Disallowed Profile Combinations in a Choice Experimental Design?A customer asked me if there was some way to disallow certain profile combinations in the Choice Design platform? Their thinking was sometimes certain attributes in an experiment and their levels might never occur in practice or from a product design point of view never be considered so there is no point in collecting data for these combinations.
Doesn't look like there's an obvious way to do this other than let JMP generate a design, then perform a profile by profile row wise examination of the design and see if the disallowed combination appears then row wise delete that profile? Or am I missing something entirely?
And from an analysis point of view are there other 'gotchas' than changing the optimality criteria and introducing parameter estimate bias from the original design?
Re: Disallowed Profile Combinations in a Choice Experimental Design?Just dawned on me. Don't really need an Add In...all I need to do is use the Rows -> Select Where and then add conditions for each attribute and the level within the disallowed profile combination. Then the rows containing the disallowed/undesired combinations are selected. From here I can delete the undesired rows. Or invert selection and subset to create a new data table with just the desired profiles.
Re: Disallowed Profile Combinations in a Choice Experimental Design?Thanks @Ryan_Lekivetz and @Jeff_Perkinson and @jiancao. The other thought I had was if I had some time I'd try and write a little JMP Add In that prompted for the disallowed combinations and then crawled a JMP data table looking for the row combinations that are not desired. Then give an option to either delete the row in place or write a new data table...kind of like the Recode in place or write a new data table.
Re: Disallowed Profile Combinations in a Choice Experimental Design?I would add that you could try Custom Design, which supports setting disallowed combinations using Define Factor Constraints. One caveat is that you would need to check for dominant profiles like "largest disk size, fastest speed, longest battery life, and lowest price " in the example of laptop designs, whereas Choice Design platform would avoid this sort of combination of factor levels.
However, if you have a small number of factors, this may not be difficult. In particular, when price is present, make sure that you don't have all other factors at the same level and price at different levels -- the lowest will always win.
Re: Disallowed Profile Combinations in a Choice Experimental Design?I ran into @Ryan_Lekivetz and asked him about this.
Support for Disallowed Combinations is a wishlist item. Not planned for JMP 14 but we know about the desire.
In the meantime, he says that the strategy that you outlined, generating a design with more choice sets that you actually need and then eliminating those that that have disallowed combinations, is a reasonable strategy. You won't introduce too much bias or destroy the optimality too much doing this.
