Type of DOE

[#1]

Hi,

My DOE has multiple factors, each with different levels ranging from 3 to 5. I cannot create a full factorial design because the total number of runs exceeds 20,000. Instead, I tried to create a custom design with only main effects and no interaction effects. The number of tests provided is 21. How can I ensure that this custom design is reliable? Additionally, is there any method where I can later add interaction effects along with the response values that have already been filled out? When I add interaction effects, I need to refill the response table. Thank you.

[#2 · Accepted solution]

Hi @faizarizmin,

Welcome in the Community !

As it seems you're novice with DoE, I would recommend taking a look at DoE resources listed in the JMP Design of Experiments Club discussions and other ressources listed in this post (and others) : Learning materials for DOE ? 

These ressources will help you create designs and understand how to compare/evaluate them.

As you seem to have a lot of factors in this screening stage, Definitive Screening Designs may also be an interesting options if the number of runs is not too high for your experimental budget.
You could also use the Design Explorer to create and compare several design in paralell, with various run size, replicates number, optimality criterion, etc...

Finally, the design evaluation and comparison will be linked to your objective (here screening). For screening designs, the objective is to identify active effects among many potential factors. The evaluation process centers on three complementary aspects:

1. Power analysis serves as the primary tool for effect detection capability. Through power analysis tables and plots, we can assess if a design can reliably detect active effects under specified conditions.
2. Standard error estimates complement power analysis by quantifying estimation precision. The relative standard error table enables direct comparison between designs, revealing their capacity to precisely estimate both main effects (and interactions if added in the assumed model). 
3. Correlation analysis, visualized through color-coded matrices, completes the screening evaluation by exposing effect confounding. These maps reveal potential aliasing structures, helping balance run economy against effect separation capabilities.

Different objectives could lead to different design strategies and different evaluations, here is a LinkedIn post I wrote about this aspect : https://www.linkedin.com/posts/victorguiller_d%C3%A9couverte-des-plans-oml-orthogonal-main-activity-...

Hope this first asnwer will help you,

[#3]

Thank you very much!