Correlate-Not-Cause Risk Audit
Before blaming the loudest variable, control for known drivers and see if the effect survives.
When something feels like the obvious cause of a problem, the brain wants to act on it. Pfeifer's framework forces a slower question: is this variable a cause, or is it a correlate of the real causes? Her case study is social media and youth mental health. The vivid story says phones are destroying a generation. The meta-analytic story says excessive social media use is linked to roughly 15% higher mental-health-problem rates — moving baseline depression risk from 20% to 23%, an effect that nearly disappears once you control for known drivers like bullying, parent mental health, and friendship quality. The framework has four moves: (1) name the variable everyone is blaming, (2) list the known higher-impact drivers of the same outcome, (3) check whether the blamed variable's effect survives controlling for those drivers, (4) reallocate intervention budget toward the survivors. The discipline protects you from spending political and parental capital on the wrong lever while the actual drivers — relationships, family mental health, bullying — go under-resourced.
- Feeling true is not evidence — science doesn't care if a claim feels right, and neither should your decision rule.
- Use meta-analyses, not single studies, to avoid 'science whiplash' from headlines that ping-pong with each new shiny paper.
- A correlate looks identical to a cause until you control for the real drivers; the real test is whether the effect survives controls.
- Effect size matters more than direction — a real but tiny effect rarely justifies a large intervention.
- When the loud variable fails the audit, redirect attention and resources to the high-impact drivers it was masking.