This action might not be possible to undo. Are you sure you want to continue?

BooksAudiobooksComicsSheet Music### Categories

### Categories

### Categories

Editors' Picks Books

Hand-picked favorites from

our editors

our editors

Editors' Picks Audiobooks

Hand-picked favorites from

our editors

our editors

Editors' Picks Comics

Hand-picked favorites from

our editors

our editors

Editors' Picks Sheet Music

Hand-picked favorites from

our editors

our editors

Top Books

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Audiobooks

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Comics

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Sheet Music

What's trending, bestsellers,

award-winners & more

award-winners & more

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

“The real voyage of discovery consists not in seeking new landscapes, but in having new eyes.” Marcel Proust

Design of Experiments

JMP, A Business Unit of SAS SAS Campus Drive Cary, NC 27513

JMP 6.0.3

JMP Design of Experiments, Release 6 Copyright © 2005, SAS Institute Inc., Cary, NC, USA ISBN 1-59047-816-9 All rights reserved. Produced in the United States of America. For a hard-copy book: No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, or otherwise, without the prior written permission of the publisher, SAS Institute Inc. For a Web download or e-book: Your use of this publication shall be governed by the terms established by the vendor at the time you acquire this publication. U.S. Government Restricted Rights Notice: Use, duplication, or disclosure of this software and related documentation by the U.S. government is subject to the Agreement with SAS Institute and the restrictions set forth in FAR 52.227-19, Commercial Computer Software-Restricted Rights (June 1987). SAS Institute Inc., SAS Campus Drive, Cary, North Carolina 27513. 1st printing, August 2005 JMP®, SAS® and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are registered trademarks or trademarks of their respective companies.

Contents

Contents

Design of Experiments 1 Introduction to Designing Experiments

About Designing Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 My First Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 The Situation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Step 1: Design the Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Step 2: Define Factor Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Step 3: Add Interaction Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Step 4: Determine the Number of Runs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Step 5: Check the Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Step 6: Gather and Enter the Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Step 7: Analyze the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2

**Examples Using the Custom Designer
**

Creating Screening Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Main-Effects-Only Screening Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Screening Design to Fit All Two-Factor Interactions . . . . . . . . . . . . . . . . . . . . . . . . A Compromise Design Between Main Effects Only and All Interactions . . . . . . . . . . . . . . . . . Creating “Super” Screening Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Screening Designs with Flexible Block Sizes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Checking for Curvature Using One Extra Run . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating Response Surface Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Exploring the Prediction Variance Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Introducing I-Optimal Designs for Response Surface Modeling . . . . . . . . . . . . . . . . . . . . . . . . A Three-Factor Response Surface Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Response Surface with a Blocking Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating Mixture Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Special-Purpose Uses of the Custom Designer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Designing Experiments with Fixed Covariate Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Design with Two Hard-to-Change Factors: Split Plot . . . . . . . . . . . . . . . . . . . . . . . Technical Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 17 19 20 22 26 29 32 32 36 37 39 42 45 45 48 52

3

**Building Custom Designs
**

Creating a Custom Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

ii

Contents

Entering Responses and Factors in the Custom Designer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Describing the Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Selecting the Number of Runs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Understanding the Design and Profilers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Understanding the Variance of Coefficients Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Understanding the Alias Matrix (Confounding Pattern) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Specifying Output Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Making the Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating Split Plot Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Special Custom Design Commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Saving Responses and Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Loading Responses and Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Loading and Saving Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Setting the Random Number Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Viewing Diagnostics Reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Saving the Design (X) Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Changing the Design Criterion (D- or I- Optimality) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Selecting the Number of Random Starts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Constraining a Design to a Hypersphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Accounting for Factor Level Restrictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Advanced Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Assigning Column Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Defining Low and High Values (DOE Coding) for Columns . . . . . . . . . . . . . . . . . . . . . . . . . Setting Columns as Factors for Mixture Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Defining Response Column Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Giving Columns a Design Role . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Identifying Factor Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . How Custom Designs Work: Behind the Scenes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

57 61 62 63 63 64 64 65 65 66 67 67 68 69 69 70 70 71 72 73 74 75 76 76 77 78 79 80

4

Screening Designs

Screening Design Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using Two Continuous and One Categorical Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using Five Continuous Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Screening Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Entering Responses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Entering Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Choosing a Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Displaying and Modifying the Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Specifying Output Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Viewing the Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Continuing the Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 83 85 87 88 89 90 93 97 97 98

Contents

iii Contents

5

**Response Surface Designs
**

A Box-Behnken Design: The Tennis Ball Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 Viewing the Geometry of a Box-Behnken Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 Creating a Response Surface Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 Entering Responses and Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 Choosing a Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 Specifying Axial Value (Central Composite Designs Only) . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 Specifying Output Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 Viewing the Design Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 Continuing the Analysis, If Needed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

6

**Full Factorial Designs
**

The Five-Factor Reactor Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Factorial Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Entering Responses and Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Selecting Output Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Making the Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 125 125 126 126

7

Mixture Designs

Choosing a Mixture Design Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 Creating an Optimal Mixture Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 Using the Simplex Centroid Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 Creating the Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 Simplex Centroid Design Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 Using the Simplex Lattice Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 Using the Extreme Vertices Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 Creating the Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 An Extreme Vertices Example with Range Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136 An Extreme Vertices Example with Linear Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 Extreme Vertices Method: How It Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 Using the ABCD Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 Creating Ternary Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140 Fitting Mixture Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141 Understanding Whole Model Tests and ANOVA Reports . . . . . . . . . . . . . . . . . . . . . . . . . . . 142 Understanding Response Surface Reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 A Chemical Mixture Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 Creating the Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 Running a Mixture Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144 Viewing the Prediction Profiler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 Plotting a Mixture Response Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Guidelines for the Analysis of Deterministic Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Replicating a Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using the Uniform Design Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Sphere Packing Design . . Creating a Latin Hypercube Design . . . . . . . . . . . . . . . Creating an Augmented Design . . . . . . . . . . . . . . . . . . . . Comparing the Sphere Packing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 151 151 153 154 154 155 157 158 159 160 161 162 163 9 Nonlinear Designs Example of Creating a Nonlinear Design . . . . . Creating a Taguchi Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results of the Borehole Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 169 174 178 178 179 179 180 180 10 Taguchi Designs The Taguchi Design Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Analyze the Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Nonlinear Design . . . . . . . . . . . . . . . . . . . . . . . . . . Choosing Inner and Outer Array Designs . . . . . . . . . . . . . . . . . . . .iv Contents 8 Space Filling Designs Introduction to Space Filling Designs . . . . . . . . . . . . . . . . . . . . . . Detailing the Response and Add Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Setting Up Factors and Parameters . . . . . . . . . . . . . . . . . . . . . Making or Augment the Table . . . . . . . . . . . . . Analyze the Augmented Design . . . . . . . . . . . . Borehole Model Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Selecting the Number of Runs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using the Minimum Potential Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Displaying and Modify the Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Visualizing the Latin Hypercube Design . . . . and Uniform Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Viewing the Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using the Sphere Packing Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Identifying the Responses/Column with Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 198 204 204 . . . . . . . . . . . Creating a Nonlinear Design with No Prior Data . . . . . . Latin Hypercube. . . . . . . . . . . . . . . . . . . . . . . . . Using the Latin Hypercube Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Taguchi Design Example . . . . . . . . . . Fitting to Get Prior Parameter Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185 185 187 189 190 191 191 192 11 Augmented Designs The Reactor Example: Using D-Optimal Augmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Making the Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Sphere Packing Design . . . . . . . . . . . . . . . . . . . . . . . . . . Visualizing the Sphere Packing Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . Adding New Runs and Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . One-Sample and Two-Sample Means . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Single-Sample Mean . . . . . Specifying the Sphere Radius Value . . . . . . . . . . . . . . . . . . . . . . . . . . One-Sample Variance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sigma Quality Level . . . . . . . . . . . . . . . . . . . . . . . . . . .Contents v 207 208 209 210 213 214 214 214 215 215 215 Contents Adding Centerpoints . . . . . . . 12 Prospective Power and Sample Size Prospective Power Analysis . . . . . . . . . . . . . . . . . k-Sample Means . . . . . . . . . . . . . . . . . . . . . . Saving the Design (X) Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .or I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Foldover Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . One-Sample and Two-Sample Proportions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Selecting the Number of Random Starts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Optimality) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Adding Axial Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Modifying the Design Criterion (D. . . . . . . . . . . . . Counts per Unit . . . . . . . . . . . Special Augment Design Commands . . . . . Disallowing Factor Combinations . . . . . . . . . . . . . . . . . . . . . . . Two-Sample Means . . . . . . . . 219 220 221 223 224 225 226 227 229 230 13 References 14 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Power and Sample Size Animation for a Single Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Viewing Diagnostics Reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.

David Iklé. to the alpha and beta testers of JMP. Frank Shen. and Padraic Neville for statistical R&D support. Credits JMP was conceived and started by John Sall. George Fraction. Lynn Vanatta. Fang Chen. Wenjun Bao.. Tom Johnson. Kathleen Kiernan. Chris Gotwalt. Sky Alibhai. Gordon Johnston. and Tonya Mauldin. Francois Mainville.H. Kyoko Keener. Zoe Jewell. Stan Young. Laura Lancaster. Homer Hegedus. Barry Hembree. Additional testing and technical support is done by Noriki Inoue. Jim Borek. Lenore Herzenberg. Erin Vang. Bob Hickey is the release engineer. Jeff Sweeney. Jack Reese. Paul Wenz. though portions of JMP were adapted from routines in the SAS System. Annie Dudley Zangi. Also. and Meredith Blackwelder. Thanks also to Georges Guirguis. Ryan Gilmore. JMP is not a part of the SAS System. David Tanaka. DaRon Huffaker. Elizabeth Edwards. Duane Hayes. Howard Yetter (d). Mike Leonard. Brian Corcoran. Russell Wolfinger. Goodman. with contributions from Annie Dudley Zangi and Brian Corcoran. . Ron Thisted. In the SAS Institute Technical Support division. Duane Hayes. Joseph Morgan. Gray McQuarrie. Design and development were done by John Sall. Nicole Jones. Bradley Jones. Andy Mauromoustakos. Joe Hockman. Ying So. Robert Stine. Special thanks to Mark Bailey and Lori Rothenberg from SAS. Cy Wegman. David Coleman. Al Fulmer. Michael Emptage. Mike Bowen. Paul Nelson. Warren Kuhfeld. Version 1 of JMP went into production in October. Michael Hecht. George MacKensie. Hui Di. Robert N. José Ramirez. NC. In particular we thank Michael Benson. and Toby Trott provide technical support and conduct test site administration. Lee Creighton. Chung-Wei Ng. Paul Dwyer. Nancy McDermott. and John Wass. Melanie Drake implemented the help system. Gudmunder Axelsson. and Kris Ghosh. Bradley Jones. Linda Blazek. Rodriguez. and Masakazu Okada from SAS Japan. and Hugh Crews provide ongoing quality assurance. Avigdor Cahaner.Credits Credits and Acknowledgments Origin JMP was developed by SAS Institute Inc. Scott McQuiggan. we thank Dick DeVeaux. Craige Hales. Al Best. Bob Lucas. Xan Gregg. Ramon Leon. Acknowledgments We owe special gratitude to the people that encouraged us to start JMP. John Sall. Kyoko Takenaka. Statistical technical support is provided by Craig DeVault. Skip Weed. and Wenjie Bao. J. Lou Valente. Georgia Morgan. Cary Tuckfield. Robert Muenchen. Creative services and production was done by SAS Publications. Veronica Czitrom. Wendy Murphrey. Lori Gates. Warren Kuhfeld. The JMP manuals were written by Ann Lehman. Tom Lange. 1989. Michael Friendly. and to the reviewers of the documentation. Pat Spagan. Warren Sarle. Dan Obermiller. Richard Potter. Randall Tobias. particularly for linear algebra and probability calculations. Mike Thompson. Cary. Jianfeng Ding. Kevin Norwood.

Chris Fehily. Eddie Routten. the Texas air control board for the pollution data. Dana C. Matthew Lay. Robert Mee for screening design generators. DATA OR PROFITS.S. Thanks to SAS Institute quality assurance by Jeanne Martin. George Milliken and Yurii Bulavski for development of mixed models. Tim Clapp. Ruth Lee. Thanks for support from Charles Shipp. Ralph O’Brien for advice on homogeneity of variance tests. For sample data. David Schlotzhauer. Will Potts and Cathy Maahs-Fladung for data mining. Also thanks to Jenny Kendall. and Maura Stokes for testing JMP Version 1. Yusuke Ono. Jeff Polzin. Kyoko Takenaka. Jim Shook. Science University of Tokyo) and Professor Hirohiko Asano (Tokyo Metropolitan University). Jason Hsu for advice on multiple comparisons methods (not all of which we were able to incorporate in JMP). Dan Chilko. Aultman. Harold Gugel (d). Meeker for advice on reliability plots. Finally. and J. Grey Piepel. Ramon Leon. and Masakazu Okada of SAS Japan were indispensable throughout the project. Dave Tilley. Additional testing for Versions 3 and 4 was done by Li Yang. Translations Erin Vang coordinated localization. Brenda Sun. Rubin Gabriel. and Tom Little. Rick Blahunka. IN NO EVENT SHALL KEITH PACKARD BE LIABLE FOR ANY SPECIAL. Stuart Hunter for advice on design of experiments. Brian Ruff. Past Support Many people were important in the evolution of JMP. Technology License Notices JMP for the Power Macintosh was compiled and built using the CodeWarrior C compiler from MetroWorks Inc. Hocking and P. and Joe Ficalora for advice in the area of Six Sigma. and William Q. ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. and Andrea Ritter. Paul Tobias. Ken Bodner. Fouad Younan. and William Fehlner. Eric Gjertsen. . Katrina Hauser. XRender is Copyright © 2002 Keith Packard. Special thanks to Professor Toshiro Haga (retired. thanks to all the members of our outstanding translation teams. Paul Wright for advice on statistical power. and David Coleman for the pollen (eureka) data. Aaron Walker. Lijian Yang and J. Keith Muller for advice in multivariate methods. Tim Christensen. Dave Trindade. and James Mulherin. Kristin Nauta. The ImageMan DLL is used with permission of Data Techniques. Peter Goos.viii Credits and Acknowledgments We also thank the following individuals for expert advice in their statistical specialties: R. thanks to Patrice Strahle for Pareto examples. Inc. Marron for bivariate smoothing design. Wayne Nelson. Annette Sanders. Ralph O’Brien and S. and Junji Kishimoto. Blanton Godfrey. Thanks to Steve Shack. Eric Wasserman. NEGLIGENCE OR OTHER TORTIOUS ACTION. Clay Thompson for advice on contour plotting algorithms. INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE. Harry Martz. Greg Weier. David DeLong. Tom Esposito. Special thanks to Jeffrey Perkinson. Mary Cole. Jim Winters. Spector for advice on effective hypotheses. and Frank Lassiter. Tim Rey. Noriki Inoue. INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS.KEITH PACKARD DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE. John Hansen. David Morganstein. Charles Soper. Susan West. Ike Walker. William Lisowski. WHETHER IN AN ACTION OF CONTRACT.

1 Introduction to Designing Experiments A Beginner’s Tutorial This tutorial chapter introduces you to the design of experiments (DOE) using JMP’s custom designer. It gives a general understanding of how to design an experiment using JMP. Refer to subsequent chapters in this book for more examples and procedures on how to design an experiment for your specific project. 1 Introduction .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Step 7: Analyze the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Step 4: Determine the Number of Runs . . . . . . . My First Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Step 3: Add Interaction Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Step 2: Define Factor Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Contents About Designing Experiments . . . . . . . . . . . . . . . . . . . . . . . Step 5: Check the Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Situation . . . . . . . . . . . . . . . . . . Step 1: Design the Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Step 6: Gather and Enter the Data . . . . . . . 3 3 3 4 6 6 7 8 10 10 . . . . . . . . . . .

you define “the best popped bag” based on the ratio of popped kernels to the total number of kernels. you can use JMP’s profiling tools to determine the optimum settings of popping time. Therefore. JMP’s custom designer determines which brand to use. After popping a bag. but some remain unpopped. The methods for determining how to increase productivity and improve quality are evolving. how long to cook each bag in the microwave and what power setting to use. enter the total number of kernels and the number of popped kernels into the appropriate row of JMP data table. My First Experiment If you have never used JMP to design an experiment. and cost-effective statistical methods that JMP provides. it is easy to decide on some reasonable ranges for the important factors: • how long to cook the popcorn (between 3 and 5 minutes) • what level of power to use on the microwave oven (between settings 5 and 10) • which brand of popcorn to use (Top Secret or Wilbur) When a bag of popcorn is popped. The Situation Your goal is to find a better procedure for microwaving a bag of popcorn.1 Introduction to Designing Experiments—About Designing Experiments 3 1 Introduction About Designing Experiments Increasing productivity and improving quality are important goals in any business. power level. elegant. and runs. After doing all the experimental runs. You prefer to have all (or nearly all) of the kernels popped and no (or very few) unpopped kernels. Because you have some experience with this. use JMP’s model fitting capabilities to do the data analysis. . Each run involves popping one bag of corn. Tip: The recommended way to create an experiment is to use the custom designer. For each experimental run. and brand. Then. They have changed from costly and time-consuming trial-and-error searches to the powerful. JMP helps you determine if and how a factor affects a response. A good way to improve any procedure is to conduct an experiment. this section shows you how to design the experiment and how to understand JMP’s output. most of the kernels pop. JMP also provides classical designs for use in other situations. Designing experiments in JMP is centered around factors. responses.

” Since you want to increase the number of popped kernels. Add the second response (total number of kernels) by clicking the Add Response button and choosing None from the menu that appears.” The completed Responses panel looks like Figure 1. Rename the response labeled Y2 by double-clicking it and typing “Total Kernels.4 1 Introduction to Designing Experiments—My First Experiment Step 1: Design the Experiment The first step is to select DOE > Custom Design (Figure 1. By default. define the responses and factors. . JMP labels this response Y2 by default.3. leave the goal at Maximize. Rename the response labeled Y by double-clicking it and typing “Number Popped. the custom designer contains one response labeled Y (Figure 1. Figure 1. so enter both in the custom designer in place of the default Y response: 1.1).2 Open and Close the Responses Panel by Clicking the Disclosure Icon You have more than one response. 3.2). After popping the bag add the number of unpopped kernels to the number of popped kernels to get the total number of kernels in the bag. Figure 1. 2. Then. you have two responses: • the number of popped kernels • the total number of kernels in the bag.1 Select the Custom Designer Define the Responses: Popped Kernels and Total Kernels In this experiment.

Rename the default levels (currently named -1 and 1) as 3 and 5 by clicking the current name and typing. and Brand In this experiment. Rename the default levels (currently named -1 and 1) as 5 and 10 by clicking the current name and typing. 8.3 Renamed Responses with Specified Goals Define the Factors: Time. 9. Change the name of the factor (currently named X1) by double-clicking it and typing Brand. 5. Rename the default levels (currently named L1 and L2) as Top Secret and Wilbur by clicking the current name and typing. Click Add Factor and select Categorical > 2 Level. Click Add Factor and select Continuous. Change the name of the factor (currently named X2) by double-clicking it and typing Time.Click Continue. 6. The completed Factors panel looks like Figure 1.1 Introduction to Designing Experiments—My First Experiment 5 1 Introduction Figure 1. Figure 1. 2. Add Time as a two-level continuous factor: 4. Power.4 Renamed Factors with Specified Values 10. the factors are: • brand of popcorn (Top Secret or Wilbur) • cooking time for the popcorn (between 3 and 5 minutes) • microwave oven power level (between settings 5 and 10) Add Brand as a two-level categorical factor: 1.4. Change the name of the factor (currently named X3) by double-clicking it and typing Power. . Click Add Factor and select Continuous. 3. Add Power as a two-level continuous factor: 7.

and the power settings on the microwave are between five and 10.5. but greater than or equal to 10: 1. Click the Interactions button and select 2nd.5 Defining Factor Constraints Step 3: Add Interaction Terms You suspect that the effect of any factor on the proportion of popped kernels may depend on the value of some other factor. JMP adds two-factor interactions to the model. as shown to the right in Figure 1. Click the Add Constraints button twice. once for each of the known constraints. Figure 1. . 2. you know that: • You don’t want to have runs in the experiment where the corn is popped for a long time on a high setting because this tends to scorch some kernels. 1. You decide to include all the possible two-factor interactions in your a priori model of the popcorn popping process. • You don’t want to have runs in the experiment where the corn is popped for a brief time on a low setting because then not many kernels pop. which avoids combinations of Power and Time that sum to less than 10 and more than 13. 3.6 1 Introduction to Designing Experiments—My First Experiment Step 2: Define Factor Constraints The popping time for this experiment is between three and five minutes. Complete the information. As illustrated on the left in Figure 1.5. Open the Constraints panel by clicking the disclosure button beside the Define Factor Constraints title bar. for the Wilbur brand of popcorn the effect of a change in time could be larger than the effect of the same change in time using the Top Secret brand. For example. the area inside the parallelogram is the allowable region for the runs.You can see the popping for 5 minutes at a power of 10 is not allowed and neither is popping for 3 minutes at a power of 5. This kind of synergistic effect of factors acting in concert is called a two-factor interaction. From experience. To limit the experiment so that the amount of popping time combined with the microwave power level is less than or equal to 13. Be sure to change <= to >= in the second constraint.

JMP has no restrictions on the number of runs you request. or you can specify your own number of runs as long as they’re above the minimum. You can use the minimum. Step 4: Determine the Number of Runs The Design Generation panel shows the minimum number of runs needed to perform the experiment based on the effects you’ve added to the model. any of the other suggested sample sizes. let’s use the default number of runs. Click the Powers button and select 2nd to add quadratic effects of the continuous factors.1 Introduction to Designing Experiments—My First Experiment 7 1 Introduction 2. For this example. You do this because you suspect the graph of the relationship between any factor and any response might be curved. 16. . power and time. Click Make Design to continue. Quadratic model terms allow the model to fit that curvature.

8 1 Introduction to Designing Experiments—My First Experiment Step 5: Check the Design When you click the Make Design button. The Run Order option lets you designate the order you want the runs to appear in the data table when it is created. Keep the selection at Randomize so the rows (runs) in the output table appear in a random order. Note that because JMP uses a random seed to generate custom designs and there is no unique optimal design for this problem. JMP generates and displays a design table. as shown below. Now click Make Table. Now look at the Output Options area. . your table may be different than the one shown here. Take a look at the table to see that the custom design requires 8 runs using each brand of popcorn.

the difference is negligible. then cook Top Secret for five minutes using a power setting of five. Tip: Before clicking Make Table in the Output Options area. You do not have infinitesimal control over the power and time settings on a microwave. and so forth. Although this altered design is slightly less optimal than the one the custom designer suggested. .1 Introduction to Designing Experiments—My First Experiment 9 1 Introduction The resulting data table shows the order in which you should do the experimental runs and provides columns for you to enter the number of popped and total kernels. you can select Sort Left to Right in the Run Order menu to have JMP present the results according to the brand. So you should feel free to round the power and time settings. cook the first bag of Wilbur for three minutes using a power setting of ten. According to the table.

start with the design data table. as shown in Figure 1. Distribution to Binomial. But because you are modeling the proportion of popped kernels it is more appropriate to assume that your responses come from binomial distribution. manually enter the numbers. it is time to model the data. shown below.6. We have conducted this experiment for you. results from experiment Step 7: Analyze the Results After conducting the experiment and entering the number of popped and total kernels into the data table. The default fitting personality in the model dialog is Standard Least Squares. One assumption of standard least squares is that your responses are normally distributed. 2. It contains a script in the top left panel labeled Model.10 1 Introduction to Designing Experiments—My First Experiment Step 6: Gather and Enter the Data Pop the popcorn according to the design JMP provided. and Link Function to Logit. Change the Personality to Generalized Linear Model. 1. Finally. You can use this assumption by changing to a generalized linear model. To do this. Right-click Model and select Run Script.jmp. When you created the design. the appropriate analysis (in this case. a standard least squares analysis) is stored in the Model script. . and placed our results in the Sample Data folder that was installed when you installed JMP. Then. into the data table’s appropriate columns. count the number of popped and unpopped kernels left in each bag. To open the results. select Sample Data > Design Experiment > Popcorn DOE Results.

7) and look in the column labeled Prob>Chisq. Figure 1. In this column. in this experiment.1 Introduction to Designing Experiments—My First Experiment 11 1 Introduction Figure 1.6 Fitting the Model 3. You have confirmed that there is a strong relationship between popping time. Select Prediction Profiler from the red triangle menu on the report title bar. all the model effects except the two-factor interaction involving Brand and Power are highly significant. power. This column lists p-values. use the Prediction Profiler to develop an understanding of how changes in the factor settings affect the numbers of popped and unpopped kernels: 1. Values with * beside them are p-values that indicate the results are statistically significant. Open it by clicking the disclosure button beside the Prediction Profiler title bar. 4. Scroll to the bottom of the report to open the Prediction Profiler. To further investigate. A low p-value (a value less than 0. The Profiler (Figure 1. and brand and the proportion of popped kernels.05) indicates that results are statistically significant.8) displays prediction traces for each factor. You can therefore conclude that.7 Investigating P-Values p-values indicate significance. Click the Run Model button. Scroll down to view the Effect Tests table (Figure 1. asterisks indicate the low p-values. .

Click the red line in the time graph and drag it right and left (Figure 1. Figure 1. the steepness of a prediction trace reveals a factor’s importance. JMP automatically adjusts the graph to display the optimal settings at which the most kernels will be popped (Figure 1. The fact that there is a substantial slope shift tells you that there is an interaction (synergistic effect) involving time and power.8 The Prediction Profiler Prediction trace for brand Prediction trace for time Prediction trace for power predicted value of the response 95% confidence interval on the mean response Factor values (here. Because the prediction trace for power is steeper than that for brand or time (see Figure 1. you can see that the microwave power setting is more important than the brand of popcorn or the amount of cooking time. Click the red triangle icon in the Prediction Profiler title bar and select Desirability Functions. 4.9).9 Moving the Time Value from 4 to 5 As time increases and decreases. Furthermore. Move the red dotted lines to see the effect that changing the factor’s value has on the response. Click the red triangle icon in the Prediction Profiler title bar and select Maximize Desirability. 3. Now for the final steps. time = 4) 2. the curved time and power prediction traces shift their slope and maximum/minimum values as you change the values of time.12 1 Introduction to Designing Experiments—My First Experiment Figure 1.10). .9).

1 Introduction to Designing Experiments—My First Experiment 13 1 Introduction Our experiment found how to cook the bag of popcorn with the greatest proportion of popped kernels: use the Wilbur brand. Figure 1. The experiment predicts that cooking at these settings will yield greater than 97% popped kernels. and use a power level of 7.8 (or 8).10 The Most Desirable Settings The best settings are: • the Wilbur brand • cooking time at 5 • power set at 7. cook for five minutes.79 .

.

JMP’s custom designer is the recommended way to describe your process and create a design that works for your situation.2 Examples Using the Custom Designer The use of statistical methods in industry is increasing. measure response for each run Key mathematical steps: appropriate computer-based tools are empowering . You can also control the number of experimental runs.To use the custom designer. then JMP tailors a design to suit your unique case. This approach is more general and requires less experience and expertise than previous tools supporting the statistical design of experiments. It shows how to drive its interface to build a design using this easy step-by-step approach: Key engineering steps: process knowledge and engineering judgement are important. which can be any number greater than or equal to the number of unknowns in the model. the most cost-beneficial of these methods for quality and productivity improvement is statistical design of experiments. Custom designs accommodate any number of factors of any type. predict. Arguably. 2 Examples Describe identify factors and responses Design compute design for maximum information from runs Collect Fit compute best fit of mathematical model to data from test runs Predict use model to find best factor settings for on-target responses and minimum variability use design to set factors. This makes custom design more flexible and more cost effective than alternative approaches. The purpose of experimental design is to characterize. This chapter presents several examples showing the use of custom designs. you first enter the process variables and constraints. and then cost-effectively improve the behavior of any system or process. A trial-and -error search for the vital few factors that most affect quality is costly and time-consuming.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Introducing I-Optimal Designs for Response Surface Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating “Super” Screening Designs . . . . . . . . . . . . . . . . . . . . . . . . Special-Purpose Uses of the Custom Designer . . . . . . . . . . . . Technical Discussion . . . . . . . . . Designing Experiments with Fixed Covariate Factors . . . . . . . . . Checking for Curvature Using One Extra Run . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A Three-Factor Response Surface Design . . . . . . . . . . . . . . . . . . Creating Mixture Experiments . . . Creating Response Surface Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 17 19 20 22 26 29 32 32 36 37 39 42 45 45 48 52 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Screening Design to Fit All Two-Factor Interactions . . . . . . . . . . . . . . . . A Compromise Design Between Main Effects Only and All Interactions . . . . . . . . . Response Surface with a Blocking Factor . . . . . . .2 Contents Creating Screening Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Exploring the Prediction Variance Surface . . . . Screening Designs with Flexible Block Sizes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Main-Effects-Only Screening Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Design with Two Hard-to-Change Factors: Split Plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Figure 2. . As such. The default model contains only the main effects. 3. The straightforward screening examples described below show that ‘custom’ is not equivalent to “exotic.” The custom designer is a general purpose design environment. 4. it can create screening designs.2 Examples Using the Custom Designer—Creating Screening Experiments 17 2 Examples Creating Screening Experiments You can use the screening designer in JMP to create screening designs. but the custom designer is more flexible and general. Using the default of eight runs. All the main effects are estimable. for details). click Make Design. Click Continue. but they are confounded with two factor interactions. Select DOE > Custom Design. Creating a Main-Effects-Only Screening Design To create a main-effects-only screening design using the custom designer: 1. Enter six continuous factors into the Factors panel (see “Step 1: Design the Experiment. 2.” p. 4. Note to DOE experts: The result is a resolution-three screening design.1 shows the six factors.

