Stat-Ease
If you are having trouble viewing this email view it online.
 
Vol: 16 | No: 1 | Jan/Feb '16
Stat-Ease
The DOE FAQ Alert
     
 

Stat-Ease Statistical Group

Dear Experimenter,
Here’s another set of frequently asked questions (FAQs) from me and the rest of our StatHelp team about design of experiments (DOE), plus alerts to timely information and free software updates. If you missed the previous DOE FAQ Alert click here.

To open another avenue of communication with fellow DOE and Stat-Ease fans, sign up for The Stat-Ease Professional Network on LinkedIn. A recent post features “Stat-Ease has a YouTube channel.

 
Stats Made Easy Blog

 

 

 

 

 

 

 

 

 

 

-

 
Topics in the body text of this DOE FAQ Alert are headlined below (the "Expert" ones, if any, delve into statistical details):

1:  Stat-Ease Academy e-learning alert: New web-based class now available for “Easier Experimenting with Factorial Split Plots”
2:  FAQ: How do I set the signal and noise ratio for a given set of data?
3:  FAQ: Is ‘verification’ the same as ‘confirmation’?
4: FAQ: Screening designs for mixtures
5: Info alert: See ASQ TV interview to find out why there’s no need to randomize all your runs
6: Events alert: Visit with Stat-Ease reps around the world: in Asia, Europe and North America
7:  Workshop alert: From New Jersey to Norway
 
 

P.S. Quote for the month: Nate Silver on being happy with what our data tells us. (Page down to the end of this e-zine to enjoy the actual quote.)


- Back to top -

1: Stat-Ease Academy e-Learning Alert: New web-based class now available for “Easier Experimenting with Factorial Split Plots”

Stat-Ease Academy

We are pleased to announce the release of a new Stat-Ease Academy class on Easier Experimenting with Factorial Split Plots. This interactive web-based course introduces experimenters to split-plot design and analysis tools that handle hard-to-change (HTC) factors. Learn how you can work around the difficulties of randomizing temperatures and the like that can be varied far more easily by grouping their settings, e.g., low versus high. Naturally taking short cuts like this come at a cost. Do not play with fire—for a modest investment of time (a few hours) and money (follow link above for details) in “Easier Experimenting with Factorial Split Plots” get a handle on these hot tools provided by highly-capable Design-Expert® DOE software.

(Applying the same delightful recipe that makes our workshops so effective, Stat-Ease Academy e-learning delivers what you need to tool up on design of experiments, all on your schedule at your desk or on your mobile device. For a complete listing of web-based DOE classes, click here.)


- Back to top -


2: FAQ: How do I set the signal and noise ratio for a given set of data?

Original question from a Pharmaceutical Formulation Scientist (R&D):“We are now using and practicing the new Design-Expert software that we purchased. I am facing some difficulty in setting the signal and noise ratio for a given set of data. Please share with me any guidance to set these parameters. Thank you for your support in advance.”

Answer from Stat-Ease Consultant Brooks Henderson:“First of all, as you probably know, the signal-to-noise ratio must be set for each of your responses as shown in the screen shot below.

Entering Signal and NoiseEntering signal (difference to detect—delta) and noise (estimated standard deviation—sigma)
For signal, the top part of the ratio, consider the smallest change in the response you want to detect. For example, let’s say you are working to reduce dissolution rate from a current level of 30 minutes to 20 minutes or less, then this implies you need to detect a change of 10 minutes. Anything less would not get where you want to be. So, in this case your entry for the signal (delta y) is 10. Keep in mind that the smaller the signal you’re trying to detect, the lower the power will be.

For the entry of the noise, the bottom part of the ratio, use past data on the response. Get this by simply calculating the standard deviation of a series of runs done at similar conditions. Another source of information on the noise might be the analysis of variance (ANOVA) of a prior DOE—Design-Expert reporting the standard deviation directly. Easy!

