Can anybody explain exactly how JMP creates D-Optimal designs? I assumed the routine used the Bayesian modification approach (DuMouchel & Jones, 1994) treating necessary terms as "primary" and if possible terms as "potential".
If I try to replicate a very simple example from the paper in JMP (using v8) with 2 factors and the interaction model (X1, X2, X1X2) as "necessary" and the quadratic terms (X1^2, X2^2) as "if possible" and ask for a 5-run D-Optimal I do not always get what I expect (the full factorial + 1 centre-point). In fact, sometimes I do and sometimes I don't.
Can somebody clarify whether it uses the Bayesian modification and if so whether the JMP implementation is a bit "buggy". I'm concerned at not getting a consistent result from such a simple 2-factor example.