. ones and negative ones. Figure 2.1 A Main-Effects-Only Screening Design 5. the columns labeled 2 6 and 3 4 in the table have a 1 in the row for X1. This means that the expected value of the main effect of X1 is actually the sum of the main effect of X1 and the two-factor interactions X2*X6 and X3*X4. For example. Click the disclosure button ( on Windows/Linux and on the Macintosh) to open the Alias Matrix.2 shows a table of zeros.18 2 Examples Using the Custom Designer—Creating Screening Experiments Figure 2. You are assuming that these interactions are negligible in size compared to the real effect of X1. Figure 2.2 The Alias Matrix The Alias Matrix shows how the coefficients of the constant and main effect terms in the model are biased by the two-factor interaction terms that are not in the model.

select Interactions > 2nd. 4. if they are strong.2 Examples Using the Custom Designer—Creating Screening Experiments 19 2 Examples Note to DOE experts: The Alias matrix is a generalization of the confounding pattern in fractional factorial designs. Two-factor interactions are estimable but are confounded with three-factor interactions. you can create experiments resolving all the two-factor interactions.” p. To avoid this risk. In the Design Generation Panel choose Minimum for Number of Runs and click Make Design. Figure 2. In the Model panel. 1. Click Continue. 3. The sample size.3 All Two-Factor Interactions .3 shows the runs of the two-factor design with all interactions. can confuse the results of such experiments. for details). The values in your table may be different from those shown below. 2. 16 (a power of two) is large enough to fit all the terms in the model. Note to DOE experts: The result in this example is a resolution-five screening design. Figure 2. Enter five continuous factors into the Factors panel (see “Step 1: Design the Experiment. 5. 4. The risk is that two-factor interactions. Creating a Screening Design to Fit All Two-Factor Interactions There is risk involved in designs for main effects only. Select DOE > Custom Design.

main effects and two-factor interactions is 22. Define six continuous factors (X1 . This example shows how to obtain such a design compromise using the custom designer. as shown in Figure 2. 1. It would be good to find a compromise between the main-effects only design and a design capable of fitting all the two-factor interactions. All the rows above the row labelled X1*X2 contain only zeros which means that the Intercept and main effect terms are not biased by any two-factor interactions. This is more than the resource budget of 16 runs. This means that the expected value of the two-factor interaction X1*X2 is not biased by any other terms. Figure 2.4 Alias Matrix Showing all Two-Factor Interactions Clear of all Main Effects A Compromise Design Between Main Effects Only and All Interactions In a screening situation.X6). Look at the column labelled 1 2. Click the disclosure button to open the Alias Matrix. All the others are 0. 3. Click Continue. 4. 5. With six factors. All the other rows and columns are similar. The first example in this section showed an eight-run design that fit all the main effects. There is only one value of 1 in that column. The default estimability of these terms is Necessary. Change Necessary to If Possible.20 2 Examples Using the Custom Designer—Creating Screening Experiments 6. there are 15 possible two-factor interactions. The 1 occurs in the row labelled X1*X2.4 shows a table of zeros and ones. Figure 2. Add all the two-factor interactions by clicking the Interactions button and choosing 2nd. suppose there are six continuous factors and resources for n = 16 runs.5. Select all the interaction terms and click the current estimability (Necessary) to reveal a menu. The minimum number of runs that could fit the constant. The model includes the main effect terms by default. . 2. Select DOE > Custom Design.

the custom designer builds a design to estimate as many two-factor interactions as possible. but with a similar pattern. .2 Examples Using the Custom Designer—Creating Screening Experiments 21 2 Examples Figure 2.6). Click Make Design. Although the desired number of runs (16) is less than the total number of model terms. 7.5 Model for Six-Variable Design with Two-Factor Interactions Designated If Possible 6. The values in your table may be different from those shown below. Type 16 in the Number of Runs edit box. click the disclosure button to open the Alias Matrix (Figure 2. After the custom designer creates your design.

the number of model terms exceeds the number of runs (Lin. Two-factor interactions are estimable but are confounded with other two-factor interactions.25 in the 1 2 column and the same value in the 5 6 column. you can (barely) fit the model. which makes them attractive for factor screening when there are many factors. Creating “Super” Screening Designs This section shows how to use the technique used in the previous example to create “super” screening (supersaturated) designs. The row labelled X1*X2 has the value 0. The experimenter expects that only a few of the factors in a screening experiment are active. The Need for Supersaturated Designs The 27–4 and the 215–11 fractional factorial designs available using the screening designer are both saturated with respect to a main effects model. 1993). Until recently. saturated designs represented the limit of efficiency in designs for screening. Factor screening relies on the sparsity principle. That means the expected value of the estimate for X1*X2 is actually the sum of X1*X2 and any real effect due to X5*X6. and experimental runs are expensive. In a supersaturated design.6 Alias Matrix All the rows above the row labelled X1*X2 contain only zeros. which means that the Intercept and main effect terms are not biased by any two-factor interactions. In a saturated design. The problem is not knowing which are the vital few factors and . but there are no degrees of freedom for error or for lack of fit. A supersaturated design can examine dozens of factors using fewer than half as many runs. Supersaturated designs have fewer runs than factors. the number of runs equals the number of model terms. Note to DOE experts: The result in this particular example is a resolution-four screening design. In the analysis of a saturated design.22 2 Examples Using the Custom Designer—Creating Screening Experiments Figure 2. as the name suggests.

as shown in Figure 2.7. Using model terms designated If Possible provides the machinery for creating a supersaturated design. • Analysis of supersaturated designs cannot yet be reduced to an automatic procedure. Change Necessary to If Possible. In a supersaturated design. A rule of thumb is that the number of runs should be at least four times larger than the number of active factors. To set this up. Figure 2. the only primary term is the intercept. and you can examine dozens of factors using less than half as many runs. Highlight all terms except the Intercept and click the current estimability (Necessary) to reveal the menu. In a supersaturated design. you should have at least 20 runs. then it is likely that these factors will be impossible to identify. Click the red triangle on the Custom Design title bar and select Simulate Responses. 5. 3. we specified two-factor interactions terms as If Possible. This is usually economically unfeasible. Add 12 continuous factors and click Continue. However. Example: Twelve Factors in Eight Runs As an example. What happens to winnow the list from dozens to ten or so? If the experimenter is limited to designs that have more runs than factors. all terms—including main effects—are If Possible. If you expect that there may be as many as five active factors. To see an example of a supersaturated design with twelve factors in eight runs: 1. consider a supersaturated design with twelve factors. as shown in . the number of model terms exceeds the number of runs. The desired number of runs is eight. type 8 in the Number of Runs edit box. It is common for brainstorming sessions to turn up dozens of factors. Yet mysteriously. However. Note that in Figure 2.7 Changing the Estimability 4. Select DOE > Custom Design.2 Examples Using the Custom Designer—Creating Screening Experiments 23 2 Examples which are the trivial many. In the last example.7. drawbacks include: • If the number of active factors approaches the number of runs in the experiment. 2. using forward stepwise regression is reasonable. The result is that the factor list is reduced without the benefit of data. screening experiments in practice rarely involve more than ten factors. then dozens of factors translates into dozens of runs.

Change the default settings of the coefficients in the Simulate Responses dialog to match those in Figure 2. then click Make Table. and X10 as active factors in the simulation. The numbers in the Y column change.8 Simulating Responses 6.8. A window named Simulate Responses and a design table appear similar to the one in Figure 2.24 2 Examples Using the Custom Designer—Creating Screening Experiments Figure 2. X9. Figure 2. Figure 2. Click Make Design.9 Simulated Responses and Design Table 7. The Y column values are controlled by the coefficients of the model in the Simulate Responses window. the analysis picks up the same three factors.10 and click Apply. . The values in your table may be different from those shown below. Because you have set X2. The values in your table may be different from those shown below. Note that random noise is added to the Y column formula. so the numbers you see might not necessarily match those in the figure.9.

and X10 as shown in Figure 2. In the resulting display click the Step button three times.11. Run the Model script in the design table by clicking the red triangle beside Model and selecting Run Script.10 Give Values to Three Main Effects and Specify the Standard Error as 0. The step history appears in the bottom part of the report.2 Examples Using the Custom Designer—Creating Screening Experiments 25 2 Examples Figure 2. Your estimates may be slightly different from those shown below. 2. you should identify three active factors: X2. X9. 3. From the report that appears. Change the Personality in the Model Specification window from Standard Least Squares to Stepwise. JMP enters the factors with the largest effects. Click Run Model. .5 To identify active factors using stepwise regression: 1.

In real situations.26 2 Examples Using the Custom Designer—Creating Screening Experiments Figure 2. After you complete the design. However. it is much less likely to have such clearly differentiated effects. After you select DOE > Custom Design and enter factors. Figure 2.12 shows that when you enter three continuous factors and one blocking factor with three runs per block. only one block appears in the Factors panel. This is because the sample size is unknown at this point.11 Stepwise Regression Identifies Active Factors Note: This example defines a few large main effects and sets the rest to zero. custom designs can have blocks of any size. JMP shows the appropriate number of blocks calculated as the sample size divided by the number of runs per block. the available block sizes for the listed designs are a power of two. instead a power of two. The blocking example shown in this section is flexible because it is using three runs per block. . For example. you see that the blocking factor shows only one level in the Values section of the Factors panel. Screening Designs with Flexible Block Sizes When you create a design using the screening designer.

13). you chose the Grid option with 24 runs.14). Figure 2. which requires three blocks with three runs each. . in the Design Generation panel. the Factors panel changes and now contains 8 blocks (Figure 2.13 Three Blocks in the Factors Panel If. This is because the default sample size is nine. there are three blocks in the Factors panel (Figure 2.12 One Block Appears in the Factors Panel The default sample size of nine requires three blocks.2 Examples Using the Custom Designer—Creating Screening Experiments 27 2 Examples Figure 2. After clicking Continue.

28 2 Examples Using the Custom Designer—Creating Screening Experiments Figure 2.14 Eight Blocks If you add all the two-factor interactions and change the number of runs to 15. three runs per block produces five blocks (as shown in Figure 2.15 Changing the Runs to 15 . Figure 2.15). so the Factors panel displays five blocks in the Values section.

2 Examples Using the Custom Designer—Creating Screening Experiments 29 2 Examples Click Make Design.16 Table of Relative Variance of the Model Coefficients The main question here is whether the relative size of the coefficient variance is acceptably small. The error variance is assumed to be 1. 63. it is also ad hoc. then click the disclosure button ( on Windows/Linux and on the Macintosh) to open the Relative Variance of Coefficients.) To take advantage of the benefits of the approach using If Possible model terms. If not. (Let q denote the potential terms designated If Possible in JMP. see “Understanding the Variance of Coefficients Table. The custom designer provides a way to improve on this ad hoc practice while supplying a theoretical foundation and an easy-to-use interface for choosing a design robust to the modeling assumptions. The values in your table may be slightly different from those shown below. Note to DOE experts: There are four rows associated with X4 (the block factor). (Let p denote the primary terms designated Necessary in JMP.16 shows a table showing the variance of each coefficient in the model relative to the unknown error variance. p < n < p+q. adding more runs (18 or more) would lower the variance of each coefficient. The formal name of the approach using If Possible model terms is Bayesian D-Optimal design. The purpose of check points in a design is to provide a detection mechanism for higher-order effects that are contained in the assumed model. Checking for Curvature Using One Extra Run In screening designs. Figure 2. Although this is good practice. That is. . Each degree of freedom is associated with one unknown coefficient in the model. For more details. Figure 2. This type of design allows the precise estimation of all of the primary terms while providing omnibus detectability (and some estimability) for the potential terms. These higher-order terms are called potential terms. 4 degrees of freedom.” p. therefore. That is because X4 has 5 blocks and. the sample size should be larger than the number of primary terms but smaller than the sum of the primary and potential terms.) The assumed model consists of the primary terms. experimenters often add center points and other check points to a design to help determine whether the assumed model is adequate. Notice that the variance of each coefficient is about one-tenth the error variance and that all the variances are roughly the same size.

30 2 Examples Using the Custom Designer—Creating Screening Experiments For a two-factor design having a model with an intercept. Define two continuous factors (X1 and X2). then click the Interactions button in the Model panel and choose 2nd. the custom designer replicates one of the four vertices. there are p = 4 primary terms. but it provides no way to check for lack of fit. Adding the two quadratic terms to the model makes a total of six terms. Select DOE > Custom Design. Highlight the two main effects. However. Figure 2. If you leave the model the same and increase the sample size.18 Second-Level Interactions . This is a way to model curvature directly. When you enter this model in the custom designer.17 Two Continuous Factors with Interaction Now suppose you can afford an extra run (n = 5). Replicating any run is the optimal choice for improving the estimates of the terms in the model. 2. 3. Click Continue. the default is a four-run design with the factor settings shown in Figure 2. Figure 2. The results appear as shown in Figure 2. which exceeds your budget of five runs. The Bayesian D-Optimal design provides a way to check for curvature while adding only one extra run. and an interaction. two main effects. To create this design: 1. to do this the custom designer requires two additional runs (at a minimum). You would like to use this point as a check point for curvature.18. 4.17.

five. Change Necessary to If Possible. Figure 2. 10. 7. and insert X1 as Y and X2 as X.20. Click the Powers button in the Model panel and choose 2nd. as shown in Figure 2.19. The resulting factor settings appear in Figure 2. 6.19 Changing the Estimability Now. . Enter 5 into the Number of Runs edit box. The desired number of runs. This adds two quadratic terms. and the interaction) are designated as Necessary while the q = 2 potential terms (the two quadratic terms) are designated as If Possible. The values in your design may be different from those shown below.20 Five-Run Bayesian D-Optimal Design 9. The overlay plot illustrates how the design incorporates the single extra run but places the factor settings at the center of the design instead of at one of the corners. Select the two quadratic terms (X1*X1 and X2*X2) and click the current estimability (Necessary) to reveal the menu.21 by selecting Graph > Overlay Plot.2 Examples Using the Custom Designer—Creating Screening Experiments 31 2 Examples 5. the p = 4 primary terms (the intercept. Click Make Design. 8. The data table appears. two main effects. is between p = 4 and p + q = 6. Figure 2. Click Make Table.Create the overlay plot in Figure 2.

In response surface experiments. the prediction variance over the range of the factors is more important than the variance of the coefficients.22). researchers use response surface experiments when they already know which factors are important.21 Overlay Plot Creating Response Surface Experiments Response surface experiments traditionally involve a small number (generally 2 to 8) of continuous factors. In screening experiments one measure of the quality of the design is the size of the relative variance of the coefficients. 3. Select DOE > Custom Design. One way to visualize the prediction variance is JMP’s prediction variance profile plot. Follow the steps below to create a design for a quadratic model with a single continuous factor. Add one continuous factor by selecting Add Factor > Continuous (Figure 2.32 2 Examples Using the Custom Designer—Creating Response Surface Experiments Figure 2. In contrast to screening experiments. 2. In the Model panel. The main goal of response surface experiments is to create a predictive model of the relationship between the factors and the response. . select Powers > 2nd to create a quadratic term (Figure 2. Exploring the Prediction Variance Surface The purpose of the example below is to generate a simple Prediction Variance Profile Plot.22) and click Continue. This plot is a powerful diagnostic tool for evaluating and comparing response surface designs. 1. The a priori model for a response surface experiment is usually quadratic. Using this predictive model allows the experimenter to find better operating settings for the process.

If the design model is quadratic. The y-axis is the relative variance of prediction of the expected value of the response. In this design. Compare profile plots by clicking the Back button and choosing Minimum in the Design Generation panel. Click Make Design and then open the Prediction Variance Profile again. on the Macintosh) to open the For continuous factors.22 Adding a Factor and a Quadratic Term 4. which gives a sample size of three. 7.2 Examples Using the Custom Designer—Creating Response Surface Experiments 33 2 Examples Figure 2. As the number of runs increases. 63. The number of runs is inversely proportional to the size of variance of the predicted response. and 1. Figure 2. the three design points are –1.24. the initial setting is the first level. When the relative prediction variance is one. 6. Make sure to use the default number of six runs (Figure 2. . 0.23 Adjusting the Number of Runs 5. then the prediction variance function is quartic. the prediction variances decrease.23) and click Make Design. The prediction variance profile shows that the variance is a maximum at each of these points on the interval –1 to 1. the initial setting is at the mid-range of the factor values. Click the disclosure button ( on Windows/Linux and Prediction Variance Profile.” p. Figure 2. the absolute variance is equal to the error variance of the regression model.24 Prediction Profile for Single Factor Quadratic Model The prediction variance is relative to the error variance. as shown in Figure 2. More details on the Prediction Variance Profiler is in “Understanding the Design and Profilers. For categorical factors.

and clicking Make Design. The results are shown in Figure 2. the plot is symmetric. as shown in Figure 2.26. the prediction plot is not symmetric. Create an unbalanced design by clicking the Back button.5 rather than one). so doubling the number of runs halves the prediction variance. the prediction variance is inversely proportional to the sample size. You can see the prediction variance increase as the sample size decreases.5.25 compares plots for a sample size of six and sample size of three for this quadratic model. entering a sample size of four in the Design Generation panel.25 Comparison of Prediction Variance Profiles Tip: Control-click (Command-click on the Mac) on the factor to set a factor level precisely. for sample sizes six (top charts) and sample size three (bottom charts). In fact.26. When the design is balanced. When the design is unbalanced. Figure 2. The symmetry of the plot is related to the balance of the factor settings. These profiles show settings for the maximum variance and minimum variance.25. Figure 2. but the maxima are at one instead of 0. as shown in Figure 2. 8.34 2 Examples Using the Custom Designer—Creating Response Surface Experiments This produces a curve that has the same shape as the previous plot. You can see that the variance of prediction at –1 is lower than the other sample points (its value is 0. .

Then select Edit > Run Script from the Edit menu.445) to see that the maximum variance on the interval decreases by more than 10%. . move the free values from the equally spaced points to the optimal points (-0.2 Examples Using the Custom Designer—Creating Response Surface Experiments 35 2 Examples Figure 2.jsl from the Sample Scripts folder that was installed when you installed JMP.26 Sample Size of Four for the One-Factor Quadratic Model Prediction variance at X1= 0 Prediction variance at X1= -1 Note to scripting experts: The following JSL code shows graphically that a one-factor design for a cubic model with unequally spaced points can have a lower prediction variance than the equally spaced design. When the plot appears. View the script in this example by opening Cubic Model.445 and 0.

text({-0. detV = det(V). n = 4. This makes I-Optimal designs most appropriate for prediction."Determinant = ". yFunction(xi=cubicx(x). // Drag the middle points to -0. V = M`*M.sqrt(xi`*C*xi).3. DragMarker(u. M = M||(x^i)).1}.36 2 Examples Using the Custom Designer—Creating Response Surface Experiments // DOE for fitting a cubic model.1.i<=2.1.99))).2) .0).y). .i<=3.u[2].2).1).445 and 0.445 // for a D-Optimal design. D-Optimal designs are most appropriate for screening experiments because the optimality criterion focuses on precise estimates of the coefficients.).1}. However.25}.99)).6. If an experimenter has precise estimates of the factor effects. NewWindow("DOE . Graph(FrameSize(500.i++.for(i=1.1).1. I-Optimal designs minimize the average prediction variance inside the region of the factors. cubicx = function({x1}.)).333 0. Introducing I-Optimal Designs for Response Surface Modeling The custom designer generates designs using a mathematical optimality criterion. M = j(n.Text({u[i]. D-Optimal designs are not as appropriate for designing experiments where the primary goal is prediction..6.rr[i+1]=x1^i).1. for(i=1.1. rr=j(4.u).i++. then it is easy to tell which factors’ effects are important and which are negligible. // number of points //Start with equally spaced points.333]. As a result I-Optimality is the recommended criterion in JMP for response surface designs.1.yScale(0.x). y = j(2.i++. Double Buffer. rr.0. x = {-1. for(i=1.u[1].char(u[i]. C = inverse(V).XScale(-1.i<=3..d.300).Variance Function of a Cubic Polynomial". show(n.char(detV. All the designs in this chapter so far have been D-Optimal designs. u = [-0.

2 Examples Using the Custom Designer—Creating Response Surface Experiments

37 2 Examples

An I-Optimal design tends to place fewer runs at the extremes of the design space than does a D-Optimal design. As a small example, consider a one-factor design for a quadratic model using n = 12 experimental runs. The D-Optimal design for this model puts four runs at each end of the range of interest and four runs in the middle. The I-Optimal design puts three runs at each end point and six runs in the middle. In this case, the D-Optimal design places two-thirds of its mass at the extremes versus one-half for the I-Optimal design. Figure 2.27 compares prediction variance profiles of the one-factor I- and D-Optimal designs for a quadratic model with (n = 12) runs. The variance function for the I-Optimal design is less than the corresponding function for the D-Optimal design in the center of the design space; the converse is true at the edges. Figure 2.27 Prediction Variance Profiles for 12-Run I-Optimal (left) and D-Optimal (right) Designs

The average variance (relative to the error variance) for the I-Optimal design is 0.178 compared to the D-Optimal design, which is 0.2. This means that confidence intervals for prediction will be nearly 10% shorter on average for the I-Optimal design. To compare the two design criteria, create a one-factor design with a quadratic model that uses the I-Optimality criterion, and another one that uses D-Optimality: 1. 2. 3. 4. 5. 6. 7. Select DOE > Custom Design. Add one continuous factor: X1. Click Continue. Click the RSM button in the Model panel to make the design I-Optimal. Change the number of runs to 12. Click Make Design. Click the disclosure button ( on Windows/Linux and on the Macintosh) to open the Prediction Variance Profile, as shown on the left in Figure 2.27. 8. Repeat the same steps to create a D-Optimal design, but select Optimality Criterion > Make D-Optimal Design from the custom design title bar’s red triangle menu. The results in the Prediction Variance Profile should look the same as those on the right in Figure 2.27.

**A Three-Factor Response Surface Design
**

In higher dimensions, the I-Optimal design continues to place more emphasis on the center of the region of the factors. The D-Optimal and I-Optimal designs for fitting a full quadratic model in three factors using 16 runs are shown in Figure 2.28.

38

2 Examples Using the Custom Designer—Creating Response Surface Experiments

To compare the two design criteria, create a three-factor design that uses the I-Optimality criterion, and another one that uses D-Optimality: 1. 2. 3. 4. Select DOE > Custom Design. Add three continuous factors: X1, X2, and X3. Click Continue. Click the RSM button in the Model panel to add interaction and quadratic terms to the model and to change the default optimality criterion to I-Optimal. 5. Use the default of 16 runs. 6. Click Make Design. 7. The design is shown in the Design panel (the left in Figure 2.28). 8. If you want to create a D-Optimal design for comparison, repeat the same steps but select Optimality Criterion > Make D-Optimal Design from the red triangle menu on the custom design title bar. The design should look similar to those on the right in Figure 2.28. The values in your design may be different from those shown below. Figure 2.28 16-run I-Optimal and D-Optimal designs for RSM Model

Profile plots of the variance function are displayed in Figure 2.29. These plots show slices of the variance function as a function of each factor, with all other factors fixed at zero. The I-Optimal design has the lowest prediction variance at the center. Note that there are two center points in this design. The D-Optimal design has no center points and its prediction variance at the center of the factor space is almost three times the variance of the I-Optimal design. The variance at the vertices of the D-Optimal design is not shown. However, it should be noted that the D-Optimal design predicts better than the I-Optimal design near the vertices.

2 Examples Using the Custom Designer—Creating Response Surface Experiments

39 2 Examples

Figure 2.29 Variance Profile Plots for 16 run I-Optimal and D-Optimal RSM Designs

**Response Surface with a Blocking Factor
**

It is not unusual for a process to depend on both qualitative and quantitative factors. For example, in the chemical industry, the yield of a process might depend not only on the quantitative factors temperature and pressure, but also on such qualitative factors as the batch of raw material and the type of reactor. Likewise, an antibiotic might be given orally or by injection, a qualitative factor with two levels. The composition and dosage of the antibiotic could be quantitative factors (Atkinson and Donev, 1992). The response surface designer (described in “Response Surface Designs,” p. 101) only deals with quantitative factors. You could use the response surface designer to produce a Response Surface Model (RSM) design with a qualitative factor by replicating the design over each level of the factor. But, this is unnecessarily time-consuming and expensive. Using custom designer is simpler and more cost-effective because fewer runs are required. 1. First, define two continuous factors (X1 and X2). 2. Now, click Add Factor and select Blocking > 4 runs per block to create a blocking factor with four runs per block (X3). The blocking factor appears with two levels, as shown below, but the number of levels adjusts later to accommodate the number of runs specified for the design.

40

2 Examples Using the Custom Designer—Creating Response Surface Experiments

3. Click Continue, and then click the RSM button in the Model panel to add the quadratic terms to the model. This automatically changes the recommended optimality criterion from D-Optimal to I-Optimal. Note that when you click RSM, a message reminds you that nominal factors (the blocking factor) are not used in quadratic terms.

4. Enter 12 in the Number of Runs text edit box in the Design Generation panel, and note that the Factors panel now shows the Blocking factor, X3, with three levels. Twelve runs defines three blocks with four runs per block.

5. Click Make Design. 6. In the Output Options, select Sort Right to Left from the Run Order list 7. Click Make Table to see an I-Optimal table similar to the one on the left in Figure 2.30. Figure 2.30 compares the results of a 12-run I-Optimal design and a 12-run D-Optimal Design. To see the D-Optimal design: 1. Click the Back button. 2. Click the red triangle icon on the Custom Design title bar and select Optimality Criterion > Make D-Optimal Design.

2 Examples Using the Custom Designer—Creating Response Surface Experiments

41 2 Examples

3. Click Make Design, then click Make Table. Figure 2.30 JMP Design Tables for 12-Run I-Optimal and D-Optimal Designs

Figure 2.31 gives a graphical view of the designs generated by this example. These plots were generated for the runs in each JMP table by choosing Graph > Overlay from the main menu and using the blocking factor (X3) as the Grouping variable. Note that there is a center point in each block of the I-Optimal design. The D-Optimal design has only one center point. Instead, three of the vertices are repeated (–1, –1 in blocks 2 and 3, –1, 1 in blocks 1 and 3, and 1, 1 in blocks 1 and 2). The values in your graph may be different from those shown below. Figure 2.31 Plots of I-Optimal (left) and D-Optimal (right) Design Points by Block.

Either of the above designs supports fitting the specified model. The D-Optimal design does a slightly better job of estimating the model coefficients. The D-efficiency of the I-Optimal design is 92.4%. The I-Optimal design is preferable for predicting the response inside the design region. Using the formulas given in “Technical Discussion,” p. 52, you can compute the relative average variance for these designs. The average variance (relative to the error variance) for the I-Optimal design is 0.61 compared

In this example: • The response is the electromagnetic damping of an acrylonitrile powder. the researchers were only interested in predictions at three discrete wavelengths. The authors do not mention how much damping is desirable. the optimal formulation could change depending on the operating environment. As a result. The mixture designer is limited. Select Sample Data > Design Experiment > Donev Mixture factors.32 Mixture Experiment Response Panel and Factors Panel .86 for the D-Optimal design. 2. • The three mixture ingredients are copper sulphate. This example from Atkinson and Donev (1992) shows how to create designs for experiments with mixtures where one or more factors are not ingredients in the mixture. You are not forced to modify your problem to conform to the restrictions of a special-purpose design approach. so right-click the goal and create Damping’s response goal to be None. Create Damping as the response. and glyoxal. add the three mixture ingredients and the categorical factor. This means confidence intervals for prediction will be more than 16% longer on average for D-Optimal designs. Though wavelength is a continuous variable. In the Factors panel. they treated it as a categorical factor with three levels. you can choose between using the custom designer or the specialized mixture designer. • The nonmixture environmental factor of interest is the wavelength of light. Rather than entering them by hand.jmp. 3. though.42 2 Examples Using the Custom Designer—Creating Mixture Experiments to 0. The custom designer can handle mixture ingredients and process variables in the same study. Figure 2. load them from the Sample Data folder that was installed when you installed JMP: click the red triangle icon on the Custom Design title bar and select Load Factors. Wavelength. The mixture ingredients have range constraints that arise from the mechanism of the chemical reaction. After all. To create this custom design: 1. Select DOE > Custom Design. because it requires all factors to be mixture components. It seems natural to vary the process settings along with the percentages of the mixture ingredients. sodium thiosulphate. Creating Mixture Experiments If you have factors that are ingredients in a mixture.

. Click OK. Figure 2. shown in Figure 2. 2. and then click Make Table.33). Click Make Design.33 Mixture Experiment Design Generation Panel 3. type 18 in the Number of Runs text edit box (Figure 2. A warning dialog appears telling you that JMP removes the main effect terms for non-mixture factors that interact with all the mixture factors. The values in your table may be different from those shown below. and choose 2nd.33 is a response surface model in the mixture ingredients along with the additive effect of the wavelength. Click Interactions. In the Design Generation panel. The resulting data table is shown in Figure 2.34. To create this model: 1. which results in six runs each for the three levels of the wavelength factor.2 Examples Using the Custom Designer—Creating Mixture Experiments 43 2 Examples The model.

To view this: Click the Back button. 2. Figure 2. Remove all the effects by highlighting them all and clicking Remove Terms. Note that there are necessarily unequal numbers of runs for each wavelength.6. The Design table to the right in Figure 2. The other two ingredients both drop.36). keeping their ratio constant. In that case.4 to 0. Because of this lack of balance it is a good idea to look at the prediction variance plot (top plot in Figure 2. In the Design Generation panel.35) and click Make Design.35 Ten-Run Mixture Response Surface Design 1. it is not possible to run a complete mixture response surface design for every wavelength. The values of the first three ingredients sum to one because they are mixture ingredients. which is a good indication that the lack of balance is not a problem.34 Mixture Experiment Design Table Atkinson and Donev also discuss the design where the number of runs is limited to 10. change the number of runs to 10 (Figure 2. the others adjust to keep the sum constant. . The bottom profiler in Figure 2. If you vary one of the values.44 2 Examples Using the Custom Designer—Creating Mixture Experiments Figure 2.35 shows the factor settings for 10 runs. The prediction variance is almost constant across the three wavelengths. Add the main effect by clicking the Main Effects button.36 shows the result of increasing the copper sulphate percentage from about 0. 3. 4.

To follow along with this example that shows column properties. . Build the custom design as follows: 1. Select DOE > Custom Design. the covariate should be a design factor. The weight of the student is known and it seems important to include this information in the experimental design. 2. The Covariate selection displays a list of the variables in the current data table. as shown in Figure 2. the designs presented in this section cannot be created without using the custom designer. The custom designer supplies a reasonable design option. open Big Class. For this example.jmp from the Sample Data folder that was installed when you installed JMP. Designing Experiments with Fixed Covariate Factors Pre-tabulated designs rely on the assumption that the experimenter controls all the factors.36 Prediction Variance Plots for Ten-Run Design Special-Purpose Uses of the Custom Designer While some of the designs discussed in previous sections can be created using other designers in JMP or by looking them up in a textbook containing tables of designs. Click Add Factor and select Covariate. naming them calories and sleep. 3. Sometimes you have quantitative measurements (a covariate) on the experimental units before the experiment begins. The pre-defined design that allows only a few discrete values is too restrictive. A physical education researcher has proposed an experiment where you vary the number of hours of sleep and the calories for breakfast and ask each student run 1/4 mile. suppose there are a group of students participating in a study. Add two continuous variables to the models by clicking Add Factors and selecting Continuous.37.2 Examples Using the Custom Designer—Special-Purpose Uses of the Custom Designer 45 2 Examples Figure 2. If this variable affects the experimental response.

and then clicking the Cross button (Figure 2.38 Design with Fixed Covariate 5. Click Make Design.46 2 Examples Using the Custom Designer—Special-Purpose Uses of the Custom Designer Figure 2. Click Continue. Figure 2. Add the interaction to the model by selecting calories in the Factors panel.40 shows the design table. Figure 2.39). .39 Design With Fixed Covariate Factor 7. The data table in Figure 2. 6.37 Add a Covariate Factor 4. selecting sleep in the Model panel. Your runs might not look the same because the order of the runs has been randomized. then click Make Table. Select weight from the variable list and click OK.

The leverage plots are nearly horizontal. that means the custom designer has successfully separated the effect of weight from the effects of calories and sleep. Click Run Model. We can see how well the custom designer did by fitting a model of weight as a function of calories and sleep. but it is not controlled. measured for each student. Figure 2.. and the analysis of variance table shows that the model sum of squares is near zero compared to the residuals (Figure 2. If that fit has a small model sum of squares.41. . The custom designer has calculated settings for calories and sleep for each student.41 Model Script 9. It would be desirable if the correlations between calories. Therefore.2 Examples Using the Custom Designer—Special-Purpose Uses of the Custom Designer 47 2 Examples Figure 2. as shown to the right in Figure 2.40 Design Table Note: Covariate factors cannot have missing values. as shown on the left in Figure 2. The values in your analysis may be a little different from those shown below. Remember that weight is the covariate factor. 8. Click the red triangle icon beside Model and select Run Script. sleep. weight is independent of calories and sleep. Arrange the dialog so weight is Y and calories. and calories*sleep are the model effects.42).41. sleep and weight were as small as possible.

Additionally. As is expected of mixture experiments. . m2. In split plot experiments. The experiment studies the effect of five factors on the thickness of vinyl used to make automobile seat covers. The example in this section is from Kowalski.43). but is now commonplace in manufacturing and engineering studies. They are plasticizers whose proportions are represented by m1. Use the default goal. They are hard to change. there is one response. and m3. Select DOE > Custom Design. To create this design in JMP: 1. By default. The whole plot is divided into subplots. Y. only in the last few years has it been possible to create optimal split plot designs (Goos 2002). Double-click the name and change it to thickness. and Vining (2002). hard-to-change factors change only between one whole plot and the next. Cornell. the component proportions always sum to one. Maximize (Figure 2. The process variables are the whole plot factors of the experiment.Enter the lower limit of 10. They are the rate of extrusion and the temperature of drying.42 Analysis to Check That Weight is Independent of Calories and Sleep Creating a Design with Two Hard-to-Change Factors: Split Plot While there is substantial research literature covering the analysis of split plot designs. The split plot design capability accessible in the custom designer is the first commercially available tool for generating optimal split plot designs. The split plot design originated in agriculture. The response and factors in the experiment are described below: • Three of the factors are ingredients in a mixture. the mixture components are the subplot factors of the experiment. and the levels of the easy-to-change factors are randomly assigned to each subplot. showing. • The response in the experiment is the thickness of the vinyl used for automobile seat covers. • Two of the factors are process variables. 2.48 2 Examples Using the Custom Designer—Special-Purpose Uses of the Custom Designer Figure 2. The response of interest (thickness) depends both on the proportions of the mixtures and on the effects of the process variables.

Click OK. and click Add Factor > Continuous. then JMP makes it impossible to designate the model terms If Possible. Next. Figure 2. click Easy and select Hard. Ensure that you are using the default levels.43. Name the three mixture factors m1. –1 and 1. To make temperature the subplot factor. Use the default levels 0 and 1 for those three factors. Add two continuous factors by typing 2 in the box beside Add N Factors. click the box next to Number of Whole Plots and type 7. JMP makes it impossible to make factor changes be Hard. 4. This causes a warning that JMP removes the main effect terms for non-mixture factors that interact with all the mixture factors. . and m3. 6. in the Values area corresponding to these two factors. To add three mixture factors. Your dialog should look like the one in Figure 2.2 Examples Using the Custom Designer—Special-Purpose Uses of the Custom Designer 49 2 Examples 3. 9. Name these factors extrusion rate and temperature. type 3 in the box beside Add N Factors. 10. add interaction terms to the model by selecting Interactions > 2nd (Figure 2. click Easy and select Hard. m2.43 Entering Responses and Factors 8.In the Design Generation panel. Click Continue. Figure 2.44). Conversely. and click Add Factor > Mixture. 5. 7. if you have factors that have their estimability set to Hard. To make extrusion rate the whole plot factor.44 Adding Interaction Terms Note that if you have model terms that have their estimability set to If Possible.

