Home / DOE / Create Designs / Three-Level Factorial Designs / Three-Level Factorial How-To
Three-Level Factorial How-To¶
This walkthrough builds a full three-level factorial for two factors and shows what the third level buys that a two-level design cannot give you.
Nothing here needs data. A design is built before the experiment is run, so every number below is determined by the design itself and you can check each one against your own sheet.
Steps¶
-
QXL DOE New > Create Design, choose 3 Level Factorial, press Next.
-
On the factor page set Select number of factors: to 2. Each factor row carries a # Levels entry and a Categorical check box; leave Categorical clear on both and leave the levels at 3. Name them Temp and Press and give each one three settings.
-
Name one output, Yield, press Next.
-
Leave Select number of replicates: at 1 and the blocking list at No blocking. Press Finish.
What you get¶
Nine runs, every combination of three levels of two factors, with the first factor changing fastest:
Press Copy for Excel and paste the block into an empty part of the worksheet to compare it against your sheet.
| Run | Temp | Press |
|---|---|---|
| 1 | level 1 | level 1 |
| 2 | level 2 | level 1 |
| 3 | level 3 | level 1 |
| 4 | level 1 | level 2 |
| 5 | level 2 | level 2 |
| 6 | level 3 | level 2 |
| 7 | level 1 | level 3 |
| 8 | level 2 | level 3 |
| 9 | level 3 | level 3 |
Nine is 3 times 3, and the run count of a full factorial is always the product of the level counts. Three factors at three levels is 27 runs, four is 81.
What the third level buys¶
A squared term becomes estimable. Two levels can only fit a straight line through a factor, so a two-level design cannot tell a flat response from a curved one. Three levels give the middle point that makes the curvature visible.
The cost is the run count, and it grows faster than a two-level design: 3^k against 2^k. At five factors that is 243 runs against 32. When curvature matters on only two or three factors, a central composite or Box-Behnken design reaches the same curved model in far fewer runs.
Things to try next¶
- Tick Categorical on one factor. Its three levels become names you type rather than numbers, and the model treats it as three unordered categories rather than a quantity with a middle.
- Add a third factor. The run count goes from 9 to 27, which is the multiplication made visible.
- Set Select number of replicates: to 2. 18 runs, and the regression gains a pure error estimate.
See Also¶
- Three-Level Factorial Designs
- Options, every control on every page of the builder
- Math Details
- Central Composite Designs, a curved model in fewer runs