This is the first presentation from the 2023 Summer webinar series Unlocking the Power of DOE.
In this presentation, learn about the fundamentals of statistical design of experiments (DOE). This session shows how this experimental method differs from traditional methods and how it enables scientists and engineers to maximize the return on investment in their experiment.
Q&A session
Q: What if the factors have a nonlinear relationship with the response?
A: You can add polynomial terms or other model terms that can approximate the nonlinear response. You could also consider using additional modeling capabilities that are not based on linear regression (decision trees, neural networks, etc.).
Q: How can you use JMP to analyze prior experiments where a one-factor-at-a-time approach was used?
A: You can use JMP to model historic data just like you would model a design. The factors, however, may be correlated and the resulting model may not be well-fit. You could also use more advanced regression methods to create a model or use factor screening (Response Screening or Predictor Screening) to identify important variables.
Q: If you can't precisely control your factors, can you still study the system with a designed experiment?
A: There are a few ways you can handle this situation:
- You could make sure you have a large enough range on your factor setting to make sure the variability in the setting doesn't cause the high and low settings to overlap.
- You could also simulate the variance in the factor settings using the Profile Simulator to see how the factor variance affects the response.
Q: If your X factor setting changes from the set point in the design, how do you handle this? For instance, what if the DOE called for a factor setting of 8, but you did the run at a setting of 8.2 due to limitations in controlling the factor.
A: If there are variables that cannot be controlled to the levels in the DOE, you can create a column in the data table for Actual X to record the actual value for the factor setting. You could then model the set point that the design required and then also model the actual value to see if there is a difference.
