Field note · February 20, 2024
We shipped a Simpson's paradox to the exec team
Every region improved. The company average got worse. Both were true.
Support satisfaction improved in every single region, quarter over quarter. AMER, EMEA, APAC, LATAM — all up.
Company-wide satisfaction went down.
The slide said satisfaction was declining and recommended an investigation into support quality. Nobody had computed it both ways, because why would you.
The mechanism: LATAM has structurally lower satisfaction (a language coverage gap we knew about) and had grown from 6% to 19% of ticket volume. The mix shifted toward the lower-scoring region faster than each region improved. Both facts are real; the aggregate is dominated by the mix change, not by quality.
The right answer is not one number. It is: every region improved, and the company number fell because of mix. That is two sentences and a completely different recommendation — the action is language coverage in LATAM, not a support quality review.
The habit we adopted, and it is cheap:
Whenever you compute an aggregate relationship, compute it within your most important grouping too. If the two disagree, that disagreement is the finding.
world-indicators has this deliberately baked in for exactly this reason — pooled across countries, internet penetration correlates negatively with life expectancy gains; within every country it is positive. It is a good five-minute exercise before you need it in anger.