Sequential procedures are better than fixed allocation procedures, but they have drawbacks including:

• one needs responses in a timely fashion
• they are more complex to carry out
• one is unable to randomize

Few-stage allocations are preferred by many practioners.

We consider the most general form: the size and relative allocations for the next stage can be based on the results of all allocations of previous stages. Some people do not allow the size of the stages to depend on the outcomes.

Determine optimal k-stage allocation procedures, for k small.

This has never been done before, neither analytically nor computationally. We also want to compare these procedures to the OS allocation.

This problem is computationally intensive - here is our paper that discusses how to compute the optimal few-stage allocation.

Few-Stage Efficiency Outline MP vs BF