Correlation vs. Causation
Explore the critical distinction between correlation and causation within data science. This lesson helps you understand how variables can be associated without one causing the other, recognize confounding factors like plan tier that influence observed relationships, and identify pitfalls like Simpson's paradox. You'll gain skills to critically analyze data patterns, interpret correlation coefficients correctly, and apply statistical reasoning to avoid misleading conclusions. This foundation prepares you to evaluate when causal claims are justified and when further experimental design is necessary.
Customers who use Feature X churn less, so the product should push every customer to use Feature X. The data may support the first clause, but not the decision in the second.
A correlation is a pattern of association between two variables. In this case, higher Feature X use appears alongside longer retention. Causation means that changing one thing produces a change in another. To claim that Feature X reduces churn, we would need evidence that increasing Feature X use itself changes retention.
The missing piece is a causal design. Customers may choose Feature X because they already have traits that make them less likely to churn. Without separating that possibility, forcing Feature X use could consume product effort without reducing churn.
Note: The plotting lines below use Matplotlib again.
plt.scatter()draws a scatter plot: one dot per customer, placed by usage minutes along the horizontal axis and retention days along the vertical axis. This lets us see the ...