Easy DOE Flexible Mode - Maximizing Your DOE Capabilities

[#1]

NOTE: Easy DOE Flexible Mode is most useful for users already familiar with using JMP DOE and/or need to go beyond the more basic Easy DOE Guided Mode.  If you are relatively new to DOE, consider first viewing the Mastering JMP Video Easy DOE - Guiding New Users Through   Designing and Analyzing an Experiment.

Often you may start in DOE Guided Mode and determine that you need more capabilites, such as handling replicates, that are found in the the more comprehensive DOE Flexible Mode.  This video covers cases like this towards the end.

Easy DOE Flexible mode provides more experienced users full access to conventional JMP platforms, such as Custom DOE and Definitive Screening Designs. Flexible Mode offers advanced capabilities through Define, Model, Design, Data Entry, Analyze, Predict and Report stages.  Flexible Mode also provides a Generalized Regression solution path not usually available to users with Standard JMP licenses.

JMP Easy DOE Flexible Mode is useful in situations when 1) you want more control over the design by specifying number of blocks, center points, replicates or runs or 2) you have factors that are hard to change. Flexible Mode is also useful when you want more control over fitting the model. Flexible Model lets you compare fits of different models plus lets you change:

• estimation method from default Best Subset
• estimation method Advanced Controls
• distribution from default Normal
• validation method from default AICc

See how to:

• Understand why to use JMP Easy DOE
  • End-to-end coverage of every step of experimentation
  • Streamlined experience through tailored elements in user interface
  • Guided mode for novice experimenters (default) and Flexible mode for more demanding situations
  • Comprehensive summary report is automatically written based on the current state of the experiment
  • Save your work at any time and return to the same point
  • Easily share experiments with others
• Understand Easy DOE Guided Mode paper airplane experiment steps and reports created by JMP SE, Peter Polito
  • Full Model has no significant terms
  • Best Model has 4 barely significant terms, 2 nearly significant terms, and drops the 5 least significant effects
• Identify the paper airplane Guided Mode results that might indicate more data or custom design is needed
• Understand how to use Flexible Mode with additional level(s) of complexity
  • Change plane width to continuous response as proportion of length
  • Add paper weight in pounds
  • Add 3 categorical throwers
  • Add 2 categorical launch sites as hard-to-change factor
  • Use Response Surface Model
  • Run 36 trials to shrink the size of the effects detected
  • Use a Lognormal distribution because of belief that same data in not normal
  • Force all main effects into model

Questions ansered by Clark Ledbetter @Clark_Ledbetter and Olivia Lippincott @O_Lippincott at this and the previous live webinar on this topic:

Q: Can you present p-values in the graph displayed during Easy DOE Guided Analyze? We are only seeing confidence intervals.

A: In Guided Mode we don’t see p-values.  Guided Mode is designed to keep it light statistically. To see more statistics, including p-values, you need to switch to Flexible Mode.  Flexible Mode also lets you do some blocking, use center points, do replicates and adjust the number of unique trials or the number of user specified trials.

Q: I am confused about difference between blocks and plots when you did that last one. What is block?

A: Blocks are about variance and let you group homogeneous factors to isolate extraneous noise associated with the factors (like time of day). For example, I may have 3 bags of raw material, and I think they're all the same. But one's got an older date code, so I want to check to see if that bag is so much out of date that it's shifted the process or the shift from day to day.  Plots address variability and let you organize units so you can estimate the noise/signals associated with factors. Plots target situations where there are hard and easy to change factors. So, if factors are hard to change, it may be an indication that you're going to have plots instead of blocks.  See this video, Using Blocking When Designing an Experiments.

Q: Could we run 3 tosses per term, and then average/median the distance?

A: You could have your response be the average distance and combine the 3 tosses per run (repeated measurements). Flexible mode also allows you to run replicate runs as another option. Question 2 in this blog has a good discussion on this topic.

Q: How are split plot designs accommodated with Guided Mode?

A: I believe Tom will cover split plot designs in a little bit. Flexible mode gives you an option to specify changes from easy, hard or very hard.

Q: Can we change the Confidence from 95%?

A: No, at this time, you can’t change CI in Easy DOE, either mode.

