-test: Which product belief on your team is built on co-occurrence alone — and how would you test it?
10Suggested reading
Suggested reading is temporarily unavailable. Please check back later.