All simulators
Blocking on Districts
Complete versus within-district randomisation of a schooling programme: the variance gain from blocking.
Day 38: Stratified & Blocked Experimentssd, complete randomisation
2.716
sd, blocked by district
0.960
variance ratio
0.12
blocked ÷ complete
equivalent sample gain
701%
complete randomisationblocked randomisation
Test scores in four districts that differ in baseline level. Randomising within each district removes the between-district variation from the estimator's variance; the gain grows with the share of variance that lies between blocks. True effect: +3 points.