Figure 2.45).47 The Vinyl Data Design Table 15. 14. .Click Make Table.The result is shown in Figure 2.From the Sample Data folder that was installed when you installed JMP.jmp.45 Assigning the Number of Whole Plots and Number of Runs 12.Click Make Design.50 2 Examples Using the Custom Designer—Special-Purpose Uses of the Custom Designer 11. Figure 2. The dialog in Figure 2. select Design Experiment > vinyl data.Click the red triangle icon next to the Model script and select Run Script.46.Give the experiment 28 runs (Figure 2.46 Final Design Structure 13. Figure 2.48 appears. The values in your table may be different from those shown below.

. JMP designates the error term by appending &Random to the name of the effect. as indicated in the menu beside Method in Figure 2.48 Creating the Model 16. For more information about REML models.48. Figure 2. REML will be used for the analysis.49. see the JMP Statistics and Graphics Guide. As shown in the dialog in Figure 2.Click Run Model to run the analysis. The results are shown in Figure 2.2 Examples Using the Custom Designer—Special-Purpose Uses of the Custom Designer 51 2 Examples The models for split plots have a random effect associated with the whole plot effect.48.

• Has runs whose purpose is to lower the variability of the coefficients of this pre-stated model. and Bayesian D-Optimal designs in JMP. By focusing on minimizing the standard errors of coefficients. a D-Optimal design may not allow for checking that the model is correct. This is appropriate for first-order models and in screening situations. a D-Optimal design may have just p distinct runs with no degrees of freedom for lack of fit. parameter estimation is key.49 Analysis Results Technical Discussion This section provides more technical information about I-. This is a limitation because in most real situations. because the experimental goal in such situations is often to identify the active factors. D-. D-Optimality: • Is the default design type produced by the custom designer (except when the model is a response surface design) • Minimizes the variance of the model coefficient estimates. .52 2 Examples Using the Custom Designer—Technical Discussion Figure 2. In the extreme. It may not include center points when investigating a first-order model. the form of the pre-stated model is not known in advance. • Is dependent on a pre-stated model.

2 Examples Using the Custom Designer—Technical Discussion 53 2 Examples • Maximizes D when D = det[XTX]. to develop feedback-control models. such as in response surface design problems. That is. p + q > n > p. • Is more appropriate than D-Optimality in some situations. The Bayesian D-Optimal design is an approach that allows the precise estimation of all of the primary terms while providing omnibus detectability (and some estimability) for the potential terms. Bayesian D-Optimality: • Is a modification of the D-Optimality criterion that effectively estimates the coefficients in a model. • Uses the potential terms to force in runs that allow for detecting any inadequacy in the model containing only the primary terms. and also is appropriate to determine regions in the design space where the response falls within an acceptable range. Precise estimation of the response therefore takes precedence over precise estimation of the parameters. and at the same time has the ability to detect and estimate some higher-order terms. then I = ∫ f T (x)( XTX) R –1 f ( x ) dx M] = Trace [ ( X T X ) where –1 M = ∫ f ( x ) f ( x )T dx R is a moment matrix that is independent of the design and can be computed in advance. If there are interactions or curvature. • D-optimal split plot designs maximize D when D = det[XTV -1X] where V -1is the block diagonal variance matrix of the responses (Goos 2002). I-Optimality: • Minimizes the average variance of prediction over the region of the data. It is more appropriate when your objective is to determine optimum operating conditions. The Bayesian D-Optimal design maximizes the determinant of (XT X + K). k. The difference between the criterion for D-Optimality and Bayesian D-Optimality is this constant added to the diagonal elements corresponding to the potential terms in the XT X matrix. . or. Using an I-Optimality criterion is more appropriate because you can predict the response anywhere inside the region of data and therefore find the factor settings that produce the most desirable response value. • Maximizes this criterion: If f T(x) denotes a row of the X matrix corresponding to factor combinations x. the Bayesian D-Optimality criterion is advantageous. Let K be the (p + q) by (p + q) diagonal matrix whose first p diagonal elements are equal to 0 and whose last q diagonal elements are the constant. If your goal is to predict the response rather than the coefficients. • Works best when the sample size is larger than the number of primary terms but smaller than the sum of the primary and potential terms.

54 2 Examples Using the Custom Designer—Technical Discussion .

and good for routine factor screening or response optimization. select DOE > Custom Design. . To create these tailor-made designs. This chapter introduces you to the steps you need to complete to build a custom design. flexible.3 Building Custom Designs The Basic Steps 3 Custom JMP can build a custom design that both matches the description of your engineering problem and remains within your budget for time and material. They are general.

. . . . . . . . . . . . . . . . . . . . . . . . . .Optimality) . . . . . . . . . . . . . . . . . . . Describing the Model . . . . . . . . . . . . . . . . . . . . . . . . . . . Defining Low and High Values (DOE Coding) for Columns . . . . . . . . . Constraining a Design to a Hypersphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Viewing Diagnostics Reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Making the Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Contents Creating a Custom Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Selecting the Number of Runs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Defining Response Column Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Understanding the Variance of Coefficients Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating Split Plot Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Understanding the Design and Profilers . . . . . Saving the Design (X) Matrix . . . Accounting for Factor Level Restrictions . . . . . Setting the Random Number Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Loading and Saving Constraints . . . . . . . . . . . . 57 57 61 62 63 63 64 64 65 65 66 67 67 68 69 69 70 70 71 72 73 74 75 76 76 77 78 79 80 . . . . . . . . . . . . Special Custom Design Commands . . . . . . . . . . . Selecting the Number of Random Starts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Advanced Options . . . . . . . . . . . . . . . . . . . . . . . . Giving Columns a Design Role . . . . . . . . . . . . . . . . . . . . Entering Responses and Factors in the Custom Designer . Loading Responses and Factors . . . . . . . . . . . . . . . . . Assigning Column Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Saving Responses and Factors . . Identifying Factor Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Understanding the Alias Matrix (Confounding Pattern) . . . . . . . . . . Setting Columns as Factors for Mixture Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Specifying Output Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . How Custom Designs Work: Behind the Scenes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Changing the Design Criterion (D. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .or I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

63 • “Specifying Output Options. if desired. select DOE > Custom Design.” p. Match Target. if desired. 64 • “Making the Table.1.” p. 62 • “Understanding the Design and Profilers. 61 • “Selecting the Number of Runs. follow the steps in Figure 3. • “Entering Responses and Factors in the Custom Designer. 57 • “Describing the Model. or None.1 Entering Responses Click to enter lower and upper limits and importance weights.” p. you can tell JMP that your goal is to obtain the maximum or minimum value possible.” p. Tip: To quickly enter multiple responses. 3 Click to change the response goal. follow the steps below. or that there is no goal. Then. Specifying Response Goal Types and Lower and Upper Limits When entering responses. Minimize.” p.” p. 65 Entering Responses and Factors in the Custom Designer How to Enter Responses To enter responses. Figure 3.3 Building Custom Designs—Creating a Custom Design 57 3 Custom Creating a Custom Design To start a custom design. . 4 To enter one 1 response at a time. or click the Custom Design button on the JMP Starter DOE page. click the N Responses button and enter the number of responses you want. click then select a goal type: Maximize. 2 Double-click to edit the response name. to match a specific value.

Understanding Response Importance Weights When computing overall desirability. Create the design. A goal of Maximize supports this objective. would be y = 28 + 4 X 1 + 5 X 2 + random noise . then specifying importance is unnecessary. Before clicking the Make Table button.1): • For responses such as strength or yield. 4. you might want to add simulated responses that help you determine whether your design can detect effects of a size that you expect: 1. JMP uses the value you enter as the importance weight (step 4 in Figure 3. With two responses you can give greater weight to one response by assigning it a higher importance value. An example of such an equation. • The Match Target goal supports the objective when the best value for a responses is a specific target value. 3. If there is only one response. and the Y column contains values for simulated responses. enter values you want to apply to the Y column in the data table and click Apply. The default target value is assumed to be midway between the lower and upper limits. Click Make Table to create the design table. such as with part dimensions. where the random noise is distributed with mean zero and standard deviation one. Adding Simulated Responses. click the red triangle icon in the title bar and select Simulate Response. such as when the response is impurity or defects. For custom and augment designs.2) appears along with the design data table.2. 2. a window (Figure 3. If Desired If you do not have specific responses to add. In this window.1) as the weight of each response. you can alter the default target after you make a table from the design. In the data table. and enter an asymmetric target value. . as shown in Figure 3. Note: If your target range is not symmetrical. the best value is usually the largest possible. open the response’s Column Info dialog by double-clicking the column name.58 3 Building Custom Designs—Creating a Custom Design The following description explains the relationship between the goal type (step 3 in Figure 3. • The Minimize goal supports an objective of having the best objective be the smallest value. The numbers you enter represent the coefficients in an equation.1) and the lower and upper limits (step 4 in Figure 3.

Mixture. . type the number of factors in the Add N Factors box. click the Add Factor button and choose the type of factor.3 Entering Factors in a Custom Design To add one factor. Click and select Add Level to 2 increase the number of levels. 5 Click to enter or change factor values. Categorical. Specify Values for Simulated Responses How to Enter Factors Figure 3. Changing to Hard will cause the resulting design to be a split plot design. press the delete key on the keyboard. Blocking. To remove a level.2 In Custom and Augment Designs. or Constant (see description below). 4 Click to indicate that changing a factor’s setting from run to run is Easy or Hard. click Add Factor and select a factor type: Continuous.3 Building Custom Designs—Creating a Custom Design 59 3 Custom Figure 3. 1 To add multiple factors. Covariate. click the Add Factor button. and select the factor type. click it. 3 Double-click to edit the factor name. then press the Return or Enter key on the keyboard. Types of Factors When adding factors.

If the values are numbers. commercial software has not supplied a general capability for the design and analysis of these experiments. If Necessary Sometimes it is impossible to vary factors simultaneously over their entire experimental range. . you have factors whose levels are constrained. The settings of a categorical factor are discrete and have no intrinsic order. • Uncontrolled Uncontrolled factors are factors whose values cannot be controlled during an experiment. In industrial experimentation this structure is desirable because certain factors may be difficult and expensive to change from one run to the next. but they are factors you wish to include in the model. Examples of categorical factors are machine. It is convenient to make several runs while keeping such factors constant. the study cannot simultaneously set high power and long time without burning all the kernels. • Covariate Covariate factors are not controllable. • Categorical Categorical data types have no implied order.60 3 Building Custom Designs—Creating a Custom Design Below is a description of factor types from which you can choose. To indicate that a factor’s settings are difficult to change. the order is the numeric magnitude. • Constant Constant factors are factors whose values are fixed during an experiment. and gender. the order is the sorting sequence. if you are studying the affect of cooking time and microwave power level on the number of kernels popped in a microwave popcorn bag. but their values are known in advance of an experiment. In theory. Changing to Hard will cause the resulting design to be a split plot design. for more details. click Easy or Hard in the Factors Panel Changes column and select from the menu that appears. • Mixture Creating Optimal Split Plot Designs Split plot experiments are performed in groups of runs where one or more factors are held constant within a group but vary between groups. For example. Mixture factors are continuous factors that are ingredients in a mixture. • Continuous Continuous factors are numeric data types only. Until now. See “Creating Split Plot Designs.” p. operator. Defining Factor Constraints. • Blocking Blocking factors are a special kind of categorical factor. 65. Factor settings for a run are the proportion of that factor in a mixture and vary between zero and one. Blocking factors differ from other categorical factors in that there is a limit to the number of runs that you can perform within one level of a blocking factor. you can set a continuous factor to any value between the lower and upper limits you supply. Therefore. If the values are character.

click the Add Constraint button again and repeat the above steps. X1*X2 is added to the list of model terms. Note that this feature is disabled if you have already controlled the design region by entering disallowed combinations or chosen a sphere radius. as shown in Figure 3.4. Click the Add Constraint button. Click the <= button to change it to >=. you can add factor interactions or powers of continuous factors to the model.5. Specify the coefficients and their limiting value in the boxes provided.5 Add Terms To add specific factor types to the model: Action Instructions Add interaction terms involving the selected factors. For example. The appropriate terms are added. 4.3 Building Custom Designs—Creating a Custom Design 61 3 Custom To define the constraints: 1. if you do not want to limit your model to main effects. or 5th. However. 2. as shown in Figure 3.4 Example of a Linear Inequality Constraint Describing the Model Initially. the Model panel lists only the main effects corresponding to the factors you entered. Click the RSM button. JMP adds all of the interactions to the specified order. click the RSM button. The design now uses I-Optimality criterion rather than D-Optimality criterion. click the blue disclosure button ( on Windows/ Linux and on the Macintosh) to open the Define Factor Constraints panel. If none are selected. . Figure 3. After adding factors and clicking Continue. 3rd. To add all the two-factor interactions and quadratic effects at once. including two-factor interactions and quadratic effects Click the Interactions button and select 2nd. Add all second-order effects. 3. if you your factors are X1 and X2 and you click Interactions > 2nd. 4th. Figure 3. To add another constraint.

6 Options for Selecting the Number of Runs The Design Generation panel has the following options for selecting the number of runs you want: • Minimum is the smallest number of terms that can create a design. 3rd. in most cases. Its value is generally between Default and Grid. Exceptions are for mixture and blocking designs.62 3 Building Custom Designs—Creating a Custom Design Action Instructions Add selected cross product terms 1. When you use Minimum. 4th. and is appropriate only when the cost of extra runs is prohibitive. This is an extreme choice that can be risky. When the Design Generation panel looks the way you want it. • Default is a custom design suggestion for the number of runs. This value is based on heuristics for creating balanced designs with a few additional runs above the minimum. • User Specified is a value that specifies the number of runs you want. Generally Grid is unnecessarily large and is only included as an option for reference and comparison. • Grid. . • Compromise is a second suggestion that is more conservative than the Default. Add powers of continuous factors to the model effects Selecting the Number of Runs The Design Generation panel (Figure 3. the resulting design is saturated (no degrees of freedom for error). Highlight term(s) in the model list. 2.6) shows the minimum number of runs needed to perform the experiment based on the effects you’ve added to the model. click Make Design. Balancing the cost of each run with the information gained by extra runs you add is a judgment call that you control. shows the number of points in a full-factorial design. Figure 3. 3. Note: Choosing Grid will not necessarily produce a full factorial design. or 5th. Click the Powers button and select 2nd. Click the Cross button. It also shows alternate numbers of runs and lets you choose your own number of runs. Highlight the factor names. Enter that value into the Number of Runs text box.

is one way to compare two designs.3 Building Custom Designs—Creating a Custom Design 63 3 Custom Understanding the Design and Profilers Before creating the design table. but the surface plot appears only if there are two or more variables. if the error variance is five and the reported prediction variance at a particular factor setting is 0. For example.7) and surface plot. Understanding the Variance of Coefficients Table Before clicking Make Table in the custom designer. drag the vertical lines in the plot to change the factor settings to different points. . Comparing the prediction variance profilers for two designs side-by-side. then the prediction variance at that point is two. In the profiler. Figure 3. the prediction variance is reported relative to whatever the error variance turns out to be.7 A Factor Design Layout The prediction variance for any factor setting is a function of the design and the unknown error variance. the error variance will drop as the sample size increases.4.8 Prediction Profile for Single Factor Quadratic Model It is ideal for the prediction variance to be small throughout the allowable regions of the factors. A design that has lower prediction variance on the average is preferred. preview the design and investigate details using the prediction profiler (Figure 3. Figure 3. Dragging the lines reveals any points that have prediction variances that are larger than you would like. Generally. The error variance is assumed to be one. click the disclosure button ( on Windows/ Linux and on the Macintosh) to open the Relative Variance of Coefficients table. That is. JMP always provides the prediction profiler.

64 3 Building Custom Designs—Creating a Custom Design The Relative Variance of Coefficients table (Figure 3. The variances are relative to the error variance. It allows you to see the confounding pattern in D-Optimal designs. Once you have an estimate for the error variance. Figure 3. on Windows/ The alias matrix only appears for custom designs with two factor interactions if all factors are two-levels. The square root of this value should match the standard error of prediction for the coefficient when you fit a model using Analyze > Fit Model.9 Table of Relative Variance of Coefficients Understanding the Alias Matrix (Confounding Pattern) Before clicking Make Table in the custom designer.10 Alias Matrix Specifying Output Options Use the Output Options panel to specify how you want the output data table to appear: . The table shows the aliasing (if any) between the model terms and all the two-factor interactions. Figure 3. you can multiply that by the relative variance to get the estimated variance of the coefficient.9) shows the relative variance of all the coefficients for a custom design. click the disclosure button ( Linux and on the Macintosh) to open the alias matrix. which is unknown before the experiment. The error variance is assumed to be one.

Figure 3. Click the red triangle icon and select Run Script to open the Fit Model dialog. Type the number of times you want to replicate the design in the associated text box. • Sort Right to Left—the rows (runs) in the output table will appear sorted from right to left. • Randomize within Blocks—the rows (runs) in the output table will appear in random order within the blocks you set up. One replicate doubles the number of runs. Model is a script. which is used to generate the analysis appropriate to the design. • Sort Left to Right—the rows (runs) in the output table will appear sorted from left to right. Click Custom Design to edit.11 Example Design Table This area identifies the design type that generated the table. Making the Table When the Design panel shows the layout you want. JMP calls these factors Hard to change because this is usually how split plotting arises in industrial practice. click Make Table. • Number of Replicates—Specify the number of times to replicate the entire design. . Usually this is because these factors are difficult or expensive to change from run to run. Parts of the table contain information you might need to continue working with the table and your design in JMP.3 Building Custom Designs—Creating Split Plot Designs 65 3 Custom • Run Order—Lets you designate the order you want the runs to appear in the data table when it is created. • Number of Center Points—Specifies additional runs placed at the center of each continuous factor’s range. Creating Split Plot Designs Split plot experiments happen when it is convenient to run an experiment in groups of runs (called whole plots) where one or more factors stay constant within each group. including centerpoints. • Randomize—the rows (runs) in the output table will appear in a random order. Choices are: • Keep the Same—the rows (runs) in the output table will appear as they do in the Design panel.

66

3 Building Custom Designs—Special Custom Design Commands

In a completely randomized design, any factor can change its setting from one run to the next. When certain factors are hard to change, the completely randomized design may require more changes in the settings of hard-to-change factors than desired. If you know that a factor or two are difficult to change, then you can set the Changes setting of a factor from the default of Easy to Hard. Before making the design, you can set the number of whole plots you are willing to run. For an example of creating a split plot design, see “Creating a Design with Two Hard-to-Change Factors: Split Plot,” p. 48. To create a split-plot design using the custom designer: 1. In the factors table there is a column called Changes. By default, changes are Easy for all factors. If, however, you click in the changes area for a factor, you can choose to make the factor Hard to change. Note that Bayesian D-optimal split plot designs are not available. If you have model terms that are If Possible, JMP makes it impossible to make factor changes be Hard. 2. Once you finish defining the factors and click continue, you will see an edit box for supplying the number of whole plots. You can supply any value as long it is above the minimum necessary to fit all the model parameters. You can also leave this field empty. In this case, JMP will choose a number of whole plots to minimize the omnibus uncertainty of the fixed parameters.

Note: If you want to create a split plot design every time you use a certain factor, save steps by

setting up that factor to be “hard” in all experiments. See “Identifying Factor Changes,” p. 79, for details.

**Special Custom Design Commands
**

After you select DOE > Custom Design, the window in Figure 3.12 appears. Click the red triangle icon on the title bar to reveal a list of commands. The following sections describe the commands and how to use them.

3 Building Custom Designs—Special Custom Design Commands

67 3 Custom

Figure 3.12 Click the Red Triangle Icon to Reveal Commands

**Saving Responses and Factors
**

If you plan to do further experiments with factors and/or responses to which you have given meaningful names and values, you can save them for later use. To save factors or responses: 1. Select a design type from the DOE menu. 2. Enter the factors and responses into the appropriate panels (see “Entering Responses and Factors in the Custom Designer,” p. 57, for details). 3. Click the red triangle icon on the title bar and select Save Responses or Save Factors. • Save Responses creates a data table containing a row for each response with a column called Response Name that identifies the responses. Four additional columns identify more information about the responses: Lower Limit, Upper Limit, Response Goal, and Importance. • Save Factors creates a data table containing a column for each factor and a row for each factor level. The columns have a column property (noted with an asterisk icon in the column panel) called Design Role that identifies them as DOE factors and lists their type (continuous, categorical, blocking, and so on). 4. Save the data table.

**Loading Responses and Factors
**

If you have saved responses and factors, you can quickly apply them to your design and avoid retyping this information each time you run an experiment. To design an experiment using responses or factors you have previously saved: 1. Open the data table that contains the factor names and levels. 2. Select a design type from the DOE menu.

68

3 Building Custom Designs—Special Custom Design Commands

3. Click the red triangle icon on the title bar (Figure 3.13) and select Load Responses or Load

Factors.

Figure 3.13 Loading Factor Names and Values from a JMP Table

.

Tip: Create a factors table by keying data into an empty table, but remember to assign each column a factor type. Do this by right-clicking the column name, selecting Column Info, and then selecting Column Properties > Design Role. Lastly, click the button in

the Design Role area and select the appropriate role.

**Loading and Saving Constraints
**

In custom, augment, and mixture designs, if you set up factor constraints and plan to do further experiments with them, you can save them for later use. You can quickly apply these constraints to your design and avoid retyping this information each time you run an experiment. To save factor constraints: 1. Select a design type from the DOE menu. 2. Enter the factor constraints into the appropriate panels (see “Entering Responses and Factors in the Custom Designer,” p. 57, for details). 3. Click the red triangle icon on the title bar (Figure 3.13) and select Save Constraints. Save Constraints creates a data table that contains the information you enter into a constraints panel. There is a column for each constraint. Each has a column property called Constraint State that identifies it as a ‘less than’ or a ‘greater than’ constraint. There is a row for each variable and an additional row that has the inequality condition for each variable. 4. Save the data table.

3 Building Custom Designs—Special Custom Design Commands

69 3 Custom

To design an experiment using factor constraints you have previously saved: 1. Open the data table that contains the constraints. 2. Select a design type from the DOE menu. 3. Click the red triangle icon on the title bar (Figure 3.13) and select Load Constraints.

**Setting the Random Number Generator
**

The design process begins with a random number. To set the random number the designer uses, click the red triangle icon in the design title bar (Figure 3.13) and select Set Random Seed. The window that appears shows the generating seed for that design (Figure 3.14). From this window, you can set a new random number and then run the design again. If you use the same seed as a previous design, you will get the same design again. Figure 3.14 Setting the Random Seed

**Viewing Diagnostics Reports
**

To display a table with relative D-, G-, and A-efficiencies, average variance of prediction, and length of time to create the design, click the red triangle icon in the Custom Design title bar (Figure 3.12) and select Show Diagnostics before clicking the Make Design button. Now when you click the Make Design button, a diagnostics report (Figure 3.15) appears beneath the model in the Model panel. 11 / p⎞ D -efficiency = 100 ⎛ ------⎝ ND X ′ X ⎠ ⎛ ⎞ p -⎟ A -efficiency = 100 ⎜ -----------------------------------------------– 1 ⎝ trace ( N ( X ′ X ) )⎠

D

⎛ p⎞ -⎟ ⎜ -----G -efficiency = 100 ⎜ ND⎟ ⎜ ----------σ ⎟ ⎝ M⎠ where • ND is the number of points in the design • p is the number of effects in the model including the intercept • σM is the maximum standard error for prediction over the design points.

70

3 Building Custom Designs—Special Custom Design Commands

These efficiency measures are single numbers attempting to quantify one mathematical design characteristic. While the maximum efficiency is 100 for any criterion, an efficiency of 100% is impossible for many design problems. It is best to use these design measures to compare two competitive designs with the same model and number of runs rather than as some absolute measure of design quality. Figure 3.15 Custom Design Showing Diagnostics

**Saving the Design (X) Matrix
**

To create a script and save the design matrix as a table property in a data table, click the red triangle icon in the Custom Design title bar (Figure 3.12) and select Save X Matrix. A script is saved as a table property called Design Matrix. When you run this script, JMP creates a global matrix called X and displays its number of rows in the log. If you do not have the log visible, select View > Log (Window > Log on the Macintosh).

**Changing the Design Criterion (D- or I- Optimality)
**

To change the design criterion, click the red triangle icon in the Custom Design title bar (Figure 3.16) and select Optimality Criterion, then choose Make D-Optimal Design or Make I-Optimal Design.

which varies depending on the design. see “Technical Discussion. all orthogonal designs are D-optimal with respect to a linear additive model.” p. the window in Figure 3.12) and select Number of Starts. . For specific information about optimality criterion. the optimization problem becomes harder. the design produced at the end is not always the same.3 Building Custom Designs—Special Custom Design Commands 71 3 Custom Figure 3. Selecting the Number of Random Starts To override the default number of random starts. It is useful to know that: • If random starts are used for the optimization. When you select this command. 52. It is easy for an optimizer to converge to a local optimum instead of a global optimum.17 appears with an edit box for you to enter the number of random starts for the design you want to build. Increasing the number of random starts tends to improve the optimality of the resulting design. The number you enter overrides the default number of starts. click the red triangle icon in the Custom Design title bar (Figure 3.16 Changing the Design Criterion The default criterion for Recommended is D-optimal for all design types unless you have used the RSM button in the Model panel to add effects that make the model quadratic. Figure 3.17 Changing the Number of Starts Why Change the Number of Starts? One difficulty with the creation of optimal designs is that the methods used to create them do not always find the globally optimal design in cases where the optimal design is known from theory. For example. As the number of factors and sample size increase.

This modification of rules puts extra effort towards finding two-level fractional factorial and Plackett-Burman designs (or their equivalents).12) and select Sphere Radius.” p. Enter the appropriate value and click OK. and c is the number of columns in the design matrix. 65). Note that hypersphere constraints do not work with other constraints. once you set the sphere radius. each factor has only two levels. the algorithm stops. 60. . n is the sample size. if the number of runs is a multiple of four. Also. The time it takes for one iteration of the algorithm (coordinate exchange) increases roughly as the product of the sample size and the number of terms in the model. Before making the design. there is a formula for the optimal determinant: If D is the determinant. and “Creating Split Plot Designs. you cannot make a factor Hard to change. The design is D-optimal and orthogonal. Conversely. The default number of starts for each design are as follows: Sample Size Number of Starts 9 or fewer from 9 to 16 from 17 to 24 from 25 to 32 more than 32 80 40 10 5 2 Finally. split plot designs cannot be generated with hypersphere constraints.” p. then the number of starts listed above is multiplied by four. and the model is linear. the total time it takes to generate a design is kept roughly constant over the range of usual sample sizes. If you have designed any factor’s changes as Hard (see “Creating Optimal Split Plot Designs.72 3 Building Custom Designs—Special Custom Design Commands • For designs with all two-level factors. By doing a large number of random starts for small sample sizes and reducing this number proportional to the square of the sample size as the designs get larger. the sphere radius item becomes unavailable. Default Choice of Number of Random Starts JMP does not start over with random designs until a jackpot is hit. the LogD = cLogn. Constraining a Design to a Hypersphere You can constrain custom and augmented designs to a hypersphere by editing the sphere radius. click the red triangle icon in the Custom Design title bar (Figure 3. If the determinants that result from the random starts match the formula above.

Enter a Boolean expression that identifies what you do not want allowed (Figure 3. use the ordinal value of the level instead of the name of the level. which tells JMP to disallow a run with the best features at the lowest price or a run with the best quality and lowest price. Figure 3. in Figure 3. you would use Feature==3 & Price==3 | Quality==3 & Price==3. 3.18 Enter a Boolean Expression 4. It excludes the combination of factors you specified. For example. Click the red triangle icon in the title bar (Figure 3.12) of the designer window and select Disallowed Combinations. 2. Note that this menu item is not available if you have already defined linear inequality constraints. You want to exclude the third Feature level (best) and the third Price level (low). In this case. JMP evaluates your expression. and low (3). To disallow a combination of factor levels: 1. as shown in Figure 3. the factor Feature has levels of worst (1). Figure 3. This feature can also be used with continuous factors or mixed continuous and categorical factors. medium. For example. Feature==3 & Price==3 will not allow a run containing the best features at the lowest price. and best (3). medium (2). and when it sees it as true.19 No Row Contains L3 for Both Price and Feature . and low. If the level names of the factor called Price are high. Make the design table. 2. and 3. Note: When forming the expression.3 Building Custom Designs—Special Custom Design Commands 73 3 Custom Accounting for Factor Level Restrictions JMP gives you the flexibility to disallow particular combinations of levels of factors. medium (2). Begin by adding the factors. You can do this for custom and augmented designs except for experiments with mixture or blocking factors. you might want to exclude a choice that allows all the best features of a product at the lowest price. If there were two disallowed combinations in this example. and the factor Price has levels of high (1). in a market research choice experiment. it disallows the specified combination.19.18.18). their associated ordinal values are 1.

2. 3. 3. Enter a positive integer and click OK. Select DOE > Custom Design.001 and 1. To reduce the number of rows considered at each iteration. Click the red triangle icon in the title bar (Figure 3. you can set k less than n.12) of the designer window and select Advanced Options > Mixture Sum. Select DOE > Custom Design. Click the red triangle icon in the title bar (Figure 3. . Click the red triangle icon in the title bar (Figure 3. then the sum of the other mixture components must be between 0. This generally speeds up design creation but tends to produce designs that are less efficient for estimating the unknowns in the model. 4. Select DOE > Custom Design. the coordinate exchange algorithm in the custom designer only considers the high and low values. Enter a positive number and click OK. 3. You may have one or more fixed ingredients so that the sum of the remaining add to less than one but more than zero. Make the design.12) of the designer window and select Advanced Options > Search Points Per Factor. To change the K exchange value: 1. if you enter 5. the coordinate exchange algorithm considers every row of factor settings for possible replacement in every iteration. you can change this so the algorithm considers more search points. Setting the K Exchange Value By default. To alter the mixture sum: 1. 2.74 3 Building Custom Designs—Special Custom Design Commands Advanced Options The following options are only for advanced design of experiment users. Make the design. then JMP considers five equally spaced settings for each factor. For example. Altering the Mixture Sum If you want to keep a component of a mixture constant throughout an experiment.12) of the designer window and select Advanced Options > K Exchange Value. Changing the Search Points Per Factor For a main effects model. 2. Enter a positive integer and click OK. 4. 4. Make the design. To change the search points: 1. However.

Click the red triangle icon in the title bar (Figure 3. If you have some prior knowledge of this variance ratio. 4.jmp.20. The following discussion covers properties specific to DOE and useful for analyzing DOE data. you can supply it by following these steps: 1.20 Column Info Window and New Property Menu All special column properties are discussed in the JMP User Guide. right-click the column name in the data table and select Column Info. Notice the column called Stretch. which has two special properties: Role and Response Limits. By default. Select DOE > Custom Design. 2.12) of the designer window and select Advanced Options > Split Plot Variance Ratio. this variance is one.3 Building Custom Designs—Assigning Column Properties 75 3 Custom Split Plot Variance Ratio The optimal split plot design depends on the ratio of the variance of the random whole plot variance to the error variance. select Help (View on the Macintosh) > Sample Data Directory > Design Experiment > Bounce Data. Assigning Column Properties Columns in a data table can contain special column properties. . To follow along with this example that shows column properties. Make the design. as shown in Figure 3. Enter a positive number and click OK. 3. Figure 3. To see these properties.

when JMP generates a design table from values entered in the Factors panel. Figure 3. Select Coding from the Column Properties drop-down menu. JMP uses those transformed data values (for continuous variables only) to compute meaningful parameter estimates in the analysis. 4. You can specify the range in which the low and high values of the column are transformed. When you select Analyze > Fit Y by X. To set up a column as a mixture factor: 1.76 3 Building Custom Designs—Assigning Column Properties Defining Low and High Values (DOE Coding) for Columns Coding transforms data in the range you specify from -1 to +1. The Column Info window appears. Click OK. By default. This transformation is called coding. it uses those values as the low and high values of the coding property. indicating the column contains a property. If a column has one or more limits missing. 2. Setting Columns as Factors for Mixture Experiments You might have a column in a data table that is one of several factors that form 100% of a mixture. Double-click the column name in the data grid. You can set up the column so JMP uses it to automatically generate a no-intercept model when you analyze the data by selecting Analyze > Fit Model. Type the values you want to use as low and high values into the text boxes. To change the range in which the low and high values are transformed: 1. as shown in Figure 3. The properties icon ( ) now appears next to the column name in the data table’s column panel.21. . Coding information appears on the right. JMP substitutes the data’s minimum and maximum for the high and low values. The Column Info window appears. Double-click the column name in the data grid.21 Coding Window 3.

