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

Showing results for

- JMP User Community
- :
- Discussions
- :
- Re: Is it possible to put a mixture variable in Sc...

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
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Jan 11, 2018 7:55 PM
(525 views)

Hi,

I see that there is an option to put mixture variables in custom design, but not screening design. I want to make a screening design with 4 mixture variables. So, do i have to choose custom design and then choose D-optimal for screening purposes? If I do that I do not get clear confounding effects. I wanted to choose screening design because I can see clear confounding effects in that.

1 ACCEPTED SOLUTION

Accepted Solutions

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Jan 12, 2018 6:36 AM
(992 views)

Solution

The Screening Design platform in JMP is for the "classical" approach to experimental design and is only relevant for independent factors, in other words, not mixture factors. There are MANY approaches to screening factors beyond those classical designs. JMP Education even has an entire course devoted to creating and analyzing screening designs! (JMP Software: Modern Screening Designs)

In screening designs, you are trying to identify the most important factors. For a mixture situation, this can easily be accomplished using custom design and just specifying a main effects model. The difference between the parameter estimate for each component and the overall average of the response is the indication of the effect size. The larger the effect size, the more important that component is. You can see a more full discussion along with other possible approaches in the book Strategies for Formulations Development by Snee and Hoerl.

As for the confounding effects, in classical designs (with independent factors), you run a screening design and will have COMPLETE confounding/aliasing. With a mixture design there is ALWAYS partical confounding even when not screening. Therefore, you will never get clear confounding. You always get a bunch of partial confoundings. Mixture designs are very different from the classical designs. JMP Education also has a course that covers the basics of mixture designs and analysis (JMP Software: Design and Analysis of Mixture Experiments).

Dan Obermiller

7 REPLIES

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Jan 12, 2018 6:36 AM
(993 views)

The Screening Design platform in JMP is for the "classical" approach to experimental design and is only relevant for independent factors, in other words, not mixture factors. There are MANY approaches to screening factors beyond those classical designs. JMP Education even has an entire course devoted to creating and analyzing screening designs! (JMP Software: Modern Screening Designs)

In screening designs, you are trying to identify the most important factors. For a mixture situation, this can easily be accomplished using custom design and just specifying a main effects model. The difference between the parameter estimate for each component and the overall average of the response is the indication of the effect size. The larger the effect size, the more important that component is. You can see a more full discussion along with other possible approaches in the book Strategies for Formulations Development by Snee and Hoerl.

As for the confounding effects, in classical designs (with independent factors), you run a screening design and will have COMPLETE confounding/aliasing. With a mixture design there is ALWAYS partical confounding even when not screening. Therefore, you will never get clear confounding. You always get a bunch of partial confoundings. Mixture designs are very different from the classical designs. JMP Education also has a course that covers the basics of mixture designs and analysis (JMP Software: Design and Analysis of Mixture Experiments).

Dan Obermiller

Highlighted

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Jan 12, 2018 3:13 PM
(487 views)

Thanks Dan!

So I guess for mixture design I will need to look at the Alias Matrix for partial confounding. So is there any particular value below which I may consider the 2 factor-interaction effect to be ** not **confounded with the main effect?

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Jan 12, 2018 5:42 PM
(482 views)

You can use the Alias Matrix and/or use the Color Map on Correlations in the Design Evaluation outline item.

Please realize that mixture designs will have correlations (partial confounding) between MAIN EFFECTS, not just interactions (after all, the last component's value is completely determined by the rest). The degree of the confounding will depend on how narrow some of the component ranges are. This can be lessened by using pseudo-component coding (which JMP does automatically), but there is not much you can do about these correlations. The best you can do is to have the component ranges as broad as practically possible.

Mixture models are about generating predictions, not cause-and-effect conclusions (for example, a higher response could be due to increasing component A or decreasing component B since you cannot change one component without changing another component). Bottom line is that the rules of the classical designs do not apply to mixtures, so there is no "desirable limit" for the correlation value.

Dan Obermiller

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Jan 12, 2018 6:07 PM
(477 views)

Thanks Dan!!

You said that the mixture designs will have correlation between main effects. But the Alias Matrix shows only the correlation between main effect and 2 factor interactions. Is there a way to find correlation between main effects?

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Jan 12, 2018 6:37 PM
(471 views)

The Alias Matrix addresses the issue of how terms that are not included in the model affect the estimation of the model terms. For a mixture design, you really want to see how correlated your model terms are. The higher that correlation, the most "unstable" your parameter estimates will be, the higher the variance will be on those parameter estimates which affects the statistical testing, etc.

Rather than the alias matrix, you should look at the color map on correlations. Notice that hovering over any of the colored squares on that matrix will also give you the correlation value.

Dan Obermiller

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Jan 12, 2018 7:35 PM
(464 views)

Thanks Dan!!

Is there a way to export the color map of correlations in the form of a table with the correlation values?

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Jan 15, 2018 6:38 AM
(437 views)

I'm not aware of a way to get the color map on correlations into a table directly from the design creation process. You could save the design and create the correlations from that resulting data table using Multivariate Methods > Multivariate if you really need these results in a table. Note that you may need to build any interaction columns in your data table for that approach.

Dan Obermiller