After completing the entries, check the power results (see screen shot below). Focus on the response that exhibits the lowest power—this being your worst-case scenario. If you build a design with enough runs for that response, the other ones should be fine.

Power results
Power results

In the case illustrated, the response for friability had the lowest signal-to-noise ratio and therefore the least power. Fortunately it just exceeded the recommended minimum of 80%. Therefore this experiment, a minimum-run characterization (resolution V) design, is good to go. If the power had come in lower, then it would have been back to the drawing board for a bigger design or other alternatives that cannot be gone into detail here.

(Learn more about signal-to-noise ratio and its effect on power by working through the free “4 Easy Steps to Effective Factorial Design” web-based class listed at the Stat-Ease Academy. Then, to further your education on this topic and much, much, more attend the two-day computer-intensive workshop Experiment Design Made Easy. Click on the title for a description of this class and link from this page to the course outline and schedule. Then, if you like, enroll online.)


- Back to top -


3: FAQ: Is 'verification’ the same as ‘confirmation’?Question from a Lead Biotechnologist:
“I see in a response surface graph that the design points called verification points show up on a graph. Is the term ‘verification’ here the same as ‘confirmation’? It must be different and, if so, are the verification points selected by the user or based on something in the software?”

Answer from Stat-Ease Consultant Wayne Adams:“We define “verification” as runs done during the experiment, i.e., embedded; whereas “confirmation” comes only after the analysis is complete. Verification runs, a relatively new feature in Design-Expert, are set by you, the user. They are conducted (and recorded) during the experiment, but the data are not used to fit the model. However, their results do show up on diagnostics and the model graphs. If they don’t seem amiss, then you can surmise that your model predicts them reasonably well.

This is all spelled out by my fellow Consultant Martin Bezener in his webinar on “Practical Strategies for Model Verification”. Find the slides and recording for this, along with all of our webinars here.”

(Learn more about confirmation and verification by attending the two-day computer-intensive workshop Response Surface Methods for Process Optimization. Click on the title for a complete description. Link from this page to the course outline and schedule. Then, if you like, enroll online or contact the Client Specialist at workshops@statease.com.)


- Back to top -


4: FAQ: Screening designs for mixtures

Original question from a Statistician:
“We were looking at doing a mixture screen on 12 components via the design provided by Design-Expert. I also ran the same information through an optimal mixture design to merely estimate linear effects. I got quite different numbers of runs between the two choices. The screening design gave me 41 runs, whereas the optimal mixture estimating linear effects only gave me 22 runs. Any idea why that is?”

Answer:
The simplex screening designs originally developed by Snee & Marquardt in 1976 lay out 3q+n points, where n is the number of centroids.  They are over built for their purpose of fitting a linear model. That’s why we make the axial check blends, constraint plane centroids and overall centroids optional. Going with none of these optional points leaves the q vertices, i.e., 12 points for your case.

The optimal design also establishes q points at a minimum (equaling the number of linear coefficients in Scheffè polynomial) with optional additional model points (default zero) lack-of-fit (5) replicate (5) and additional center points—aka centroids (default zero). So in your case Design-Expert recommends 22 runs (=12+5+5).

If runs are dear (costly, time-consuming or limited by the supply of materials), I recommend as a compromise going with simplex screening (being built to a template not so mysterious as optimal and thus less daunting to chemists) with the q constraint plane centroids (a bit over the top for those who are geometry-challenged) turned off. In your case of 12 components that would lead to a 29-component design.

P.S. FYI see Wayne’s elaboration below. Design-Expert will suggest up to 2q vertices if this many are available from the geometry based on inputted constraints. So for 12 components it defaults to 29 points, including 5 additional centroid reps. Wayne says:

“The screening design isn’t limited to a simplex. If the experimenter doesn’t change anything, and leaves the total set to 1, and the range for all components 0 to 1, then a simplex is the default. The type of design you get depends on the component entries. This is different behavior than optimal designs. For optimal designs you get the number of runs from the model and nothing else.”

(Learn more about screening designs for mixtures by attending the new computer-intensive three-day workshop Mixture and Combined Designs for Optimal Formulations. Click on the title for a complete description of this class. Link from this page to the course outline and schedule. Then, if you like, enroll online.)


- Back to top -

5: Info alert: See ASQ TV interview to find out why there’s no need to randomize all your runs

Find out why you need not randomize all runs by viewing this interview of me by ASQ TV.


- Back to top -


6: Events alert: Visit with Stat-Ease reps around the world: in Asia, Europe and North America

(Last notice.) Register now for the 2nd Asian DOE User Meeting in Udaipur, India March 3-5. Udaipur has a romantic and historic past and is known for its culture, palaces and scenic areas. It is often called the “Venice of East” and was voted the best city in the world in 2009 by the Travel + Leisure magazine. Don't miss this opportunity to meet with other Design-Expert users and increase your DOE skills in this popular tourist destination. For all the details, click here.

(Third notice.) Save the Date and a Call for Papers for the Sixth European Design of Experiments (DOE) User Meeting and Workshops in lovely Leuven, Belgium on May 18 through 20. This biannual get-together is co-sponsored by Stat-Ease and the on-site host CQ Consultancy. Learn more about this fun and informative conference by clicking this link.

Click here for other 2016 appearances by Stat-Ease professionals. We hope to see you sometime this year.


- Back to top -

7: Workshop alert: From New Jersey to Norway

You can do no better for quickly advancing your DOE skills than attending a Stat-Ease workshop. In these intensive classes, our expert instructors provide you with a lively and extremely informative series of lectures interspersed by valuable hands-on exercises with one-on-one coaching. Enroll at least 6 weeks prior to the date so your place can be assured—plus get a 10% “early-bird” discount. Also, take advantage of a $400 discount when you take two complementary workshops that are offered on consecutive days.

*Take both EDME and RSM to earn $400 off the combined tuition!

See this web page for complete schedule and site information on all Stat-Ease workshops open to the public. To enroll, scroll down to the workshop of your choice and click on it, or contact the Client Specialist at workshops@statease.com or 612-746-2030. If spots remain available, bring along several colleagues and take advantage of quantity discounts in tuition. Or, consider bringing in an expert from Stat-Ease to teach a private class at your site.**

**Once you achieve a critical mass of about 6 students, it becomes very economical to sponsor a private workshop, which is most convenient and effective for your staff. For a quote, e-mail workshops@statease.com.


- Back to top -


I hope you learned something from this issue. Address your general questions and comments to me at: mark@statease.com.

Please do not send me requests to subscribe or unsubscribe—follow the instructions at the end of this message.
Sincerely,

Mark

Mark J. Anderson, PE, CQE
Principal, Stat-Ease, Inc.
2021 East Hennepin Avenue, Suite 480
Minneapolis, Minnesota 55413 USA

P.S. Quote for the month—Nate Silver on being happy with what our data tells us:


"The story data tells us is often the one we'd like to hear, and we usually make sure it has a happy ending.”
  —Nate Silver, The signal and the noise, Penguin Press, 2012

Trademarks: Stat-Ease, Design-Ease, Design-Expert and Statistics Made Easy are registered trademarks of Stat-Ease, Inc.

Acknowledgements to contributors:
—Students of Stat-Ease training and users of Stat-Ease software
Stat-Ease consultants Pat Whitcomb, Shari Kraber, Wayne Adams, Brooks Henderson and Martin Bezener
—Statistical advisor to Stat-Ease: Dr. Gary Oehlert
Stat-Ease programmers led by Neal Vaughn
—Heidi Hansel Wolfe, Stat-Ease sales and marketing director, and all the remaining staff that provide such supreme support!

Twitter-SmileyFor breaking news from Stat-Ease go to our Twitter feed.

DOE FAQ Alert ©2016 Stat-Ease, Inc.
Circulation: Over 8900 worldwide
All rights reserved.

 
 

Subscribe