rates
Simpson's paradox
Better in each group. Together it flips.
A treatment can look helpful inside every group, yet look harmful when all the groups are combined. The overall numbers reverse the pattern seen in the separate groups.
Look at the success rates inside each hospital, then look at the combined rate.
In Hospital A the treatment succeeds more often than no treatment.
In Hospital B the treatment also succeeds more often.
When the hospitals are added together the treatment looks worse.
This happens because the hospitals have very different numbers of patients and different overall success rates. The treatment was used more often in the hospital that had harder cases.
The group-level pattern and the overall pattern point in opposite directions.