You should not manually control for residual differences between randomized groups; by the law of large numbers random differences are computable, so the correct approach is to let the data reveal whether the observed difference exceeds what chance alone could produce, reported as a statistical confidence (e.g., 95% probability the effect exceeds zero), with the sample size built into that calculation.
definitionpending
Speaker
Abhijit BanerjeeEvidence Quote
“with 95% probability the effect of this is greater than zero... that 95% takes into account the fact that these things can happen by chance in any study”
Created: 6/13/2026, 7:00:20 PM
Your notes
Loading notes...