Balanced repeated replications based on orthogonal multi-arrays

SUMMARY To obtain variance estimates in stratified sampling with possibly unequal numbers of primary sampling units per stratum, one can either randomly divide the units in each stratum into two groups and apply the balanced half-samples method to these groups, or use mixed orthogonal arrays to form a set of balanced subsamples. With the first technique, resulting variance estimates can be very inefficient, and with the second technique it can be difficult to construct orthogonal arrays with a small number of replicates for a given stratified design. The proposed extension of the orthogonal array method to orthogonal multi-arrays retains all of the orthogonal array method's desirable properties but allows greater flexibility in construction, and thus can handle more complex stratified designs using fewer replicates.