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

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

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

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

.

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

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

Refer to subsequent chapters in this book for more examples and procedures on how to design an experiment for your specific project. 1 Introduction .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.

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

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

Rename the response labeled Y by double-clicking it and typing “Number Popped.2 Open and Close the Responses Panel by Clicking the Disclosure Icon You have more than one response. so enter both in the custom designer in place of the default Y response: 1. 2.3. . you have two responses: • the number of popped kernels • the total number of kernels in the bag. By default. JMP labels this response Y2 by default. define the responses and factors.” The completed Responses panel looks like Figure 1.4 1 Introduction to Designing Experiments—My First Experiment Step 1: Design the Experiment The first step is to select DOE > Custom Design (Figure 1.1 Select the Custom Designer Define the Responses: Popped Kernels and Total Kernels In this experiment. 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. the custom designer contains one response labeled Y (Figure 1. Add the second response (total number of kernels) by clicking the Add Response button and choosing None from the menu that appears. Figure 1. leave the goal at Maximize. Rename the response labeled Y2 by double-clicking it and typing “Total Kernels. Figure 1. Then. 3.2).1).” Since you want to increase the number of popped kernels.

Add Power as a two-level continuous factor: 7. Click Add Factor and select Continuous. 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. Add Time as a two-level continuous factor: 4. Change the name of the factor (currently named X2) by double-clicking it and typing Time. . Click Add Factor and select Continuous.4. 8. Change the name of the factor (currently named X3) by double-clicking it and typing Power.Click Continue. Power. 3. Rename the default levels (currently named -1 and 1) as 5 and 10 by clicking the current name and typing.3 Renamed Responses with Specified Goals Define the Factors: Time. 6. and Brand In this experiment. Click Add Factor and select Categorical > 2 Level. Change the name of the factor (currently named X1) by double-clicking it and typing Brand. 9. Rename the default levels (currently named -1 and 1) as 3 and 5 by clicking the current name and typing. The completed Factors panel looks like Figure 1. 2.4 Renamed Factors with Specified Values 10. Figure 1.1 Introduction to Designing Experiments—My First Experiment 5 1 Introduction Figure 1. 5. Rename the default levels (currently named L1 and L2) as Top Secret and Wilbur by clicking the current name and typing.

. For example. 2. Figure 1. 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. • 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. You decide to include all the possible two-factor interactions in your a priori model of the popcorn popping process. 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. 3. but greater than or equal to 10: 1. Be sure to change <= to >= in the second constraint. Click the Add Constraints button twice.5. as shown to the right in Figure 1. once for each of the known constraints.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. This kind of synergistic effect of factors acting in concert is called a two-factor interaction. Complete the information. which avoids combinations of Power and Time that sum to less than 10 and more than 13. JMP adds two-factor interactions to the model. 1.5. From experience. the area inside the parallelogram is the allowable region for the runs. To limit the experiment so that the amount of popping time combined with the microwave power level is less than or equal to 13. and the power settings on the microwave are between five and 10.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. Click the Interactions button and select 2nd. As illustrated on the left in Figure 1.6 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. Open the Constraints panel by clicking the disclosure button beside the Define Factor Constraints title bar.

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

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

and so forth. According to the table. You do not have infinitesimal control over the power and time settings on a microwave. 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. Tip: Before clicking Make Table in the Output Options area.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. the difference is negligible. you can select Sort Left to Right in the Run Order menu to have JMP present the results according to the brand. Although this altered design is slightly less optimal than the one the custom designer suggested. then cook Top Secret for five minutes using a power setting of five. .

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

Scroll to the bottom of the report to open the Prediction Profiler.7 Investigating P-Values p-values indicate significance. You can therefore conclude that. Select Prediction Profiler from the red triangle menu on the report title bar. 4. Figure 1. .05) indicates that results are statistically significant. asterisks indicate the low p-values. To further investigate. A low p-value (a value less than 0.6 Fitting the Model 3. Values with * beside them are p-values that indicate the results are statistically significant. This column lists p-values. In this column.7) and look in the column labeled Prob>Chisq.1 Introduction to Designing Experiments—My First Experiment 11 1 Introduction Figure 1. use the Prediction Profiler to develop an understanding of how changes in the factor settings affect the numbers of popped and unpopped kernels: 1. Open it by clicking the disclosure button beside the Prediction Profiler title bar. and brand and the proportion of popped kernels. all the model effects except the two-factor interaction involving Brand and Power are highly significant. You have confirmed that there is a strong relationship between popping time. in this experiment. power. Scroll down to view the Effect Tests table (Figure 1. Click the Run Model button. The Profiler (Figure 1.8) displays prediction traces for each factor.

Figure 1. Now for the final steps.12 1 Introduction to Designing Experiments—My First Experiment Figure 1. the steepness of a prediction trace reveals a factor’s importance. The fact that there is a substantial slope shift tells you that there is an interaction (synergistic effect) involving time and power. . JMP automatically adjusts the graph to display the optimal settings at which the most kernels will be popped (Figure 1. 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.9).9 Moving the Time Value from 4 to 5 As time increases and decreases. Because the prediction trace for power is steeper than that for brand or time (see Figure 1. Click the red triangle icon in the Prediction Profiler title bar and select Desirability Functions. 4. you can see that the microwave power setting is more important than the brand of popcorn or the amount of cooking time.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.9). 3. Click the red line in the time graph and drag it right and left (Figure 1. the curved time and power prediction traces shift their slope and maximum/minimum values as you change the values of time.10). time = 4) 2.

79 .10 The Most Desirable Settings The best settings are: • the Wilbur brand • cooking time at 5 • power set at 7.8 (or 8). The experiment predicts that cooking at these settings will yield greater than 97% popped kernels. Figure 1. and use a power level of 7.1 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. cook for five minutes.

.

the most cost-beneficial of these methods for quality and productivity improvement is statistical design of experiments. then JMP tailors a design to suit your unique case. measure response for each run Key mathematical steps: appropriate computer-based tools are empowering . This chapter presents several examples showing the use of custom designs.To use the custom designer. predict. Custom designs accommodate any number of factors of any type. A trial-and -error search for the vital few factors that most affect quality is costly and time-consuming. This approach is more general and requires less experience and expertise than previous tools supporting the statistical design of experiments. you first enter the process variables and constraints. JMP’s custom designer is the recommended way to describe your process and create a design that works for your situation. You can also control the number of experimental runs. 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. which can be any number greater than or equal to the number of unknowns in the model.2 Examples Using the Custom Designer The use of statistical methods in industry is increasing. Arguably. This makes custom design more flexible and more cost effective than alternative approaches. 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. and then cost-effectively improve the behavior of any system or process. The purpose of experimental design is to characterize.

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

4.” p. click Make Design. 3. but they are confounded with two factor interactions. .1 shows the six factors. 4. Figure 2. Using the default of eight runs. 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. The straightforward screening examples described below show that ‘custom’ is not equivalent to “exotic. Click Continue. Select DOE > Custom Design. All the main effects are estimable. 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. 2. Note to DOE experts: The result is a resolution-three screening design. but the custom designer is more flexible and general. As such. The default model contains only the main effects. for details).” The custom designer is a general purpose design environment.

.18 2 Examples Using the Custom Designer—Creating Screening Experiments Figure 2. Figure 2.1 A Main-Effects-Only Screening Design 5. Click the disclosure button ( on Windows/Linux and on the Macintosh) to open the Alias Matrix. 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. the columns labeled 2 6 and 3 4 in the table have a 1 in the row for X1.2 shows a table of zeros.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. ones and negative ones. Figure 2. For example. You are assuming that these interactions are negligible in size compared to the real effect of X1.

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

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. 2. There is only one value of 1 in that column. 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. The 1 occurs in the row labelled X1*X2. Select DOE > Custom Design. . there are 15 possible two-factor interactions. All the other rows and columns are similar. Look at the column labelled 1 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). Add all the two-factor interactions by clicking the Interactions button and choosing 2nd. With six factors. The default estimability of these terms is Necessary. Select all the interaction terms and click the current estimability (Necessary) to reveal a menu.5. main effects and two-factor interactions is 22. 5. as shown in Figure 2. The model includes the main effect terms by default. Figure 2. 4. Click Continue. This means that the expected value of the two-factor interaction X1*X2 is not biased by any other terms. Define six continuous factors (X1 . The first example in this section showed an eight-run design that fit all the main effects. 3. 1. The minimum number of runs that could fit the constant. Click the disclosure button to open the Alias Matrix.20 2 Examples Using the Custom Designer—Creating Screening Experiments 6.4 shows a table of zeros and ones. This is more than the resource budget of 16 runs. All the others are 0. This example shows how to obtain such a design compromise using the custom designer. Change Necessary to If Possible. Figure 2. suppose there are six continuous factors and resources for n = 16 runs.

The values in your table may be different from those shown below. but with a similar pattern.2 Examples Using the Custom Designer—Creating Screening Experiments 21 2 Examples Figure 2. Type 16 in the Number of Runs edit box. the custom designer builds a design to estimate as many two-factor interactions as possible. After the custom designer creates your design. Although the desired number of runs (16) is less than the total number of model terms. click the disclosure button to open the Alias Matrix (Figure 2. 7.5 Model for Six-Variable Design with Two-Factor Interactions Designated If Possible 6.6). Click Make Design. .

The row labelled X1*X2 has the value 0. which means that the Intercept and main effect terms are not biased by any two-factor interactions. In a saturated design. 1993). the number of runs equals the number of model terms. but there are no degrees of freedom for error or for lack of fit. 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. Creating “Super” Screening Designs This section shows how to use the technique used in the previous example to create “super” screening (supersaturated) designs. which makes them attractive for factor screening when there are many factors. Factor screening relies on the sparsity principle. In a supersaturated design. The problem is not knowing which are the vital few factors and . and experimental runs are expensive. In the analysis of a saturated design. as the name suggests. you can (barely) fit the model. saturated designs represented the limit of efficiency in designs for screening.6 Alias Matrix All the rows above the row labelled X1*X2 contain only zeros. Two-factor interactions are estimable but are confounded with other two-factor interactions.22 2 Examples Using the Custom Designer—Creating Screening Experiments Figure 2.25 in the 1 2 column and the same value in the 5 6 column. Note to DOE experts: The result in this particular example is a resolution-four screening design. The experimenter expects that only a few of the factors in a screening experiment are active. Until recently. the number of model terms exceeds the number of runs (Lin. Supersaturated designs have fewer runs than factors. 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. A supersaturated design can examine dozens of factors using fewer than half as many runs.

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

9. the analysis picks up the same three factors.24 2 Examples Using the Custom Designer—Creating Screening Experiments Figure 2.10 and click Apply. Note that random noise is added to the Y column formula. Figure 2. so the numbers you see might not necessarily match those in the figure. The values in your table may be different from those shown below. Because you have set X2.8 Simulating Responses 6. then click Make Table. The values in your table may be different from those shown below. The numbers in the Y column change.8. Figure 2. Click Make Design. Change the default settings of the coefficients in the Simulate Responses dialog to match those in Figure 2. . A window named Simulate Responses and a design table appear similar to the one in Figure 2.9 Simulated Responses and Design Table 7. X9. The Y column values are controlled by the coefficients of the model in the Simulate Responses window. and X10 as active factors in the simulation.

Click Run Model. JMP enters the factors with the largest effects. you should identify three active factors: X2.5 To identify active factors using stepwise regression: 1. . The step history appears in the bottom part of the report. From the report that appears. X9.10 Give Values to Three Main Effects and Specify the Standard Error as 0. 2.11. and X10 as shown in Figure 2. Change the Personality in the Model Specification window from Standard Least Squares to Stepwise. 3.2 Examples Using the Custom Designer—Creating Screening Experiments 25 2 Examples Figure 2. Run the Model script in the design table by clicking the red triangle beside Model and selecting Run Script. Your estimates may be slightly different from those shown below. In the resulting display click the Step button three times.

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

13 Three Blocks in the Factors Panel If.14). in the Design Generation panel. . the Factors panel changes and now contains 8 blocks (Figure 2. Figure 2.2 Examples Using the Custom Designer—Creating Screening Experiments 27 2 Examples Figure 2. there are three blocks in the Factors panel (Figure 2. which requires three blocks with three runs each. you chose the Grid option with 24 runs. This is because the default sample size is nine.12 One Block Appears in the Factors Panel The default sample size of nine requires three blocks.13). After clicking Continue.

15).14 Eight Blocks If you add all the two-factor interactions and change the number of runs to 15. Figure 2. so the Factors panel displays five blocks in the Values section.28 2 Examples Using the Custom Designer—Creating Screening Experiments Figure 2. three runs per block produces five blocks (as shown in Figure 2.15 Changing the Runs to 15 .

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. p < n < p+q. The formal name of the approach using If Possible model terms is Bayesian D-Optimal design. For more details. Note to DOE experts: There are four rows associated with X4 (the block factor).” p. 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. That is. therefore. If not. Each degree of freedom is associated with one unknown coefficient in the model. (Let p denote the primary terms designated Necessary in JMP. 4 degrees of freedom. Notice that the variance of each coefficient is about one-tenth the error variance and that all the variances are roughly the same size. (Let q denote the potential terms designated If Possible in JMP. The error variance is assumed to be 1. experimenters often add center points and other check points to a design to help determine whether the assumed model is adequate. 63. then click the disclosure button ( on Windows/Linux and on the Macintosh) to open the Relative Variance of Coefficients.2 Examples Using the Custom Designer—Creating Screening Experiments 29 2 Examples Click Make Design.16 shows a table showing the variance of each coefficient in the model relative to the unknown error variance. These higher-order terms are called potential terms. That is because X4 has 5 blocks and. Figure 2. adding more runs (18 or more) would lower the variance of each coefficient. see “Understanding the Variance of Coefficients Table.) The assumed model consists of the primary terms. Figure 2.16 Table of Relative Variance of the Model Coefficients The main question here is whether the relative size of the coefficient variance is acceptably small. .) To take advantage of the benefits of the approach using If Possible model terms. The values in your table may be slightly different from those shown below. Although this is good practice. Checking for Curvature Using One Extra Run In screening designs. the sample size should be larger than the number of primary terms but smaller than the sum of the primary and potential terms. 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. it is also ad hoc.

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

2 Examples Using the Custom Designer—Creating Screening Experiments 31 2 Examples 5. 10. Enter 5 into the Number of Runs edit box. 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. 6. the p = 4 primary terms (the intercept. Figure 2. The resulting factor settings appear in Figure 2. and the interaction) are designated as Necessary while the q = 2 potential terms (the two quadratic terms) are designated as If Possible. Figure 2. Click Make Table. Click Make Design. 8.21 by selecting Graph > Overlay Plot. The values in your design may be different from those shown below. is between p = 4 and p + q = 6. and insert X1 as Y and X2 as X. Click the Powers button in the Model panel and choose 2nd.20 Five-Run Bayesian D-Optimal Design 9. The desired number of runs. 7. This adds two quadratic terms.20.19.19 Changing the Estimability Now. The data table appears. . five.Create the overlay plot in Figure 2. Change Necessary to If Possible. Select the two quadratic terms (X1*X1 and X2*X2) and click the current estimability (Necessary) to reveal the menu. as shown in Figure 2. two main effects.

.22). Select DOE > Custom Design. The a priori model for a response surface experiment is usually quadratic.32 2 Examples Using the Custom Designer—Creating Response Surface Experiments Figure 2. 2. This plot is a powerful diagnostic tool for evaluating and comparing response surface designs. In screening experiments one measure of the quality of the design is the size of the relative variance of the coefficients. researchers use response surface experiments when they already know which factors are important. 1. In the Model panel. Add one continuous factor by selecting Add Factor > Continuous (Figure 2. Using this predictive model allows the experimenter to find better operating settings for the process. 3. select Powers > 2nd to create a quadratic term (Figure 2. In response surface experiments.21 Overlay Plot Creating Response Surface Experiments Response surface experiments traditionally involve a small number (generally 2 to 8) of continuous factors.22) and click Continue. the prediction variance over the range of the factors is more important than the variance of the coefficients. Exploring the Prediction Variance Surface The purpose of the example below is to generate a simple Prediction Variance Profile Plot. The main goal of response surface experiments is to create a predictive model of the relationship between the factors and the response. Follow the steps below to create a design for a quadratic model with a single continuous factor. One way to visualize the prediction variance is JMP’s prediction variance profile plot. In contrast to screening experiments.

The y-axis is the relative variance of prediction of the expected value of the response. 63.” p. the initial setting is the first level. As the number of runs increases. Figure 2. More details on the Prediction Variance Profiler is in “Understanding the Design and Profilers. 7. 0. and 1. In this design. Click Make Design and then open the Prediction Variance Profile again. The prediction variance profile shows that the variance is a maximum at each of these points on the interval –1 to 1. as shown in Figure 2. then the prediction variance function is quartic. Make sure to use the default number of six runs (Figure 2. Figure 2. on the Macintosh) to open the For continuous factors.22 Adding a Factor and a Quadratic Term 4. . For categorical factors. Click the disclosure button ( on Windows/Linux and Prediction Variance Profile.23) and click Make Design. the initial setting is at the mid-range of the factor values.24 Prediction Profile for Single Factor Quadratic Model The prediction variance is relative to the error variance. the three design points are –1.2 Examples Using the Custom Designer—Creating Response Surface Experiments 33 2 Examples Figure 2. which gives a sample size of three. When the relative prediction variance is one. the prediction variances decrease.23 Adjusting the Number of Runs 5. Compare profile plots by clicking the Back button and choosing Minimum in the Design Generation panel. 6. the absolute variance is equal to the error variance of the regression model. The number of runs is inversely proportional to the size of variance of the predicted response.24. If the design model is quadratic.

Figure 2. 8. and clicking Make Design.26.5 rather than one). for sample sizes six (top charts) and sample size three (bottom charts). When the design is balanced. Create an unbalanced design by clicking the Back button. as shown in Figure 2. You can see that the variance of prediction at –1 is lower than the other sample points (its value is 0.34 2 Examples Using the Custom Designer—Creating Response Surface Experiments This produces a curve that has the same shape as the previous plot. . When the design is unbalanced. the prediction variance is inversely proportional to the sample size.25. the plot is symmetric. These profiles show settings for the maximum variance and minimum variance. The results are shown in Figure 2. but the maxima are at one instead of 0. so doubling the number of runs halves the prediction variance.26.25 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. Figure 2. In fact. as shown in Figure 2. You can see the prediction variance increase as the sample size decreases. the prediction plot is not symmetric. The symmetry of the plot is related to the balance of the factor settings.5.

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.445) to see that the maximum variance on the interval decreases by more than 10%. Then select Edit > Run Script from the Edit menu. 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. .445 and 0. View the script in this example by opening Cubic Model.jsl from the Sample Scripts folder that was installed when you installed JMP.

1}.333 0. then it is easy to tell which factors’ effects are important and which are negligible.. for(i=1.333]. // Drag the middle points to -0. D-Optimal designs are not as appropriate for designing experiments where the primary goal is prediction.99))).99)).)).d.1). If an experimenter has precise estimates of the factor effects.i<=3. As a result I-Optimality is the recommended criterion in JMP for response surface designs.0. rr=j(4.XScale(-1.0).u).for(i=1.). C = inverse(V). for(i=1.i++.6.1. Graph(FrameSize(500. // number of points //Start with equally spaced points.sqrt(xi`*C*xi).u[2].1. show(n.rr[i+1]=x1^i).1.1."Determinant = ". M = M||(x^i)).i++. rr. .y).. D-Optimal designs are most appropriate for screening experiments because the optimality criterion focuses on precise estimates of the coefficients. y = j(2.x). cubicx = function({x1}.1}.yScale(0.char(detV.i<=3. yFunction(xi=cubicx(x).1.6. This makes I-Optimal designs most appropriate for prediction. DragMarker(u.Text({u[i].i<=2. M = j(n. x = {-1. V = M`*M. I-Optimal designs minimize the average prediction variance inside the region of the factors. Double Buffer.300).u[1].1).36 2 Examples Using the Custom Designer—Creating Response Surface Experiments // DOE for fitting a cubic model.i++. All the designs in this chapter so far have been D-Optimal designs.char(u[i]. However. detV = det(V).2). u = [-0.445 // for a D-Optimal design. text({-0. NewWindow("DOE . Introducing I-Optimal Designs for Response Surface Modeling The custom designer generates designs using a mathematical optimality criterion.3.2) .Variance Function of a Cubic Polynomial".25}. n = 4.1.445 and 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

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

The values in your table may be different from those shown below.33).2 Examples Using the Custom Designer—Creating Mixture Experiments 43 2 Examples The model. In the Design Generation panel. type 18 in the Number of Runs text edit box (Figure 2.34. shown in Figure 2. and then click Make Table. Figure 2. . which results in six runs each for the three levels of the wavelength factor. To create this model: 1. Click OK. Click Interactions. 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.33 Mixture Experiment Design Generation Panel 3.33 is a response surface model in the mixture ingredients along with the additive effect of the wavelength. and choose 2nd. The resulting data table is shown in Figure 2. Click Make Design.

it is not possible to run a complete mixture response surface design for every wavelength. The bottom profiler in Figure 2. 4.34 Mixture Experiment Design Table Atkinson and Donev also discuss the design where the number of runs is limited to 10. In that case. To view this: Click the Back button. Figure 2.4 to 0.35 Ten-Run Mixture Response Surface Design 1. Add the main effect by clicking the Main Effects button.36). Because of this lack of balance it is a good idea to look at the prediction variance plot (top plot in Figure 2. the others adjust to keep the sum constant. The other two ingredients both drop. which is a good indication that the lack of balance is not a problem. 2. If you vary one of the values. keeping their ratio constant. The Design table to the right in Figure 2. Note that there are necessarily unequal numbers of runs for each wavelength. In the Design Generation panel. change the number of runs to 10 (Figure 2.6. Remove all the effects by highlighting them all and clicking Remove Terms.35) and click Make Design.35 shows the factor settings for 10 runs.44 2 Examples Using the Custom Designer—Creating Mixture Experiments Figure 2. The values of the first three ingredients sum to one because they are mixture ingredients. The prediction variance is almost constant across the three wavelengths.36 shows the result of increasing the copper sulphate percentage from about 0. . 3.

2. as shown in Figure 2. . open Big Class. To follow along with this example that shows column properties. Add two continuous variables to the models by clicking Add Factors and selecting Continuous. suppose there are a group of students participating in a study. If this variable affects the experimental response.2 Examples Using the Custom Designer—Special-Purpose Uses of the Custom Designer 45 2 Examples Figure 2. Build the custom design as follows: 1.jmp from the Sample Data folder that was installed when you installed JMP. Sometimes you have quantitative measurements (a covariate) on the experimental units before the experiment begins. The Covariate selection displays a list of the variables in the current data table. naming them calories and sleep.37. Click Add Factor and select Covariate. The weight of the student is known and it seems important to include this information in the experimental design. 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. The pre-defined design that allows only a few discrete values is too restrictive. For this example. Designing Experiments with Fixed Covariate Factors Pre-tabulated designs rely on the assumption that the experimenter controls all the factors. Select DOE > Custom Design. the designs presented in this section cannot be created without using the custom designer. The custom designer supplies a reasonable design option. 3.36 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. the covariate should be a design factor.

Add the interaction to the model by selecting calories in the Factors panel. and then clicking the Cross button (Figure 2. The data table in Figure 2. Figure 2. Select weight from the variable list and click OK. selecting sleep in the Model panel.46 2 Examples Using the Custom Designer—Special-Purpose Uses of the Custom Designer Figure 2. then click Make Table. Click Make Design.40 shows the design table.37 Add a Covariate Factor 4. . Figure 2. 6. Your runs might not look the same because the order of the runs has been randomized. Click Continue.39 Design With Fixed Covariate Factor 7.38 Design with Fixed Covariate 5.39).

but it is not controlled. that means the custom designer has successfully separated the effect of weight from the effects of calories and sleep. Arrange the dialog so weight is Y and calories. weight is independent of calories and sleep. Therefore.41. It would be desirable if the correlations between calories. Figure 2. Click the red triangle icon beside Model and select Run Script..40 Design Table Note: Covariate factors cannot have missing values. sleep. and calories*sleep are the model effects. as shown to the right in Figure 2. and the analysis of variance table shows that the model sum of squares is near zero compared to the residuals (Figure 2. The leverage plots are nearly horizontal. Remember that weight is the covariate factor. measured for each student. as shown on the left in Figure 2. 8.42). The values in your analysis may be a little different from those shown below. If that fit has a small model sum of squares.2 Examples Using the Custom Designer—Special-Purpose Uses of the Custom Designer 47 2 Examples Figure 2. We can see how well the custom designer did by fitting a model of weight as a function of calories and sleep.41. sleep and weight were as small as possible. The custom designer has calculated settings for calories and sleep for each student.41 Model Script 9. Click Run Model. .

the mixture components are the subplot factors of the experiment. By default. The process variables are the whole plot factors of the experiment. The whole plot is divided into subplots. and Vining (2002). • The response in the experiment is the thickness of the vinyl used for automobile seat covers. To create this design in JMP: 1. m2.Enter the lower limit of 10.42 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. • Two of the factors are process variables. Additionally. They are the rate of extrusion and the temperature of drying. The experiment studies the effect of five factors on the thickness of vinyl used to make automobile seat covers. . Use the default goal. Y.48 2 Examples Using the Custom Designer—Special-Purpose Uses of the Custom Designer Figure 2. and m3. hard-to-change factors change only between one whole plot and the next. Select DOE > Custom Design. Maximize (Figure 2. The split plot design capability accessible in the custom designer is the first commercially available tool for generating optimal split plot designs. In split plot experiments. The example in this section is from Kowalski. As is expected of mixture experiments. The response of interest (thickness) depends both on the proportions of the mixtures and on the effects of the process variables. only in the last few years has it been possible to create optimal split plot designs (Goos 2002). there is one response. Double-click the name and change it to thickness. showing. but is now commonplace in manufacturing and engineering studies. They are hard to change. They are plasticizers whose proportions are represented by m1. 2. the component proportions always sum to one. Cornell. 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. and the levels of the easy-to-change factors are randomly assigned to each subplot.43).

click Easy and select Hard. This causes a warning that JMP removes the main effect terms for non-mixture factors that interact with all the mixture factors. Your dialog should look like the one in Figure 2. Ensure that you are using the default levels. if you have factors that have their estimability set to Hard. . type 3 in the box beside Add N Factors. then JMP makes it impossible to designate the model terms If Possible. in the Values area corresponding to these two factors. Use the default levels 0 and 1 for those three factors. Click Continue. and click Add Factor > Continuous. Next. Figure 2.44).2 Examples Using the Custom Designer—Special-Purpose Uses of the Custom Designer 49 2 Examples 3.43. Conversely. and m3.43 Entering Responses and Factors 8. Add two continuous factors by typing 2 in the box beside Add N Factors. and click Add Factor > Mixture. 9. click Easy and select Hard. Name the three mixture factors m1. Click OK. m2. To make temperature the subplot factor.44 Adding Interaction Terms Note that if you have model terms that have their estimability set to If Possible.In the Design Generation panel. 6. To add three mixture factors. 10. Figure 2. 4. click the box next to Number of Whole Plots and type 7. 5. To make extrusion rate the whole plot factor. Name these factors extrusion rate and temperature. 7. add interaction terms to the model by selecting Interactions > 2nd (Figure 2. JMP makes it impossible to make factor changes be Hard. –1 and 1.

The result is shown in Figure 2.46. The values in your table may be different from those shown below. Figure 2. . Figure 2.Click Make Table.50 2 Examples Using the Custom Designer—Special-Purpose Uses of the Custom Designer 11.45). Figure 2. select Design Experiment > vinyl data.Give the experiment 28 runs (Figure 2.Click Make Design.jmp. The dialog in Figure 2.47 The Vinyl Data Design Table 15.45 Assigning the Number of Whole Plots and Number of Runs 12. 14.46 Final Design Structure 13.48 appears.Click the red triangle icon next to the Model script and select Run Script.From the Sample Data folder that was installed when you installed JMP.

Click Run Model to run the analysis. JMP designates the error term by appending &Random to the name of the effect. as indicated in the menu beside Method in Figure 2. The results are shown in Figure 2.48.48. see the JMP Statistics and Graphics Guide. REML will be used for the analysis. As shown in the dialog in Figure 2. Figure 2.49. . For more information about REML models.48 Creating the Model 16.2 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.

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

• Uses the potential terms to force in runs that allow for detecting any inadequacy in the model containing only the primary terms. • Maximizes this criterion: If f T(x) denotes a row of the X matrix corresponding to factor combinations x. 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. . the Bayesian D-Optimality criterion is advantageous. 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. I-Optimality: • Minimizes the average variance of prediction over the region of the data. 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. and also is appropriate to determine regions in the design space where the response falls within an acceptable range. 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. • 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. such as in response surface design problems. That is.2 Examples Using the Custom Designer—Technical Discussion 53 2 Examples • Maximizes D when D = det[XTX]. The Bayesian D-Optimal design maximizes the determinant of (XT X + K). and at the same time has the ability to detect and estimate some higher-order terms. to develop feedback-control models. If there are interactions or curvature. 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. • 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). Bayesian D-Optimality: • Is a modification of the D-Optimality criterion that effectively estimates the coefficients in a model. If your goal is to predict the response rather than the coefficients. or. k. • Is more appropriate than D-Optimality in some situations. Precise estimation of the response therefore takes precedence over precise estimation of the parameters. p + q > n > p. It is more appropriate when your objective is to determine optimum operating conditions.

54 2 Examples Using the Custom Designer—Technical Discussion .

This chapter introduces you to the steps you need to complete to build a custom design.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. flexible. To create these tailor-made designs. and good for routine factor screening or response optimization. select DOE > Custom Design. .

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

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

where the random noise is distributed with mean zero and standard deviation one. as shown in Figure 3. such as when the response is impurity or defects. would be y = 28 + 4X1 + 5X2 + random noise . you can alter the default target after you make a table from the design. In the data table. the best value is usually the largest possible. Create the design.1) and the lower and upper limits (step 4 in Figure 3. A goal of Maximize supports this objective. 2. With two responses you can give greater weight to one response by assigning it a higher importance value. and the Y column contains values for simulated responses. open the response’s Column Info dialog by double-clicking the column name. 4.58 3 Building Custom Designs—Creating a Custom Design The following description explains the relationship between the goal type (step 3 in Figure 3. Note: If your target range is not symmetrical. • The Match Target goal supports the objective when the best value for a responses is a specific target value. The default target value is assumed to be midway between the lower and upper limits. Adding Simulated Responses.2. The numbers you enter represent the coefficients in an equation. For custom and augment designs. a window (Figure 3. you might want to add simulated responses that help you determine whether your design can detect effects of a size that you expect: 1.1): • For responses such as strength or yield. JMP uses the value you enter as the importance weight (step 4 in Figure 3. Understanding Response Importance Weights When computing overall desirability. If Desired If you do not have specific responses to add. click the red triangle icon in the title bar and select Simulate Response. and enter an asymmetric target value. . • The Minimize goal supports an objective of having the best objective be the smallest value. then specifying importance is unnecessary. such as with part dimensions. 3. In this window.2) appears along with the design data table. Click Make Table to create the design table. An example of such an equation.1) as the weight of each response. If there is only one response. Before clicking the Make Table button. enter values you want to apply to the Y column in the data table and click Apply.

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

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

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

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

4. Comparing the prediction variance profilers for two designs side-by-side. preview the design and investigate details using the prediction profiler (Figure 3. Generally. The error variance is assumed to be one. . Understanding the Variance of Coefficients Table Before clicking Make Table in the custom designer. Dragging the lines reveals any points that have prediction variances that are larger than you would like.3 Building Custom Designs—Creating a Custom Design 63 3 Custom Understanding the Design and Profilers Before creating the design table. if the error variance is five and the reported prediction variance at a particular factor setting is 0. Figure 3.8 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. then the prediction variance at that point is two. Figure 3. JMP always provides the prediction profiler. click the disclosure button ( on Windows/ Linux and on the Macintosh) to open the Relative Variance of Coefficients table. the prediction variance is reported relative to whatever the error variance turns out to be. but the surface plot appears only if there are two or more variables. For example.7 A Factor Design Layout The prediction variance for any factor setting is a function of the design and the unknown error variance.7) and surface plot. drag the vertical lines in the plot to change the factor settings to different points. That is. In the profiler. is one way to compare two designs. the error variance will drop as the sample size increases.

The table shows the aliasing (if any) between the model terms and all the two-factor interactions. Figure 3. The error variance is assumed to be one. 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. It allows you to see the confounding pattern in D-Optimal designs.64 3 Building Custom Designs—Creating a Custom Design The Relative Variance of Coefficients table (Figure 3. you can multiply that by the relative variance to get the estimated variance of the coefficient. on Windows/ The alias matrix only appears for custom designs with two factor interactions if all factors are two-levels. Once you have an estimate for the error variance.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. The variances are relative to the error variance. Figure 3.9 Table of Relative Variance of Coefficients Understanding the Alias Matrix (Confounding Pattern) Before clicking Make Table in the custom designer. which is unknown before the experiment.10 Alias Matrix Specifying Output Options Use the Output Options panel to specify how you want the output data table to appear: .

Model is a script. • Number of Replicates—Specify the number of times to replicate the entire design. • Number of Center Points—Specifies additional runs placed at the center of each continuous factor’s range. click Make Table. JMP calls these factors Hard to change because this is usually how split plotting arises in industrial practice. Usually this is because these factors are difficult or expensive to change from run to run. • Sort Right to Left—the rows (runs) in the output table will appear sorted from right to left. 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. including centerpoints. .11 Example Design Table This area identifies the design type that generated the table. • Randomize within Blocks—the rows (runs) in the output table will appear in random order within the blocks you set up. Figure 3. Choices are: • Keep the Same—the rows (runs) in the output table will appear as they do in the Design panel. One replicate doubles the number of runs. Type the number of times you want to replicate the design in the associated text box. • Sort Left to Right—the rows (runs) in the output table will appear sorted from left to right. Making the Table When the Design panel shows the layout you want. • Randomize—the rows (runs) in the output table will appear in a random order. 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. Click the red triangle icon and select Run Script to open the Fit Model dialog. Click Custom Design to edit. which is used to generate the analysis appropriate to the design.

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. 1D-efficiency = 100 ⎛ ------- X′X 1 / p⎞ ⎝ ND ⎠ ⎛ ⎞ 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.

the window in Figure 3. which varies depending on the design. The number you enter overrides the default number of starts. the design produced at the end is not always the same.12) and select Number of Starts.” p. When you select this command. For specific information about optimality criterion. Increasing the number of random starts tends to improve the optimality of the resulting design. 52. click the red triangle icon in the Custom Design title bar (Figure 3.17 appears with an edit box for you to enter the number of random starts for the design you want to build. For example.3 Building Custom Designs—Special Custom Design Commands 71 3 Custom Figure 3. all orthogonal designs are D-optimal with respect to a linear additive model. Selecting the Number of Random Starts To override the default number of random starts. It is easy for an optimizer to converge to a local optimum instead of a global optimum.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. see “Technical Discussion.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. the optimization problem becomes harder. It is useful to know that: • If random starts are used for the optimization. As the number of factors and sample size increase. Figure 3. .

there is a formula for the optimal determinant: If D is the determinant. 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. . Also. n is the sample size.” p. if the number of runs is a multiple of four. Before making the design. 65).” p. then the number of starts listed above is multiplied by four. Note that hypersphere constraints do not work with other constraints. Enter the appropriate value and click OK. The design is D-optimal and orthogonal. click the red triangle icon in the Custom Design title bar (Figure 3.12) and select Sphere Radius. and c is the number of columns in the design matrix. the algorithm stops. If you have designed any factor’s changes as Hard (see “Creating Optimal Split Plot Designs. If the determinants that result from the random starts match the formula above.72 3 Building Custom Designs—Special Custom Design Commands • For designs with all two-level factors. split plot designs cannot be generated with hypersphere constraints. once you set the sphere radius. and “Creating Split Plot Designs. each factor has only two levels. Conversely. Default Choice of Number of Random Starts JMP does not start over with random designs until a jackpot is hit. and the model is linear. 60. the sphere radius item becomes unavailable. the LogD = cLogn. This modification of rules puts extra effort towards finding two-level fractional factorial and Plackett-Burman designs (or their equivalents). you cannot make a factor Hard to change. 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. 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. the total time it takes to generate a design is kept roughly constant over the range of usual sample sizes. Constraining a Design to a Hypersphere You can constrain custom and augmented designs to a hypersphere by editing the sphere radius.

18). you would use Feature==3 & Price==3 | Quality==3 & Price==3. You can do this for custom and augmented designs except for experiments with mixture or blocking factors.18. and when it sees it as true. This feature can also be used with continuous factors or mixed continuous and categorical factors. and the factor Price has levels of high (1). In this case.19 No Row Contains L3 for Both Price and Feature . use the ordinal value of the level instead of the name of the level. For example. If the level names of the factor called Price are high.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. and low. medium. Enter a Boolean expression that identifies what you do not want allowed (Figure 3. as shown in Figure 3. To disallow a combination of factor levels: 1.19. Make the design table. For example. in Figure 3. their associated ordinal values are 1. 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. you might want to exclude a choice that allows all the best features of a product at the lowest price. and 3. Click the red triangle icon in the title bar (Figure 3. Figure 3. medium (2). If there were two disallowed combinations in this example. in a market research choice experiment. You want to exclude the third Feature level (best) and the third Price level (low). 2. the factor Feature has levels of worst (1). It excludes the combination of factors you specified. it disallows the specified combination. Figure 3. and best (3). Note: When forming the expression. and low (3). Feature==3 & Price==3 will not allow a run containing the best features at the lowest price.12) of the designer window and select Disallowed Combinations. Note that this menu item is not available if you have already defined linear inequality constraints. medium (2). JMP evaluates your expression. 3.18 Enter a Boolean Expression 4. 2. Begin by adding the factors.

To change the search points: 1. you can change this so the algorithm considers more search points. Select DOE > Custom Design. 3.12) of the designer window and select Advanced Options > K Exchange Value. For example. 2. Click the red triangle icon in the title bar (Figure 3. Changing the Search Points Per Factor For a main effects model. you can set k less than n. Make the design. the coordinate exchange algorithm considers every row of factor settings for possible replacement in every iteration.74 3 Building Custom Designs—Special Custom Design Commands Advanced Options The following options are only for advanced design of experiment users. Enter a positive integer and click OK. 4. 3. then JMP considers five equally spaced settings for each factor. Make the design. You may have one or more fixed ingredients so that the sum of the remaining add to less than one but more than zero. Enter a positive integer and click OK. Enter a positive number and click OK. Click the red triangle icon in the title bar (Figure 3. Setting the K Exchange Value By default. 3. 4. 2.12) of the designer window and select Advanced Options > Mixture Sum. Select DOE > Custom Design. Altering the Mixture Sum If you want to keep a component of a mixture constant throughout an experiment. the coordinate exchange algorithm in the custom designer only considers the high and low values. However. Select DOE > Custom Design. To reduce the number of rows considered at each iteration.001 and 1. then the sum of the other mixture components must be between 0. 4.12) of the designer window and select Advanced Options > Search Points Per Factor. if you enter 5. Make the design. To change the K exchange value: 1. To alter the mixture sum: 1. Click the red triangle icon in the title bar (Figure 3. . This generally speeds up design creation but tends to produce designs that are less efficient for estimating the unknowns in the model. 2.

. To see these properties. 4. Enter a positive number and click OK. Assigning Column Properties Columns in a data table can contain special column properties.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. Make the design. Click the red triangle icon in the title bar (Figure 3.20. this variance is one.jmp. 3. The following discussion covers properties specific to DOE and useful for analyzing DOE data. which has two special properties: Role and Response Limits.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. you can supply it by following these steps: 1. select Help (View on the Macintosh) > Sample Data Directory > Design Experiment > Bounce Data. Notice the column called Stretch. By default. If you have some prior knowledge of this variance ratio. Figure 3.12) of the designer window and select Advanced Options > Split Plot Variance Ratio. To follow along with this example that shows column properties. 2. Select DOE > Custom Design. as shown in Figure 3.

When you select Analyze > Fit Y by X. Select Coding from the Column Properties drop-down menu. Type the values you want to use as low and high values into the text boxes. The properties icon ( ) now appears next to the column name in the data table’s column panel. 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. . Figure 3. The Column Info window appears. 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.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. Coding information appears on the right. You can specify the range in which the low and high values of the column are transformed. By default. To change the range in which the low and high values are transformed: 1. If a column has one or more limits missing. 2.21. Double-click the column name in the data grid. as shown in Figure 3. Click OK. it uses those values as the low and high values of the coding property. Double-click the column name in the data grid. JMP uses those transformed data values (for continuous variables only) to compute meaningful parameter estimates in the analysis.21 Coding Window 3. This transformation is called coding. To set up a column as a mixture factor: 1. JMP substitutes the data’s minimum and maximum for the high and low values. The Column Info window appears. indicating the column contains a property. 4. when JMP generates a design table from values entered in the Factors panel.

middle. you can enter an importance value. then come back later and change the data. if you run an analysis that employs these limits.3 Building Custom Designs—Assigning Column Properties 77 3 Custom 2. Select Response Limits from the New Property drop-down menu. . Mixture information appears on the right. Crossed effects are not labeled. and desirability values. Minimize. 4.22. Defining Response Column Values You can save response limits in a column. In the Fit Model report. To enter response limits: 1. Click OK. where the mixture sum value is 1. which lets JMP know how to weigh one response’s importance against another’s. or None.22. 3. If you do this. the Fit Model platform uses the L coding if (1 – L) < (U – 1). Complete the settings in the window. as shown in Figure 3. you can run a new analysis on the new data using the same limits. Using the example in Figure 3.23. Enter the upper and lower limits as well as the sum of terms. If you have two responses. The Column Info window appears. Double-click the column name in the data grid.22 Mixture Information 3.23. indicating the column contains a property. You can also select Maximize. You can specify values for the lower. The properties icon ( ) now appears next to the column name in the data table’s column panel. 5. but coding values are used. Saving these limits in a column facilitates consistency from use to use. the main effects are labeled with the coding transformation. as shown in Figure 3. you can run analyses without having to re-specify response limits each time. For example. if desired. 2. as shown in Figure 3. which are possible goals for a DOE response variable. and the U coding otherwise. If you check both in this example. Check the boxes beside L and U PseudoComponent Coding. Figure 3. All the features of fitting. L is the sum of Li and U is the sum of Ui. such as the profilers and saved formulas. Match Target. respect the pseudocomponent coding but present the un-coded values in the tables and plots. and upper limits. Select Mixture from the New Property drop-down menu. 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.

To give a column a design role: 1.78 3 Building Custom Designs—Assigning Column Properties Figure 3. Double-click the column name in the data grid. enter a number to indicate the amount of weight you want this response to have when JMP computes the overall desirability Enter the lower.24. The properties icon ( ) now appears next to the column name in the data table’s column panel. For example. Giving Columns a Design Role You can tell JMP how to use a factor column in a model to design an experiment. 2. middle and upper limits as well as the desirability values 4. The Column Info window appears. Click OK. constant. Select Design Role from the New Property drop-down menu. if you are in the prediction profiler and want the desired value to be close to zero. or noise factor. You can tell JMP to use it as a continuous. categorical. covariate.23 Defining Response Column Values Click to select a goal for the response variable. as shown in Figure 3. Design role information appears on the right. select Match Target When you have two responses. indicating the column contains a property. . mixture. signal. blocking.

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

If so.Optimality). the new value replaces the old. The computational method is an iterative algorithm called coordinate exchange (Meyer and Nachtsheim. the design algorithm may generate different (but equivalent) designs when you click the Back and Make Design buttons repeatedly.80 3 Building Custom Designs—How Custom Designs Work: Behind the Scenes Figure 3. When this is true. Click OK. Sometimes a design problem can have several equivalent solutions. 1995).or I. Click the Factor Change drop-down menu and select how you want JMP to use the factor column: Easy or Hard. . For more details.25 Choosing a Factor Change 3. indicating the column contains a property. 70. To avoid converging to a local optimum. The properties icon ( ) now appears next to the column name in the data table’s column panel. the whole process is repeated several times using a different random start. 4. How Custom Designs Work: Behind the Scenes The custom designer starts with a random set of points inside the range of each factor.” p. The custom designer displays the best of these designs. Equivalent solutions are designs with equal precision for estimating the model coefficients as a group. see the section “Changing the Design Criterion (D. 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. Iteration continues until no replacement occurs in an entire iterate.

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

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

causing them to fuse. Change the default response as follows: 1. Click the Goal text edit area and choose Match Target. 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. who is the technician operating the welding machine • Rotation Speed. 2. This reveals an area where the two parts have fused. The Length of this fused area is the depth of penetration of the weld. The first section consists of two examples using screening designs. The default goal for the single default response is Maximize. but the goal of this process is to get a target value of 0.17 inches with a lower bound of 0. For this study. there is a single default response called Y. The second section outlines the procedures to follow to create a screening design to match your needs.4 Screening Designs—Screening Design Examples 83 4 Screening Screening Design Examples This chapter is divided into two sections.). 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. Note that in the Responses panel. The ideal depth of the fused region is 0. which is a current that affects the intensity of the beam After each processing run.1 Screening Design Response With Match Target Goal 3.17 inches. the engineer cuts the part in half. 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. Figure 4.1. Double-click the response name and change it to Depth (In. the engineer wants to explore the following three inputs.12 and an upper bound of 0.22. Click the Lower Limit text edit area and enter 0. The engineer fits the two parts into a welding fixture that holds them snugly together. which is the speed at which the part rotates under the beam • Beam Current. which are the factors for the study: • Operator.12 as the lower limit (minimum acceptable value). as shown in Figure 4. select DOE > Screening Design from the main menu. .17 inches To begin this example. A voltage applied to a beam generator creates a stream of electrons that heats the two parts. This depth of penetration is the response for the study.

View the table produced in this example by selecting Help (View on the Macintosh) > Sample Data Directory > Design Experiment > DOE Example 1.2. factors. Double-click the factor names and rename them Operator. 7. 5. 6. Figure 4. This example has one categorical factor (Operator) and two continuous factors (Speed and Current). Speed. click Make Table to create a JMP table that contains the specified design. Set high and low values for Speed to 3 and 5 rpm. as shown in Figure 4. Add two continuous factors by typing 2 in the Continuous box and clicking the associated Add button. 10.22 as the upper limit (maximum acceptable value).4 appears. Add the categorical factor by clicking the Add button beside 2-Level Categorical.3. as shown in Figure 4. Set high and low values for Current to 150 and 165 amps. 9.2 Screening Design with Two Continuous and One Categorical Factor 8. and then click Continue.3 List of Screening Designs for Two Continuous and One Categorical Factors When the design details are complete. The Pattern variable shows the coded design runs. 4. Figure 4. Click Continue.jmp. The table uses the names for responses. Select Full Factorial in the list of designs. and Current. and assign Mary and John as values for the categorical factor called Operator.84 4 Screening Designs—Screening Design Examples Then click the Upper Limit text edit area and enter 0. and levels you specified. . The table in Figure 4.

4 The Design Data Table Using Five Continuous Factors As illustrated in the previous section. Temperature. 2. 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. Catalyst. Enter the high and low values. In this study. Change the default response name (Y) to Percent Reacted. and Concentration. a chemical engineer investigates the effects of five factors on the percent reaction of a chemical process. This allows the examination of many factors with a minimum number of runs. The factors are: • feed rate. Stir Rate. as shown in Figure 4. and importance to 1 (Figure 4. . et al.4 Screening Designs—Screening Design Examples 85 4 Screening Figure 4. 6. 3. the amount of raw material added to the reaction chamber in liters per minute • percentage of catalyst • stir rate.5). 4. 5. adapted from Meyer.5. Select DOE > Screening Design. the RPMs of a propeller in the chamber • reaction temperature in degrees Celsius • concentration of reactant To start the example: 1. Add five continuous factors. Change the default factor names (X1-X5) to Feed Rate. (1996). You see one default response called Y. upper limit to 99. The following example. demonstrates how to use JMP’s screening designer when you have many factors.

86 4 Screening Designs—Screening Design Examples Figure 4. Select the first item in the list.6.6. . as shown to the left in Figure 4. which is an 8-run fractional factorial design. Figure 4. JMP lists the designs for the number of factors you specified.5 Screening for Many Factors 7.7 that modify the final design table. 9. Now.6 Two-level Screening Design (left) and Design Output Options (right) The design dialog has options shown in Figure 4. Click Continue. 8. Click Continue again to see the Design Output Options panel on the right in Figure 4.

“Entering Responses. 93 .4 Screening Designs—Creating a Screening Design 87 4 Screening Figure 4. Figure 4.” p. right-click the Model script and select Run Script. open Reactor 8 Runs.8 that lists the runs for the design you selected. 11. “Displaying and Modifying the Design. Then. you expect a few to stand out in comparison to the others. select DOE > Screening Design. as shown previously in Figure 4. or choose Analyze > Fit Model to analyze the data. follow the steps below: 1. “Choosing a Design. or click the Screening Design button on the JMP Starter DOE page.To run the model generated by the data.In the design data table.” p.jmp from the Design Experiment folder inside the Sample Data folder that was installed when you installed JMP. 90 3. 12.Click Make Table to create the data table shown in Figure 4. Creating a Screening Design To start a screening design.8 JMP Table of Runs for Screening Example Of the five factors in the reaction percentage experiment. Note that it also has a column called Percent Reacted for recording experimental results. as shown to the right of the data table.8.” p. 88 2. Let’s take an approach to the analysis that looks for active effects.7 Output Options for Design Table 10.

or None. 97 5. • The Minimize goal supports an objective of having the best objective be the smallest value. if desired. the best value is usually the largest possible. click the N Responses button and enter the number of responses you want. if desired.9): • For responses such as strength or yield. “Specifying Output Options.9 Entering Responses Click to enter lower and upper limits and importance weights. Match Target.” p. A goal of Maximize supports this objective. to match a specific value. click then select a goal type: Maximize. “Viewing the Table. such as when the response is impurity or defects. Minimize. 4 To enter one 1 response at a time. follow the steps in Figure 4. 2 Double-click to edit the response name.9. 98 Entering Responses To enter responses. Tip: To quickly enter multiple responses.” p. Figure 4. Specifying Goal Types and Lower and Upper Limits When entering responses. 97 6. 3 Click to change the response goal. you can tell JMP that your goal is to obtain the maximum or minimum value possible. The following description explains the relationship between the goal type (step 3 in Figure 4.88 4 Screening Designs—Creating a Screening Design 4. or that there is no goal.” p. .9) and the lower and upper limits (step 4 in Figure 4. “Continuing the Analysis.

10.10 Entering Factors To enter factors. when designing experiments. and enter an asymmetric target value. in Figure 4. then press the Return or Enter key on your keyboard. The default target value is assumed to be midway between the lower and upper limits. For information on entering factors in Taguchi designs. If there is only one response. Space Filling Design.” p. click it. Entering factors is the same in Screening Design.9) as the weight of each response. Types of Factors In general. type the number of factors and click Add.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. open the response’s Column Info dialog by double-clicking the column name. and “Creating a Custom Design. you can alter the default target after you make a table from the design. Understanding Importance Weights When computing overall desirability. Highlight the factor and click the Remove Selected button to remove a factor in the list.” p. you can enter different types of factors in the model. “Full Factorial Designs. Click to enter factor values. then specifying importance is unnecessary. press the delete key on your keyboard. JMP uses the value you enter as the importance weight (step 4 in Figure 4. and Response Surface Design. 57. Mixture Design. With two responses you can give greater weight to one response by assigning it a higher importance value. full factorial designs. enter factors. Below is a description of factor types from which you can choose when creating screening designs: . Note: If your target range is not symmetric around the target value. This process is described below. Double-click to edit the factor name. In the data table. such as with part dimensions. 117. Entering Factors After entering responses. The Factors panel’s appearance depends on the design you select. see “Taguchi Designs. and custom designs. To remove a level. 183. Figure 4.” p.

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

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. –1 -1. 1. This is different from resolution-three fractional factorial where two-factor interactions are indistinguishable from main effects. -1. An effect is nonestimable when it is confounded with another effect. -1. This provides a good estimate of everything. One useful characteristic is that the sample size is a multiple of four rather than a power of two. The main effects are orthogonal and two-factor interactions are only partially confounded with main effects. or between main effects and two-factor interactions. -1 1. This causes some of the higher-order effects in a model to become nonestimable. fractional factorials are designed by deciding in advance which interaction effects are confounded with the other interaction effects. 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.12 is geometric representation of a two-level fractional factorial design. In fact. • Resolution = 4 means that main effects are not confounded with other main effects or two-factor interactions. There are no two-level fractional factorial designs with sample sizes between 16 and 32 runs. two-factor interactions are confounded with other two-factor interactions. which must be assumed to be zero for the main effects to be meaningful. 1. 1 . you typically only run a fraction of all possible levels. –1. you can run several replicates of all possible combinations of m factor levels. and 28-run Plackett-Burman designs. Figure 4.12 Representation of Full Factorial (Left) and Two-Level Fractional Factorial (Right) Designs –1. 1. 1 Plackett-Burman Designs Plackett-Burman designs are an alternative to fractional factorials for screening. • Resolution ≥ 5 means there is no confounding between main effects. If price is no object. The figure on the right in Figure 4. the minimum aberration design of a given resolution minimizes the number of words in the defining relation that are of minimum length. between two-factor interactions. But because running experiments costs time and money. few experimenters worry about interactions higher than two-way interactions. For DOE experts. These higher-order interactions are assumed to be zero. All the fractional factorial designs are minimum aberration designs. However. there are 20-run. -1 1. including interaction effects to the mth degree. Resolution Number: The Degree of Confounding In practice. 24-run. -1 1. However.

A. a stepwise regression approach can allow for removing some insignificant main effects while adding highly significant and only somewhat correlated two-factor interactions.92 4 Screening Designs—Creating a Screening Design In cases of effect sparsity. Some of these designs are not balanced. Suppose you believe in effect sparsity— that very few effects are truly nonzero. Cotter designs are easy to set up. and you believe there may be interactions. and the others high. if you have three factors. This completes the Cotter design. Then follow k runs each with one factor in turn at its low level. 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. Cotter Designs Cotter designs are used when you have very few resources and many factors. listed below. For pure three-level factorials. The danger is that several active effects with mixed signs will cancel and still sum to near zero and give a false negative. When you use JMP to generate a Cotter design. A Cotter design begins with a run having all factors at their high level. You believe in this so strongly that you are willing to bet that if you add up a number of effects. JMP offers complete factorials and specialized orthogonal-array designs. 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 . subject to randomizing the runs. Mixed-Level Designs If you have qualitative factors with three values. For k factors. These are of the form factorOdd and factorEven where factor is a factor name. 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. JMP offers fractional factorials. JMP also includes a set of extra columns to use as regressors. B. For mixed two-level and three-level designs. The design is similar to the “vary one factor at a time” approach many books call inefficient and naive. even though they are all orthogonal. then none of the classical designs discussed previously are appropriate. They are constructed by adding up all the odd and even interaction terms for each factor. For example.

How to Run a Cotter Design By default. Select File > Preferences. Changing the setting via the red triangle menu applies only to the current design. Figure 4. 2. click the disclosure button ( on Windows/Linux and on the Macintosh) to reveal the Aliasing of Effects panel. AEven estimates the sum of the even terms involving A. Select Supress Cotter Designs.4 Screening Designs—Creating a Screening Design 93 4 Screening Because these columns in a Cotter design make an orthogonal transformation. It shows effects and confounding up to two-factor interactions (Figure 4. and so forth.13). click the red triangle icon in the Screening Design title bar. .13 Display and Modification Options • Change Generating Rules—Controls the choice of different fractional factorial designs for a given number of factors. Aliasing of Effects To see which effects are confounded with which other effects. Click DOE to highlight it. AOdd estimates the sum of odd terms involving A. In the example of factors listed above. 3. However. Click the Platform icon.14).11). 4. many statisticians discourage their use. Uncheck the box beside Supress Cotter Designs. Displaying and Modifying the Design After you select a design type. Immediately after entering responses and factors (and before clicking Continue). To alter the setting for all screening designs: 1. JMP does not include a Cotter design in the list of available screening designs (Figure 4. • Aliasing of Effects—Shows the confounding pattern for fractional factorial designs. if you would like to make a Cotter design: 1. 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. 2. Because Cotter designs have a false-negative risk. testing the parameters on these combinations is equivalent to testing the combinations on the original effects. • Coded Design—Shows the pattern of high and low values for the factors in each run.

This means that you can’t tell the difference of the effect of Temperature and the synergistic (interactive) effect of Feed Rate and Concentration. For example. all the main effects are confounded with two-factor interactions. Confounding is the direct result of the assignment of new factor values to products of the coded design columns.14 Generating Rules and Aliasing of Effects Panel For example. Click the red triangle at the bottom of the Aliasing of Effects area.14. the values for Temperature are the product of the values for Feed Rate and Concentration. It is necessary to construct the two additional factors in terms of the first three factors. In the example shown in Figure 4. Select Show Confounding Pattern (Figure 4.15). 2.94 4 Screening Designs—Creating a Screening Design Figure 4. Eight runs can only accommodate a full factorial with three two-level factors. Viewing the Confounding Pattern JMP can create a data table that shows the aliasing pattern for a specified level. a full factorial with five factors requires 25 = 32 runs. This is characteristic of resolution-three designs. The price of reducing the number of runs from 32 to eight is effect aliasing (confounding). . To create this table: 1.

Catalyst. Thus. rows for the first three columns of the coded design. Figure 4. Enter the order of confounding you want to see (Figure 4.15 Show Confounding Patterns 3. and Stir Rate are all combinations of high and low values (a full . with second order aliasing as “1 5” (Feed Rate and Concentration). and Stir Rate). Click OK. As shown in Figure 4. Temperature appears as “4”. which represent Feed Rate. The effect names begin with C (Constant) and are shown by their order number in the design. and third order confounding as “1 2 3” (Feed Rate.17 shows the third level alias for the five-factor reactor example. Figure 4.17 The Third Level Alias for the Five-Factor Reactor Example Understanding the Coded Design In the coded design panel.16 Enter Order 4. each row represents a run.18. Catalyst.4 Screening Designs—Creating a Screening Design 95 4 Screening Figure 4. Plus signs designate high levels and minus signs represent low levels.16). Figure 4.

the last column (Concentration) is the product of the second and third columns. and Stir Rate). Temperature is now the product of Feed Rate and Catalyst. Figure 4. The check marks for Temperature show it is a function of Feed Rate. Similarly. changing the checkmarks and clicking Apply changes the coded design. Catalyst.19 and click Apply. Catalyst. so the fourth column of the coded design is the element by element product of the first two columns.96 4 Screening Designs—Creating a Screening Design factorial design). the Coded Design panel changes. and Stir Rate. The fourth column (Temperature) of the coded design is the element-by-element product of the first three columns. Concentration is a function of Feed Rate and Stir Rate. The checkmarks for Concentration show it is a function of Catalyst and Stir Rate. it changes the choice of different fractional factorial designs for a given number of factors. Figure 4.18 Default Coded Designs Changing the Coded Design In the Change Generating Rules panel. If you check the options as shown in Figure 4.19 shows how the last two columns are constructed in terms of the first three columns. The Change Generating Rules table in Figure 4.19 Modified Coded Designs and Generating Rules . The first three columns of the coded design remain a full factorial for the first three factors (Feed Rate.

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

22).98 4 Screening Designs—Creating a Screening Design Figure 4. you can now run analyses on the data. This script allows you to easily fit a model using the values in the design table. Continuing the Analysis After creating and viewing the data table. The data table contains a script labeled Model. Figure 4. Pattern is especially generated it. .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 “+”. useful as a label variable in plots. Right-click it and select Run Script to run a fit model analysis (Figure 4.22 Running the Model Script The next sections describe some of the parts of the analysis report that appears when you click Run Model.

The last column of the table has the p-values for each effect. the mean line falls inside the bounds of the 95% confidence curves.4 Screening Designs—Creating a Screening Design 99 4 Screening Viewing an Actual-by-Predicted Plot When the model contains no interactions. JMP displays a Scaled Estimates report (Figure 4.23. it is impossible to determine which two-factor interactions are important without performing more experimental runs. Because of the confounding between two-factor interactions and main effects in this design. shown on the left in Figure 4. right-click the graph. which is much larger than expected. None of the factor effects are significant.23). The Scaled Estimates report displays a bar chart of the individual effects embedded in a table of parameter estimates. In this case.199. but this does not mean that the experiment has failed.23. The model p-value. The RMSE is an estimate of the standard deviation of the process noise assuming that the unestimated effects are negligible. The pattern variable displayed in the data table serves as the label for each point. At this stage the results are not clear. Figure 4.24) as a part of the Fit Model report. select all points. an actual-by-predicted plot. which tells you that the model is not significant.23 An Actual-by-Predicted Plot To show labels in the graph (on the right in Figure 4. the RMSE is 14. This suggests that effects other than the main effects of each factor are important. and RMSE appear below the plot. R2. appears at the top of the Fit Model report.24 Example of a Scaled Estimates Report . 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. In Figure 4. and select Row Label. It means that some follow-up runs are necessary.

Open the outline node. such as described in “Augmented Designs. see the chapter “Full Factorial Designs. 117.25 Example of a Power Analysis Refer to the JMP Statistics and Graphics Guide for details.” p. 193. JMP displays an Effect Details panel. Then run the power analysis by clicking the red triangle icon in a factor’s title bar and selecting Power Analysis. you might also want to have complete 32-run factorial experimental data and analysis. . you would then want to augment the design.100 4 Screening Designs—Creating a Screening Design If this scaled estimates report were not merely an example. For comparison.” p. Figure 4. For this. Running a Power Analysis When the model contains interactions.

is an alternative to central composite designs. a response surface model can pinpoint it. The Box-Behnken design. illustrated in the figure to the left below. illustrated in the figure on the right below. 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. for which all the factor values are at the zero (or midrange) value.5 Response Surface Designs 5 Surface Response surface designs are useful for modeling a curved quadratic surface to continuous factors. The price of this characteristic is the higher uncertainty of prediction near the vertices compared to 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. It combines a two-level fractional factorial and two other kinds of points: • Center points. so the standard two-level designs cannot fit curved surfaces. This is sometimes useful when it is desirable to avoid these points due to engineering considerations. If a minimum or maximum response exists inside the factor region. Central Composite Design fractional factorial points Box-Behnken Design axial points center points . The most popular response surface design is the central composite design. • Axial (or star) points. for which all but one factor are set at zero (midrange) and that one factor is set at outer (axial) values.

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.

The sliders for each factor set values for Current X and Current Y.09 raised the predicted response from 396 to the target value of 450. This example shows the profile to Silica and Sulphur for a fixed value of Silane. The contour profiler is useful for viewing response surfaces graphically. Click the red triangle on the Response title bar and select Factor Profiling > Contour Profiler to display the interactive contour profiler.3 to 2.9.5 Response Surface Designs—A Box-Behnken Design: The Tennis Ball Example 107 5 Surface Figure 5.2 to 0. and Sulfur from 2. Changing the settings of Silica from 1.9.8 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. Using a Response Surface Plot (Contour Profiler) Another way to look at the response surface is to use the Contour Profiler.8 shows the result of the most desirable settings.5. Options on the Contour Profiler title bar can be used to set the grid density. as shown in Figure 5. especially when there are multiple responses. Figure 5.98. and add contours at specified intervals. Silane from 50 to 41. request a surface plot (mesh plot). . as shown in the contour plot in Figure 5.

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. . Figure 5.9 Contour Profiler for a Response Surface Analysis Figure 5. drawing shaded regions on the contour.10 shows the Contour profile when the Current x values have Lo and Hi limits in effect.

11 illustrates the three Box-Behnken design columns. inscribe the points in a box. You can clearly see the 12 points midway between the vertices. . and suppress the x. see the JMP Statistics and Graphics Guide. and z axes. Options available on the spin platform draw rays from the center to each point. y. leaving three points in the center.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. Figure 5.

Response surface designs are useful for modeling and analyzing curved surfaces. and Full Factorial Design. Mixture Design. RSM designs involve a small number of factors. . select DOE > Response Surface Design.” p. or click the Response Surface Design button on the JMP Starter DOE page. 113 • “Continuing the Analysis. Some of these designs also allow blocking. These steps are outlined in “Entering Responses and Factors in the Custom Designer.” p. 110 • “Choosing a Design. follow the step in Figure 5. 57 The steps for entering factors in a response surface design are unique to this design. In industrial applications. RSM can find it.” p. This is because the required number of runs increases dramatically with the number of factors. you choose to use well-known RSM designs for two to eight continuous factors. If Needed.” p. Response Surface Design. Using the response surface designer.12.110 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. If a maximum or minimum exists inside the factor region. These designs are capable of fitting a second-order prediction equation for the response. To start a response surface design. 111 • “Specifying Axial Value (Central Composite Designs Only). Space Filling Design. Custom Design. To add factors. 112 • “Specifying Output Options. Then. 113 • “Viewing the Design Table.” p.” p. follow the steps below: • “Entering Responses and Factors. 114 Entering Responses and Factors The steps for entering responses are the same in Screening Design. The quadratic terms in these equations model the curvature in the true response function.” p.

Figure 5.5 Response Surface Designs—Creating a Response Surface Design 111 5 Surface Figure 5. The price of this characteristic is the higher uncertainty of prediction near the vertices compared to the central composite design. This is sometimes useful when it is desirable to avoid extreme points due to engineering considerations.12 Entering Factors into a Response Surface Design Click Continue to proceed to the next step.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. . Choosing a Design Highlight the type of response surface design you would like to use and click Continue.

which controls how far out the axial points are. it is recommended that you choose On Face or specify your own value. Specifying Axial Value (Central Composite Designs Only) When you select a central composite (CCD-Uniform Precision) design and then click Continue. This causes the axial points to be more extreme than the –1 or 1 representing the range of the factor. Enter that value into the Axial Value text box. • Orthogonal makes the effects orthogonal in the analysis.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. It supplies default axial scaling information. If you would like to inscribe the design. If this factor range cannot be practically achieved. • On Face leaves the axial points at the end of the -1 and 1 ranges. You have the flexibility to enter the values you want to use. This causes the axial points to be more extreme than the range of the factor. • User Specified uses the value entered by the user. • For orthogonal designs. Figure 5.112 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. you see the panel in Figure 5. 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). Entering 1.0 in the text box instructs JMP to place the axial value on the face of the cube defined by the factors. it is recommended that you choose On Face or specify your own value. When checked. which can be any value greater than zero. click the box beside Inscribe. . If this factor range cannot be practically achieved. the number of center points is chosen so that the second order parameter estimates are minimally correlated with the other parameter estimates.14. 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.

One replicate doubles the number of runs. When the options are correctly set up. as described in Figure 5. Viewing the Design Table Now you have a data table that outlines your experiment. • Number of Center Points—Specifies additional runs placed at the center points. • Sort Right to Left—the rows (runs) in the output table will appear sorted from right to left. Choices are: • Keep the Same—the rows (runs) in the output table will appear as they do in the Design panel. click Make Table. Figure 5. • Randomize within Blocks—the rows (runs) in the output table will appear in random order within the blocks you set up. • Number of Replicates—Specify the number of times to replicate the entire design.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. Type the number of times you want to replicate the design in the associated text box. .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.16. • Sort Left to Right—the rows (runs) in the output table will appear sorted from left to right. including centerpoints. • Randomize—the rows (runs) in the output table will appear in a random order.

16 The Design Data Table The column called Pattern identifies the coding of the factors. Continuing the Analysis. Figure 5. the pattern value shows the factor coding of the point.17 Running the Script After clicking Run Model in the dialog. running the experiment. It shows all the codings with “+” for high. “–” for low factor. There are two center points per replicate. . review the analysis. The name of the table is the design type that generated it. The data table contains a script labeled Model. If Needed After creating and viewing the design table. (Figure 5.18).17) to fit the model. and “0” for midrange.114 5 Response Surface Designs—Creating a Response Surface Design Figure 5. “a” and “A” for low and high axial values. The Y column is for recording experimental results. Pattern is suitable to use as a label variable in plots because when you hover over a point in a plot of the factors. and recording your results in the design table’s Y column. This script allows you to easily fit a model using the values in the design table. Runs are in a random order. Right-click it and select Run Script (Figure 5. run a fit model analysis on the data.

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 .

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

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

To follow along with this example. Stir Rate. The values in your table may be different from those shown below. adapted from Meyer et al. Since there are five variables. The design data table (Figure 6. the table has an empty Y column named Percent Reacted for entering response values when the experiment is complete. The factors (Feed Rate. 3. Click the red triangle icon on the Full Factorial Design title bar and select Load Factors.jmp to load the responses by opening the Sample Data folder that was installed with JMP.jmp to load the responses by opening the Sample Data folder that was installed with JMP. Click Make Table. open the folder Sample Data that was installed when you installed JMP. Hunter. In the Sample Data folder.jmp.jmp. giving 32 runs. Catalyst. The completed dialog is shown in Figure 6. 2. Temperature. Click the red triangle icon on the Full Factorial Design title bar and select Load Responses. (1996) and Box.6 Full Factorial Designs—The Five-Factor Reactor Example 119 6 Factorial The Five-Factor Reactor Example The following example. 1.1. Open Reactor Factors. open Design Experiment > Reactor 32 Runs.jmp. 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. Click Continue to see Output Options panel. and Hunter (1978).3) contains a run for every combination of high and low values for the five variables. open Design Experiment > Reactor Factors. This covers all combinations of a five factors with two levels each. Initially. 7. and Concentration) are all two-level continuous factors. In the Sample Data folder. Open Reactor Response. Within this folder. there are 25=32 runs. 4. . shows a five-factor reactor example. open Design Experiment > Reactor Response. 5. Figure 6. Select DOE > Full Factorial Design.” p. A full factorial design includes runs for all combinations of high and low factors for the five variables. 81).1 Full-Factorial Example Response and Factors Panels 6.

open the folder Sample Data that was installed when you installed JMP. The results are shown in Figure 6. .jmp.2 25 Factorial Reactor Data To see the completed experiment and continue following this example. 3. 1. Columns.120 6 Full Factorial Designs—The Five-Factor Reactor Example Figure 6. Within this folder.4. Click the red triangle icon on the Percent Reacted title bar and select Normal Quantile Plot.3 Reactor 32 Runs. open Design Experiment > Reactor 32 Runs. Highlight Percent Reacted and click Y. Figure 6.jmp Begin the analysis with a quick look at the data before fitting the factorial model. Select Analyze > Distribution. Then click OK. 2.

5.05. The probability to remove a factor (Prob to Leave) should be 0.1. To do this. and brings up the Stepwise control panel. 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. 3.6. Figure 6.4 Distribution of Response Variable for Reactor Data Start the formal analysis with a stepwise regression. A useful way to use the Stepwise platform is to check all the main effects in the Current Estimates table. The probability to enter a factor (Prob to Enter) in the model should be 0. Click the red triangle icon next to the Fit Model script and select Run Script. 2. .6 Full Factorial Designs—The Five-Factor Reactor Example 121 6 Factorial Figure 6.5 Run JSL Script for Stepwise Regression 4. 1. Click Go. Check the boxes for the main effects of the factors as shown in Figure 6. make sure the menu beside Direction specifies Mixed.

Figure 6.7 Model After Mixed Stepwise Regression 7. The end result is shown in Figure 6.6 Starting Model For Stepwise Process The mixed stepwise procedure removes insignificant main effects and adds important interactions. .122 6 Full Factorial Designs—The Five-Factor Reactor Example Figure 6. Click the Make Model button.7. The Model Specification window that appears is automatically set up with the appropriate effects (Figure 6.8). Note that the Feed Rate and Stir Rate factors are no longer in the model.

10. The figure on the right in Figure 6.9 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.9). which is more than an order of magnitude smaller than the range of predictions. . The figure on the left in Figure 6. The predicted model covers a range of predictions from 40% to 95% reacted. 1.9 shows a table of model coefficients and their standard errors (labeled Parameter Estimates). Figure 6. Open the Prediction Profiler by clicking the red triangle on the Response Percent Reacted title bar and selecting Factor Profiling > Profiler. as shown in Figure 6. This is strong evidence that the model has good predictive capability.6 Full Factorial Designs—The Five-Factor Reactor Example 123 6 Factorial Figure 6.9 shows the actual by predicted plot for the model.8 Fitting a Prediction Model 8. The size of the random noise as measured by the RMSE is only 3.3311%. All effects selected by the stepwise process are highly significant. Click Run Model to see the analysis for a candidate prediction model (Figure 6.

Click the red triangle on the Prediction Profiler title bar and select Maximize Desirability to see the profiler in Figure 6. 101.12 Viewing the Maximum Desirability . Figure 6.11 shows the profiler’s initial display.” p.11 Viewing the Profiler 2.10 Selecting the Profiler Figure 6. The Prediction Profiler is discussed in more detail in the chapter “Response Surface Designs. Figure 6. 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 6.12.

Figure 6. press the delete key on the keyboard. see Figure 6. click it. follow the steps below: • “Entering Responses and Factors. To remove a level. Percent Reacted increases from 65. 125 • “Selecting Output Options. select DOE > Full Factorial Design.” p.875 at the most desirable setting. Then.” p.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. 126 • “Making the Table.13.” p. Double-click to edit the factor name. 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. then press the Return or Enter key on the keyboard. click either the Continuous button or the Categorical button and select a factor type. Creating a Factorial Design To start a full factorial design. or click the Full Factorial Design button on the JMP Starter DOE page. To add factors.13 Entering Factors in a Full Factorial Design To enter factors.9. level 2 . click Continue. The reaction is unfeasible economically unless the Percent Reacted is above 90%.945 when Catalyst and Temperature are at their highest settings and Concentration is at its lowest setting.5 at the center of the factor ranges to 95. 126 Entering Responses and Factors The steps for entering responses are outlined in “Entering Responses and Factors in the Custom Designer. . The maximum Desirability is 0. Desirability increases linearly as the Percent Reacted increases. therefore the Desirability for values less than 90% decreases and finally becomes zero.” p. When you finish adding factors.

Figure 6.14 Factorial Design Table The name of the table is the design type that generated it. values in the Pattern column describe the run each row represents for continuous factors. Type the number of times you want to replicate the design in the associated text box. Choices are: • Keep the Same—the rows (runs) in the output table will appear as they do in the Design panel. • 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.14 appears. Making the Table When you click Make Table. including centerpoints. . the table shown in Figure 6. • Sort Left to Right—the rows (runs) in the output table will appear sorted from left to right. 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. • Randomize—the rows (runs) in the output table will appear in a random order. a plus sign represents high levels for continuous factors. One replicate doubles the number of runs.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. • Sort Right to Left—the rows (runs) in the output table will appear sorted from right to left.

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

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

• Simplex Lattice You specify how many levels you want on each edge of the grid.1 Mixture Design Selection Dialog The following sections show examples of each mixture design type. click Continue. 5. shown Figure 7.1. 4.” p. following instructions regarding the Design Generation panel in “Selecting the Number of Runs. you see the mixture designer. • Simplex Centroid You specify the degree up to which the factor combinations are to be made. 2. Then click Make Design and complete the remainder of the design. From here. After your responses and factors are set up. Click the Optimal button. • Extreme Vertices You specify linear constraints or restrict the upper and lower bounds to be within the 0 to 1 range. 57. To create an optimal mixture design: 1.” p.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. Add effects to the model using the instructions below. 62 and the prediction variance profile in “Understanding the Design and . The extreme vertices design is the most flexible. The steps for entering responses are outlined in “Entering Responses and Factors in the Custom Designer. Figure 7. 3. • ABCD Design This approach by Snee (1975) generates a screening design for mixtures. select one of the following types of design: • Optimal Choosing Optimal invokes the custom designer with all the mixture variables already defined. Creating an Optimal Mixture Design Creating an optimal mixture design invokes the custom designer with all the mixture variables already defined. since it handles constraints on the values of the factors. Select DOE > Mixture Design. Enter factors and responses.

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

Click Make Table.” p. then mix equal proportions of K ingredients at a time to the specified limit. View factor settings. 57. 5. 6. Choices are: • Keep the Same—the rows (runs) in the output table will appear as they do in the Design panel. The first three runs are for each single ingredient. • Randomize—the rows (runs) in the output table will appear in a random order. Specify run order. Simplex Centroid Design Examples The table of runs for a design of degree 1 with three factors (left in Figure 7. the second set shows each combination of two ingredients in equal parts. • Sort Right to Left—the rows (runs) in the output table will appear sorted from right to left. Enter factors and responses. Select DOE > Mixture Design. A center point run with equal amounts of all the ingredients is always included. The table of runs to the right is for three factors of degree 2. 9. and the last run is the center point. 3. Click the Simplex Centroid button. .3) shows runs for each single ingredient followed by the center point. The steps for entering responses are outlined in “Entering Responses and Factors in the Custom Designer. 4. Creating the Design To create a simplex centroid design: 1. Specify number of replicates. 7. 8. After your responses and factors are set up. • Randomize within Blocks—the rows (runs) in the output table will appear in random order within the blocks you set up. click Continue. • Sort Left to Right—the rows (runs) in the output table will appear sorted from left to right. 2. which is the order you want the runs to appear in the data table when it is created. JMP will run each ingredient without mixing. Enter the number of ingredients in the box labeled K.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.

in the K box. and click Simplex Centroid to see the design on the left in Figure 7. Enter 1 in the K box.132 7 Mixture Designs—Using the Simplex Centroid Design Figure 7. 5. 4. Enter three mixture factors. Then click Simplex Centroid to see the design on the right in Figure 7. Choose DOE > Mixture Design. 2. Click Continue. 5. 2. Use the default value. .3 Three-Factor Simplex Centroid Designs of Degrees 1 and 2 To generate the set of runs in Figure 7. Click Simplex Centroid. 3. Click Make Table to see the 31-run JMP data table shown in Figure 7. 4.4 Create Simplex Centroid Designs of Degrees 1 and 2 As another example: 1.3: 1. 3. Enter five factors and click Continue. Figure 7.4. Click the Back button and enter 2 in the K box. 4. Choose DOE > Mixture Design.4.5.

and 5. 3. To create simplex lattice designs: 1. . with their corresponding geometric representations.1) and click Simplex Lattice. In contrast to the simplex centroid design. Specify the number of levels you want in the Mixture Design Type dialog (Figure 7. Select DOE > Mixture Design. 4.” p. The design is the set of all combinations where the factors’ values are i / m. Figure 7. Enter factors and responses. 2. the simplex lattice design does not necessarily include the center point. 4. 57. Click Continue.7 Mixture Designs—Using the Simplex Lattice Design 133 7 Mixture Figure 7. The steps for entering responses are outlined in “Entering Responses and Factors in the Custom Designer.6 shows the runs for three-factor simplex lattice designs of degrees 3. where i is an integer from 0 to m such that the sum of the factors is 1.5 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.

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

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

4. Enter 4 in the Degree text box (Figure 7. (Optional) Alter the sample size by typing into the text box beside Choose desired sample size. 9. Add two factors and given them the values shown in Figure 7. 10. Click Continue. (Optional) Click Find Subset to generate the optimal subset having the number of runs specified in sample size box described in Step 8. Select DOE > Mixture Design.Click Make Table. • Randomize within Blocks—the rows (runs) in the output table will appear in random order within the blocks you set up. 2.8. An Extreme Vertices Example with Range Constraints The following example design table is for five factors with the range constraints shown in Figure 7. 3. 8. 1. where the ranges are smaller than the default 0 to 1 range.8 Example of Five-factor Extreme Vertices . Figure 7.8).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.

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

140. Click the Linear Constraints button three times.11. 6. Click Continue. Change the run order to Sort Right to Left.1 X1 ≤ 0. 2.4 ≤ 0. Click Make Table.” p. 3. as shown in Figure 7.7 X3 ≤ 0. 5. with five individual factor bound constraints and three additional linear constraints: X1 ≥ 0. See “Creating Ternary Plots. 4. and keep the sample size at 13 to see the 13-run factor settings like those shown on the right in Figure 7. X2.7*X1 + X3 Enter the upper and lower limits in the factors panel. X1. This example has three factors. for details. This example is best understood by viewing the design as a ternary plot.7 90 ≤ 85*X1 + 90*X2 + 100*X3 85*X1 + 90*X2 + 100*X3 ≤ 95 .11.5 X2 ≥ 0.13. Click the Extreme Vertices button. Enter the constraints as shown in Figure 7.138 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). 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.1 X2 ≤ 0. . Figure 7. and X3.11 Constraints 1.

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

There are six active constraints. Complete the window and click OK. 1/2. 0) (1/2.8) X2 (0. 1/3) (1. Specify number of replicates. and six centroid points shown on the plot. Create the design by choosing DOE > Mixture Design and making the design data table. 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. as well as two inactive (redundant) constraints.13 shows how close to one a given component is by how close it is to the vertex of that variable in the triangle. 1) X2 X1 X3 To view your mixture design as a ternary plot: 1.13. six vertices.12 illustrates a ternary plot. 1.140 7 Mixture Designs—Creating Ternary Plots 7. 138 is best understood by the ternary plot shown in Figure 7.12 Ternary Plot (left) and Tetrary Plot (right) for Mixture Design X1 (1. 8. The ternary plot in Figure 7. The Piepel (1979) example referenced in “An Extreme Vertices Example with Linear Constraints.” p. The ternary plot in Figure 7. Figure 7. The area that satisfies all constraints is the shaded feasible area. 3. Click Make Table. 0. 0. Choose Graph > Ternary Plot. Each constraint is a line. 0. 0) (1/3.13 shows how close to one a given component is by how close it is to the vertex of that variable in the triangle. The plot to the left in Figure 7. 1/3. 0) X3 (0. 2. 1. .

called the Scheffé polynomial (Scheffé 1958). This table property. 2. The design data table stores the model in the data table as a table property. 3. 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. and thus a traditional full linear model will not be fully estimable.13 Ternary Plot Showing Piepel Example Constraints Fitting Mixture Designs When fitting a model for mixture designs. Right-click the model and select Run Script to launch the Model Specification dialog. 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. is located in the left panel of the table. the Y column in the data table must contain values. called Model. which is . Remember that to fit the model. so either assign responses or click the red triangle menu and select Simulate Responses before making the table.

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

Creating the Design To create Cornell’s mixture design in JMP: 1.15. and p3) comprise 79. and enter the high and low constraints as shown in Figure 7. The contour plot feature of these reports does not fold to handle mixtures. if JMP encounters a no-intercept model and finds a hidden intercept with linear response surface terms. 1990).849. name them p1. p2.409 ≤ x1 ≤ 0. but no square terms. 2. Import the factors by opening the Sample Data folder that was installed when you installed JMP and selecting Design Experiment > Plastifactors. A Chemical Mixture Example Three plasticizers (p1. 0 ≤ x2 ≤ 0. collapsing on the last response surface term to calculate critical values for the optimum. Or. However. and 0. Select DOE > Mixture Design. and p3.5%.5% of the vinyl used for automobile seat covers (Cornell. load the factors by clicking the red triangle on the Mixture Design title bar and selecting Load Factors. not mixture models. and add points to make the plot less granular. In the Factors panel.” it creates additional reports that analyze the resulting fitted response surface.252. p2. These reports were originally designed for full response surfaces.jmp.151 ≤ x3 ≤ 0. It can do this for any combination yielding a constant and involving the last response surface term. use the three default factors. the individual plasticizers are restricted by the following constraints: 0. If you want a contour plot of the surface.274. Figure 7. you can do any of the following: • Save the model prediction formula and then create a ternary plot by selecting Graph > Ternary Plot.15 Setting Up Factors . • Create a separate contour plot by selecting Graph > Contour Plot.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. then it folds its calculations. • Refit the model using a full response surface that omits the last factor. Within this 79.

This column was added by clicking the .16 Mixture Data for Analysis 7. which runs a script that creates a completed dialog. 6.jmp to see the results (Y values) that Cornell obtained (Figure 7.17 Plasticizer. To do this. Figure 7. Click Exreme Vertices. select Rows > Add Rows.144 7 Mixture Designs—A Chemical Mixture Example 3.17). This gives a total of 14 rows in the design table. Figure 7. 4. 5. and click OK. Plasticizer.16). enter 5 as the number of rows to add.jmp contains a column called Pred Formula Y. Click Make Table. select Edit > Copy. Then click Run Model. Open the Sample Data folder that was installed when JMP was installed. Click Continue. highlight the rows you want to duplicate. Add an extra five design runs by duplicating the vertex points and center point.jmp Running a Mixture Model Now run the mixture model: 1. Enter 3 in the Degree text box. Then select Edit > Paste. 2. 8. JMP uses the 9 factor settings to generate a JMP table (Figure 7. Open Plasticizer. Right-click the table property named Mixture RSM and select Run Script.

To see the prediction formula. shown in Figure 7.570299 occurs at point (0. The Response Surface Solution report (Figure 7.1982*p2 – 911. displays optimal settings of 0. Figure 7. and the other two move in the opposite direction maintaining their ratio. The crossed effects show as curvature in the prediction traces. 0. . Drag one of the vertical reference lines.7 Mixture Designs—A Chemical Mixture Example 145 7 Mixture red triangle in the Response Y title bar and selecting Save Columns > Prediction Formula.6615 for p1.63505. click the red triangle in the Response Y title bar and select Factor Profiling > Profiler.26923.18) shows that a maximum predicted value of 19.2177 Your values may be different from those shown below.1465*p1 – 282. which give an estimated response of 19.18 Mixture Response Surface Analysis Viewing the Prediction Profiler The report contains a prediction profiler.20927). If the profiler is not visible. 0. right-click (Ctrl+click on the Mac) the column name and select Formula: 0–50. and 0.21225 for p3. Note: These results correct the coefficients reported in Cornell (1990).6484*p3 + p2*p1317.126 for p2.3298 + p3*p2*1846.15568. 0.363 + p3*p1*1464. The profiler.19.

Click OK and view the results. . as shown by the ternary plots at the bottom in Figure 7.20. the ternary plot only shows points.21. p2. 2.19 Profiler for Mixture Analysis Example Plotting a Mixture Response Surface To plot a mixture response surface: 1. Figure 7.21. The contour variable must have a prediction formula to form the contour lines. Plottting button. by selecting p1. Identify the contour variable (the prediction equation) by selecting Pred Formula Y and clicking the Contour Formula button. and p3 and clicking the Y. as shown in Figure 7. If there is no prediction formula. as shown in Figure 7. Choose Graph > Ternary Plot. Specify plot variables.146 7 Mixture Designs—A Chemical Mixture Example Figure 7.20 Creating a Ternary Plot 3.

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

.

randomization and blocking are irrelevant. Replication is undesirable because repeating the same run yields the same result. The second objective is to space the points uniformly. 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) . In space filling designs. One objective is to spread the design points out to the maximum distance possible between any two points to prevent replicate points. there are two objectives. For this case. any variability is small enough to be ignored. and any mechanistic or deterministic modeling problem. Sensitivity studies of computer-simulations is one such situation.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. For systems with no variability.

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

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

. Click Continue. In design specification dialog. Figure 8.2. shown on the left in Figure 8.2 Space-Filling Design Dialog 7. as shown in Figure 8. specify a sample size.518 as the Minimum Distance.1. In the example shown in Figure 8. 6. Click the Sphere Packing button.2. Figure 8. JMP creates this design and displays the design runs and the design diagnostics (if specified in step 4).1 Space Filling Dialog for Two Factors 5.152 8 Space Filling Designs—Using the Sphere Packing Method click the red triangle icon on the Space Filling title bar and select Show Diagnostics. a sample size of eight was entered.3 shown the Design Diagnostics panel open with 0. Figure 8.

5. . 0. This script draws a circle centered at each design point with radius 0. 2.259 (half the diameter.518 in the graphics script to the minimum distance produced by 10 runs. Adjust the frame size so that the frame is square by right-clicking the plot and selecting Size/Scale > Size to Isometric. Specify X1 as X and X2 as Y. create an overlay plot. X2}.518 is the minimum distance number you noted in the Design Diagnostics panel. This plot shows the efficient way JMP packs the design points. select Add Graphics Script and enter the following script: For Each Row(Circle({X1. Select Graph > Overlay Plot. Sphere-Packing Design Settings and Design Diagnostics 8. 8. where 0. 9. Click the red triangle icon on the Space Filling title bar and select Show Diagnostics. Now repeat the above procedure exactly. and add circles using the minimum distance from the diagnostic report as the radius for the circle: 1.4. as shown on the left in Figure 8. Right-click the plot. Create a space filling design using the default X1 and X2 factors. 6.3 Space-Filling Design Dialog. then click Make Table. but change the values so they are 0 and 1. 7. adjust the plot’s frame size.8 Space Filling Designs—Using the Sphere Packing Method 153 8 Space Filling Figure 8. Visualizing the Sphere Packing Design To visualize the nature of the sphere packing technique. then click OK. Click the Sphere Packing button.518/2)). Remember to change 0. 4. 0. 3.518). but with a sample size of 10 instead of eight (step 4). Specify a sample size of eight. Click Make Table.

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

create a space filling design using the default X1 and X2 factors. adjust the plot’s frame size. 2. Figure 8. Visualizing the Latin Hypercube Design To visualize the nature of the sphere packing technique.6 Latin Hypercube Design for Eight Runs with Eight Levels 8. 3. Click Continue.8 Space Filling Designs—Using the Latin Hypercube Method 155 8 Space Filling Figure 8.7. 6. and each column is a different permutation of the levels. Click the red triangle icon on the Space Filling title bar and select Show Diagnostics. 7. Factor settings and design diagnostics results appear similar to those in Figure 8. and add circles using the minimum distance from the diagnostic report as the radius for the circle: 1. which shows the Latin hypercube design with eight runs. First. . Click Continue. In design specification dialog. Click Make Table. specify a sample size. create an overlay plot. but change the values so they are 0 and 1.5 Space-Filling Dialog for Four Factors 5. Note that each column (factor) is assigned each level only once. Click the Latin Hypercube button.

but with 10 runs instead of eight (step 4).202 (half the diameter.404/2)). X2}. This plot shows the efficient way JMP packs the design points. 10. Factor settings and design diagnostics are shown in Figure 8. 8.6).8. select Edit > Add Graphics Script from the menu. In fact.8. Specify a sample size of eight. You should see a graph similar to the one on the right in Figure 8. Specify X1 as X and X2 as Y. Right-click the plot and select Size/Scale > Frame Size to Isometric to adjust the frame size so that the frame is square. 0. as shown on the left in Figure 8. Select Graph > Overlay Plot.404).Repeat the above procedure exactly.8 Comparison of Latin Hypercube Designs with Eight Runs (left) and 10 Runs (right) . where 0. Figure 8. and enter the following script: For Each Row(Circle({X1.404 in the graphics script to the minimum distance produced by 10 runs. Click the Latin Hypercube button.156 8 Space Filling Designs—Using the Latin Hypercube Method 4. 9. Remember to change 0. you can again repeat the process to get a slightly different picture because the arrangement is dependent on the random starting point. 7.7 Latin Hypercube Design with Eight Runs The minimum distance 6. Note the irregular nature of the sphere packing.404 is the minimum distance number you noted in the Design Diagnostics panel (Figure 8. This script draws a circle centered at each design point with radius 0.Right-click the plot. Click Make Table. Figure 8. 11.7. 5. 0. then click OK.

These designs are most useful for getting a simple and precise estimate of the integral of an unknown function. Note that the emphasis of the Uniform Design method is not to spread out the points.” p. Also. and factors. Click Continue. a uniform design does not guarantee even spacing of the factor levels. . Click Make Table. 2. 57. if necessary. increasing the number of runs and running a distribution on each factor (by selecting Analyze > Distribution) shows nearly flat histograms. specify a sample size. The estimate is the average of the observed responses from the experiment. 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. To view the discrepancy value and the minimum distance from each point to its closest neighbor. (See “Entering Responses and Factors in the Custom Designer. if necessary. In design specification dialog. Figure 8. 4.8 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.9 Diagnostics for Uniform Space Filling Designs with Eight Runs 8. Alter the factor level values. 7. 1. click the red triangle icon on the Space Filling title bar and select Show Diagnostics.9 vary substantially. Enter responses. The sphere packing design maximizes the minimum distance without any constraints. JMP creates this design and displays the design runs and the design diagnostics (if specified in step 4). However. 5. Click the Uniform button.) 3. Select DOE > Space Filling. This is because the Latin hypercube design constrains the levels of each factor to be evenly spaced. The ‘minimum distances in Figure 8.1. 6.

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

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

15m . Figure 8. In design specification dialog. rw = radius of borehole. The response variable y is the flow rate through the borehole in m3/year and is determined by the equation 2πT u ( H u – H l ) y = -------------------------------------------------------------------------------------------2LT u Tu ln ( r ⁄ r w ) 1 + --------------------------------------. 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. 0. Click the Minimum Potential button.13 Minimum Potential Design Dialog 7.160 8 Space Filling Designs—Borehole Model Example 4. Figure 8. 5. enter a sample size.12 Minimum Potential Design Dialog 6.+ -----2 Tl ln ( r ⁄ r w ) r w K w There are eight inputs to this model. Click Continue. Click Make Table.05 to 0.

5. Specify a sample size of 32 runs (Figure 8. 4.045 m/year This example is atypical of most computer experiments because the response can be expressed as a simple. Open the Sample Data folder that was installed when you installed JMP.1 to 116 m2/year Hl = potentiometric head of lower aquifier.jmp to load the factors (Figure 8. 1120 to 1680 m Kw = hydraulic conductivity of borehole. 700 to 820 m L = length of borehole.8 Space Filling Designs—Borehole Model Example 161 8 Space Filling r = radius of influence. explicit function of the input variables. 2.000 m Tu = transmissivity of upper aquifier.14).600 m2/year Hu = potentiometric head of upper aquifier. 990 to 1100 m Tl = transmissivity of lower aquifier. this simplicity is useful for explaining the design methods.15). Click the red triangle icon on the Space Filling Design title bar and select Load Factors 3. 63. Creating a Sphere Packing Design To create a sphere packing design for the borehole problem: 1. 9855 to 12. 63. In that folder. . Figure 8. However. Click Continue. Select DOE > Space Filling. Design Experiment > Borehole Factors. 100 to 50.14 Loaded Factors for Borehole Example Note: The logarithm of r and rw are used in the following discussion.070 to 115.

and the results of the analysis are saved as columns in the table. 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. 7.16) This table also has a table variable that contains a script to analyze the data. p-values from fitted statistical models do not have their usual meanings. A large F statistic (low p-value) is an indication of an effect due to a model term. Click the Sphere Packing button to produce the design. Click Make Table to make a table showing the design settings for the experiment.162 8 Space Filling Designs—Borehole Model Example Figure 8.jmp (see Figure 8. To view a completed data table of this example. open the Sample Data folder that was installed when you installed JMP.15 Space Filling Design Method 6. you cannot make valid confidence intervals about the size of the effects or about predictions made using the model.16 Borehole Sample Data Guidelines for the Analysis of Deterministic Data It is important to remember that deterministic data has no random component. Open Design Experiment > Borehole Sphere Packing. Figure 8. . As a result. However.

• The prediction bias is relatively small for each of the experimental points. the true model column contains a relatively simple formula. the true model is generally not available in a simple analytical form.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. This is an indication that the model fits the data well. led to the prediction formula column. To understand the prediction bias in this example: 1.17. because the prediction bias formula is a function of columns that are also created by formulas. Select Graph > Profiler. The profile plots at the bottom in Figure 8. In this case. • The prediction bias column is the difference between the true model column and the prediction formula column. as shown at the bottom of the upper box in Figure 8. the profile traces would be constant between the value ranges of each factor. In real world examples. If there were no bias. Distinct patterns in the residuals indicate new terms to add to the model to reduce model bias. Check the Expand Intermediate Formulas box. . it is impossible to know the prediction bias at points other than the observed data without doing additional runs. Rather a residual shows the model bias for the current model at the current point. Click OK. Prediction Formula button. 4. log y. Highlight the prediction bias column and click the Y.17 show the prediction bias at the center of the design region. the variables Hu and Hl show nonlinear effects. 3. which allows profiling the prediction bias to find its value anywhere in the region of the data. As a result. Results of the Borehole Experiment The example described in the previous sections produced the following results: • A stepwise regression of the response. versus the full quadratic model in the eight factors. 2. In this example.

This is more than three times the size of the range over the observed data. Obtain this analysis by using Analyze > Distribution. This gives a range of 7.19 shows the maximum bias (2.684 (the range is 2. Note that the maximum bias is 1. To see this.51).91) over the entire domain of the factors.826 and the minimum is –0.17 Profile of the Prediction Bias in the Borehole Sphere Packing data.75. The range of the prediction bias on the data is smaller than the range of the prediction bias over the entire domain of interest. Figure 8. The plot at the bottom shows the comparable minimum bias (–4.18.84). look at the distribution analysis of the prediction bias in Figure 8.18 Distribution of the Prediction Bias The top plot in Figure 8. .164 8 Space Filling Designs—Borehole Model Example Figure 8.

The prediction bias over the experimental data underestimates the bias throughout the design domain. the response at any factor setting is unknown. the true model is given. 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. .8 Space Filling Designs—Borehole Model Example 165 8 Space Filling Figure 8. in this example.19 Prediction Plots showing Maximum and Minimum Bias Over Factor Domains Keep in mind that. • Verification runs— new runs performed at different settings to assess the lack of fit of the empirical model. 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.

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

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. Response button. Such models. Predictor Formula button (Figure 9. Highlight Velocity (y) and click the Y.1 Chemical Kinetics. you first need a data table that has one column for each factor. This is the same format for a table you would supply to the nonlinear platform for modeling. can yield far more accurate prediction of process behavior than is possible with the standard polynomial models. Figure 9. The second example describes how to approach creating the design when you have no data. To follow along with this example. Highlight Model (x) and click the X. open Chemical Kinetics.jmp You first fit the data to the model using nonlinear least squares to get better parameter values.jmp from Sample Data > Nonlinear Examples > Chemical Kinetics. . That table can be used to create a nonlinear design for improving the estimates of the model’s nonlinear parameters.2).jmp. but you have reasonable starting values for the parameter estimates. 3. 2. To use the Nonlinear designer. