DOE Analysis¶
Analysis has three stages: fit a model, read the report, then use the model.
Fit a model¶
- Regression, the command you run. It picks the engine from each output's type, so there is nothing to choose.
The three engines, for reference:
| Engine | For | Page |
|---|---|---|
| Ordinary Least Squares | quantitative outputs | Ordinary Least Squares Regression |
| Binary Logistic | two-outcome outputs | Binary Logistic Regression |
| Nominal Logistic | outputs with three or more unordered outcomes | Nominal Logistic Regression |
There is no ordinal engine. See the note on Regression.
Read the report¶
- Regression Results for OLS
- Regression Results for Binary Logistic
- Regression Results for Nominal Logistic
- The DOE Advisor, the block of findings the regression writes below itself, and the runs it marks red on the design sheet
- Uncoded Coefficients, the same fit reported in each factor's own units, for when you want the equation rather than the significance
Measure the design itself¶
- Evaluate Design, the diagnostics of a design sheet measured against its own model: the efficiencies, the variance inflation factors, the alias matrix, the column correlations, the leverage of each run and the power of each term. It reads the design only, so no response values are needed and it can be run before the experiment
Use the model¶
- Optimize, search the model for settings that meet a goal
- Charts, ten ways to look at the fit, including the two under the menu's Regression Diagnostics heading
Nothing in the software imposes an order on these. The only prerequisite is a regression sheet, which Optimize and the regression-source charts need and the design-source charts do not. See Chart Source.
The mathematics¶
- Sums of Squares, the ANOVA table and the per-term tests
- Prediction Equation, how a predicted value is formed
- Design Coding, what a coefficient's units are