turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

- JMP User Community
- :
- Discussions
- :
- Discussions
- :
- Re: Applying definitive screening design on catego...

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Aug 10, 2017 11:34 AM
(817 views)

Hi all,

I plan to run a DOE with three categorical factors. Is Definitive Screening Design is good choice to start with?

Thanks.

Solved! Go to Solution.

1 ACCEPTED SOLUTION

Accepted Solutions

Highlighted

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Aug 12, 2017 4:22 AM
(1565 views)

Solution

Any screening design depends on the key principles of screening: *sparsity of effects*, *hierarchy of effects*, *heredity of effects*, and *projection of design*. With only three factors, it is unlikely that the first principle will hold. Do you expect that only one of the factors is important?

Also, all screening designs have the attraction of *economy* because some other design methods produce larger designs for the screening case (large number of candidate factors and their associated large number of potential effects). The definitive screening design for three categorical factors includes 14 runs at a minimum and 18 runs with the strongly suggested minimum addition of 4 runs. The size of the DSD is determined by the method, not by you, except for the choice of the number of additional runs. What about the custom design for three categorical factors? Including all three two-factor interactions, the minimum number of runs is 7. A custom design with 12 runs can *estimate all the effects* without fail (even if screening principles do not hold) and *provides high power*: 97% if the effect is at least 3-fold larger than the response standard deviation. (I assume that power for tests of the parameter estimates is important to you because this discussion is about screening.)

An advantage of JMP is that it covers the broad spectrum of situations with the best methods for designing experiments. If you decide that one design method is not the best choice for your situation, then you have the others to fall back on.

Learn it once, use it forever!

1 REPLY

Highlighted
Solution

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Aug 12, 2017 4:22 AM
(1566 views)

Any screening design depends on the key principles of screening: *sparsity of effects*, *hierarchy of effects*, *heredity of effects*, and *projection of design*. With only three factors, it is unlikely that the first principle will hold. Do you expect that only one of the factors is important?

Also, all screening designs have the attraction of *economy* because some other design methods produce larger designs for the screening case (large number of candidate factors and their associated large number of potential effects). The definitive screening design for three categorical factors includes 14 runs at a minimum and 18 runs with the strongly suggested minimum addition of 4 runs. The size of the DSD is determined by the method, not by you, except for the choice of the number of additional runs. What about the custom design for three categorical factors? Including all three two-factor interactions, the minimum number of runs is 7. A custom design with 12 runs can *estimate all the effects* without fail (even if screening principles do not hold) and *provides high power*: 97% if the effect is at least 3-fold larger than the response standard deviation. (I assume that power for tests of the parameter estimates is important to you because this discussion is about screening.)

An advantage of JMP is that it covers the broad spectrum of situations with the best methods for designing experiments. If you decide that one design method is not the best choice for your situation, then you have the others to fall back on.

Learn it once, use it forever!