In many randomized trials, subjects enter the sample sequentially. Because the covariates for all units are
not known in advance, standard methods of stratification do not apply. We describe and assess the
method of DA-optimal sequential allocation (Atkinson, 1982) for balancing stratification covariates across
treatment arms. We provide simulation evidence that the method can provide substantial improvements
in precision over commonly employed alternatives. We also describe our experience implementing the
method in a field trial of a clean water and handwashing intervention in Dhaka, Bangladesh, the first time
the method has been used. We provide advice and software for future researchers.