Understand the idea
Two variables can move together because one affects the other, a third factor affects both, or the direction runs opposite to the story being told. A causal question asks what would change under an intervention. Random assignment can make treatment groups comparable in expectation and reduce systematic confounding, but it does not excuse poor measurement or execution. Observational evidence can still be informative, especially when experiments are impossible; it requires explicit assumptions and careful alternative explanations. [S04]
Worked example
Students who attend a study club score 15 points higher in a fictional survey. Motivation may influence both attendance and scores. Randomly inviting students addresses a different question from forcing attendance: distinguish the effect of an invitation from the effect of participation.
Where the idea stops
Randomization does not guarantee balance in every finite sample or generalizability to every population. Some causal questions cannot ethically or practically be randomized. [S04]
Connect it
Biology, psychology, and business: the common logic of evidence.
Go deeper
Why can self-selection create an apparent effect even when the feature does nothing?
Sources
UC Berkeley, Sense & Sensibility & Science · Research reference · Date: Date not established
Introduces interventions, confounding, and randomized experiments.
Link checked 2026-10-17 · Public webpage accessed; no specialist peer review of this brief.