Factorial
Change every factor at once, and still see what each one does.
Design an experiment, run it, and analyse the results, with the process-improve Python package running inside this browser tab. Nothing to install, and your data never leave your machine.
Your designs (0), kept in this browser
Each design and its results are saved here as you work, on this device only: nothing is uploaded. The 30 most recent are kept. To keep a copy elsewhere, download its workbook or copy a link to it, in step 2.
1 Design
| Factor name | Type | Low | High | Units | Remove |
|---|
A factor is numeric, from a low to a high value, or categorical, with two labels such as supplier A and B; the first label is coded -1 and the second +1. Choose each factor's type, or type two labels as its levels. Only the two-level designs take a categorical factor: full factorial, fractional factorial and Plackett-Burman.
The factors have changed since this design was generated. Press Generate design to use them in the run sheet, the simulation and the analysis.
Show the Python code
The process-improve code for this step. Copy it and it runs as it is.
2 Run the experiments
Generate a design first: its runs appear here.
Run the experiments in the order listed and type each result into the table, or simulate the results from a process whose true model is known, to see what the analysis recovers.
Show the true process
In coded units: each factor runs from -1 at its low value to +1 at its high value. The analysis estimates these coefficients from the noisy results.
Type each result into the table below as it comes in, or download the workbook, fill it in, and drop it here. A partly filled sheet is fine: the page analyses the completed runs.
Show the Python code
Steps 1 and 2 together, so it runs as it is.
3 Analyse
Simulate the results, or enter them, and the analysis appears here.
Show the Python code
Steps 1 to 3 together, so it runs as it is.
4 Compare designs
Which design suits these factors? Here the model is fixed first, and the designs differ only in where they place their runs. Each design is generated as step 1 would make it and scored at its own size, on the nine measures that Process Improvement using Data defines; the book's comparison of design families reads such a table row by row.
Show the Python code
The process-improve code for this step. Copy it and it runs as it is.