23. middle. as shown in Figure 3. Mixture information appears on the right. and desirability values. Check the boxes beside L and U PseudoComponent Coding. if desired. . 4. and the U coding otherwise. For example. the Fit Model platform uses the L coding if (1 – L) < (U – 1). If you do this. which lets JMP know how to weigh one response’s importance against another’s. but coding values are used. Select Response Limits from the New Property drop-down menu. Using the example in Figure 3. you can run a new analysis on the new data using the same limits.22. Defining Response Column Values You can save response limits in a column. where the mixture sum value is 1. the terms are coded as: XiL = (Xi – Li)/(1 – L) for the L pseudocomponent XiU = (Ui – Xi)/(U – 1) for the U pseudocomponent where Li and Ui are the lower and upper bounds. then come back later and change the data. Click OK.22 Mixture Information 3. The Column Info window appears. respect the pseudocomponent coding but present the un-coded values in the tables and plots. All the features of fitting. In the Fit Model report. If you check both in this example. 2. you can enter an importance value.22. The properties icon ( ) now appears next to the column name in the data table’s column panel. and upper limits. if you run an analysis that employs these limits. You can specify values for the lower. If you have two responses. Enter the upper and lower limits as well as the sum of terms. or None. which are possible goals for a DOE response variable. Figure 3. you can run analyses without having to re-specify response limits each time.23. as shown in Figure 3. indicating the column contains a property. Double-click the column name in the data grid. L is the sum of Li and U is the sum of Ui. such as the profilers and saved formulas. Minimize. Complete the settings in the window.3 Building Custom Designs—Assigning Column Properties 77 3 Custom 2. To enter response limits: 1. Crossed effects are not labeled. Select Mixture from the New Property drop-down menu. Match Target. You can also select Maximize. 5. the main effects are labeled with the coding transformation. as shown in Figure 3. 3. Saving these limits in a column facilitates consistency from use to use.

or noise factor. categorical.23 Defining Response Column Values Click to select a goal for the response variable. The properties icon ( ) now appears next to the column name in the data table’s column panel. indicating the column contains a property. Giving Columns a Design Role You can tell JMP how to use a factor column in a model to design an experiment. as shown in Figure 3.78 3 Building Custom Designs—Assigning Column Properties Figure 3. Click OK. Double-click the column name in the data grid. Select Design Role from the New Property drop-down menu. constant. covariate. select Match Target When you have two responses. middle and upper limits as well as the desirability values 4. 2. blocking.24. To give a column a design role: 1. The Column Info window appears. Design role information appears on the right. . mixture. if you are in the prediction profiler and want the desired value to be close to zero. signal. You can tell JMP to use it as a continuous. enter a number to indicate the amount of weight you want this response to have when JMP computes the overall desirability Enter the lower. For example.

if you know that every time you use that factor. . Double-click the column name in the data grid. However.” meaning that the factor would be difficult to change. 4.” p.25. Identifying a factor as “hard” can be done in the DOE design panel (see “Creating Split Plot Designs. Covariate. Categorical. you must identify a factor as “hard. you want to create a split plot design.3 Building Custom Designs—Assigning Column Properties 79 3 Custom Figure 3. and Random Block. Blocking. Select Factor Change from the Column Properties drop-down menu. Identifying Factor Changes To create a split plot design using JMP’s DOE (Design of Experiments) commands. Constant. indicating the column contains a property. Noise.24 Choosing a Design Role 3. Click OK. as shown in Figure 3. Factor change information appears on the right. Click the Design Role drop-down menu and select how you want JMP to use the factor column: Continuous. Signal. Uncontrolled. Mixture. you can save yourself steps by setting up a column to be “hard” in all experiments using that factor. 2. To identify a factor as “hard” and create a split plot design when using that factor: 1. for details) each time you design an experiment. The Column Info window appears. The properties icon ( ) now appears next to the column name in the data table’s column panel. 65.

How Custom Designs Work: Behind the Scenes The custom designer starts with a random set of points inside the range of each factor.Optimality). Equivalent solutions are designs with equal precision for estimating the model coefficients as a group. the whole process is repeated several times using a different random start.80 3 Building Custom Designs—How Custom Designs Work: Behind the Scenes Figure 3.25 Choosing a Factor Change 3. . When this is true. For more details. 70.or I. see the section “Changing the Design Criterion (D. The properties icon ( ) now appears next to the column name in the data table’s column panel. Each iteration of the algorithm involves testing every value of every factor in the design to determine if replacing that value increases the optimality criterion. the design algorithm may generate different (but equivalent) designs when you click the Back and Make Design buttons repeatedly. The custom designer displays the best of these designs.” p. Click OK. the new value replaces the old. indicating the column contains a property. Iteration continues until no replacement occurs in an entire iterate. To avoid converging to a local optimum. The computational method is an iterative algorithm called coordinate exchange (Meyer and Nachtsheim. Sometimes a design problem can have several equivalent solutions. 4. Click the Factor Change drop-down menu and select how you want JMP to use the factor column: Easy or Hard. If so. 1995).

and a four-level factor has 2 x 3 x 4 = 24 runs. which is why they are cheap. To understand the importance of effect sparsity. Compared to other design methods. These factors can be continuous or categorical. • By contrast. The number of runs is the product of the factor levels. This chapter discusses the screening approach in detail. with two or three levels. For example. a factorial experiment with a two-level factor. you can contrast screening designs to full factorial designs: • Full factorial designs consist of all combinations of the levels of the factors. It also describes how to use JMP’s screening designer. you can restrict the factors to two levels. So the screening approach reduces the 24-run experiment to four runs. Further. Each factor in a screening design is usually set at two levels to economize on the number of runs needed. The efficiency of screening designs depends on the critical assumption of effect sparsity. there is a price for this reduction. which yield 2 x 2 x 2 = 8 runs. screening designs reduce the number of runs by restricting the factors to two (or three) levels and by performing only a fraction of the full factorial design. Thus. You can then use the results to guide further study. Effect sparsity results because real-world processes usually have only a few driving factors. and response measurements are taken for only a fraction of the possible combinations of levels. They examine many factors to see which have the greatest effect on the results of a process. It is wise to spend only about a quarter of your resource budget on an initial screening experiment. other factors are relatively unimportant. In the case described above. Of course.4 Screening Designs 4 Screening Screening designs are arguably the most popular designs for industrial experimentation. which supplies a list of popular screening designs for two or more factors. . Often screening designs are a prelude to further experiments. a three-level factor. showing both pros and cons. The list of screening designs you can use includes designs that group the experimental runs into blocks of equal sizes where the size is a power of two. they are attractive because they are a cheap and efficient way to begin improving a process. screening designs require fewer experimental runs. by doing half of these eight combinations you can still assess the separate effects of the three factors.

. . . . . Creating a Screening Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using Five Continuous Factors . . Choosing a Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Continuing the Analysis . . . . . . . . . . . . . . . . . Entering Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using Two Continuous and One Categorical Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Displaying and Modifying the Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Contents Screening Design Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Entering Responses . . . . . . . . 83 83 85 87 88 89 90 93 97 97 98 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Specifying Output Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Viewing the Table . . . . . . . . . . . . . . . . . . . . . .

Change the default response as follows: 1.12 as the lower limit (minimum acceptable value). The default goal for the single default response is Maximize. which is the speed at which the part rotates under the beam • Beam Current. Click the Lower Limit text edit area and enter 0. Click the Goal text edit area and choose Match Target.4 Screening Designs—Screening Design Examples 83 4 Screening Screening Design Examples This chapter is divided into two sections.1 Screening Design Response With Match Target Goal 3.17 inches To begin this example. A voltage applied to a beam generator creates a stream of electrons that heats the two parts.12 and an upper bound of 0. . The engineer wants to study the welding process to determine the best settings for the beam generator to produce the desired depth in the fused region. The first section consists of two examples using screening designs. causing them to fuse. This depth of penetration is the response for the study. the engineer wants to explore the following three inputs.17 inches. Double-click the response name and change it to Depth (In. This reveals an area where the two parts have fused. The second section outlines the procedures to follow to create a screening design to match your needs.).17 inches with a lower bound of 0. 2. For this study. but the goal of this process is to get a target value of 0. select DOE > Screening Design from the main menu.22. who is the technician operating the welding machine • Rotation Speed. The ideal depth of the fused region is 0. which are the factors for the study: • Operator. The goals of the study are to: • find which factors affect the depth of the weld • quantify those effects • find specific factor settings that predict a weld depth of 0. The engineer fits the two parts into a welding fixture that holds them snugly together. Using Two Continuous and One Categorical Factor Suppose an engineer wants to investigate a process that uses an electron beam welding machine to join two parts. there is a single default response called Y. Note that in the Responses panel.1. the engineer cuts the part in half. as shown in Figure 4. Figure 4. The Length of this fused area is the depth of penetration of the weld. which is a current that affects the intensity of the beam After each processing run.

Double-click the factor names and rename them Operator.3. Figure 4. and assign Mary and John as values for the categorical factor called Operator. as shown in Figure 4. and then click Continue. Add the categorical factor by clicking the Add button beside 2-Level Categorical. 7. 4. Figure 4.3 List of Screening Designs for Two Continuous and One Categorical Factors When the design details are complete. 10. 6.4 appears. The Pattern variable shows the coded design runs. as shown in Figure 4. factors. click Make Table to create a JMP table that contains the specified design.84 4 Screening Designs—Screening Design Examples Then click the Upper Limit text edit area and enter 0. and levels you specified. This example has one categorical factor (Operator) and two continuous factors (Speed and Current). The table uses the names for responses. Set high and low values for Current to 150 and 165 amps. Speed. and Current.jmp. Click Continue.2 Screening Design with Two Continuous and One Categorical Factor 8.View the table produced in this example by selecting Help (View on the Macintosh) > Sample Data Directory > Design Experiment > DOE Example 1.2.22 as the upper limit (maximum acceptable value). 5. 9. Select Full Factorial in the list of designs. Add two continuous factors by typing 2 in the Continuous box and clicking the associated Add button. . Set high and low values for Speed to 3 and 5 rpm. The table in Figure 4.

et al. . the amount of raw material added to the reaction chamber in liters per minute • percentage of catalyst • stir rate. upper limit to 99. Change the default factor names (X1-X5) to Feed Rate. Stir Rate. 5. Catalyst. Temperature. 2.4 The Design Data Table Using Five Continuous Factors As illustrated in the previous section.4 Screening Designs—Screening Design Examples 85 4 Screening Figure 4. as shown in Figure 4.5). a chemical engineer investigates the effects of five factors on the percent reaction of a chemical process.5. adapted from Meyer. Select DOE > Screening Design. and Concentration. and importance to 1 (Figure 4. Change the minimum acceptable reaction percentage (Lower Limit) to 90. experiments for screening the effects of many factors usually consider only two levels of each factor. (1996). demonstrates how to use JMP’s screening designer when you have many factors. You see one default response called Y. In this study. 3. The factors are: • feed rate. 4. the RPMs of a propeller in the chamber • reaction temperature in degrees Celsius • concentration of reactant To start the example: 1. 6. This allows the examination of many factors with a minimum number of runs. Enter the high and low values. Change the default response name (Y) to Percent Reacted. The following example. Add five continuous factors.

Click Continue again to see the Design Output Options panel on the right in Figure 4.5 Screening for Many Factors 7. 8. JMP lists the designs for the number of factors you specified. 9. Click Continue.6. which is an 8-run fractional factorial design. as shown to the left in Figure 4.7 that modify the final design table. Now. .6 Two-level Screening Design (left) and Design Output Options (right) The design dialog has options shown in Figure 4.6.86 4 Screening Designs—Screening Design Examples Figure 4. Select the first item in the list. Figure 4.

” p.4 Screening Designs—Creating a Screening Design 87 4 Screening Figure 4. “Entering Responses. open Reactor 8 Runs.7 Output Options for Design Table 10.” p.8 JMP Table of Runs for Screening Example Of the five factors in the reaction percentage experiment. 11. follow the steps below: 1. 88 2. select DOE > Screening Design. as shown to the right of the data table. 90 3.8 that lists the runs for the design you selected. Let’s take an approach to the analysis that looks for active effects. or click the Screening Design button on the JMP Starter DOE page. you expect a few to stand out in comparison to the others. Figure 4. Creating a Screening Design To start a screening design. 93 . as shown previously in Figure 4.To run the model generated by the data.jmp from the Design Experiment folder inside the Sample Data folder that was installed when you installed JMP. “Choosing a Design. “Displaying and Modifying the Design. Then.8. Note that it also has a column called Percent Reacted for recording experimental results. 12.Click Make Table to create the data table shown in Figure 4.In the design data table.” p. or choose Analyze > Fit Model to analyze the data. right-click the Model script and select Run Script.

such as when the response is impurity or defects. • The Minimize goal supports an objective of having the best objective be the smallest value. Specifying Goal Types and Lower and Upper Limits When entering responses.88 4 Screening Designs—Creating a Screening Design 4. 97 6. to match a specific value.9) and the lower and upper limits (step 4 in Figure 4. “Continuing the Analysis. or None. A goal of Maximize supports this objective. 4 To enter one 1 response at a time. if desired. follow the steps in Figure 4.9): • For responses such as strength or yield.” p. click the N Responses button and enter the number of responses you want. Match Target. the best value is usually the largest possible. “Specifying Output Options. Minimize.9. 3 Click to change the response goal. or that there is no goal.9 Entering Responses Click to enter lower and upper limits and importance weights. Tip: To quickly enter multiple responses. if desired. Figure 4. 2 Double-click to edit the response name.” p. click then select a goal type: Maximize. “Viewing the Table. you can tell JMP that your goal is to obtain the maximum or minimum value possible. 98 Entering Responses To enter responses. 97 5.” p. . The following description explains the relationship between the goal type (step 3 in Figure 4.

183. and “Creating a Custom Design. With two responses you can give greater weight to one response by assigning it a higher importance value. To remove a level.10 Entering Factors To enter factors. Space Filling Design. and Response Surface Design.4 Screening Designs—Creating a Screening Design 89 4 Screening • The Match Target goal supports the objective when the best value for a responses is a specific target value. Types of Factors In general. see “Taguchi Designs. This process is described below. press the delete key on your keyboard. Figure 4.9) as the weight of each response. you can alter the default target after you make a table from the design. The default target value is assumed to be midway between the lower and upper limits.” p. Understanding Importance Weights When computing overall desirability. you can enter different types of factors in the model. Entering Factors After entering responses. If there is only one response. JMP uses the value you enter as the importance weight (step 4 in Figure 4. when designing experiments. type the number of factors and click Add. and custom designs. “Full Factorial Designs. Below is a description of factor types from which you can choose when creating screening designs: . click it. such as with part dimensions. 57. then specifying importance is unnecessary. Mixture Design. open the response’s Column Info dialog by double-clicking the column name. Highlight the factor and click the Remove Selected button to remove a factor in the list. For information on entering factors in Taguchi designs. Click to enter factor values. in Figure 4. The Factors panel’s appearance depends on the design you select. Double-click to edit the factor name.10.” p. enter factors. and enter an asymmetric target value. full factorial designs. In the data table.” p. 117. Note: If your target range is not symmetric around the target value. Entering factors is the same in Screening Design. then press the Return or Enter key on your keyboard.

.11 Choosing a Type of Screening Design The screening designer provides the following types of designs: Two-Level Full Factorial A full factorial design contains all combinations of the levels of the factors. This means that the estimates of the effects are uncorrelated. After your responses and factors are entered. Two-Level Fractional Factorial A fractional factorial design also has a sample size that is a power of two. If the values are character. If you remove an effect in the analysis. If the values are numbers. click Continue. If k is the number of factors. this is 2k where k is the number of factors. The figure to the left in Figure 4. Highlight the type of screening design you would like to use and click Continue. the order is the sorting sequence. you can set a continuous factor to any value between the lower and upper limits you supply.90 4 Screening Designs—Creating a Screening Design • Continuous Continuous factors have numeric data types only. fractional factorial designs are orthogonal. These designs are orthogonal. the number of runs is 2k – p where p < k. Figure 4.or second-order approximations to the true functional relationship between the factors and the responses. Examples of categorical factors are machine. Choosing a Design The list of screening designs you can use includes designs that group the experimental runs into blocks of equal sizes where the size is a power of two. The samples size is the product of the levels of the factors. Their p-values change slightly.12 is a geometric representation of a two-level factorial. In theory. The fraction of the full factorial is 2-p. For two-level designs. the order is the numeric magnitude. This can be expensive if the number of factors is greater than 3 or 4. • Categorical Categorical factors (either numerical or categorical data types) have no implied order. Most empirical modeling involves first. the values of the other estimates remain the same. Like the full factorial. Full factorial designs allow the estimation of interactions of all orders up to the number of factors. because the estimate of the error variance and the degrees of freedom are different. operator. The settings of a categorical factor are discrete and have no intrinsic order. and gender.

you typically only run a fraction of all possible levels. -1. Experiments can therefore be classified by resolution number into three groups: • Resolution = 3 means that main effects are confounded with one or more two-way interactions. -1. including interaction effects to the mth degree. • Resolution ≥ 5 means there is no confounding between main effects. However. The figure on the right in Figure 4. Resolution Number: The Degree of Confounding In practice.4 Screening Designs—Creating a Screening Design 91 4 Screening The trade-off in screening designs is between the number of runs and the resolution of the design. 1. One useful characteristic is that the sample size is a multiple of four rather than a power of two. -1 1. -1 1. fractional factorials are designed by deciding in advance which interaction effects are confounded with the other interaction effects. there are 20-run. This causes some of the higher-order effects in a model to become nonestimable. 1. the minimum aberration design of a given resolution minimizes the number of words in the defining relation that are of minimum length. All the fractional factorial designs are minimum aberration designs. –1.12 Representation of Full Factorial (Left) and Two-Level Fractional Factorial (Right) Designs –1. Figure 4. between two-factor interactions. These higher-order interactions are assumed to be zero. But because running experiments costs time and money. 1 Plackett-Burman Designs Plackett-Burman designs are an alternative to fractional factorials for screening. and 28-run Plackett-Burman designs. or between main effects and two-factor interactions. you can run several replicates of all possible combinations of m factor levels. However. -1 1. This is different from resolution-three fractional factorial where two-factor interactions are indistinguishable from main effects. The main effects are orthogonal and two-factor interactions are only partially confounded with main effects. which must be assumed to be zero for the main effects to be meaningful. -1.12 is geometric representation of a two-level fractional factorial design. In fact. two-factor interactions are confounded with other two-factor interactions. 1. 1 . If price is no object. For DOE experts. An effect is nonestimable when it is confounded with another effect. This provides a good estimate of everything. 24-run. There are no two-level fractional factorial designs with sample sizes between 16 and 32 runs. –1 -1. few experimenters worry about interactions higher than two-way interactions. • Resolution = 4 means that main effects are not confounded with other main effects or two-factor interactions.

JMP also includes a set of extra columns to use as regressors. They are constructed by adding up all the odd and even interaction terms for each factor. For k factors.92 4 Screening Designs—Creating a Screening Design In cases of effect sparsity. even though they are all orthogonal. When you use JMP to generate a Cotter design. Then follow k runs each with one factor in turn at its low level. JMP offers complete factorials and specialized orthogonal-array designs. the sum will show an effect if it contains an active effect. The next run sets all factors at their low level and sequences through k more runs with one factor high and the rest low. The design is similar to the “vary one factor at a time” approach many books call inefficient and naive. The danger is that several active effects with mixed signs will cancel and still sum to near zero and give a false negative. and you believe there may be interactions. For mixed two-level and three-level designs. Suppose you believe in effect sparsity— that very few effects are truly nonzero. For example. subject to randomizing the runs. JMP offers fractional factorials. This completes the Cotter design. listed below. Design Two–Level Factors Three–Level Factors L18 John L18 Chakravarty L18 Hunter L36 1 3 8 11 7 6 4 12 If you have fewer than or equal to the number of factors for a design listed in the table. there are 2k + 2 runs. Mixed-Level Designs If you have qualitative factors with three values. A. A Cotter design begins with a run having all factors at their high level. Cotter designs are easy to set up. B. if you have three factors. These are of the form factorOdd and factorEven where factor is a factor name. and the others high. For pure three-level factorials. a stepwise regression approach can allow for removing some insignificant main effects while adding highly significant and only somewhat correlated two-factor interactions. You believe in this so strongly that you are willing to bet that if you add up a number of effects. you can use that design by selecting an appropriate subset of columns from the original design. and C: AOdd = A + ABC BOdd = B + ABC COdd = C + ABC AEven = AB + AC BEven = AB + BC CEven = BC + AC . Cotter Designs Cotter designs are used when you have very few resources and many factors. then none of the classical designs discussed previously are appropriate. Some of these designs are not balanced.

4. • Aliasing of Effects—Shows the confounding pattern for fractional factorial designs. To alter the setting for all screening designs: 1. Immediately after entering responses and factors (and before clicking Continue).13 Display and Modification Options • Change Generating Rules—Controls the choice of different fractional factorial designs for a given number of factors. many statisticians discourage their use. click the disclosure buttons ( on Windows/Linux and on the Macintosh) to display the design and show modification options using the Display and Modify Design panel to tailor the design (Figure 4. Uncheck the box beside Supress Cotter Designs. Aliasing of Effects To see which effects are confounded with which other effects. However.4 Screening Designs—Creating a Screening Design 93 4 Screening Because these columns in a Cotter design make an orthogonal transformation. Click DOE to highlight it. 3. Select File > Preferences.13). testing the parameters on these combinations is equivalent to testing the combinations on the original effects. Figure 4. click the red triangle icon in the Screening Design title bar. click the disclosure button ( on Windows/Linux and on the Macintosh) to reveal the Aliasing of Effects panel. AOdd estimates the sum of odd terms involving A. AEven estimates the sum of the even terms involving A. It shows effects and confounding up to two-factor interactions (Figure 4. if you would like to make a Cotter design: 1. JMP does not include a Cotter design in the list of available screening designs (Figure 4.11). Displaying and Modifying the Design After you select a design type. • Coded Design—Shows the pattern of high and low values for the factors in each run. Changing the setting via the red triangle menu applies only to the current design.14). How to Run a Cotter Design By default. 2. 2. Select Supress Cotter Designs. Click the Platform icon. Because Cotter designs have a false-negative risk. . and so forth. In the example of factors listed above.

To create this table: 1. Confounding is the direct result of the assignment of new factor values to products of the coded design columns. 2.94 4 Screening Designs—Creating a Screening Design Figure 4. This means that you can’t tell the difference of the effect of Temperature and the synergistic (interactive) effect of Feed Rate and Concentration. Select Show Confounding Pattern (Figure 4. Eight runs can only accommodate a full factorial with three two-level factors. For example.14. This is characteristic of resolution-three designs. a full factorial with five factors requires 25 = 32 runs. the values for Temperature are the product of the values for Feed Rate and Concentration. In the example shown in Figure 4. . Viewing the Confounding Pattern JMP can create a data table that shows the aliasing pattern for a specified level.14 Generating Rules and Aliasing of Effects Panel For example.15). It is necessary to construct the two additional factors in terms of the first three factors. The price of reducing the number of runs from 32 to eight is effect aliasing (confounding). Click the red triangle at the bottom of the Aliasing of Effects area. all the main effects are confounded with two-factor interactions.

18. Figure 4. Thus. and Stir Rate). Enter the order of confounding you want to see (Figure 4.16 Enter Order 4.17 shows the third level alias for the five-factor reactor example.17 The Third Level Alias for the Five-Factor Reactor Example Understanding the Coded Design In the coded design panel. Temperature appears as “4”. Click OK. The effect names begin with C (Constant) and are shown by their order number in the design. Figure 4. which represent Feed Rate. with second order aliasing as “1 5” (Feed Rate and Concentration). each row represents a run. Figure 4. rows for the first three columns of the coded design.15 Show Confounding Patterns 3. Catalyst. As shown in Figure 4.4 Screening Designs—Creating a Screening Design 95 4 Screening Figure 4. and third order confounding as “1 2 3” (Feed Rate.16). and Stir Rate are all combinations of high and low values (a full . Catalyst. Plus signs designate high levels and minus signs represent low levels.

Temperature is now the product of Feed Rate and Catalyst.19 and click Apply.96 4 Screening Designs—Creating a Screening Design factorial design).18 Default Coded Designs Changing the Coded Design In the Change Generating Rules panel. Catalyst. Concentration is a function of Feed Rate and Stir Rate. Figure 4.19 shows how the last two columns are constructed in terms of the first three columns. The fourth column (Temperature) of the coded design is the element-by-element product of the first three columns. The checkmarks for Concentration show it is a function of Catalyst and Stir Rate. The first three columns of the coded design remain a full factorial for the first three factors (Feed Rate. Similarly. so the fourth column of the coded design is the element by element product of the first two columns. and Stir Rate. it changes the choice of different fractional factorial designs for a given number of factors. the last column (Concentration) is the product of the second and third columns. and Stir Rate). The Change Generating Rules table in Figure 4. Figure 4. Catalyst. the Coded Design panel changes. changing the checkmarks and clicking Apply changes the coded design. The check marks for Temperature show it is a function of Feed Rate.19 Modified Coded Designs and Generating Rules . If you check the options as shown in Figure 4.

One replicate doubles the number of runs. In the table. When the options are correctly set up. • Sort Left to Right—the rows (runs) in the output table appear sorted from left to right. including centerpoints. • Randomize—the rows (runs) in the output table appear in a random order. . click Make Table. Type the number of times you want to replicate the design in the associated text box. • Sort Right to Left—the rows (runs) in the output table appear sorted from right to left. the high and low values you specified are displayed for each run. • Number of Replicates—Specify the number of times to replicate the entire design. • Number of Center Points—Specifies additional runs placed at the center points. • Randomize within Blocks—the rows (runs) in the output table will appear in random order within the blocks you set up.4 Screening Designs—Creating a Screening Design 97 4 Screening Specifying Output Options Use the Output Options panel to specify how you want the output data table to appear. you have a data table that outlines your experiment.20 Select the Output Options • Run Order—Lets you designate the order you want the runs to appear in the data table when it is created. Viewing the Table After clicking Make Table. Figure 4. Choices are: • Keep the Same—the rows (runs) in the output table appear as they do in the Design panel.

The data table contains a script labeled Model. you can now run analyses on the data.22 Running the Model Script The next sections describe some of the parts of the analysis report that appears when you click Run Model. Continuing the Analysis After creating and viewing the data table.22). useful as a label variable in plots. .21 The Design Data Table The name of the table is The column called Pattern shows the pattern of low values the design type that denoted “–” and high values denoted “+”. Figure 4. Pattern is especially generated it. Right-click it and select Run Script to run a fit model analysis (Figure 4.98 4 Screening Designs—Creating a Screening Design Figure 4. This script allows you to easily fit a model using the values in the design table.

The RMSE is an estimate of the standard deviation of the process noise assuming that the unestimated effects are negligible. it is impossible to determine which two-factor interactions are important without performing more experimental runs.23).24) as a part of the Fit Model report. At this stage the results are not clear. Figure 4. which tells you that the model is not significant. the RMSE is 14. R2. None of the factor effects are significant. Because of the confounding between two-factor interactions and main effects in this design. and RMSE appear below the plot. which is much larger than expected.23. The last column of the table has the p-values for each effect. but this does not mean that the experiment has failed. but the Catalyst effect is large enough to be interesting if it is real. Figure 4. Viewing a Scaled Estimates Report When you fit the model. appears at the top of the Fit Model report.23 An Actual-by-Predicted Plot To show labels in the graph (on the right in Figure 4.199. and select Row Label. right-click the graph. In this case. JMP displays a Scaled Estimates report (Figure 4. In Figure 4.23. select all points. It means that some follow-up runs are necessary. This suggests that effects other than the main effects of each factor are important.24 Example of a Scaled Estimates Report . an actual-by-predicted plot.4 Screening Designs—Creating a Screening Design 99 4 Screening Viewing an Actual-by-Predicted Plot When the model contains no interactions. The Scaled Estimates report displays a bar chart of the individual effects embedded in a table of parameter estimates. shown on the left in Figure 4. The model p-value. the mean line falls inside the bounds of the 95% confidence curves. The pattern variable displayed in the data table serves as the label for each point.

For this. you would then want to augment the design. 193.25 Example of a Power Analysis Refer to the JMP Statistics and Graphics Guide for details. JMP displays an Effect Details panel. Running a Power Analysis When the model contains interactions. Then run the power analysis by clicking the red triangle icon in a factor’s title bar and selecting Power Analysis. such as described in “Augmented Designs. . see the chapter “Full Factorial Designs.” p. Open the outline node.” p. Figure 4.100 4 Screening Designs—Creating a Screening Design If this scaled estimates report were not merely an example. For comparison. you might also want to have complete 32-run factorial experimental data and analysis. 117.

5 Response Surface Designs 5 Surface Response surface designs are useful for modeling a curved quadratic surface to continuous factors. illustrated in the figure on the right below. • Axial (or star) points. so the standard two-level designs cannot fit curved surfaces. illustrated in the figure to the left below. for which all the factor values are at the zero (or midrange) value. This is sometimes useful when it is desirable to avoid these points due to engineering considerations. Another important difference between the two design types is that the Box-Behnken design has no points at the vertices of the cube defined by the ranges of the factors. is an alternative to central composite designs. The Box-Behnken design. for which all but one factor are set at zero (midrange) and that one factor is set at outer (axial) values. The price of this characteristic is the higher uncertainty of prediction near the vertices compared to the central composite design. Central Composite Design fractional factorial points Box-Behnken Design axial points center points . a response surface model can pinpoint it. The most popular response surface design is the central composite design. Three distinct values for each factor are necessary to fit a quadratic function. One distinguishing feature of the Box-Behnken design is that there are only three levels per factor. If a minimum or maximum response exists inside the factor region. It combines a two-level fractional factorial and two other kinds of points: • Center points.

5

Contents

A Box-Behnken Design: The Tennis Ball Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Viewing the Geometry of a Box-Behnken Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Response Surface Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Entering Responses and Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Choosing a Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Specifying Axial Value (Central Composite Designs Only) . . . . . . . . . . . . . . . . . . . . . . . . . . . Specifying Output Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Viewing the Design Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Continuing the Analysis, If Needed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 109 110 110 111 112 113 113 114

5 Response Surface Designs—A Box-Behnken Design: The Tennis Ball Example

103 5 Surface

**A Box-Behnken Design: The Tennis Ball Example
**

The Bounce Data.jmp sample data file has response surface data inspired by the tire tread data described in Derringer and Suich (1980). To see this example data table, open the Sample Data folder that was installed when you installed JMP, and select Design Experiment > Bounce Data.jmp. The objective of this experiment is to match a standardized target value (450) of tennis ball bounciness. The bounciness varies with amounts of Silica, Silane, and Sulfur used to manufacture the tennis balls. The experimenter wants to collect data over a wide range of values for these variables to see if a response surface can find a combination of factors that matches a specified bounce target. To follow this example: 1. Select DOE > Response Surface Design. 2. Load factors by clicking the red triangle icon on the Response SurfaceDesign title bar and selecting Load Factors. Navigate to the Sample Data folder that was installed when you installed JMP, and select Design Experiment > Bounce Factors.jmp. 3. Load the responses by clicking the red triangle icon on the Response Surface Design title bar and selecting Load Responses. Navigate to the Sample Data folder that was installed when you installed JMP, and select Design Experiment > Bounce Response.jmp. Figure 5.1 shows the completed Response panel and Factors panel. Figure 5.1 Response and Factors For Bounce Data

After the response data and factors data are loaded, the Response Surface Design Choice dialog lists the designs in Figure 5.2. Figure 5.2 Response Surface Design Selection

The Box-Behnken design selected for three effects generates the design table of 15 runs shown in Figure 5.3.

104 5 Response Surface Designs—A Box-Behnken Design: The Tennis Ball Example

In real life, you would conduct the experiment and then enter the responses into the data table. Let’s pretend this happened and use a finalized data table called Bounce Data.jmp. 1. Open to the Sample Data folder that was installed when you installed JMP, and select Design Experiment > Bounce Data.jmp (Figure 5.3). Figure 5.3 JMP Table for a Three-Factor Box-Behnken Design

After obtaining the Bounce Data.jmp data table, run a fit model analysis on the data. The data table contains a script labeled Model. 2. Right-click Model and select Run Script to start a fit model analysis. 3. Click Run Model. The standard Fit Model analysis results appear in tables shown in Figure 5.4, with parameter estimates for all response surface and crossed effects in the model. The prediction model is highly significant with no evidence of lack of fit. All main effect terms are significant as well as the two interaction effects involving Sulfur. Figure 5.4 JMP Statistical Reports for a Response Surface Analysis of Bounce Data

5 Response Surface Designs—A Box-Behnken Design: The Tennis Ball Example

105 5 Surface

See the chapter “Standard Least Squares: Introduction” of JMP Statistics and Graphics Guide for more information about interpretation of the tables in Figure 5.4. The Response Surface report also has the tables shown in Figure 5.5. Figure 5.5 Statistical Reports for a Response Surface Analysis

Provides a summary of the parameter estimates

Lists the critical values of the surface factors and tells the kind of solution (maximum, minimum, or saddlepoint). The solution for this example is a saddlepoint. The table also warns that the critical values given by the solution are outside the range of data values. Shows eigenvalues and eigenvectors of the effects. The eigenvector values show that the dominant negative curvature (yielding a maximum) is mostly in the Sulfur direction. The dominant positive curvature (yielding a minimum) is mostly in the Silica direction. This is confirmed by the prediction profiler in Figure 5.8. See the chapter “Standard Least Squares: Exploring the Prediction Equation” of JMP Statistics and Graphics Guide for details about the response surface analysis tables in Figure 5.5. Using the Prediction Profiler Now use the response Prediction Profiler to get a closer look at the response surface to find the best settings that produce the response target. It is a way of changing one variable at a time and looking at the effects on the predicted response. 1. If not open already, open the Prediction Profiler by clicking the red triangle on the Response title bar and selecting Factor Profiling > Profiler (Figure 5.6).

