Simulations were conducted to examine statistical efficiency and potential bias for designs with varying amounts of cutoff-based assignment and randomization.

As in any seemingly unrelated regression context, the hope is to achieve gains in statistical efficiency compared to fitting each model separately.