Q: Besides a judgment call by the analyst or subject matter export, is there any systemic way to say to know if something will clearly be resolved by adding more runs. I get seeing the confidence intervals just grazing 0 is a thing, but if runs cost $$$, that could be a hard argument to justify to an executive.

A: It depends.  If you had a long baseline of a process or system where you have a good estimate of noise/RMSE I guess you could argue more systematically to say more or less runs would be helpful or save money. 

Q: Is the Generalized Regression option also available in Flexible mode without Pro edition?

A: If you have Standard JMP, a limited version of Generalized Regression is available in Easy DOE Flexible mode. However, those capabilities are only available from Easy DOE Flexible Mode and if you have standard JMP, you can’t create Generalized Regression models outside of Easy DOE.

Q: Can you just manually update a variable if it changes during an experiment outside of what was planned for a run?  Also how do you integrate multiple responses?

A: I'm assuming you the range of an input, say plane length in the example.  If you go back to edit that in the Define tab, the experiment will be cleared. It looks like you will just need to run back through the workflow to build your model and analyze.  Adding responses is easier.  You can add responses on the Define tab since that isn't related to the modeling.

Q: When he went to Lognormal the AICc improved but not the R-Squared.  So, was the model better or not?

 A: Both AICc and R-Square are metrics to compare models. It is a subject matter call on the tradeoffs looking at both metrics.

Q: Can you retroactively put a completed DOE into the easy DOE interface?

A: Not in the same manner as when Tom showed like "Load Response".  There is no "Load Previous Design".  However, you could run through the Easy DOE workflow and enter in all the info to build out the final DOE report. 

Q: How are whole plots different from blocks?

A: Blocking is a restriction of the randomization process that results a balance of numbers of patients on each treatment after a prescribed number of randomizations.

Q: When JMP is 'computing design' in Flexible mode, does it computing an I-Optimal design if RSM is selected?

A: Yes ,an I-optimal design is created for RSM. A table that shows which type of design is created with the model options in Easy DOE.

Q: Instead of changing continuous variable to discrete numerical to fix the final problem, can you specify the number of center points, or am I misunderstanding what center points mean?

A: No. 1 is really mid-level of the range.  When we say Center Point, we mean the center of a design.

Q: Can you use discrete numeric to give a factor say, 4 levels or more?

A: Yes, under the red triangle option when you can factors you can specify 4 levels to 8 and even there is an "other" option.

Q: Can we evaluate signal to noise in Guided or Flexible Mode?

A: Custom DOE does this, but not through Easy DOE Flexible Mode.

Q: In the results, which plot would identify the factors that are most critical vs the remaining factors?

A: In the Generalized Regression report, look under the "Parameter Estimates for Original Predictors" .  Right click on column and sort by p-value when in Flexible mode.

Resources

• JMP SE  @Peter_Polito Peter Polito’s paper airplane data set used in this video
• Mastering JMP Video: Easy DOE - Guiding New Users Through   Designing and Analyzing an Experiment
• JMP Developer Tutorial:  Easy DOE – Expertly Guiding Users Through Designing an Experiment
• JMP Developer Clay Barker’s Tutorials on Generalized Regression
  • Using JMP Pro Generalized Regression to Analyze Designed Experiments
  • Using JMP Pro Generalized Regression to Better Understand Observational Data
  • Selecting the Appropriate JMP Pro Generalized Regression Distribution for Your Response
• JMP Systems Engineer Phil Kay’s Linked in DOE series
• JMP Series on Smarter Innovation with Design of Experiments

[gail_massari]

A question came up bout the difference between blocks and plots. Blocking is a way  At a high level, blocks Advanced Analytics Managre (and one of my go-to people for DOE questions)  @Ryan_Lekivetz reminded me what Brad Jones and Chris Nachsteim say in their paper:

 “In simple terms, a split-plot experiment is a blocked experiment, where the blocks themselves serve as experimental units for a subset of the factors.’  

The traditional view of blocking would be to want to account for the differences between blocks.  Wikipedia says:  blocking is the arranging of experimental units;in groups (blocks) that are similar to one another. Typically, a blocking factor is a source of variability that is not of primary interest to the experimenter. An example of a blocking factor might be the sex of a patient; by blocking on sex, this source of variability is controlled for, thus leading to greater accuracy.