# Correlation and Causation

Correlation describes an observed association, while causation claims that changing one factor changes another under a defined intervention and context.

- Type: principle
- Domain: Data, experimentation & growth
- Level: Foundational
- Last reviewed: 2026-10-05
- Original: [Soroush’s Product Wiki](<https://wiki.soroush.uk/topic/correlation-and-causation/>)

## What it means

Correlation is co-variation between measurements; causation concerns what would happen to an outcome if an intervention changed. Correlated product behaviors can arise from shared causes, selection, reverse direction, measurement choices, or chance, so an association alone does not establish that changing one behavior will cause the other.

## Why it matters

Analytics can show that retained users invite teammates or that purchasers view a page, but those patterns do not tell a team which intervention will improve retention or purchase. The distinction directs teams toward experiments, causal designs, and careful claims.

## When to use it

Use this distinction whenever dashboards, segments, or observational studies inform a product action. Association can be valuable for description, prediction, and hypothesis generation; seek a credible causal design when deciding whether to change a policy, feature, or message.

## How it works

1. State whether the question is descriptive, predictive, or causal.
2. Define the intervention, outcome, population, and time horizon for causal questions.
3. Identify plausible shared causes, reverse direction, and selection mechanisms.
4. Choose a randomized experiment or a justified observational design.
5. Report assumptions and uncertainty rather than upgrading association into proof.

## A practical example

Users who configure integrations renew more often. The team treats this as a hypothesis because mature accounts are both more likely to configure integrations and renew; it tests guided setup rather than assuming integrations themselves cause renewal.

## Where it breaks

Randomization and observational methods also have assumptions, implementation failures, and generalizability limits. Some causal questions cannot be answered ethically or practically by experiments.

## Common mistakes

- Using correlation as evidence that a feature caused an outcome.
- Dismissing useful descriptive associations because they are not causal.
- Ignoring reverse causation or common causes.

## Visualization

[View the visualization on the website](<https://wiki.soroush.uk/topic/correlation-and-causation/#visualization-correlation-and-causation>)

## Sources

1. [Causal Inference: What If](<https://www.hsph.harvard.edu/miguel-hernan/wp-content/uploads/sites/1268/2024/01/hernanrobins_WhatIf_2jan24.pdf>) — Harvard T.H. Chan School of Public Health · Foundational textbook
2. [Causal Models](<https://plato.stanford.edu/entries/causal-models/>) — Stanford Encyclopedia of Philosophy · Scholarly reference

## Related topics

- [Causal Inference](<https://wiki.soroush.uk/topic/causal-inference/>)
- [Confounding Variable](<https://wiki.soroush.uk/topic/confounding-variable/>)
- [Randomization](<https://wiki.soroush.uk/topic/randomization/>)
