Number 20/Measurement & inference/Causality
The Attribution Ceiling
The quantity
The ratio of intervention magnitude to natural variability in the outcome. It sets an upper bound on the share of variance attributable to the intervention, however sophisticated the method.
01Why it matters
It separates a limit of the world from a limit of the analysis, which is the difference between a hard ceiling and a technique to be improved.
02Expected value
Attributable share should track this ratio across studies of the same intervention.
03How it would be measured
For any body of studies, tabulate effect estimates against baseline variability.
04What would refute it
If attributed share is independent of the ratio, no ceiling exists.