106 5 Response Surface Designs—A Box-Behnken Design: The Tennis Ball Example

Figure 5.6 The Profiler

The Profiler (Figure 5.8) displays prediction traces for each x variable. A prediction trace is the predicted response as one variable is changed while the others are held constant at the current values (Jones 1991). The current predicted value of Stretch, 396, is based on the default factor setting. It is represented by the horizontal dotted line that shows slightly below the desirability function target value (Figure 5.8). The profile shows desirability settings for the factors Silica, Silane, and Sulfur, which result in a value for Stretch of 396, which is close to the specified target of 450. The row has a plot for each factor, showing its desirability trace. The profiler also contains a Desirability column, which graphs desirability on a scale from 0 to 1 and has an adjustable desirability function for each y variable. The overall desirability measure appears to the left of the row of desirability traces. Figure 5.7 The Prediction Profiler

2. Adjust the prediction traces of the factors and find a Stretch value that is closer to the target. Drag the middle handle on the desirability function vertically to change the target value. Or, find good factor settings by clicking the red triangle on the Prediction Profiler title bar and selecting Maximize Desirability. This command adjusts the profile traces to produce the response value closest to the specified target (the target given by the desirability function). The range of acceptable values is determined by the positions of the upper and lower handles.

Silane from 50 to 41.2 to 0.8 Prediction Profiler for a Response Surface Analysis See the chapter “Standard Least Squares: Exploring the Prediction Equation” of JMP Statistics and Graphics Guide for further discussion of the Prediction Profiler.3 to 2. Using a Response Surface Plot (Contour Profiler) Another way to look at the response surface is to use the Contour Profiler.5 Response Surface Designs—A Box-Behnken Design: The Tennis Ball Example 107 5 Surface Figure 5. as shown in Figure 5. and add contours at specified intervals.09 raised the predicted response from 396 to the target value of 450.8 shows the result of the most desirable settings. This example shows the profile to Silica and Sulphur for a fixed value of Silane. . Click the red triangle on the Response title bar and select Factor Profiling > Contour Profiler to display the interactive contour profiler. especially when there are multiple responses. Options on the Contour Profiler title bar can be used to set the grid density.9. The sliders for each factor set values for Current X and Current Y. as shown in the contour plot in Figure 5. Figure 5.5.9. request a surface plot (mesh plot). and Sulfur from 2. The contour profiler is useful for viewing response surfaces graphically. Changing the settings of Silica from 1.98.

Figure 5.10 shows the Contour profile when the Current x values have Lo and Hi limits in effect.108 5 Response Surface Designs—A Box-Behnken Design: The Tennis Ball Example Figure 5. .10 Contour Profiler with High and Low Limits The Prediction Profiler and the Contour Profiler are discussed in more detail in the chapter “Standard Least Squares: Exploring the Prediction Equation” of JMP Statistics and Graphics Guide.9 Contour Profiler for a Response Surface Analysis Figure 5. drawing shaded regions on the contour.

You can clearly see the 12 points midway between the vertices. Options available on the spin platform draw rays from the center to each point. leaving three points in the center. and suppress the x. see the JMP Statistics and Graphics Guide. inscribe the points in a box. Figure 5. . and z axes.11 illustrates the three Box-Behnken design columns. y.5 Response Surface Designs—A Box-Behnken Design: The Tennis Ball Example 109 5 Surface Viewing the Geometry of a Box-Behnken Design The geometric structure of a design with three effects is seen by using the Spinning Plot platform. The spinning plot shown in Figure 5.11 Spinning Plot of a Box-Behnken Design for Three Effects For details on how to create a spinning plot.

” p. follow the step in Figure 5. . Custom Design.” p. To add factors. or click the Response Surface Design button on the JMP Starter DOE page. RSM can find it. If a maximum or minimum exists inside the factor region. 114 Entering Responses and Factors The steps for entering responses are the same in Screening Design. If Needed. These steps are outlined in “Entering Responses and Factors in the Custom Designer. 111 • “Specifying Axial Value (Central Composite Designs Only). 112 • “Specifying Output Options. 113 • “Viewing the Design Table. Then.” p. Using the response surface designer. 57 The steps for entering factors in a response surface design are unique to this design.” p. In industrial applications.12. RSM designs involve a small number of factors. select DOE > Response Surface Design.” p. Some of these designs also allow blocking.” p. Space Filling Design. To start a response surface design. Response surface designs are useful for modeling and analyzing curved surfaces. follow the steps below: • “Entering Responses and Factors. and Full Factorial Design.110 5 Response Surface Designs—Creating a Response Surface Design Creating a Response Surface Design Response Surface Methodology (RSM) is an experimental technique invented to find the optimal response within specified ranges of the factors. Mixture Design. This is because the required number of runs increases dramatically with the number of factors. 110 • “Choosing a Design. The quadratic terms in these equations model the curvature in the true response function. Response Surface Design. 113 • “Continuing the Analysis. These designs are capable of fitting a second-order prediction equation for the response.” p. you choose to use well-known RSM designs for two to eight continuous factors.

This is sometimes useful when it is desirable to avoid extreme points due to engineering considerations. .5 Response Surface Designs—Creating a Response Surface Design 111 5 Surface Figure 5. Choosing a Design Highlight the type of response surface design you would like to use and click Continue.12 Entering Factors into a Response Surface Design Click Continue to proceed to the next step. Figure 5. The price of this characteristic is the higher uncertainty of prediction near the vertices compared to the central composite design.13 Choose a Design Type The Response Surface designer provides the following types of designs: Box-Behnken Designs The Box-Behnken design has only three levels per factor and has no points at the vertices of the cube defined by the ranges of the factors.

0 in the text box instructs JMP to place the axial value on the face of the cube defined by the factors. If this factor range cannot be practically achieved. • User Specified uses the value entered by the user. JMP re-scales the whole design so that the axial points are at the low and high ends of the range (the axials are –1 and 1 and the factorials are shrunken based on that scaling). • For orthogonal designs. Enter that value into the Axial Value text box. • Orthogonal makes the effects orthogonal in the analysis. which can be any value greater than zero. You have the flexibility to enter the values you want to use. It supplies default axial scaling information. it is recommended that you choose On Face or specify your own value. it is recommended that you choose On Face or specify your own value. which controls how far out the axial points are.14 Display and Modify the Central Composite Design • Rotatable makes the variance of prediction depend only on the scaled distance from the center of the design. you see the panel in Figure 5. This causes the axial points to be more extreme than the range of the factor. Entering 1.14. . This causes the axial points to be more extreme than the –1 or 1 representing the range of the factor. When checked. If this factor range cannot be practically achieved. These properties of central composite designs relate to the number of center points in the design and to the axial values: • Uniform precision means that the number of center points is chosen so that the prediction variance at the center is approximately the same as at the design vertices. Specifying Axial Value (Central Composite Designs Only) When you select a central composite (CCD-Uniform Precision) design and then click Continue.112 5 Response Surface Designs—Creating a Response Surface Design Central Composite Designs The response surface design list contains two types of central composite designs: uniform precision and orthogonal. • On Face leaves the axial points at the end of the -1 and 1 ranges. the number of center points is chosen so that the second order parameter estimates are minimally correlated with the other parameter estimates. Figure 5. click the box beside Inscribe. If you would like to inscribe the design.

Choices are: • Keep the Same—the rows (runs) in the output table will appear as they do in the Design panel.15 Select the Output Options • Run Order—Lets you designate the order you want the runs to appear in the data table when it is created. Figure 5.5 Response Surface Designs—Creating a Response Surface Design 113 5 Surface Specifying Output Options Use the Output Options panel to specify how you want the output data table to appear. • Number of Replicates—Specify the number of times to replicate the entire design. . • Randomize—the rows (runs) in the output table will appear in a random order. • Sort Right to Left—the rows (runs) in the output table will appear sorted from right to left. When the options are correctly set up. Type the number of times you want to replicate the design in the associated text box. • Number of Center Points—Specifies additional runs placed at the center points. including centerpoints. • Randomize within Blocks—the rows (runs) in the output table will appear in random order within the blocks you set up. click Make Table. • Sort Left to Right—the rows (runs) in the output table will appear sorted from left to right.16. Viewing the Design Table Now you have a data table that outlines your experiment. as described in Figure 5. One replicate doubles the number of runs.

It shows all the codings with “+” for high.17 Running the Script After clicking Run Model in the dialog. Right-click it and select Run Script (Figure 5.16 The Design Data Table The column called Pattern identifies the coding of the factors. running the experiment. Runs are in a random order.18). The name of the table is the design type that generated it. the pattern value shows the factor coding of the point. There are two center points per replicate. The Y column is for recording experimental results. . and recording your results in the design table’s Y column. Figure 5. The data table contains a script labeled Model. This script allows you to easily fit a model using the values in the design table.17) to fit the model. Continuing the Analysis. “a” and “A” for low and high axial values. run a fit model analysis on the data. “–” for low factor. Pattern is suitable to use as a label variable in plots because when you hover over a point in a plot of the factors. review the analysis. and “0” for midrange. If Needed After creating and viewing the design table. (Figure 5.114 5 Response Surface Designs—Creating a Response Surface Design Figure 5.

18 Fitting the Model .5 Response Surface Designs—Creating a Response Surface Design 115 5 Surface Figure 5.

116 5 Response Surface Designs—Creating a Response Surface Design .

Full factorial designs are the most conservative of all design types. . so full factorial designs are too expensive to run for most practical purposes. and a four-level factor has 2 x 3 x 4 = 24 runs. When there are three factors. the design points are the vertices of a hypercube. a three-level factor. you perform an experimental run at every combination of the factor levels.6 Full Factorial Designs 6 Factorial A full factorial design contains all possible combinations of a set of factors. Factorial designs with only two-level factors have a sample size that is a power of two (specifically 2f where f is the number of factors). Unfortunately. For more factors. In full factorial designs. The full factorial designer supports both continuous factors and categorical factors with up to nine levels. a factorial experiment with a two-level factor. For example. There is little scope for ambiguity when you are willing to try all combinations of the factor settings. the factorial design points are at the vertices of a cube as shown in the diagram below. This is the most conservative design approach. the sample size grows exponentially in the number of factors. but it is also the most costly in experimental resources. The sample size is the product of the numbers of levels of the factors.

. . . . . . . . . . . . . . . . . . . . . . . . . . Selecting Output Options . .6 Contents The Five-Factor Reactor Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Entering Responses and Factors . . . . . . . . . . . . . . 119 125 125 126 126 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Factorial Design . . . . . . . . . . . . . . . . . . . . . . . . Making the Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3) contains a run for every combination of high and low values for the five variables.jmp. the table has an empty Y column named Percent Reacted for entering response values when the experiment is complete. The completed dialog is shown in Figure 6. Catalyst. and Concentration) are all two-level continuous factors. Hunter. Figure 6. In the Sample Data folder. 81). 5.jmp.jmp. giving 32 runs. Click Make Table. . open Design Experiment > Reactor Factors. open the folder Sample Data that was installed when you installed JMP. The design data table (Figure 6. In the Sample Data folder. Suppose you have used the screening designer to investigate the effects of five factors on the percent reaction of a chemical process (see “Screening Designs. 7.1 Full-Factorial Example Response and Factors Panels 6. Within this folder. Initially.1. 1. Click Continue to see Output Options panel. Click the red triangle icon on the Full Factorial Design title bar and select Load Factors. there are 25=32 runs. 4. The factors (Feed Rate. adapted from Meyer et al. Temperature. Stir Rate. The values in your table may be different from those shown below.jmp to load the responses by opening the Sample Data folder that was installed with JMP. Open Reactor Response.” p. open Design Experiment > Reactor Response. This covers all combinations of a five factors with two levels each. Click the red triangle icon on the Full Factorial Design title bar and select Load Responses. shows a five-factor reactor example. 2. open Design Experiment > Reactor 32 Runs.6 Full Factorial Designs—The Five-Factor Reactor Example 119 6 Factorial The Five-Factor Reactor Example The following example. To follow along with this example. Open Reactor Factors. 3. (1996) and Box. and Hunter (1978). Since there are five variables.jmp to load the responses by opening the Sample Data folder that was installed with JMP. A full factorial design includes runs for all combinations of high and low factors for the five variables. Select DOE > Full Factorial Design.

3 Reactor 32 Runs.jmp Begin the analysis with a quick look at the data before fitting the factorial model. 2. Select Analyze > Distribution. open Design Experiment > Reactor 32 Runs. Click the red triangle icon on the Percent Reacted title bar and select Normal Quantile Plot. Highlight Percent Reacted and click Y. The results are shown in Figure 6. Within this folder.jmp. 3. . Figure 6.4.120 6 Full Factorial Designs—The Five-Factor Reactor Example Figure 6.2 25 Factorial Reactor Data To see the completed experiment and continue following this example. open the folder Sample Data that was installed when you installed JMP. 1. Then click OK. Columns.

5. Check the boxes for the main effects of the factors as shown in Figure 6. The probability to enter a factor (Prob to Enter) in the model should be 0. 2.1.6. and brings up the Stepwise control panel. Click Go. . Figure 6.4 Distribution of Response Variable for Reactor Data Start the formal analysis with a stepwise regression. 6. 1. The data table has a script stored with it that automatically defines an analysis of the model that includes main effects and all two-factor interactions.6 Full Factorial Designs—The Five-Factor Reactor Example 121 6 Factorial Figure 6. 3.05. To do this. Click the red triangle icon next to the Fit Model script and select Run Script. The probability to remove a factor (Prob to Leave) should be 0.5 Run JSL Script for Stepwise Regression 4. make sure the menu beside Direction specifies Mixed. A useful way to use the Stepwise platform is to check all the main effects in the Current Estimates table.

Figure 6. The end result is shown in Figure 6.7 Model After Mixed Stepwise Regression 7. The Model Specification window that appears is automatically set up with the appropriate effects (Figure 6.6 Starting Model For Stepwise Process The mixed stepwise procedure removes insignificant main effects and adds important interactions. Note that the Feed Rate and Stir Rate factors are no longer in the model. .7. Click the Make Model button.8).122 6 Full Factorial Designs—The Five-Factor Reactor Example Figure 6.

. as shown in Figure 6. The figure on the right in Figure 6. Open the Prediction Profiler by clicking the red triangle on the Response Percent Reacted title bar and selecting Factor Profiling > Profiler. Figure 6.9 shows the actual by predicted plot for the model. The figure on the left in Figure 6. The predicted model covers a range of predictions from 40% to 95% reacted.9 Actual by Predicted Plot and Prediction Model Estimates The factor Prediction Profiler also gives you a way to compare the factors and find optimal settings.10.3311%. which is more than an order of magnitude smaller than the range of predictions. All effects selected by the stepwise process are highly significant. Click Run Model to see the analysis for a candidate prediction model (Figure 6.8 Fitting a Prediction Model 8.9). This is strong evidence that the model has good predictive capability. 1.6 Full Factorial Designs—The Five-Factor Reactor Example 123 6 Factorial Figure 6. The size of the random noise as measured by the RMSE is only 3.9 shows a table of model coefficients and their standard errors (labeled Parameter Estimates).

12. 101.12 Viewing the Maximum Desirability . Click the red triangle on the Prediction Profiler title bar and select Maximize Desirability to see the profiler in Figure 6. The Prediction Profiler is discussed in more detail in the chapter “Response Surface Designs. and in the chapter “Standard Least Squares: Exploring the Prediction Equation” of JMP Statistics and Graphics Guide.124 6 Full Factorial Designs—The Five-Factor Reactor Example Figure 6. Figure 6.11 Viewing the Profiler 2. Figure 6.” p.11 shows the profiler’s initial display.10 Selecting the Profiler Figure 6.

Percent Reacted increases from 65. 125 • “Selecting Output Options.6 Full Factorial Designs—Creating a Factorial Design 125 6 Factorial The plot of Desirability versus Percent Reacted shows that the goal is to maximize Percent Reacted. To remove a level.” p. level 2 . Then. then press the Return or Enter key on the keyboard. To add factors. see Figure 6.5 at the center of the factor ranges to 95.9. Double-click to edit the factor name.” p. click Continue. Desirability increases linearly as the Percent Reacted increases.” p. or click the Full Factorial Design button on the JMP Starter DOE page. Figure 6. click either the Continuous button or the Categorical button and select a factor type. select DOE > Full Factorial Design. The reaction is unfeasible economically unless the Percent Reacted is above 90%. press the delete key on the keyboard. 126 Entering Responses and Factors The steps for entering responses are outlined in “Entering Responses and Factors in the Custom Designer. click it. 126 • “Making the Table.875 at the most desirable setting. therefore the Desirability for values less than 90% decreases and finally becomes zero. Creating a Factorial Design To start a full factorial design. Click to enter values or change the level names. 57 The steps for entering factors in a full factorial design are unique to this design. .945 when Catalyst and Temperature are at their highest settings and Concentration is at its lowest setting.13 Entering Factors in a Full Factorial Design To enter factors. follow the steps below: • “Entering Responses and Factors. When you finish adding factors. The maximum Desirability is 0.” p.13.

14 appears. a minus sign represents low levels level numbers represent values of categorical factors This script allows you to easily fit a model using the values in the design table. . Figure 6. One replicate doubles the number of runs. values in the Pattern column describe the run each row represents for continuous factors. • Randomize—the rows (runs) in the output table will appear in a random order. • Number of Center Points—Specifies additional runs placed at the center of each continuous factor’s range. • Number of Replicates—Specify the number of times to replicate the entire design. Type the number of times you want to replicate the design in the associated text box.126 6 Full Factorial Designs—Creating a Factorial Design Selecting Output Options Use the Output Options panel to specify how you want the output data table to appear: • Run Order—Lets you designate the order you want the runs to appear in the data table when it is created.14 Factorial Design Table The name of the table is the design type that generated it. including centerpoints. Choices are: • Keep the Same—the rows (runs) in the output table will appear as they do in the Design panel. a plus sign represents high levels for continuous factors. Making the Table When you click Make Table. • Sort Right to Left—the rows (runs) in the output table will appear sorted from right to left. the table shown in Figure 6. • Sort Left to Right—the rows (runs) in the output table will appear sorted from left to right.

With mixtures. For the extreme vertices approach you can supply a set of linear inequality constraints limiting the geometry of the mixture factor space. a factor's value is its proportion in the mixture. The properties of a mixture are almost always a function of the relative proportions of the ingredients rather than their absolute amounts. For example. and the interpretation of the results. which falls between zero and one. You can rotate the plane to see the triangle face-on and see the points in the form of a ternary plot. extreme vertices. The interpretation of screening experiments is simple. it is impossible to vary one factor independently of all the others. and lattice. consider three factors in a 3-D graph. This simple fact has a profound effect on every aspect of experimentation with mixtures: the factor space. The sum of the proportions in any mixture recipe is one (100%). x3 triangular feasible region x2 x1 . You choose among several classical mixture design approaches. Designs for mixture experiments are fundamentally different from those for screening.7 Mixture Designs 7 Mixture The mixture designer supports experiments with factors that are ingredients in a mixture. When you change the proportion of one ingredient. over the course of an experiment. the design properties. Screening experiments are orthogonal. Because the proportions sum to one. In experiments with mixtures. such as simplex. The plane where the sum of the three factors sum to one is a triangle-shaped slice. because the effects of the factors on the response are separable. the proportion of one or more other ingredients must also change to compensate. the setting of one factor varies independently of any other factor. The feasible region for a mixture takes the form of a simplex. mixture designs have an interesting geometry. That is.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating the Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fitting Mixture Designs . . . . . . . . . . . . . . . . . . . Simplex Centroid Design Examples . . . . . . . . . . . . . . . . . . . . . . 129 129 130 131 131 133 135 135 136 138 139 139 140 141 142 143 143 143 144 145 146 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . An Extreme Vertices Example with Linear Constraints . . . . . . . . . . . . . . . . . Creating Ternary Plots . . . . . . . . . . . . . . Plotting a Mixture Response Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Extreme Vertices Method: How It Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using the Simplex Centroid Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 Contents Choosing a Mixture Design Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating the Design . . . . . . . . . . . . . . . . . . . . . Using the ABCD Design . . . . . . . . . . . . . . . A Chemical Mixture Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using the Simplex Lattice Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Running a Mixture Model . . . . . . . . . . . . . Creating the Design . . . . . Viewing the Prediction Profiler . . . . . . Using the Extreme Vertices Design . . . . . An Extreme Vertices Example with Range Constraints . . . . . . . . Understanding Response Surface Reports . Understanding Whole Model Tests and ANOVA Reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating an Optimal Mixture Design . .

3. 2.7 Mixture Designs—Choosing a Mixture Design Type 129 7 Mixture Choosing a Mixture Design Type After selecting DOE > Mixture Design and setting up the responses and factors. following instructions regarding the Design Generation panel in “Selecting the Number of Runs. Select DOE > Mixture Design. since it handles constraints on the values of the factors.” p. Figure 7. After your responses and factors are set up. Enter factors and responses. shown Figure 7. Click the Optimal button. To create an optimal mixture design: 1. you see the mixture designer.” p. 57. 5. • Simplex Lattice You specify how many levels you want on each edge of the grid. • ABCD Design This approach by Snee (1975) generates a screening design for mixtures. From here.1. • Extreme Vertices You specify linear constraints or restrict the upper and lower bounds to be within the 0 to 1 range. 4.1 Mixture Design Selection Dialog The following sections show examples of each mixture design type. 62 and the prediction variance profile in “Understanding the Design and . Add effects to the model using the instructions below. Then click Make Design and complete the remainder of the design. Creating an Optimal Mixture Design Creating an optimal mixture design invokes the custom designer with all the mixture variables already defined. select one of the following types of design: • Optimal Choosing Optimal invokes the custom designer with all the mixture variables already defined. The extreme vertices design is the most flexible. click Continue. The steps for entering responses are outlined in “Entering Responses and Factors in the Custom Designer. • Simplex Centroid You specify the degree up to which the factor combinations are to be made.

Add powers of continuous factors to the model effects Using the Simplex Centroid Design A simplex centroid design of degree k with n factors is composed of mixture runs with • all one factor • all combinations of two factors at equal levels . The appropriate value is added. 63. as shown in Figure 7.” p. the Model panel lists only the main effects corresponding to the factors you entered. Highlight the factors and effects you want to cross. Click the Cross button.2.2 The Model Panel However.130 7 Mixture Designs—Using the Simplex Centroid Design Profilers. For example. if you do not want to limit your model to main effects. if you want to cross the factors X1 and X2 with the effect X3. --or. 4th. 3rd. if you your factors are X1 and X2 and you click Interactions > 2nd. The appropriate value is added. or 5th. then X1*X3 and X2*X3 are added to the list of model effects. 2. you can add factor interactions or powers of continuous factors to the model. or 5th. to cross effects-1. 4th. Figure 7. To add specific factor types to the model: Action Instructions Add interaction terms to the model effects Click the Interactions button and select 2nd. X1*X2 is added to the list of model effects. For example. Click the Powers button and select 2nd. These are also known as additional effects you want JMP to estimate. 3rd. Adding Effects to the Model Initially.

Simplex Centroid Design Examples The table of runs for a design of degree 1 with three factors (left in Figure 7. Click Make Table. 2. 5. then mix equal proportions of K ingredients at a time to the specified limit. JMP will run each ingredient without mixing. • Sort Right to Left—the rows (runs) in the output table will appear sorted from right to left. 9.3) shows runs for each single ingredient followed by the center point. After your responses and factors are set up. 4. View factor settings. The table of runs to the right is for three factors of degree 2. Enter factors and responses. Choices are: • Keep the Same—the rows (runs) in the output table will appear as they do in the Design panel. Click the Simplex Centroid button. Specify run order. 3. • Randomize—the rows (runs) in the output table will appear in a random order. Creating the Design To create a simplex centroid design: 1. • Randomize within Blocks—the rows (runs) in the output table will appear in random order within the blocks you set up. . The first three runs are for each single ingredient. Select DOE > Mixture Design. the second set shows each combination of two ingredients in equal parts. 8. Enter the number of ingredients in the box labeled K. 57. A center point run with equal amounts of all the ingredients is always included. Specify number of replicates. 6. • Sort Left to Right—the rows (runs) in the output table will appear sorted from left to right. The steps for entering responses are outlined in “Entering Responses and Factors in the Custom Designer. click Continue. and the last run is the center point. which is the order you want the runs to appear in the data table when it is created.7 Mixture Designs—Using the Simplex Centroid Design 131 7 Mixture • all combinations of three factors at equal levels • and so on up to k factors at a time combined at k equal levels.” p. 7.

Click Continue.132 7 Mixture Designs—Using the Simplex Centroid Design Figure 7. 4. 3. 2. Use the default value. 4. Choose DOE > Mixture Design. Enter three mixture factors. Figure 7.5. in the K box.4.3 Three-Factor Simplex Centroid Designs of Degrees 1 and 2 To generate the set of runs in Figure 7. 5. Enter 1 in the K box. Enter five factors and click Continue. 3. Choose DOE > Mixture Design. 4. and click Simplex Centroid to see the design on the left in Figure 7. . Click the Back button and enter 2 in the K box. 5. Click Make Table to see the 31-run JMP data table shown in Figure 7.4 Create Simplex Centroid Designs of Degrees 1 and 2 As another example: 1. 2. Click Simplex Centroid. Then click Simplex Centroid to see the design on the right in Figure 7.4.3: 1.

7 Mixture Designs—Using the Simplex Lattice Design 133 7 Mixture Figure 7.1) and click Simplex Lattice. the simplex lattice design does not necessarily include the center point. 4. Select DOE > Mixture Design. The steps for entering responses are outlined in “Entering Responses and Factors in the Custom Designer. with their corresponding geometric representations. In contrast to the simplex centroid design. 57.” p. 3. Enter factors and responses. Figure 7. .5 List of Factor Settings for Five-Factor Simplex Centroid Design Using the Simplex Lattice Design The simplex lattice design is a space filling design that creates a triangular grid of runs. and 5. 4. Click Continue. The design is the set of all combinations where the factors’ values are i / m.6 shows the runs for three-factor simplex lattice designs of degrees 3. Specify the number of levels you want in the Mixture Design Type dialog (Figure 7. To create simplex lattice designs: 1. where i is an integer from 0 to m such that the sum of the factors is 1. 2.

the runs creep across the hyper-triangular region and fill the space in a grid-like manner.7 lists the runs for a simplex lattice of degree 3 for five effects. In the five-level example.134 7 Mixture Designs—Using the Simplex Lattice Design Figure 7. Order (Degree) 3 .7 JMP Design Table for Simplex Lattice with Five Variables. and 5 Figure 7.6 Three-Factor Simplex Lattice Designs for Factor Levels 3. Figure 7. 4.

Use the text boxes that appear to define a linear combination of factors to be greater or smaller than some constant. Select DOE > Mixture Design. You usually want to add points between the vertices. 4. • Sort Left to Right—the rows (runs) in the output table will appear sorted from left to right. Remember to enter the upper and lower limits in the factors panel.” p. Sometimes other ingredients need range constraints that confine their values to be greater than a lower bound or less than an upper bound. The centroid point for two neighboring vertices joined by a line is a second degree centroid because a line is two dimensional.7 Mixture Designs—Using the Extreme Vertices Design 135 7 Mixture Using the Extreme Vertices Design The extreme vertices design accounts for factor limits and selects vertices and their averages (formed by factor limits) as design points. enter the degree at which you would like the centroid point added. If you specify an extreme vertices design but give no constraints. and so on. 57. The centroid point for vertices sharing a plane is a third degree centroid because a plane is three dimensional. The extreme vertices design finds the corners (vertices) of a factor space constrained by limits specified for one or more of the factors. and centroid points of various degrees can be added. Click Continue. which defines a linear combination of factors to be greater or smaller than some constant. Enter factors and responses. The property that the factors must be non-negative and must add up to one is the basic mixture constraint that makes a triangular-shaped region. If you have linear constraints. 3. and lower bounds on the factor values. Range constraints chop off parts of the triangular-shaped (simplex) region to make additional vertices. The factor settings are displayed. Additional limits in this type of design are usually in the form of range constraints. The centroid point is the average of points that share a constraint boundary. (Optional) Change the run order to the order you want the runs to appear in the data table when it is created. The average of points that share a constraint boundary is called a centroid point. click the Linear Constraints button for each constraint you want to add. In the Degree text box. . they are often the best places to use as design points in an experiment. 6. It is also possible to have a linear constraint. Choices are: • Keep the Same—the rows (runs) in the output table will appear as they do in the Design panel. and because the vertices represent extreme conditions of the operating environment. The geometric shape of a region bound by linear constraints is called a simplex. 2. These steps are outlined in “Entering Responses and Factors in the Custom Designer. a simplex centroid design results. 5. Creating the Design 1. Click Extreme Vertices. 7. • Randomize—the rows (runs) in the output table will appear in a random order. upper bounds.

3. 4. Select DOE > Mixture Design.8. 1.Click Make Table. • Randomize within Blocks—the rows (runs) in the output table will appear in random order within the blocks you set up. Figure 7. Add two factors and given them the values shown in Figure 7. 9.8). (Optional) Click Find Subset to generate the optimal subset having the number of runs specified in sample size box described in Step 8. 10. (Optional) Alter the sample size by typing into the text box beside Choose desired sample size. An Extreme Vertices Example with Range Constraints The following example design table is for five factors with the range constraints shown in Figure 7.8 Example of Five-factor Extreme Vertices . 2. where the ranges are smaller than the default 0 to 1 range. Click Continue.136 7 Mixture Designs—Using the Extreme Vertices Design • Sort Right to Left—the rows (runs) in the output table will appear sorted from right to left.8. Enter 4 in the Degree text box (Figure 7. 8.

10.7 Mixture Designs—Using the Extreme Vertices Design 137 7 Mixture 5. Enter 4 in the Degree text box and click the Exreme Vertices button.10 JMP Design Table for 10-Run Subset of the 13 Current Runs . Figure 7.In the sample size text box. 9. Now the resulting design (Figure 7.10) will be the optimal 10-run subset of the 116 current runs. Figure 7. Click the Exreme Vertices button. 6.9 JMP Design Table for Extreme Vertices with Range Constraints Now let’s go back and enter a different sample size (number of runs). enter 10 as the desired size. This is useful when the extreme vertices design generates a large number of vertices.9 shows a partial listing of the resulting table. Figure 7. then click Continue. 11.Click Find Subset to generate the optimal subset having the number of runs specified. Click Make Table. Click Back. Select Sort Left to Right from the Run Order menu. 7. 8.

. Click Continue.7 90 ≤ 85*X1 + 90*X2 + 100*X3 85*X1 + 90*X2 + 100*X3 ≤ 95 .138 7 Mixture Designs—Using the Extreme Vertices Design An Extreme Vertices Example with Linear Constraints Consider the classic example presented by Snee (1979) and Piepel (1988). X2. for details. 5. Change the run order to Sort Right to Left. 3. and keep the sample size at 13 to see the 13-run factor settings like those shown on the right in Figure 7.7*X1 + X3 Enter the upper and lower limits in the factors panel. 4. Click the Linear Constraints button three times. This example is best understood by viewing the design as a ternary plot. as shown in Figure 7. The ternary plot shows how close to one a given component is by how close it is to the vertex of that variable in the triangle. Click Make Table.4 ≤ 0. X1.11 Constraints 1. 2. See “Creating Ternary Plots.11. 6.” p. This example has three factors. and X3.11.5 X2 ≥ 0.1 X2 ≤ 0.1 X1 ≤ 0. Figure 7. Click the Extreme Vertices button.13. Enter the constraints as shown in Figure 7. with five individual factor bound constraints and three additional linear constraints: X1 ≥ 0. 140.7 X3 ≤ 0.

Then. Wheeler. • Sort Right to Left—the rows (runs) in the output table will appear sorted from right to left. Snee (1975) has shown that it can also be useful to have the centroids of the edges and faces of the feasible region. . The method for doing centroid points is by Piepel (1988). JMP uses the CONSIM algorithm developed by R. • Randomize—the rows (runs) in the output table will appear in a random order. The steps for entering responses are outlined in “Entering Responses and Factors in the Custom Designer. Select DOE > Mixture Design. The above algorithm creates the vertices of the feasible region in the simplex defined by the factor constraints. 3. click Continue. If not. To create an ABCD design: 1. which is the order you want the runs to appear in the data table when it is created. the extreme vertices design is constructed using the XVERT method developed by Snee and Marquardt (1974) and Snee (1975). Click the ABCD button.7 Mixture Designs—Using the ABCD Design 139 7 Mixture Extreme Vertices Method: How It Works If there are linear constraints. named CONAEV. 57. After your responses and factors are set up. View factor settings. If so. Enter factors and responses. The XVERT method first creates a full 2nf – 1 design using the given low and high values of the nf – 1 factors with smallest range. The algorithm generates all possible combinations of the boundary conditions and then averages over the vertices generated on the first step.” p. it drops the vertices and calculates new ones. Using the ABCD Design This approach by Snee (1975) generates a screening design for mixtures. Choices are: • Keep the Same—the rows (runs) in the output table will appear as they do in the Design panel. and decrements or increments each of the other factors in turn by the same amount. • Randomize within Blocks—the rows (runs) in the output table will appear in random order within the blocks you set up. 5. a simplex centroid method generates combinations of vertices up to a specified order. 4. It keeps the point if it is in that factor’s range. After the vertices are found. If there are no linear constraints (only range constraints). • Sort Left to Right—the rows (runs) in the output table will appear sorted from left to right.E. Specify run order. it increments or decrements it to bring it within range. described in Snee (1979) and presented by Piepel (1988) as CONVRT. A generalized n-dimensional face of the feasible region is defined by nf–n of the boundaries and the centroid of a face defined to be the average of the vertices lying on it. The method is also described in Cornell (1990. it computes the value of the one factor left out based on the restriction that the factors’ values must sum to one. Appendix 10a). keeping the points that still satisfy the initial restrictions. 6. However. 2. The method combines constraints and checks to see if vertices violate them.

Each constraint is a line. Create the design by choosing DOE > Mixture Design and making the design data table. The ternary plot in Figure 7. . as well as two inactive (redundant) constraints. Figure 7. 0) (1/2. Click Make Table. 8. 1. 0) (1/3. 2. 0. Specify number of replicates. The ternary plot in Figure 7. Complete the window and click OK. 0. The plot to the left in Figure 7. six vertices.12 illustrates a ternary plot. 1/3. 0. 0) X3 (0.13. Choose Graph > Ternary Plot. The Piepel (1979) example referenced in “An Extreme Vertices Example with Linear Constraints. 1. 1) X2 X1 X3 To view your mixture design as a ternary plot: 1. 1/2. The area that satisfies all constraints is the shaded feasible area.140 7 Mixture Designs—Creating Ternary Plots 7.12 Ternary Plot (left) and Tetrary Plot (right) for Mixture Design X1 (1. Creating Ternary Plots A mixture problem in three components can be represented in two dimensions because the third component is a linear function of the others. There are six active constraints. 1/3) (1.13 shows how close to one a given component is by how close it is to the vertex of that variable in the triangle.8) X2 (0. and six centroid points shown on the plot. 138 is best understood by the ternary plot shown in Figure 7.” p. 3.13 shows how close to one a given component is by how close it is to the vertex of that variable in the triangle.

is located in the left panel of the table.13 Ternary Plot Showing Piepel Example Constraints Fitting Mixture Designs When fitting a model for mixture designs. called the Scheffé polynomial (Scheffé 1958). Remember that to fit the model. you must take into account that all the factors add up to a constant. Create the design by choosing DOE > Mixture Design and making the design data table. The recommended response surface model. This table property. and thus a traditional full linear model will not be fully estimable. called Model. Right-click the model and select Run Script to launch the Model Specification dialog. 2. does the following: • suppresses the intercept • includes all the linear main-effect terms • excludes all the square terms (such as X1*X1) • includes all the cross terms (such as X1*X2) To fit a model: 1.7 Mixture Designs—Fitting Mixture Designs 141 7 Mixture Figure 7. which is . 3. so either assign responses or click the red triangle menu and select Simulate Responses before making the table. the Y column in the data table must contain values. The design data table stores the model in the data table as a table property.

Figure 7. the parameters are easy to interpret (Cornell 1990). The hidden-intercept property also causes the R2 to be reported with respect to the intercept model rather than reported as missing. and testing that they are equal. . The coefficients on the cross terms indicate the curvature across each edge of the factor space. If it finds a hidden intercept. In this model. which are described below. JMP traditionally tests that all the parameters are zero except for the intercept. Click Run Model. The coefficients on the linear terms are the fitted response at the extreme points where the mixture is all one factor. it does the whole model test with respect to the intercept model rather than a zero-intercept model.14 Fitting a Mixture Design The model’s report may contain several sections of interest. in the sense that a linear combination of effects is a constant. 4. Your values may be different from those shown below. ANOVA reports. JMP looks for a hidden intercept.142 7 Mixture Designs—Fitting Mixture Designs automatically filled with the saved model. Understanding Whole Model Tests and ANOVA Reports In a whole-model ANOVA table. including the whole model tests. and response surface reports. This test is equivalent to testing that all the parameters are zero except the linear parameters. In a mixture model without an intercept.

A Chemical Mixture Example Three plasticizers (p1. If you want a contour plot of the surface. and enter the high and low constraints as shown in Figure 7. use the three default factors.252.15 Setting Up Factors . but no square terms.15. Import the factors by opening the Sample Data folder that was installed when you installed JMP and selecting Design Experiment > Plastifactors. It can do this for any combination yielding a constant and involving the last response surface term.274. not mixture models.5%. collapsing on the last response surface term to calculate critical values for the optimum. Creating the Design To create Cornell’s mixture design in JMP: 1. Or.jmp. 2. These reports were originally designed for full response surfaces. and p3. the individual plasticizers are restricted by the following constraints: 0. p2. and p3) comprise 79. 1990).5% of the vinyl used for automobile seat covers (Cornell. The contour plot feature of these reports does not fold to handle mixtures. • Create a separate contour plot by selecting Graph > Contour Plot. Within this 79. and 0.409 ≤ x1 ≤ 0. name them p1. In the Factors panel. if JMP encounters a no-intercept model and finds a hidden intercept with linear response surface terms. Select DOE > Mixture Design.” it creates additional reports that analyze the resulting fitted response surface. you can do any of the following: • Save the model prediction formula and then create a ternary plot by selecting Graph > Ternary Plot. Figure 7. • Refit the model using a full response surface that omits the last factor. then it folds its calculations.151 ≤ x3 ≤ 0. and add points to make the plot less granular. However.849. p2. 0 ≤ x2 ≤ 0. load the factors by clicking the red triangle on the Mixture Design title bar and selecting Load Factors.7 Mixture Designs—A Chemical Mixture Example 143 7 Mixture Understanding Response Surface Reports When JMP encounters effects that are marked as response surface effects “&RS.

17 Plasticizer. Enter 3 in the Degree text box. which runs a script that creates a completed dialog. Figure 7. Right-click the table property named Mixture RSM and select Run Script. select Rows > Add Rows.jmp contains a column called Pred Formula Y. Add an extra five design runs by duplicating the vertex points and center point.17). 6. JMP uses the 9 factor settings to generate a JMP table (Figure 7. and click OK. This gives a total of 14 rows in the design table. Click Make Table.16).144 7 Mixture Designs—A Chemical Mixture Example 3. Figure 7. enter 5 as the number of rows to add. 4. 8. Then select Edit > Paste. Open Plasticizer. highlight the rows you want to duplicate. Plasticizer. Then click Run Model. To do this. Open the Sample Data folder that was installed when JMP was installed. 5. This column was added by clicking the . Click Exreme Vertices.jmp Running a Mixture Model Now run the mixture model: 1. Click Continue.jmp to see the results (Y values) that Cornell obtained (Figure 7. select Edit > Copy.16 Mixture Data for Analysis 7. 2.

The crossed effects show as curvature in the prediction traces.20927). To see the prediction formula.363 + p3*p1*1464. click the red triangle in the Response Y title bar and select Factor Profiling > Profiler.570299 occurs at point (0. If the profiler is not visible.2177 Your values may be different from those shown below.26923. Drag one of the vertical reference lines.6615 for p1. 0.7 Mixture Designs—A Chemical Mixture Example 145 7 Mixture red triangle in the Response Y title bar and selecting Save Columns > Prediction Formula.126 for p2.19. Figure 7.63505.18) shows that a maximum predicted value of 19.1982*p2 – 911. displays optimal settings of 0.18 Mixture Response Surface Analysis Viewing the Prediction Profiler The report contains a prediction profiler. 0.3298 + p3*p2*1846. and the other two move in the opposite direction maintaining their ratio.1465*p1 – 282. The profiler. 0. Note: These results correct the coefficients reported in Cornell (1990).6484*p3 + p2*p1317. right-click (Ctrl+click on the Mac) the column name and select Formula: 0–50.15568. and 0.21225 for p3. . The Response Surface Solution report (Figure 7. shown in Figure 7. which give an estimated response of 19.

20 Creating a Ternary Plot 3. by selecting p1. Identify the contour variable (the prediction equation) by selecting Pred Formula Y and clicking the Contour Formula button.21. as shown in Figure 7. 2. and p3 and clicking the Y. Choose Graph > Ternary Plot. the ternary plot only shows points. Click OK and view the results. .20. The contour variable must have a prediction formula to form the contour lines. If there is no prediction formula. as shown in Figure 7. as shown by the ternary plots at the bottom in Figure 7.21.146 7 Mixture Designs—A Chemical Mixture Example Figure 7. Specify plot variables. p2. Figure 7. Plottting button.19 Profiler for Mixture Analysis Example Plotting a Mixture Response Surface To plot a mixture response surface: 1.

21 Ternary Plot of a Mixture Response Surface .7 Mixture Designs—A Chemical Mixture Example 147 7 Mixture Figure 7.

.

any variability is small enough to be ignored. In space filling designs. Four methods are implemented for these types of designs: • Sphere Packing (emphasizes spread of points) • Latin Hypercube (compromise between spread of points and uniform spacing) • Uniform (mimics the uniform probability distribution) • Minimum Potential (minimizes energy designs in a hyperspherical) . One objective is to spread the design points out to the maximum distance possible between any two points to prevent replicate points. For this case. Sensitivity studies of computer-simulations is one such situation. and any mechanistic or deterministic modeling problem. Replication is undesirable because repeating the same run yields the same result. For systems with no variability. randomization and blocking are irrelevant.8 Space Filling Designs 8 Space Filling Space filling designs are useful in situations where run-to-run variability is of far less concern than the form of the model. The second objective is to space the points uniformly. there are two objectives.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Visualizing the Latin Hypercube Design . . . . . . . . . . . . . . . . . Borehole Model Example . . . . . Creating a Sphere Packing Design . . . . . . Creating a Sphere Packing Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparing the Sphere Packing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results of the Borehole Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8 Contents Introduction to Space Filling Designs . . . . . . . . . . . . . . . . Using the Latin Hypercube Method . . . . . . . . . . . . . . . . . Visualizing the Sphere Packing Design . . . . . . . . . . . . . . . . . . . . . . . and Uniform Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using the Minimum Potential Method . . . . . . . . . . . . Creating a Latin Hypercube Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Latin Hypercube. . . . . . . . . . . . . . . . Guidelines for the Analysis of Deterministic Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using the Sphere Packing Method . . . . . . . . Using the Uniform Design Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 151 151 153 154 154 155 157 158 159 160 161 162 163 . . . . . . . . . . . . . . . . . . . . . . . . .

4. One approach is to spread the design points out as far from each other as possible consistent with staying inside the experimental boundaries. Figure 8. • Minimum Potential method Spreads points out inside a sphere around the center. The effect of this maximization is to spread the points out as much as possible inside the design region.1 shows the two existing factors. X1 and X2. There are two schools of thought on how to bound the bias. and factors. The space filling designer supports four design methods: • Sphere packing design method Maximizes the minimum distance between pairs of design points. To view the discrepancy value and the minimum distance from each point to its closest neighbor. A goal of designed experiments on these systems is to find a simpler empirical model that adequately predicts the behavior of the system over limited ranges of the factors. Alter the factor level values. if necessary. the goal is to minimize the variance of prediction. there is no variance but there is bias. The Latin Hypercube method is a compromise between the sphere packing method and the uniform design method. Enter responses. In experiments on deterministic systems. 2. The goal of space filling designs is to bound the bias.” p. values ranging of 0 and 1 (instead of the default 0 to 1).8 Space Filling Designs—Introduction to Space Filling Designs 151 8 Space Filling Introduction to Space Filling Designs Space filling designs are useful for modeling systems that are deterministic or near deterministic. Such simulations can be very complex involving many variables with complicated interrelationships. In experiments on systems where there is substantial random noise. Select DOE > Space Filling. Creating a Sphere Packing Design To use the sphere packing method: 1. Bias is the difference between the approximation model and the true mathematical function. The other approach is to space the points out evenly over the region of interest. if necessary. This method produces designs that mimic the uniform distribution. For example. • Uniform design method Minimizes the discrepancy between the design points (which have an empirical uniform distribution) and a theoretical uniform distribution. .) 3. (See “Entering Responses and Factors in the Custom Designer. 57. Using the Sphere Packing Method The sphere packing design method maximizes the minimum distance between pairs of design points. • Latin Hypercube method Maximizes the minimum distance between design points but requires even spacing of the levels of each factor. One example of a deterministic system is a computer simulation.

Figure 8.3 shown the Design Diagnostics panel open with 0. JMP creates this design and displays the design runs and the design diagnostics (if specified in step 4).152 8 Space Filling Designs—Using the Sphere Packing Method click the red triangle icon on the Space Filling title bar and select Show Diagnostics. . as shown in Figure 8.2.2. In the example shown in Figure 8. specify a sample size. Figure 8. In design specification dialog. 6. Click Continue. shown on the left in Figure 8. Figure 8. a sample size of eight was entered. Click the Sphere Packing button.1.518 as the Minimum Distance.1 Space Filling Dialog for Two Factors 5.2 Space-Filling Design Dialog 7.

Select Graph > Overlay Plot. Create a space filling design using the default X1 and X2 factors. Now repeat the above procedure exactly. Specify a sample size of eight. 0. 6. This script draws a circle centered at each design point with radius 0.518).518 is the minimum distance number you noted in the Design Diagnostics panel. then click OK. then click Make Table. select Add Graphics Script and enter the following script: For Each Row(Circle({X1. Sphere-Packing Design Settings and Design Diagnostics 8.3 Space-Filling Design Dialog. where 0. Visualizing the Sphere Packing Design To visualize the nature of the sphere packing technique. create an overlay plot.8 Space Filling Designs—Using the Sphere Packing Method 153 8 Space Filling Figure 8. as shown on the left in Figure 8. Click the Sphere Packing button. 4.259 (half the diameter. Click Make Table.518 in the graphics script to the minimum distance produced by 10 runs. but change the values so they are 0 and 1. 2. 3. 0. This plot shows the efficient way JMP packs the design points. . adjust the plot’s frame size. and add circles using the minimum distance from the diagnostic report as the radius for the circle: 1. Specify X1 as X and X2 as Y. Click the red triangle icon on the Space Filling title bar and select Show Diagnostics.4. X2}. 7. Remember to change 0. Adjust the frame size so that the frame is square by right-clicking the plot and selecting Size/Scale > Size to Isometric. but with a sample size of 10 instead of eight (step 4).518/2)). 5. 8. Right-click the plot. 9.

The levels are spaced evenly from the lower bound to the upper bound of the factor. 4. 2. if necessary. again set the frame size and create a graphics script using the minimum distance from the diagnostic report as the radius for the circle. Note the irregular nature of the sphere packing. Figure 8. . To view the discrepancy value and the minimum distance from each point to its closest neighbor. the Latin Hypercube method chooses points to maximize the minimum distance between design points. you can repeat the process a third time to get a slightly different picture because the arrangement is dependent on the random starting point. click the red triangle icon on the Space Filling title bar and select Show Diagnostics.5 shows adding two factors to the two existing factors and changing their values. You should see a graph similar to the one on the right in Figure 8. 57. but with a constraint.4. For example. In fact. Creating a Latin Hypercube Design To use the Latin Hypercube method: 1. each factor has as many levels as there are runs in the design. Select DOE > Space Filling. Alter the factor level values. and factors. (See “Entering Responses and Factors in the Custom Designer. if necessary.154 8 Space Filling Designs—Using the Latin Hypercube Method When the plot appears. Like the sphere packing method.4 Sphere Packing Example with Eight Runs (left) and 10 Runs (right) Using the Latin Hypercube Method In a Latin Hypercube.) 3. Figure 8. The constraint involves maintaining the even spacing between factor levels. Enter responses.” p.

Click Continue. Visualizing the Latin Hypercube Design To visualize the nature of the sphere packing technique. create an overlay plot. 6. adjust the plot’s frame size. which shows the Latin hypercube design with eight runs. specify a sample size. and each column is a different permutation of the levels.5 Space-Filling Dialog for Four Factors 5. 3. 7. Click Continue. but change the values so they are 0 and 1.7. Note that each column (factor) is assigned each level only once. create a space filling design using the default X1 and X2 factors. and add circles using the minimum distance from the diagnostic report as the radius for the circle: 1.6 Latin Hypercube Design for Eight Runs with Eight Levels 8. Click Make Table. Factor settings and design diagnostics results appear similar to those in Figure 8. . 2. Click the red triangle icon on the Space Filling title bar and select Show Diagnostics.8 Space Filling Designs—Using the Latin Hypercube Method 155 8 Space Filling Figure 8. Figure 8. In design specification dialog. Click the Latin Hypercube button. First.

Remember to change 0. Specify a sample size of eight.156 8 Space Filling Designs—Using the Latin Hypercube Method 4. you can again repeat the process to get a slightly different picture because the arrangement is dependent on the random starting point. In fact.8 Comparison of Latin Hypercube Designs with Eight Runs (left) and 10 Runs (right) .404/2)).202 (half the diameter. 10.404 is the minimum distance number you noted in the Design Diagnostics panel (Figure 8.8. but with 10 runs instead of eight (step 4). Select Graph > Overlay Plot. Right-click the plot and select Size/Scale > Frame Size to Isometric to adjust the frame size so that the frame is square. Note the irregular nature of the sphere packing. 0. then click OK.7.6).Right-click the plot. 8.Repeat the above procedure exactly. This plot shows the efficient way JMP packs the design points. 5.7 Latin Hypercube Design with Eight Runs The minimum distance 6. Figure 8. 9. where 0. Specify X1 as X and X2 as Y. as shown on the left in Figure 8.404). 11.404 in the graphics script to the minimum distance produced by 10 runs. 7. Click Make Table. Figure 8. select Edit > Add Graphics Script from the menu. 0. X2}. Click the Latin Hypercube button. You should see a graph similar to the one on the right in Figure 8. and enter the following script: For Each Row(Circle({X1.8. This script draws a circle centered at each design point with radius 0. Factor settings and design diagnostics are shown in Figure 8.

increasing the number of runs and running a distribution on each factor (by selecting Analyze > Distribution) shows nearly flat histograms. The estimate is the average of the observed responses from the experiment. . Enter responses. Click the Uniform button. Alter the factor level values. This is because the Latin hypercube design constrains the levels of each factor to be evenly spaced. click the red triangle icon on the Space Filling title bar and select Show Diagnostics. Figure 8. if necessary.) 3. (See “Entering Responses and Factors in the Custom Designer.9 vary substantially. 4.1.8 Space Filling Designs—Using the Uniform Design Method 157 8 Space Filling Note that the minimum distance between each pair of points in the Latin hypercube design is smaller than that for the sphere packing design. if necessary. 6. Note that the emphasis of the Uniform Design method is not to spread out the points.” p. 1. Click Continue. 57. The ‘minimum distances in Figure 8. JMP creates this design and displays the design runs and the design diagnostics (if specified in step 4). Also. 5. and factors. These designs are most useful for getting a simple and precise estimate of the integral of an unknown function. Select DOE > Space Filling. Click Make Table. 2.9 Diagnostics for Uniform Space Filling Designs with Eight Runs 8. The sphere packing design maximizes the minimum distance without any constraints. In design specification dialog. specify a sample size. However. as shown in Figure 8. Using the Uniform Design Method The Uniform Design method minimizes the discrepancy between the design points (empirical uniform distribution) and a theoretical uniform distribution. a uniform design does not guarantee even spacing of the factor levels. 7. To view the discrepancy value and the minimum distance from each point to its closest neighbor.

11 shows a comparison of the design diagnostics for three eight-run space filling designs. or Uniform button. For both spread and discrepancy. the Latin Hypercube design represents a healthy compromise solution. The discrepancy for the Latin Hypercube takes an intermediate value that is closer to the optimal value. and Uniform Methods Figure 8. . Note that the discrepancy for the uniform design is the smallest (best). Also note that the minimum distance between pairs of points is largest (best) for the sphere packing method. The discrepancy for the sphere packing design is the largest (worst). Latin Hypercube. The Latin Hypercube design behaves more like the sphere packing design in spreading the points out.10 Histograms are Flat for Each Factor When Number of Runs is Increased to 16 Comparing the Sphere Packing. and Uniform Methods For a comparison of three space filling design methods. click the red triangle icon in the Space Filling Design title bar and select Show Diagnostics before clicking the Sphere Packing.158 8 Space Filling Designs—Comparing the Sphere Packing. Latin Hypercube. Latin Hypercube. This value is the integrated difference between the design points and the uniform distribution. The uniform design has pairs of points that are only about half as far apart. Figure 8. This displays the minimum distance from each point to its closest neighbor and the discrepancy value.

Select DOE > Space Filling. The coulomb force pushes the points apart. but the springs pull them together. To use the minimum potential method: 1. The uniform design method requires the most time to compute.8 Space Filling Designs—Using the Minimum Potential Method 159 8 Space Filling Figure 8. the time increases rapidly with the number of points. and factors. Enter responses. if necessary. 57. Using the Minimum Potential Method The Minimum Potential method spreads points out inside a sphere. (See “Entering Responses and Factors in the Custom Designer. To understand how this design is created. if necessary. all the space filling design methods take longer to compute than the D-optimal designs in the custom designer. .11 Comparison of Diagnostics for Three Eight-Run Space Filling Methods Sphere Packing Latin Hypercube Uniform Another point of comparison is the time it takes to compute a design. imagine the points as electrons with springs attached to every other point. Also. 2. For comparable problems. The design is the spacing of points that minimizes the potential energy of the system.” p. Alter the factor level values.) 3.

13 Minimum Potential Design Dialog 7. In design specification dialog. Figure 8. Click Continue. JMP creates this design and displays the design runs and the design diagnostics (if specified). Borehole Model Example Worley (1987) presented a model of the flow of water through a borehole that is drilled from the ground surface through two aquifiers. enter a sample size. Click Make Table.15m . Figure 8.05 to 0. 0.12 Minimum Potential Design Dialog 6.160 8 Space Filling Designs—Borehole Model Example 4. 5. rw = radius of borehole. Click the Minimum Potential button. The response variable y is the flow rate through the borehole in m3/year and is determined by the equation 2 π Tu ( Hu – Hl ) y = -------------------------------------------------------------------------------------------2 LT u Tu ln ( r ⁄ r w ) 1 + --------------------------------------.+ -----2 Tl ln ( r ⁄ r w ) r w K w There are eight inputs to this model.

63.070 to 115. In that folder.600 m2/year Hu = potentiometric head of upper aquifier.1 to 116 m2/year Hl = potentiometric head of lower aquifier. 2. However. .8 Space Filling Designs—Borehole Model Example 161 8 Space Filling r = radius of influence. 4. explicit function of the input variables. 63. Click Continue.000 m Tu = transmissivity of upper aquifier.045 m/year This example is atypical of most computer experiments because the response can be expressed as a simple. Select DOE > Space Filling. Figure 8. Creating a Sphere Packing Design To create a sphere packing design for the borehole problem: 1. 5. 1120 to 1680 m Kw = hydraulic conductivity of borehole. Click the red triangle icon on the Space Filling Design title bar and select Load Factors 3. Design Experiment > Borehole Factors.15). Specify a sample size of 32 runs (Figure 8.jmp to load the factors (Figure 8. Open the Sample Data folder that was installed when you installed JMP. 9855 to 12. 990 to 1100 m Tl = transmissivity of lower aquifier.14 Loaded Factors for Borehole Example Note: The logarithm of r and rw are used in the following discussion. 100 to 50. this simplicity is useful for explaining the design methods.14). 700 to 820 m L = length of borehole.

open the Sample Data folder that was installed when you installed JMP. However.15 Space Filling Design Method 6. A large F statistic (low p-value) is an indication of an effect due to a model term. 8. Open Design Experiment > Borehole Sphere Packing.jmp (see Figure 8. As a result.162 8 Space Filling Designs—Borehole Model Example Figure 8.16 Borehole Sample Data Guidelines for the Analysis of Deterministic Data It is important to remember that deterministic data has no random component. Figure 8. The factor settings in the example table might not have the same ones you see when generating the design because the designs are generated from a random seed. p-values from fitted statistical models do not have their usual meanings. Click the Sphere Packing button to produce the design. . Click Make Table to make a table showing the design settings for the experiment.16) This table also has a table variable that contains a script to analyze the data. To view a completed data table of this example. 7. and the results of the analysis are saved as columns in the table. you cannot make valid confidence intervals about the size of the effects or about predictions made using the model.

2. 4. In this case. The profile plots at the bottom in Figure 8. Check the Expand Intermediate Formulas box. the profile traces would be constant between the value ranges of each factor. Rather a residual shows the model bias for the current model at the current point. it is impossible to know the prediction bias at points other than the observed data without doing additional runs. the true model column contains a relatively simple formula.17 show the prediction bias at the center of the design region. Prediction Formula button.17. In real world examples. • The prediction bias column is the difference between the true model column and the prediction formula column. • The prediction bias is relatively small for each of the experimental points. versus the full quadratic model in the eight factors. as shown at the bottom of the upper box in Figure 8. the true model is generally not available in a simple analytical form. Select Graph > Profiler. In this example. log y. which allows profiling the prediction bias to find its value anywhere in the region of the data. Distinct patterns in the residuals indicate new terms to add to the model to reduce model bias. . led to the prediction formula column. the variables Hu and Hl show nonlinear effects. If there were no bias. As a result. Click OK. This is an indication that the model fits the data well. Highlight the prediction bias column and click the Y. Results of the Borehole Experiment The example described in the previous sections produced the following results: • A stepwise regression of the response. To understand the prediction bias in this example: 1. 3. because the prediction bias formula is a function of columns that are also created by formulas.8 Space Filling Designs—Borehole Model Example 163 8 Space Filling Residuals from any model fit to deterministic data are not a measure of noise.

To see this. This is more than three times the size of the range over the observed data.18. Figure 8.91) over the entire domain of the factors.75.84).17 Profile of the Prediction Bias in the Borehole Sphere Packing data.684 (the range is 2. The plot at the bottom shows the comparable minimum bias (–4. This gives a range of 7. Note that the maximum bias is 1.826 and the minimum is –0. The range of the prediction bias on the data is smaller than the range of the prediction bias over the entire domain of interest. look at the distribution analysis of the prediction bias in Figure 8. .164 8 Space Filling Designs—Borehole Model Example Figure 8.19 shows the maximum bias (2. Obtain this analysis by using Analyze > Distribution.51).18 Distribution of the Prediction Bias The top plot in Figure 8.

8 Space Filling Designs—Borehole Model Example 165 8 Space Filling Figure 8. . The prediction bias over the experimental data underestimates the bias throughout the design domain. the true model is given. • Verification runs— new runs performed at different settings to assess the lack of fit of the empirical model.19 Prediction Plots showing Maximum and Minimum Bias Over Factor Domains Keep in mind that. There are two ways to assess the extent of this underestimation: • Cross validation—refits the data to the model while holding back a subset of the points and looks at the error in estimating those points. in this example. the response at any factor setting is unknown. In any meaningful application.

.

as shown below.9 Nonlinear Designs 9 Nonlinear Design for experiments with models that are nonlinear in the parameters is available using either the DOE menu or the JMP Starter DOE category. .

. . . Creating a Nonlinear Design with No Prior Data . . . . . . . . . . . . . . . . .9 Contents Example of Creating a Nonlinear Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 169 174 178 178 179 179 180 180 . . . . . . . . . . . . . . . . . . Identifying the Responses/Column with Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Setting Up Factors and Parameters . . . . . . Selecting the Number of Runs . . . . . . . . Creating a Nonlinear Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Viewing the Design . . . . . Making or Augment the Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fitting to Get Prior Parameter Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.9 Nonlinear Designs—Example of Creating a Nonlinear Design 169 9 Nonlinear Example of Creating a Nonlinear Design Nonlinear designs allows scientists to generate optimal designs and optimally augment data for fitting models that are nonlinear in their parameters. 3. Fitting to Get Prior Parameter Estimates Suppose you have already collected experimental data and placed it in a JMP data table. but you have reasonable starting values for the parameter estimates.jmp You first fit the data to the model using nonlinear least squares to get better parameter values. when they are descriptive of the underlying process. To use the Nonlinear designer.1) contains a column (Model (x))whose values are formed by a formula with parameter values that represent a poor initial guess. can yield far more accurate prediction of process behavior than is possible with the standard polynomial models. open Chemical Kinetics. Highlight Velocity (y) and click the Y.1 Chemical Kinetics.jmp. . The first example in this section describes how to approach creating a nonlinear design when you have prior data. Select Analyze > Modeling > Nonlinear. you first need a data table that has one column for each factor. Response button. To follow along with this example. then click OK.jmp from Sample Data > Nonlinear Examples > Chemical Kinetics.jmp (Figure 9. and a column that contains a formula showing the functional relationship between the factor(s) and the response. The second example describes how to approach creating the design when you have no data. This is the same format for a table you would supply to the nonlinear platform for modeling. Such models. 1.2). Figure 9. That table can be used to create a nonlinear design for improving the estimates of the model’s nonlinear parameters. one column for the response. Chemical Kinetics. 4. Highlight Model (x) and click the X. Predictor Formula button (Figure 9.

.2 Running an Analysis 5.170 9 Nonlinear Designs—Example of Creating a Nonlinear Design Figure 9.3). Click the Go button on the Nonlinear control panel (Figure 9.

Select DOE > Nonlinear. 1.3 Results After Clicking Go These ranges are the intervals for Vmax and k.jmp’s column formula. Identify the responses by highlighting the column that contains the response values and click OK. For this example. 7. The formula looks like: Now create a nonlinear design for fitting the model’s nonlinear parameters. named Model (x). They are asymptotically normal. 6. 2. Use these limits when creating a nonlinear design in JMP. .9 Nonlinear Designs—Example of Creating a Nonlinear Design 171 9 Nonlinear Figure 9. Click the Save Estimates button to have JMP place the fitted parameters in Chemical Kinetics. use Velocity (y). Click the Confidence Limits button to produce confidence intervals.

4 Choosing Response Columns Note that in Chemical Kinetics. Type the desired number of runs into the text box and click Make Design to preview the design (Figure 9. Describe the a priori distribution of the parameters by changing the distribution of both Vmax and k to Normal.172 9 Nonlinear Designs—Example of Creating a Nonlinear Design Figure 9. Figure 9. Look back at the analysis report you created above (Figure 9. Change the values for Vmax and k to correspond to those limits.3) and locate in the Solution table the upper and lower confidence limits for Vmax and k. as shown in Figure 9.417 to 6.jmp (Figure 9.6 Changing the Distribution 5. Figure 9.5 Changing the Factor Values 4. those values appear as the high and low values in the Factors control panel.1). Change the range to a broader interval: from 0.7 Changing the Values Now you have described your current level of uncertainty of the two parameters.8). .5).7. 6. Therefore.25. Figure 9. 3. the range of data for Concentration goes from 0.3 to 7 (Figure 9.

Click Make Table.9) whose rows are the runs you defined.8 Selecting the Number of Runs 7. however. Figure 9. You created a new table to avoid altering the sample data table Chemical Kinetics. This creates a new JMP design table (Figure 9.9 Nonlinear Designs—Example of Creating a Nonlinear Design 173 9 Nonlinear Figure 9. In most cases. .9 Making the Table The new runs make use of the wider interval of allowed concentration. This will lead to more precise estimates of k and Vmax. you will want to augment the original table by clicking Augment Table rather than making a new table.jmp.

2. 5. To follow along with this example. but you have a guess for the unknown parameters.3. The formula looks like: This model was taken from Box and Draper (1987). use the values for Reaction Temperature of 520 and 540.1 and 0. Identify the response by highlighting the column that contains the response (predictor) values and click OK (Figure 9.jmp. Change the two factors’ values to be a reasonable range of values. (This may be an educated guess. Open Sample Data > Design Experiment> Reaction Kinetics Start.174 9 Nonlinear Designs—Example of Creating a Nonlinear Design Creating a Nonlinear Design with No Prior Data This example describes how to approach creating the design when you have not yet collected data. Select DOE > Nonlinear Design. . 3. 1. Leave each parameter’s distribution as at the default Uniform. Use values for Reaction Time of 0. open the Sample Data folder that was installed when you installed JMP.jmp. use observed yield.11).10 Choosing Response Columns 4. Figure 9. The formula in the yield model column will be used to create a nonlinear design for fitting the model’s nonlinear parameters. and change the number of runs to 12 (Figure 9. Open Reaction Kinetics Start. as described above.10). For this example. It is as a function of time and temperature. The formula arises from the fractional yield of the intermediate product in a consecutive chemical reaction.) For this example.

Click Make Design.11 Changing the Number of Runs 6.13 Reactor Kinetics.jmp Now examine the design region. Your results should look similar to those in Figure 9.12. Figure 9. .12 Design Table You have now completed the design without data. then Make Table.9 Nonlinear Designs—Example of Creating a Nonlinear Design 175 9 Nonlinear Figure 9. Figure 9.

14 Create an Overlay Plot Notice that the points are not at the corners of the design region. and highlight Reaction Time and click the X button (Figure 9.14). The maximum yield is 60. 12. then click OK. 9. 14. Figure 9. 10. Predictor Formula button.Click the Go button on the Nonlinear control panel.Now you can choose Profilers >Profiler from the red triangle menu. 13.Select Analyze > Modeling > Nonlinear. Create an overlay plot by selecting Graph > Overlay Plot. Click OK. 11. 8.Highlight yield model and click the X. In particular.Highlight observed yield and click the Y. there are no points at low temperature and high time—the lower right corner of the graph. Response button. Maximize Desirability.1126 minutes. . Highlight Reaction Temperature and click the Y button.69% at a reaction temperature of 540 degrees Kelvin and a reaction time of 0.176 9 Nonlinear Designs—Example of Creating a Nonlinear Design 7.

15 Time and Temperature Settings for Maximum Yield .9 Nonlinear Designs—Example of Creating a Nonlinear Design 177 9 Nonlinear Figure 9.

Open a data table that contains a column whose values are formed by a formula with parameters. This column cannot contain any missing values.” p.178 9 Nonlinear Designs—Creating a Nonlinear Design Creating a Nonlinear Design To start a nonlinear design.16). If your data table contains more than one column with formulas. see the JMP User Guide). JMP asks you to identify the column that contains the nonlinear model formula (Figure 9. 180 Identifying the Responses/Column with Formula 1. Figure 9. . 179 • “Selecting the Number of Runs. 2.” p. follow the steps below: • “Identifying the Responses/Column with Formula.16 Example of Identifying the Column with the Nonlinear Formula 4. This formula must have parameters.” p. 179 • “Viewing the Design.” p. open a data table that contains a column whose values are formed by a formula (for details about formulas.” p. Then. or click the Nonlinear Design button on the JMP Starter DOE page. Then. select DOE > Nonlinear Design. 178 • “Setting Up Factors and Parameters. 3. Highlight the column name and click OK. JMP asks you to identify the column that contains the response values and click OK. Select DOE > Nonlinear Design. 180 • “Making or Augment the Table.

Type the desired number of runs into the text box and click Make Design. Normal. Selecting the Number of Runs The Design Generation panel shows the minimum number of runs needed to perform the experiment. or Exponential.18 Example of Setting Up Factors and Parameters Double-click to edit the factor or parameter name.9 Nonlinear Designs—Creating a Nonlinear Design 179 9 Nonlinear Figure 9.17 Example of Choosing Response Columns Setting Up Factors and Parameters Use Figure 9.18 to understand how to set up factor and parameter names and values. Figure 9. . Click to enter or change factor values. Click to edit the distribution: Uniform. Lognormal.

Clicking Augment Table adds the new runs to the existing table. . Clicking Make Table creates a new table with all runs included. Figure 9.180 9 Nonlinear Designs—Creating a Nonlinear Design Figure 9.19 Selecting the Number of Runs Viewing the Design Before creating the design table.20 Example Preview Design Making or Augment the Table After previewing the design. click either Make Table or Augment Table. JMP lets you preview the design.

21 Example of a Nonlinear Design Table .9 Nonlinear Designs—Creating a Nonlinear Design 181 9 Nonlinear Figure 9.

.

You choose from inner and outer array designs. Its formula depends on whether the experimental goal is to maximize. The noise factors. and mixed-level fractional factorial designs. form the outer array.10 Taguchi Designs 10 T aguchi Quality was the watchword of 1980s. Noise factors are variables that are typically difficult or expensive to control. which use the traditional Taguchi orthogonal arrays. minimize or match a target value of the quality characteristic of interest. three-level. it is the signal-to-noise ratio. associated with process or environmental variability. Taguchi’s design approach employs two designs called the inner and outer array. The purpose of the Taguchi method was to develop products that worked well in spite of natural variation in materials. The inner array is a design in the signal factors and the outer array is a design in the noise factors. and Genichi Taguchi was a leader in the growth of quality consciousness. A Taguchi experiment repeats the outer array design for each run of the inner array. His orthogonal arrays are two-level. inner and outer arrays. and L16. Much of the Taguchi method is traditional. The Taguchi experiment is the cross product of these two arrays. and signal-to-noise ratios as Taguchi specifies. Signal factors are system control inputs. The goal of the Taguchi method is to find control factor settings that generate acceptable responses despite natural environmental and process variability. Dividing system variables according to their signal and noise factors is a key ingredient in robust engineering. The response variable in the data analysis is not the raw response or quality characteristic. The unique aspects of his approach are the use of signal and noise factors. One of Taguchi’s technical contributions to the field of quality control was a new approach to industrial experimentation. L8. inner and outer arrays. The Taguchi designer in JMP supports signal and noise factors. In each experiment. form the inner array. used to tweak the process. suppliers. . The control factors. and signal-to-noise ratios. operators. A signal-to-noise ratio is a statistic calculated over an entire outer array. This is robust engineering. The Taguchi designer supports all these features of the Taguchi method. and environmental change. Taguchi’s signal-to-noise ratios are functions of the observed responses over an outer array. such as L4.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Displaying and Modify the Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185 185 187 189 190 191 191 192 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Taguchi Design . . . . . . . . . . . . . .10 Contents The Taguchi Design Approach . . . . . . . . . . . . . . . . . . . . . . . Detailing the Response and Add Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Taguchi Design Example . . . Analyze the Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . Making the Table . . . . . . . . . . . . . . . . . . . . . Choosing Inner and Outer Array Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Factor Name Interfer Wall Depth Adhesive Type Levels Comment control control control control 3 3 3 3 tubing and connector interference the wall thickness of the connector insertion depth of the tubing into the connector percent adhesive . The experiment is performed on all combinations of the inner and outer design runs.table found in the Design Experiment folder of the Sample Data. Open the Byrne Taguchi Data.= – 10 log ⎛ 1 .∑Y 2⎟ --⎜ -i n N ⎝ i ⎠ nominal is best larger-is-better (maximize) smaller-is-better (minimize) Taguchi Design Example The following example is an experiment done at Baylock Manufacturing Corporation and described by Byrne and Taguchi (1986). To follow along with this example. Or. A performance statistic is calculated across the outer runs for each inner run. This becomes the response for a fit across the inner design runs. An inner design constructed over the control factors finds optimum settings. open the Sample Data folder that was installed when you first installed JMP. An outer design over the noise factors looks at how the response behaves for a wide range of noise conditions. The table below lists the recommended performance statistics. The signal and noise factors for this example appear in the table below.jmp. Goal S/N Ratio Formula ⎛ 2⎞ S-------⎟ = 10 log ⎜ Y N ⎝ s2 ⎠ ⎛ ⎞ S1-⎟ 1 ------= – 10 log ⎜ -∑ ⎜ n i Y 2⎟ N i ⎠ ⎝ ⎞ S. The objective of the experiment is to find settings of predetermined control factors that simultaneously maximize the adhesiveness (pull-off force) and minimize the assembly costs of nylon tubing.10 Taguchi Designs—The Taguchi Design Approach 185 10 T aguchi The Taguchi Design Approach The Taguchi method defines two types of factors: control factors and noise factors. generate the original design table on your own using DOE > Taguchi Arrays.

However. and Signal and Noise Factors for the Byrne-Taguchi Example 4. Click Make Table to create the design table shown in Figure 10.1.186 10 Taguchi Designs—Taguchi Design Example Factor Name Time Temp Humidity Type Levels Comment noise noise noise 2 2 2 the conditioning time temperature the relative humidity To start this example: 1. 5. Figure 10. Select DOE > Taguchi. Open Design Experiment > Byrne Taguchi Factors. The factors panel then shows the four three-level control (signal) factors and three noise factors. The outer design has three two-level factors. A full factorial in eight runs is generated. Click the red triangle icon on the Taguchi Design title bar and select Load Factors.1 Response. Click Continue. Highlight L9-Taguchi to give the L9 orthogonal array for the inner design. as shown in Figure 10. Open the factors table by opening the Sample Data folder that was installed when you first installed JMP. 6. 3. 7. 2. Highlight L8 to give an eight-run outer array design.jmp. it is only used as a guide to identify a new set of eight columns in the final JMP data table—one for each combination of levels in the outer design. .2.

.2 y +++2 This expression is large when all of the individual y values are small. Click the red triangle icon next to the Model script and select Run Script as shown in Figure 10.2 y+ . 1..10 Taguchi Designs—Taguchi Design Example 187 10 T aguchi Figure 10. . as shown in Figure 10.is the column of measurements taken with the three noise factors set at their low levels. To see the completed experiment and continue following this example.+2 y ++. . . .. 8. which is –10 times the common logarithm of the average squared reciprocal: –10Log10 Mean 1 1 1 1 1 1 1 1 .jmp) The column named SN Ratio Y is the performance statistic computed with the formula shown below.jmp are now ready to analyze. y . The missing data column names are appended with the letter Y before the levels (+ or –) of the noise factors for that run. The goal of the analysis is to find factor settings that maximize both the mean and the signal-to-noise ratio. .2 y .3.4. . Y--.3 Complete Taguchi Design Table (Byrne Taguchi Data.+ 2 y . .jmp. Analyze the Data The data in Byrne Taguchi Data. Figure 10. it is the “larger–the–better” (LTB) formula... open the table shown in Figure 10.2 Taguchi Design Before Data Entry Now suppose the pull-off adhesive force measures are collected and entered into the columns containing missing data.+ +2 y+ .2 y .3 by selecting Help (View on the Macintosh) > Sample Data Directory > Design Experiment > Byrne Taguchi Data.+ . In this case. For example.

Figure 10. 3. Open the Prediction Profiler by clicking the red triangle on the Response Mean Y title bar and selecting Factor Profiling > Profiler.4 Fitting a Prediction Model 2. The prediction profiler is a quick way to find settings that give the highest signal-to-noise ratio for this experiment.5) indicate that different settings of the first three factors would increase SN Ratio Y.5 The Prediction Profiler . Figure 10. The two responses are the mean and signal-to-noise ratio over the outer array. The profile traces (Figure 10. It includes the main effects of the four signal factors. Click Run Model.188 10 Taguchi Designs—Taguchi Design Example The window that appears is automatically set up with the appropriate effects.

” p. the settings for Interfer and Wall changed from L1 to L2 (Figure 10. In this example. follow the steps below: • “Detailing the Response and Add Factors. There was no change in Adhesive. 192 . 191 • “Displaying and Modify the Design. The Depth setting changed from L1 to L3. These new settings increased the signal-to-noise ratio from 24.” p. This adds the row of traces and column of function settings to the profiler. or click the Taguchi Arrays button on the JMP Starter DOE page. 191 • “Making the Table. Again click the red triangle on the Prediction Profiler title bar and select Maximize Desirability to automatically set the prediction traces to give the best results according to the desirability functions. To find optimal settings. select DOE > Taguchi Arrays. Figure 10. click the red triangle on the Prediction Profiler title bar and select Desirability Functions. Then.6 Best Factor Settings for Byrne Taguchi Data Creating a Taguchi Design To start a Taguchi design. 5. 190 • “Choosing Inner and Outer Array Designs. The default desirability functions are set to larger-is-better.” p.0253 to 26.9075.” p.10 Taguchi Designs—Creating a Taguchi Design 189 10 T aguchi 4. which is what you want in this experiment.6). See the chapter “Standard Least Squares: Perspectives on the Estimates” of JMP Statistics and Graphics Guide for more details about the prediction profiler.

7. 2 Click to change the response goal: Larger Is Better. For information on importance weights and lower and upper limits.8. 58. The steps for setting up the details of this response are outlined in Figure 10. then select a signal type: 2 Level. The steps for setting up the factors are outlined in Figure 10. or None.190 10 Taguchi Designs—Creating a Taguchi Design Detailing the Response and Add Factors The Responses panel has a single default response.8 Entering Factors Click to add a signal. click and then type the new value. Nominal is Best.7 Setting Up the Response Click to enter lower and upper limits and importance weights. Smaller is Better. Figure 10. 3 1 Double-click to edit the response name. or 3 Level. 1 To change a signal or noise’s value. . Figure 10. see “Understanding Response Importance Weights. 3 2 Double-click to edit the factor name. or Click to add a noise.” p.

10). • Coded Design—Shows the pattern of high and low values for the factors in each run.” p. click Continue. click the disclosure buttons ( on Windows/Linux and on the Macintosh) to display the design and show modification options to tailor the design (Figure 10.9 Selecting a Design Using Eight Factors Displaying and Modify the Design After you select a design type. 96. see “Understanding the Coded Design. Figure 10.” p. see “Aliasing of Effects. For more details on the coded design.9 shows your choices when you have eight factors. 95 and “Changing the Coded Design. Figure 10. Choosing Inner and Outer Array Designs Depending on the number of factors you have. For more details on aliasing of effects.10 Display and Modification Options • Aliasing of Effects—Shows the confounding pattern for fractional factorial designs. 93. Highlight the design(s) you would like to use and click Continue.10 Taguchi Designs—Creating a Taguchi Design 191 10 T aguchi When you finish adding factors. . Figure 10.” p. JMP will ask you to select an inner and outer array design.

plus signs designate high levels and minus signs represent low levels. For details on codes and how to change them. and “Changing the Coded Design. 95. each row represents a run. Figure 10. see “Understanding the Coded Design. 96. a table similar to that shown in Figure 10.11 Taguchi Design Table . In the data table.11 appears. In the values for the Pattern variable.” p.” p.192 10 Taguchi Designs—Creating a Taguchi Design Making the Table When you click Make Table.

It also presents a case study of design augmentation. That way. The augment designer modifies an existing design data table. . supporting your iterative process.11 Augmented Designs 11 Augmented It is best to treat experimentation as an iterative process. you can master the temptation to assume that one successful screening experiment has optimized your process. You can also avoid disappointment if a screening experiment leaves behind some ambiguities. It gives the following five choices: • replicate the design a specified number of times • add center points • create a foldover design • add axial points together with center points to transform a screening design to a response surface design • add runs to the design using a model that can have more terms than the original model This chapter provides an overview of the augment designer.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Adding Axial Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11 Contents The Reactor Example: Using D-Optimal Augmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . Adding New Runs and Terms . . . . . . . . . . . . . . . . Modifying the Design Criterion (D. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Replicating a Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating an Augmented Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Foldover Design . . . . . . . . . . Selecting the Number of Random Starts . . . . . . . . . . . . . . . . . . . . . . Special Augment Design Commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Viewing Diagnostics Reports . . . . . . . . . . . . . . . . . . . . . Disallowing Factor Combinations . . . . . . . . . . . . . . . .Optimality) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 198 204 204 207 208 209 210 213 214 214 214 215 215 215 . . . . . . . . . . . . . . . . . . . . . . . . . . . . Analyze the Augmented Design . . . . . . . . . . . . . . . . . . . . . . . Specifying the Sphere Radius Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Saving the Design (X) Matrix . . . . . . . . . . . . . . . . . . . . .or I. . . . . . . . . . . . . . . Adding Centerpoints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

the Augment Design dialog shown in Figure 11. a chemical engineer investigates the effects of five factors on the percent reaction of a chemical process. the purpose of this example is to estimate all two-factor interactions in 16 runs.2. Then select Augment Design from the DOE menu. which can’t be done when there is the additional blocking factor in the model.1 appears.1 Augment Design Dialog for the Reactor Example Now click Augment on the Augment Design dialog to see the display in Figure 11. . Open Design Experiment > Reactor 8 Runs.11 Augmented Designs—The Reactor Example: Using D-Optimal Augmentation 195 11 Augmented The Reactor Example: Using D-Optimal Augmentation This example. In this study. However. To begin. Figure 11. open the Sample Data folder that was installed when you installed JMP. demonstrates how to use JMP’s augment designer to resolve ambiguities left by a screening design.jmp. et al. (1996). However. After you identify the factors and response and click OK. Note: You can check Group New Runs into Separate Blocks to add a blocking factor to any design. This model is the result of the model stored with the data table when it was created by the Custom designer. the augmented design is to have 16 runs in order to estimate all two-factor interactions. adapted from Meyer.

The Minimum number of runs given for the specified model is 16.196 11 Augmented Designs—The Reactor Example: Using D-Optimal Augmentation Figure 11.3. as shown in the Design Generation text edit box. . This adds all the two-factor interactions to the model.2 Initial Augmented Model To continue with the augmented reactor design. choose 2nd from the Interactions popup menu as shown in Figure 11.

(or the most recent design you generated).4. The dialog in Figure 11.4.5 shows the random number (12834729) used to generate the runs in Figure 11.11 Augmented Designs—The Reactor Example: Using D-Optimal Augmentation 197 11 Augmented Figure 11. JMP computes D-optimally augmented factor settings. or Cancel to generate a design based on a new random number. To reproduce the exact factor settings table in Figure 11. choose Set Random Seed from the popup menu on the Augment Design title bar. Figure 11.3 Augmented Model with All Two-Factor Interactions When you click Make Design. similar to the design shown in Figure 11. A dialog shows the most recently used random number.4 D-Optimally Augmented Factor Settings Note: The resulting design is a function of an initial random number seed. Click OK to use that number again.4. .

5 Specifying a Random Number Click Make Table.jmp sample data file.jmp. suppose you augmented that data table . run the Model script stored as a table property with the data table by right-clicking the script name and selecting Run. This table property contains the JSL commands that display the Fit Model dialog with all main effects and two-factor interactions as effects.jmp) To start the analysis. Also.7. To see these results. These folders were installed when you installed JMP. Analyze the Augmented Design Suppose you have already run the experiment and recorded results in the Percent Reacted column of the data table. Figure 11.6 Completed Augmented Experiment (Reactor Augment Data. open the Reactor 8 Runs. . Change the fitting personality from Standard Least Squares to Stepwise. found in the Design Experiment folder of the Sample Data .6 shows the Reactor Augment Data.198 11 Augmented Designs—The Reactor Example: Using D-Optimal Augmentation Figure 11. as shown in Figure 11. Figure 11.

Then. choose Mixed from the Direction menu and make sure Prob to Enter is 0.11 Augmented Designs—The Reactor Example: Using D-Optimal Augmentation 199 11 Augmented Figure 11.050 and Prob to Leave is 0. You should see the dialog shown in Figure 11.7 Fit Model Dialog for Stepwise Regression on Generated Model When you click Run Model. Click the check boxes for all the main effect terms.100. . the stepwise regression control panel appears.8.

9. Note that Feed Rate is out of the model while the Catalyst*Temperature. Stir Rate*Temperature. . and the Temperature*Concentration interactions have entered the model.8 Initial Stepwise Model o Click Go to start the stepwise regression and watch it continue until all terms are entered into the model that meet the Prob to Enter and Prob to Leave criteria in the Stepwise Regression Control panel.200 11 Augmented Designs—The Reactor Example: Using D-Optimal Augmentation Figure 11. Figure 11. shows the result of this example analysis.

10 New Prediction Model Dialog . click Make Model on the Stepwise control panel to generate this reduced model. make predictions. as shown in Figure 11.10. Figure 11. You can now fit the reduced model to do additional diagnostic work. and find the optimal factor settings.9 Completed Stepwise Model After Stepwise is finished.11 Augmented Designs—The Reactor Example: Using D-Optimal Augmentation 201 11 Augmented Figure 11.

. This is the reason that the initial screening experiment. the significant two-factor interactions are of the same order of magnitude as the main effects.202 11 Augmented Designs—The Reactor Example: Using D-Optimal Augmentation The ANOVA and Lack of Fit Tests in Figure 11.11 Prediction Model Analysis of Variance and Lack of Fit Tests The Scaled Estimates table in Figure 11. Figure 11. shows that maximum occurs at the high levels of Catalyst. shown in the chapter “Screening Designs.12 Prediction Model Estimates Plot It is desirable to maximize the percent reaction.12.1 to 81. and Temperature and the low level of Concentration. Figure 11. When you drag the prediction traces for each factor to their extreme settings. shows that Catalyst has the largest main effect. 81. Stir Rate.” p.11. However.13. the estimate of Percent Reacted increases from 65. had ambiguous results.8. indicate a highly significant regression model with no evidence of Lack of Fit. The prediction profile plot in Figure 11.

32 runs. Contentration*Temperature) as significant effects. • By using a D-optimal augmentation of 8 runs to produce 8 additional runs. .13 Maximum Percent Reacted To summarize.” p. This approach identified three main effects (Catalyst. was first subjected to a stepwise regression.11 Augmented Designs—The Reactor Example: Using D-Optimal Augmentation 203 11 Augmented Figure 11. using an iterative approach to DOE can save time and money. • “Full Factorial Designs. None of the factors effects were significant. a stepwise analysis returned the same results as the analysis of 32 runs. The bottom line is that only half as many runs yielded the same information. Temperature. 117. a full factorial of the five two-level reactor factors.” p. the analysis of a screening design with only 8 runs produced a model with the five main effects and two interaction effects with confounding. Thus. although the Catalyst factor was large enough to encourage collecting data for further runs. compare the analysis of 16 runs with the analyses of reactor data from previous chapters: • “Screening Designs. and Concentration) and two interactions (Temperature*Catalyst. 81.

207. This example uses Reactor 8 Runs. 210. • Fold Over—creates a foldover design. Identify the responses you want to use for the augmented design by highlighting the column name(s) and clicking OK (Figure 11. which is located in the Sample Data folder’s Design Experiment folder. which can have more terms than the original model. See “Replicating a Design.14) and clicking OK. See “Creating a Foldover Design. • Add Centerpoints—adds centerpoints. Open a data table that contains a design you’d like to augment. Identify the factors you want to use for the augmented design by highlighting the column names (Figure 11. 3. 209. It gives the following five choices: • Replicate—replicates the design a specified number of times.15).” p. These folders were installed when you installed JMP. 2. • Augment—adds runs to the design (augment) using a model. Replicating a Design Replication provides a direct check on the assumption that the error variance is constant. It also reduces the variability of the regression coefficients in the presence of large process or measurement variability. 208. To replicate the design a specified number of times: 1.” p. Select DOE > Augment Design. Figure 11.204 11 Augmented Designs—Creating an Augmented Design Creating an Augmented Design The augment designer modifies an existing design data table.” p.” p.jmp. • Add Axial—adds axial points together with center points to transform a screening design to a response surface design. 204. See “Adding Centerpoints.14 Identify Factors 4. . See “Adding New Runs and Terms. See “Adding Axial Points.” p.

Enter the number of times you want JMP to perform each run and click OK.17 Enter the Number of Runs 8. check the box beside Group New Runs into Separate Block (Figure 11. This is the same as one replicate (Figure 11. entering 2 specifies that you want each run to appear twice in the resulting design. 6. as show in Figure 11. Click the Replicate button. For example.17).16 Choose an Augmentation Type 7. Figure 11. . View the design.16).18. Figure 11. If you want the original runs and the resulting augmented runs to be identified by a blocking factor.15 Identify Responses 5.11 Augmented Designs—Creating an Augmented Design 205 11 Augmented Figure 11.

19) by clicking the disclosure icon to the left of the associated title bars. If desired. Figure 11.Click Make Table. The design table appears (Figure 11. view the prediction variance profile and the prediction variance surface plots (Figure 11.19 Prediction Profiler and Surface Plot 10.20).18 Reactor Data Design Augmented With Two Replicates 9.206 11 Augmented Designs—Creating an Augmented Design Figure 11. .

The table below shows two center points appended to the end of the design. which can be used in a lack-of-fit test. Enter the number of centerpoints you want and click OK. Click Make Table.15). If you want the original runs and the resulting augmented runs to be identified by a blocking factor. Centerpoints are usually replicated points that allow for an independent estimate of pure error.20 The Replicated Design Adding Centerpoints Adding centerpoints is useful for checking for curvature and reducing the prediction error in the center of the factor region. 8. check the box beside Group New Runs into Separate Block (Figure 11. Open a data table that contains a design you’d like to augment. Select DOE > Augment Design. for example. To add centerpoints: 1. 5. . which is located in the Sample Data folder’s Design Experiment folder. 6. These folders were installed when you installed JMP.16).11 Augmented Designs—Creating an Augmented Design 207 11 Augmented Figure 11. 4.jmp. Click the Add Centerpoints button and add two centerpoints. 2. Identify the responses you want to use for the augmented design by highlighting the column name(s) and clicking OK (Figure 11. 3. 7. 9. Identify the factors you want to use for the augmented design by highlighting the column names (Figure 11. This example uses Reactor 8 Runs.14) and clicking OK.

6. 2. If you choose a subset of factors to fold over. 3. which is located in the Sample Data folder’s Design Experiment folder. This is especially useful as a follow-up to saturated or near-saturated fractional factorial or Plackett-Burman designs. To create a foldover design: 1. . These folders were installed when you installed JMP. The design data table that results lists the original set of runs as block 1 and the new (foldover) runs are block 2. Select DOE > Augment Design.22 folds on all five factors and includes a blocking factor. Identify the responses you want to use for the augmented design by highlighting the column name(s) and clicking OK (Figure 11. The default. 8. is to fold on all design factors. the remaining factors are replicates of the original runs. Identify the factors you want to use for the augmented design by highlighting the column names (Figure 11. Click the Fold Over button. The example in Figure 11. 4. To identify the original runs and the resulting augmented runs by a blocking factor.16). the remaining factors are replicates of the original runs. 5.21 Design with Two Center Points Added Creating a Foldover Design A foldover design removes the confounding of two-factor interactions and main effects. if you choose no factors. 7.15).jmp. Choose which factors to fold. This example uses Reactor 8 Runs. If you choose a subset of factors to fold over.208 11 Augmented Designs—Creating an Augmented Design Figure 11. check the box beside Group New Runs into Separate Block (Figure 11.14) and clicking OK. Click Make Table. Open a data table that contains a design you’d like to augment. Note: Adding centerpoints or replicating the design also generates an additional Block column in the JMP design table.

To do this: 1. the augmented design includes the number of centerpoints specified and constructs two axial points for each variable in the original design. This example uses Reactor 8 Runs. Identify the factors you want to use for the augmented design by highlighting the column names (Figure 11.14) and clicking OK.jmp. If you want the original runs and the resulting augmented runs to be identified by a blocking factor. 3. then enter the number of centerpoints you want. 6. which is located in the Sample Data folder’s Design Experiment folder. check the box beside Group New Runs into Separate Block (Figure 11. 4. Enter the axial values in units of the factors scaled from –1 to +1.16). Select DOE > Augment Design. 7.15). Click Add Axial. Identify the responses you want to use for the augmented design by highlighting the column name(s) and clicking OK (Figure 11. These folders were installed when you installed JMP. 5.22 Listing of a Foldover Design On All Factors Adding Axial Points You can add axial points together with centerpoints.11 Augmented Designs—Creating an Augmented Design 209 11 Augmented Figure 11. Open a data table that contains a design you’d like to augment. When you click OK. 2. therefore transforming a screening design to a response surface design. .

The design table appears. You can achieve the objectives of response surface methodology by changing a linear model to a full quadratic model and adding the necessary number of runs. You can also group the two sets of experimental runs into separate blocks. two-level. Figure 11. To add new runs and terms to the original model: . which optimally blocks the second set with respect to the first. suppose you start with a two-factor.24 Design Augmented With Two Center and Ten Axial Points center points axial points Adding New Runs and Terms Using the augment designer.210 11 Augmented Designs—Creating an Augmented Design Figure 11. Using this feature you can add terms to the original model and find optimal new test runs with respect to this expanded model. Click Make Table. For example.24 shows a table augmented with two center points and two axial points for five variables. If you add quadratic terms to the model and five new points. four-run design.23 Entering Axial Values 8. JMP generates the 3 by 3 full factorial as the optimal augmented design. D-optimal augmentation is a powerful tool for sequential design. you can add runs using a model that can have more terms than the original model. Figure 11. Adding runs to a design is powerful.

Identify the factors you want to use for the augmented design by highlighting the column names and clicking OK.25) appear in the Factor Design panel.25 Viewing the Existing Design 7. which is located in the Sample Data folder’s Design Experiment folder. 2. If you want the original runs and the resulting augmented runs to be identified by a blocking factor. Enter the number of total runs you want this design to contain. This number is the original number . 4.11 Augmented Designs—Creating an Augmented Design 211 11 Augmented 1. The original number of runs (Figure 11. Identify the responses you want to use for the augmented design by highlighting the column name(s) and clicking OK. This example uses Reactor Augment Data. Click the Augment button. check the box beside Group New Runs into Separate Block. 5.jmp. 3. These folders were installed when you installed JMP. 6. Figure 11. Select DOE > Augment Design. Open a data table that contains a design you’d like to augment.

. If desired.212 11 Augmented Designs—Creating an Augmented Design of runs plus the additional runs.Click the Make Table button.27) appears with the additional runs. The augmented design table (Figure 11. 8. 10. The new number of runs (Figure 11. Click the Make Design button.26 24 Total Runs 9. Figure 11.26) appear in the Design panel. view the prediction variance profile and the prediction variance surface.

66. Most of these commands are for saving and loading information about variables.28 appears. Click the red triangle icon on the Augment Design title bar to reveal a list of commands.27 The Augmented Design Table with New Runs additional runs Special Augment Design Commands After you select DOE > Augment Design and identify factors and responses. The following sections describe commands found in this menu that are specific to augment designs. .” p.11 Augmented Designs—Special Augment Design Commands 213 11 Augmented Figure 11. they are available in all designs and more information is in “Special Custom Design Commands. the window in Figure 11.

click the red triangle icon in the Augment Design title bar (Figure 11. The script is saved as a table property called Design Matrix. G and A efficiencies in augmented designs.28) and select Save X Matrix.or I. If you do not have the log visible. select View > Log.Optimality) .28 Click the Red Triangle Icon to Reveal Commands Viewing Diagnostics Reports To display a table with relative D. Modifying the Design Criterion (D.214 11 Augmented Designs—Special Augment Design Commands Figure 11. When you run this script. Saving the Design (X) Matrix To create a script and save it as a table property in the JMP design data table. click the red triangle icon in the Augment Design title bar and select Show Diagnostics before clicking the Make Design button. JMP creates a global matrix called X and displays its number of rows in the log.

then choose Make D-Optimal Design or Make I-Optimal Design.” p. the window in Figure 11.28) and select Sphere Radius. you can define general factor constraints on the factors. In this statement you can replace 1.28) and select Optimality Criterion. You can disallow any combination of levels of categorical factors if you have not already defined linear inequality constraints. Or. which varies depending on the design. . Specifying the Sphere Radius Value Augment designs can be constrained to a hypersphere. Selecting the Number of Random Starts To override the default number of random starts.28) and select Number of Starts.29 Changing the Number of Starts For additional information on the number of starts.0. click the red triangle icon in the Augment Design title bar (Figure 11. Disallowing Factor Combinations In addition to linear inequality constraints on continuous factors and constraining a design to a hypersphere.0 with any positive number. 1). click the red triangle icon in the Augment Design title bar (Figure 11. using JSL. The default criterion for Recommended is D-optimal for all design types unless you have used the RSM button in the Model panel to add effects that make the model quadratic. submit the following command before you build a custom design: DOE Sphere Radius = 1. click the red triangle icon in the Augment Design title bar (Figure 11. 71. To edit the sphere radius for the design in units of the coded factors (–1. The number you enter overrides the default number of starts. see “Why Change the Number of Starts?. 73.” p.11 Augmented Designs—Special Augment Design Commands 215 11 Augmented To modify the design criterion. Figure 11. When you select this command. see “Accounting for Factor Level Restrictions. For information on how to do this. Enter the appropriate value and click OK.29 appears with an edit box for you to enter the number of random starts for the design you want to build.

216 11 Augmented Designs—Special Augment Design Commands .

You can supply either one or two of the three values. the result is a plot of the other two. two-sample. and a number-of-parameters adjustment is specified. The calculations assume that there are equal numbers of units in each group. you must give JMP an estimate of the group means and sample sizes in a data table as well as an estimate of the within-group standard deviation (σ). and power. The Sample Size and Power command determines how large of a sample is needed to be reasonably likely that an experiment or sample will yield a significant result. given that the true effect size is at least a certain size. sample size. . If you supply only one of these values. You can apply this platform to more general experimental designs. JMP computes the third. Using the Sample Size and Power command when doing a prospective analysis helps answer the question. This capability is available for the single sample. If you supply two values. and the magnitude of the effect. given my proposed sample size. It requires that you enter any two of three quantities. and alpha level?” In this type of analysis. These depend on the significance level—alpha—of the hypothesis test for the effect and the standard deviation of the noise in the response. difference to detect. “Will I detect the group differences I am looking for. if they are balanced. estimate of within-group variance. and computes the third for the following cases: • difference between one sample's mean and a hypothesized value • difference between two samples means • differences in the means among k samples • difference between a variance and a hypothesized value • difference between one sample proportion and a hypothesized value • difference between two sample proportions • difference between counts per unit in a Poisson-distributed sample and a hypothesized value. power. and k-sample situations.12 Prospective Power and Sample Size 12 Power Use the DOE > Sample Size and Power command to answer the question “How many runs do I need?” The important quantities are sample size.

. . . . . . . . . . . . . . . . . . . . . Sigma Quality Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . k-Sample Means . . . One-Sample Variance . . . . . . . . . . . . . . . . . . . . . One-Sample and Two-Sample Means . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Two-Sample Means . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Single-Sample Mean . . . . . . . . . . . . . . . . . . . . . . . . . . . . Counts per Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Power and Sample Size Animation for a Single Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12 Contents Prospective Power Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219 220 221 223 224 225 226 227 229 230 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . One-Sample and Two-Sample Proportions . . . . . . . . . .

1 appears with button selections for experimental situations. The following sections describe each of these selections and explains how to enter estimated parameter values and the desired computation. The Sample Size and Power command in JMP helps estimate in advance either the sample size needed. or the effect size expected in the experimental situation where there is a single mean comparison. • Sample Size Sample size is how many experimental units (runs. When you select DOE > Sample Size and Power. • Error Standard Deviation Error Standard Deviation is the unexplained random variation around the means. Figure 12. power expected.1 Sample Size and Power Choices . • Effect Size Effect size is how different the means are from each other or from the hypothesized value. or samples) are involved in the experiment. a two sample comparison. or when comparing k sample means.12 Prospective Power and Sample Size—Prospective Power Analysis 219 12 Power Prospective Power Analysis The following five values have an important relationship in a statistical test on means: • Alpha Alpha is the significance level that prevents declaring a zero effect significant more than alpha portion of the time. the panel shown in Figure 12. • Power Power is the probability of declaring a significant result.

The power and sample size platform then calculates the missing item. in addition to fitting the means described in the situation. or Two Sample Means in the initial Sample Size selection list (Figure 12. given specific effect size • power as a function of effect size. and Power. and one or two of the other three values: Difference to detect. there are other factors with the extra parameters that can be specified here. the power calculations are an exercise in probability that calculates what might happen if the true values were as specified. • Error Std Deviation Error Std (Standard) Deviation is the true residual error.2 appears and asks for the anticipated experimental values. usually 0. Leave this field zero in simple cases. the number of extra parameters is . in a three-factor two-level design with all three two-factor interactions. for a given power. consider the two-sample situation. as shown here. If there are two unspecified fields. The values you enter depend on your initial choice.05. The Power and Sample Size dialog asks for the values depending the first choice of design: • Alpha Alpha is the significance level. the Power and Sample Size dialog in Figure 12.220 12 Prospective Power and Sample Size—One-Sample and Two-Sample Means One-Sample and Two-Sample Means After you click either One Sample Mean. given a sample size • effect size as a function of sample size. Even though the true error is not known. • Extra Params Extra Params (Parameters) is only for multi-factor designs. As an example. Sample Size. Figure 12.2 Initial Power and Sample Size Dialogs for Single Mean (left) and Two Means (right) The Two Sample Means choice in the initial Power and Sample Size dialog always requires values for Alpha and the error standard deviation (Error Std Dev). For example. In a multi-factor balanced design.1). the power and sample size platform constructs a plot that shows the relationship between those two values: • power as a function of sample size. This implies willingness to accept (if the true difference between groups is zero) that 5% (alpha) of the time a significant difference will be incorrectly declared.

experimental units. • Back Goes back to the previous dialog. but the cost is higher in sample size. or samples). See the section.90 or 0. rounding to 1. The dialog on the right in Figure 12. • Power Power is the probability of getting a statistic that will be declared statistically significant. Bigger power is better. • Animation Script The Animation Script button runs a JSL script that displays an interactive plot showing power or sample size.9.12 Prospective Power and Sample Size—One-Sample and Two-Sample Means 221 12 Power five—two parameters for the extra main effects. • Sample Size Sample size is the total number of observations (runs. • Difference to Detect Difference to detect is the smallest detectable difference (how small a difference you want to be able to declare statistically significant). To calculate the power when the sample size is 10.” p. Sample size is not the number per group. for an illustration of this animation script. but the total over all groups. Single-Sample Mean Suppose there is a single sample and the goal is to detect a difference of 2 where the error standard deviation is 0. 223. when there is one.95 if you can afford it.99998. Power is equal to alpha when the specified effect size is zero. it isn’t very important what values you enter here unless the experiment is in a range where there is very few degrees of freedom for error. and three parameters for the interactions. plan so that the experimental design has the power to detect a sizable effect.3. . “Power and Sample Size Animation for a Single Sample. For single sample problems this is the difference between the hypothesized value and the true value. You should go for powers of at least 0. shows the power is calculated to be 0. Computed sample size numbers can have fractional values. leave Power missing in the dialog and click Continue. In practice. which you need to adjust to real units.3. • Continue Evaluates at the entered values. If an experiment requires considerable effort. This is usually done by increasing the estimated sample size to the smallest number evenly divisible by the number of groups. as shown in the left-hand dialog in Figure 12.

95. The left-hand graph in Figure 12.3 A One-Sample Example To see a plot of the relationship of power and sample size.4.222 12 Prospective Power and Sample Size—One-Sample and Two-Sample Means Figure 12.4. leave both Sample Size and Power empty and click Continue. . a plot of power by difference appears. has the horizontal axis scaled from 1 to 8. Change the range of the curve by changing the range of the horizontal axis.4 A One-Sample Example Plot When only Sample Size is specified (Figure 12. the plot on the right in Figure 12.5) and Difference to Detect and Power are empty. which gives a more typical looking power curve. Double click on the horizontal axis to get any desired scale. Figure 12. For example.2 to 0. shows a range of sample sizes for which the power varies from about 0.

05.4.5 Plot of Power by Difference to Detect for a Given Sample Size Power and Sample Size Animation for a Single Sample The Animation Script button on the Power and Sample Size dialog for the single mean displays an interactive plot that illustrates the effect that changing the sample size has on power. Also. You can drag the handles over the curves drag to see how their positions affect power. The probability of committing a Type II error (not detecting a difference when there is a difference). In the example shown in Figure 12. Sample Size is 10. .4 (the difference to be detected). Alpha is 0.6. The animation begins showing a normal curve positioned with mean at zero (representing the estimated mean and the true mean). often represented as β in literature.12 Prospective Power and Sample Size—One-Sample and Two-Sample Means 223 12 Power Figure 12. and another with mean at 0. and the Difference to Detect is set to 0. is shaded in blue on this plot. you can click on the values for sample size and alpha showing beneath the plot to change them.

9. the Difference to Detect is the difference between two means. Suppose the error standard deviation is 0. the desired detectable difference is 1. .7. To increase the power requires a larger sample. and the sample size is 16.7. The crosshair tool estimates that a sample size of about 35 is needed to obtain a power of 0.6 Example of Animation Script to Illustrate Power Two-Sample Means The dialogs work similarly for two samples. leave both Sample Size and Power blank and examine the resulting plot. The comparison is between two random samples instead of one. To find out how large. This is considerably lower than in the single sample because each mean has only half the sample size.9 (as before).5433. 0. as shown in the dialog on the left in Figure 12. shown on the right in Figure 12. Leave Power blank and click Continue to see the power calculation.224 12 Prospective Power and Sample Size—One-Sample and Two-Sample Means Figure 12.

5)2+(0.9.7 Plot of Power by Sample Size to Detect for a Given Difference k-Sample Means The k-Sample Means situation can examine up to 10 kinds of means.5)2. as shown in the dialog on the left in Figure 12. When a sample size of 16 is entered the power calculation is 0. the power and sample size calculations produce the power curve shown on the right in Figure 12.8. Notice that the difference in means is 2. calculated as square root of the sum of squared deviations from the grand mean.95.8.236. In this case it is the square root of (–1.5)2+(1. .5)2+ (–0. which is the square root of 5.12 Prospective Power and Sample Size—k-Sample Means 225 12 Power Figure 12. If both Sample Size and Power are left blank. This confirms that a sample size of 16 looks acceptable. and the Error Std Dev is 0. The next example considers a situation where 4 levels of means are expected to be about 10 to 13.

gov/div898/handbook. . An example is when the variance for resistivity measurements on a lot of silicon wafers is claimed to be 100 ohm-cm and a buyer is unwilling to accept a shipment if variance is greater than 55 ohm-cm for a particular lot. The examples throughout the rest of this chapter use engineering examples from the online manual of The National Institute of Standards and Technology (NIST). As with previous dialogs. entered as the Baseline Variance.nist. specify a baseline variance. To indicate direction of change. you enter two of the items and the Power and Sample Size calculations determines the third. select either Larger or Smaller from the Guarding a change menu.226 12 Prospective Power and Sample Size—One-Sample Variance Figure 12.8 Prospective Power for k-Means and Plot of Power by Sample Size One-Sample Variance The One-Sample Variance choice on the Power and Sample Size dialog (Figure 12. In the dialog. You can access the NIST manual examples at http://www.The usual purpose of this option is to compute a large enough sample size to guarantee that the risk of accepting a false hypothesis (β) is small. alpha level.itl. and direction of change you want to detect.1) determines sample size for detection of a change in variance. The computations then show whether the true variance is larger or smaller than its hypothesized value.

as indicated on the dialogs (Figure 12. When there are two samples. the Baseline Proportion is the average of a known baseline proportion and the single sample proportion. The dialogs are the same except you enter Baseline Proportion and also specify either a one-sided or two-sided test. . the computed result shows that you need a sample size of 170. enter a negative amount in the Difference to Detect box.05 and power of 0. The sampling distribution for proportions is actually binomial. Note: Remember to enter the variance in the Baseline Variance box. with an alpha of 0. When you click Continue. the Baseline Proportion you enter is the average of the two sample proportions.9 Sample Size and Power Dialog To Compare Single-Direction One-Sample Variance One-Sample and Two-Sample Proportions The dialogs and computations to test power and sample sizes for proportions are similar to those for testing sample means. Enter these items as shown in Figure 12. Figure 12. For a one-sample situation. not the standard deviation.99. If you want to detect a change to a small variance.9.10). but the computations to determine sample size and test proportions use a normal approximation.12 Prospective Power and Sample Size—One-Sample and Two-Sample Proportions 227 12 Power Suppose you want to detect an increase of 55 for a baseline variance of 100.

Figure 12.11 shows the entries in the Sample Size and Power dialog to detect a given difference between an observed proportion and a baseline proportion. or average of two-sample proportions.10 Default Power and Sample Dialog for One-Sample and Two-Sample Proportions Enter average of baseline and one-sample proportions. The Baseline Proportion in this example is 0. leave both Difference to Detect and Sample Size blank. Use the grabber tool (hand) to move the x-axis and show a specific range of differences and sample sizes.228 12 Prospective Power and Sample Size—One-Sample and Two-Sample Proportions Figure 12. The example process is monitored with a one-sided test at 5% alpha and a 10% risk (90% power) of failing to detect a change of that magnitude. Figure 12. which is the average of the baseline (10%) and and the proportion above the baseline (6%).The manager does not want to stop the process unless it has degenerated greater than 16% defects (6% above the 10% known baseline defective). To see the plot on the right in Figure 12.or two-sided) Testing proportions is useful in production lines.11. where proportion of defects is part of process control monitoring. and the computed sample size of approximately 77. Enter 1 or 2 to indicate the type of test (one. For example. suppose a line manager wants to detect a change in defective units that is 6% above a known baseline of approximately 10% defective.08.11 Dialog To Compare One Proportion to a Baseline and Sample Size Plot .

Click the Counts per Unit button. 2. A unit can be an area and the counts can be fractions or large numbers. Click Continue to see the computed sample size of 8. Enter the Difference to Detect in terms of the baseline count per unit (defects per unit).9. Figure 12.12 Prospective Power and Sample Size—Counts per Unit 229 12 Power Counts per Unit The Counts per Unit selection calculates sample size for the Poisson-distributed counts typical when you can measure more than one defect per unit. Questions of interest are: • Is the defect density within prescribed limits? • Is the defect density greater than or less than a prescribed limit? Enter alpha and the baseline count per unit. Select DOE > Sample Size and Power. Although the number of defects observed in an area of a given size is often assumed to have a Poisson distribution. Then enter two of the remaining fields to see the calculation of the third. alpha is sometimes called the producer’s risk and beta is called the consumer’s risk.12 Dialog For Counts Per Unit Example . As an example. The test is for one-sided (one-tailed) change. Enter a power of 0.1 to be the chance of failing the test if the new process is as good as the target. the area and count are assumed to be large enough to support a normal approximation. Enter an alpha of 0. 5. 4. consider a wafer manufacturing process with a target of 4 defects per wafer and you want to verify that a new process meets that target. The computed sample size is expressed in those units.128 (Figure 12. which is the chance of detecting a change larger than 2 (6 defects per wafer). 1. 3. The process meets the target if there are less than 48 defects (6 defects per wafer in a sample of 8 wafers). In this kind of situation.12).

000 opportunities. are a Sigma quality level of 5. 3. 4. Click Continue. 2. Select DOE > Sample Size and Power.5.13 Sigma Quality Level Example 1 If you want to know how many defects reduce the Sigma Quality Level to “six-sigma” for 1.000 opportunities: 1.14). by entering any two of the following three quantities: • number of defects • number of opportunities • Sigma quality level When you click Continue.13. the sigma quality calculator computes the missing quantity using the formula Sigma Quality Level = NormalQuantile(1 – defects/opportunities) + 1.000. enter 6 as the Sigma Quality Level and leave the Number of Defects blank (window to the left in Figure 12. as shown in the window on the right in Figure 12. Enter 50 for the number of defects and 1. Figure 12. as shown in the window to the left in Figure 12.14) shows that the Number of Defects cannot be more than approximately 3. As an example. .000.000.4. use the Sample Size and Power feature to compute the Sigma quality level for 50 defects in 1. The computation (window to the right in Figure 12.230 12 Prospective Power and Sample Size—Sigma Quality Level Sigma Quality Level Use the Sigma Quality Level feature. accessed by selecting DOE > Sample Size and Power.13.3.000 as the number of opportunities. Click the Sigma Quality Level button. The results.

12 Prospective Power and Sample Size—Sigma Quality Level 231 12 Power Figure 12.14 Sigma Quality Level Example 2 .

.

J.C. Optimum Experimental Designs Clarendon Press. 1.W. 214–219. Empirical Model–Building and Response Surfaces. Cook. Bose. “Simultaneous Optimization of Several Response Variables. and Suich. Box. and Draper.” Technometrics 28. (1986).” Journal of the Royal Statistical Society. 1–40. D. (1988). R.D. Daniel. C.” Technometrics 2. (1993). N.S. R. Box. New York: John Wiley and Sons. G. and Wood.. Revised Edition. C. (1980). R. Statistics for Experimenters. Milwaukee. p131-145. pp. 1980. D.” International Statistical Review. Cornell.” Technometrics. D. “An Analysis of Unreplicated Fractional Factorials.G.F. Oxford (1992) p.” Technometrics 32.E. (1990). 311–314.E. Vol. 107-166. Box..E. and Nachtsheim. (1987). Part 2.P. Byrne.W. 61.X. Second Edition. Box. G. G. Series B. 11–18.E. “Some New Three-Level Designs for the Study of Quantitative Variables.P. New York: John Wiley and Sons. 13. (1990).References References Atkinson. J. Box. A. and Cox. N. 1. D.E. “A Catalogue of Two-level and Three-Level Fractional Factorial Designs with Small Runs. C. WI: American Society of Quality Control. and Meyer. New York: John Wiley and Sons. Sun. . “On the Experimental Attainment of Optimum Conditions. (1959). and Wu. “Signal–to–Noise Ratio. “Use of Half–normal Plots in Interpreting Factorial Two–level Experiments. and Donev. and Taguchi. G. Oct. and Transformations. ASQC 40th Anniversary Quality Control Congress Transactions. F.D. G. G. G. (1960). Daniel C.E. and Wilson.” Journal of Quality Technology. 12:4. C.B. and Hunter.P. (1947) “Mathematical Theory of the Symmetrical Factorial Design” Sankhya: The Indian Journal of Statistics. Cochran. W. Second Edition New York: John Wiley & Sons.A.” Technometrics 30. (1951). 455–475. New York: John Wiley and Sons.J. 168–177.R.G. “Letter to the Editor: Response to James M. 8. Chen.. Inc. K. Box.M. (1986).P. (1957). G. Fitting Equations to Data.P. (1980).P. (1978). J. Derringer. R. J. Lucas. Hunter. A. and Behnken. International Statistical Institute.. 1-45. Performance Criteria. Experiments with Mixtures.M.. Experimental Designs.148. Inc. 363-364.

D. Mitchell. 469–473. Meeker. “A New Class of Supersaturated Design. Boston: Allyn and Bacon. (1990). Experimental Design in Biotechnology. “An algorithm for the Construction of D-Optimal Experimental Designs. (1974).K. 8.M. P. R. (1989). 16:2. Technometrics. K. Mahalanobis. J. P.” Technometrics 25. N.” 1991 Joint Statistical Meetings. Response Surface Methodology. Bradley (1991).J. and Nachtsheim. Steinberg. “Some Guidelines for Constructing Exact D–Optimal Designs on Convex Design Spaces. (1972).. and Draper. G. Myers.P. and Vining. J. (2002) The Optimal Design of Blocked and Split-Plot Experiments. New York: Springer Haaland. 125-139.(1996).. Georgia Khuri..M. “D-Optimality for Regression Designs: A Review. and Cornell.D.C. R.C. Vol. (1995). (1975). “Bayesian Design and Analysis of Computer Experiments: Use of Derivatives in Surface Prediction. G. I. pp. “Sankhya.L. (1976). and Ylvisaker.A. R. C. 36. R. M. Morris. 305–325. Hahn.” Technometrics. (1976) Response Surface Methodology. “A Simple Bayesian Modification of D-Optimal Designs to Reduce Dependence on an Assumed Model.J. p307. (1989). “The Design of Optimum Multifactorial Experiments. D.. Follow-up Designs to Resolve Confounding in Multifactor Experiments. G. (1994). J. G. C. Kowalski. Lucas.” Biometrika.R. 38:4. and Burman. “An Interactive Graph For Exploring Multidimensional Response Surfaces. R. Virginia Polytechnic and State University. The Coordinate Exchange Algorithm for Constructing Exact Optimal Designs.M. P..A. New York: Marcel Dekker. P. and Jones. 35. Meyer. New York: Macmillan Publishing Company.” Technometrics 44: 72-79. John. John.” Technometrics. Myers.” Journal of Quality Technology.” The Indian Journal of Statistics.. 17 pp 15-23. Johnson. W. New York: Marcel Dekker. Plackett. 271–277. 33. Jones.W.G. 140-157.E. (1993).” Technometrics.I.J.” Technometrics. (1947). Piepel. #3. “Programs for Generating Extreme Vertices and Centroids of Linearly Constrained Experimental Regions. P.. Meyer. Atlanta. “Letter to the Editor: Comments on Cook and Nachtsheim (1989). 32. T.. (1983). D.” Journal of Quality Technology 20:2. J.” Technometrics. Statistical Design and Analysis of Experiments.V. Inc.. (1947).” Technometrics. St. (1988). Cornell.J. Vol 8. Lin. “Quick and Easy Analysis of Unreplicated Fractional Factorials. R. 243-255.” Technometrics. and Box. Part 2. Inc. (1988). 31. “The Evaluation and Comparison of Experimental Designs for Fitting Regression Relationships. 363–364.H. . T. J. and Nachtsheim.M. M.Q.F. Goos. J.D. pp. W. (1993).H.” Technometrics 35:2. R.203-210.234 References DuMouched. Vol 37. 37–47. S. 28-31. Mitchell. 60-69. A. (1987) Response Surfaces: Design and Analysis. (2002) “Split Plot Designs and Estimation Methods for Mixture Experiments with Process Variables. Lenth. B. and Feder.. D. April. pp.

(1974). 26. 149–159. Japan: Central Japan Quality Control Association. (1979). 344-360.J. and Marquardt. Snee. G. 391–408. W. 17:2.” Technometrics.D. R. Snee.W.D. (1985)Computer Aided Design of Experiments . Ronald D. R. (1975). Snee.D. Taguchi. 4 October 1985 p. . H. No. A8(4). D. 16. 303–326.” Nagoya.References 235 References Sheffé. (1976). (1984).” Commun. “Computer-Aided Design of Experiments for Response Estimation. 217–224. Journal of Quality Technology.” Technometrics. “Extreme Vertices Designs for Linear Mixture Models.” Technometrics. “Experimental Designs for Quadratic Models in Constrained Mixture Spaces. R. Snee. JRSS B 20.231. Vol 17. (1958) Experiments with Mixtures.Some Practical Experiences. “Experimental Designs for Mixture Systems with Multicomponent Constraints. “An Introduction to Quality Control. Welch. Statistics.

.

central composite designs 112 Bounce Data.Index Index Design of Experiments A ABCD designs 139 aberration designs 91 acceptable values See lower limits and upper limits Actual-by-Predicted plots 99 adding center points in augment designs 207. 103.jmp 169 chemical mixture. 59 linear constraints 135 runs and terms in augment designs 211 additional runs 65.jmp 186 C Categorical factors 60. 97. 190 augment designs choices provided 204 extending experiments 204 augmentation 210 axial points 101 scaling. column property 76 coefficients. 64 aliasing effects 93 Alpha 219–220 animation scripts 221 Anova reports 142 assigning importances (of responses) 57. relative variance 63–64 column properties coding 76 constrained state 68 design roles 67. 209 factors 39. 113. 215 CONAEV method 139 confounding 94.jmp 103 Bounce Factors. 53 Big Class. 209 central composite designs 101 response surface designs 101 simplex centroid designs 131 central composite designs 101. 99 resolution numbers 91 confounding pattern 64.jmp 45 block sizes 26 Borehole Factors. examples 143 coding. 78 mixture 76 responses limits 77 columns design roles 78 factors for mixture experiments 76 combinations disallowed 73. 90 CCD See central composite designs center points augment designs 207. 126 aias matrix 18 algorithms CONSIM 139 coordinate exchange 80 alias matrix 18.jmp 161 Borehole Sphere Packing. 111 See also Response Surface designs Byrne Taguchi Factors. 111–112 See also response surface designs centroid points 135 Chakravarty 92 Chemical Kinetics. 94 CONSIM algorithm 139 constraints B balanced designs 62 Bayesian D-Optimal designs 52 Bayesian D-optimal designs 29.jmp 103–104 Box-Behnken designs 101. 88.jmp 162 .

133 space-filling 149–165 Sphere Packing 161 uniform precision 112 desirability functions 106. 125 full factorials 90 Latin Hypercube 154 minimum aberration 91 mixed-level 92 mixture 129. as column properties 77 determinant formula 72 diagnostics for custom designs 69. 52–53 optimality criteria 70. 62 examples 15. 111 central composite 101 foldover 208 fractional factorials 90 full factorial 81.jmp 42 D-optimal augmentation 210 designs 29. See DOE 76 design roles 78 designers augment 204 designs ABCD 139 aberration 91 augment 204 balanced 62 Bayesian D-optimal 29. 130 design matrix table properties 70. 53 Box-Behnken 101. 26 factors. 215 Disallowing factor combinations 215 disallowing factor combinations 73 DOE simple examples 83 DOE (Design of Experiments) coding 76 DOE Sphere Radius 215 Donev Mixture factors. 90 contour plots 143 control factors 183. 103. 185 CONVRT method 139 coordinate exchange algorithms 80 Cotter designs 92 counts per unit (power and sample size) 229 creating factors tables 68 criterion optimality 70. 88 desirability values. 143 orthogonal screening designs 90 screening experiments 127 surface designs 112 orthogonal arrays 92 Plackett-Burman 91 replicating 204–205 response surface 101 saturated 62 screening 81 simplex centroids 130 simplex lattice 130. 189 maximizing 106 traces 106 values 58. 215 entering 68 linear 138 saving 68 continuous factors 60. defining 57 how they work 80 models. 214 resolutions 91 roles 67 Design of Experiments. 224 disallowed combinations 73. 117. describing 61 screening 17 D defaults number of random starts 71 defects 229 D-efficiencies 41 describing models 61. 215 Cubic Model.238 Index adding 135 disallowing combinations 73. 215 E effect aliasing 94 eigenvalue 105 eigenvector 105 . 214 Difference to Detect option 220–221.jsl 35 custom designs data tables 65 Design Generation panel 7.

90 continuous 60. 190 importance values. 117. 117. 188 factorial designs fractionals 90 full 81. entering 77 importance weight 67 inner arrays. 125 three level 92 factors adding 39. 59 categorical 60. 90 control factors 183. 89 minimizing and maximizing 58.Index 239 size 219 sparsity 81. 107. inner designs 183. 135. 215 I identifying key factors 81 importance of responses 57. 145. 183 efficiencies D. 215 F factor column 79–80 for mixture experiments 76 factor combinations. disallowing 73 Factor Profiling option 105. 113. 131. 97. 123. 90. 185 key factors 81 nonmixture 42 tables. 126. 189 J JSL (JMP Scripting Language) animation scripts 221 augmented designs 198 examples for DOE 35 sphere radius 215 K Keep the Same command 65. 125 Design Generation panel 62 examples 119 functions desirability 106. 214 eigenvalue of effect 105 eigenvector of effect 105 equivalent solutions 80 error standard deviation 219–220 error variance 33 excluding factor combinations 73 extra parameters 220 extreme vertices 135 finding subsets 136 G global optimum 71 goal types 57. and A 69. 90. 88 goals for DOE (Design of Experiments) response variables 77 Group New Runs into Separates Block 207 Index H hidden intercepts 142 hyperspheres 72. 88 goals matching targets 58. 185 Inscribe option 112 interactions 92 high-order 91 intercepts hidden 142 I-optimal designs 52 optimality criteria 70. G. 139 k-Sample Means (power and sample size) 225 L L18 Chakravarty 92 L18 Hunter 92 L18 John 92 L36 92 . 92 effects nonestimable 91 orthogonal 112. 88. creating 68 false negatives 92 finding subsets (extreme vertices) 136 fitting mixture designs 141 fixed covariates example 45 folding the calculation 143 foldover designs 208 fractional factorial designs 90 full factorial designs 81.

113. 104. outer designs 183. 97. 140 points axial 101 center See center points centroid 135 Poisson-distributed counts 229 potential terms (DOE) 29 power N N factors. 131. 185 P parameters. 185 nonestimable effects 91 nonlinear design 178 nonmixture factors 42 number of runs 15. 126 patterns confounding 94 performance statistics 185 Plackett-Burman designs 91 Plastifactors. 135. 138 local optimum 71 no-intercept model 76 noise factors 183. 62 screening designs 91 number of starts 71. 114. column properties 76 Model script Model Specification dialog 98. 126. responses 77 linear constraints 135. 114 models custom design 61 describing 61. adding 59 N Responses button 57. 215 M matching targets 77 matching target goals 57–58. 133 with nonmixture factors 42 mixture. 214 matrix. 98. 215 Optimality Criterion 214 optimality criterion 70 order of runs 65. 88–89 matrix 70. 88 means one and two sample 220 methods CONAEV 139 minimize responses 57. 154–157 limits. 88 minimum aberration designs 91 Minimum Potential 159 mixed-level designs 92 mixture experiments (column property) 76 mixture designs 129 compared to screening designs 127 definition 127 examples 143–145 fitting 141 linear constraints 138 response surfaces 145 simplex centroids 130. 114 larger–the–better formulas (LTB) 187 Latin Hypercube space filling design 149. 88 minimizing goals 58. 139 orthogonal array designs 92.jmp 143 plots Actual-by-Predicted 99 contour 143 spinning 109 ternary 127. 88 maximizing desirability 106 goals 58. 183 orthogonal designs screening designs 90 screening experiments 127 surface designs 112 Orthogonal option 112 outer arrays. 130 O On Face option 112 one-sample and two-sample means 220 one-sample proportion (power and sample size) 227 one-sample variance (power and sample size) 226 optimal determinant formula 72 optimality criteria 70. extra 220 Pattern column 84. alias 64 maximize responses 57.240 Index L9-Taguchi 186 Label column 98. 88 .

139 requesting additional 65.jmp 119–120 Reactor Response. 126 order they appear in table 65. 113. 37 traces 106 variances 33. 131. 88 goals 57. 97. design 78 Rotatable option 112 RSM (Response Surface Methodology) 110 runs additional 65. 107–108 introduction 110 purpose 101 reports 105 with blocking factors 39 with categorical factors 39 Response Surface Methodology (RSM) 110 response surfaces effects 143 mixture designs 145 responses custom designs 57. 136.jmp 195 Reactor Augment Data. 113. 135. 139 range constraints 135 Reaction Kinetics Start. 88 responses. 88 saving 67 upper limits 58.jmp 198 Reactor Factors. 113. 97. 131. 97. 97. 88 goals. 97.jmp 119 regressor columns 92 relative proportions See mixture designs Relative Variance of Coefficients 29 relative variance of coefficients 63–64 replicating designs 204–205 requesting additional runs 65. 113. 126. 126. 88.Index 241 Index analyses 219 in statistical tests on means 219 one-sample and two-sample means 220–221 power and sample size calculations 217–231 animation 223 counts per unit 229 k-sample means 225 one-sample and two sample proportions 227 one-sample mean 221 one-sample variance 226 sigma quality level 230 two-sample means 224 prediction profilers 33. 113.jmp 119 Reactor 8 Runs. 215 random starts 71 Randomize within Blocks 65.jmp 174 Reactor 32 Runs. columns 75 proportions (power and sample size) 227 pseudocomponent (mixture column property) 77 pure error 207 Q quadratic model 32–64 R radius. 139 randomizing runs 65. 126 screening designs 91 S sample means 220 Sample Size and Power command 217 sample sizes example comparing one proportion to baseline and sample size plot 228 example comparing single-direction one-sam- . multiple 77 RMSE 99. 112 prediction formulas saving 145 Prediction Profiler 105 prediction variance profile plot 32 primary terms (Bayesian D-optimal design) 29 profilers mixture response surface 145 prediction profilers 33. 126 rescaling designs 112 resolution numbers 91 resolutions of designs 91 response limits. sphere 72. 131. 97. 110 desirability values 58. desirability functions 106 lower limits 58. 135. 113. 37 properties. column property 77 response surface designs examples 103. 123 robust engineering 183 roles design roles 78 roles.

126. 140 traces desirability 106 trade-off in screening designs 91 tutorial examples augment designs 195–203 custom designs 15. 97. 139 Sort Right to Left 65. 139 space-filling designs 149–165 simplex lattice 133 simplex centroids 130 sparsity. 97. 126. 131. 224 two-sample proportion (power and sample size) 227 U Uniform (space filling design) 149. error 219 star points 101 starts.242 Index ple variances 226 example with counts per unit 229 one and two sample means 221 prospective power analysis 219 screening designs 117 saturated designs 62 saving constraints 68 prediction formulas 145 responses 67 X Matrix 70. 26 DOE 83 full factorial designs 119 mixture designs 143–145 response surface designs 103. 131. 215 starts. 89 variance error and prediction 33 . 92 sphere packing Borehole problem 161 methods 151 space filling designs 149 sphere radius 72. creating 68 making in custom designs 65 Taguchi designs 183–189 description 183 methods 183 target value 77 target values 58. 157 uniform precision designs 112 User Defined option 112 V values high and low for columns (in DOE) 76 target 58. finding 136 supersaturated designs 22 surface designs See response surface designs T tables factors table. 135. 107–108 screening designs 85 two-level categorical 84 two-level fractional factorials 90 two-level full factorials 90 two-sample and one-sample means 220. performance 185 subsets. number of 71. 89 ternary plots 127. 113. 214 scaling axial 112 designs 112 Scheffé polynomial 141 screening designs 81 custom designs 17 design types 90 dialogs 85 examples 85–113 scripts animation 221 generating the analysis model Model script See Model table property scripting See JSL sigma quality level (power and sample size) 230 signal factors 183 signal-to-noise ratios 183 simplex 127 centroid designs 130 lattice designs 130. random 71 statistics. 215 spinning plots Box-Behnken designs 109 Split plot designs 60 split plot designs 65 standard deviation. 136. effect 81. 113. 133 single-sample means (power and sample sizes) 221 solutions equivalent 80 Sort Left to Right 65.

190 weight. importance 57. 88. saving 70.Index 243 variance of prediction 112 vertices extreme 135 extreme. 214 XVERT method 139 . response importance 78 whole model test 142 X Matrix. finding subsets 136 Index W-Z weight importance 67 weight.

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd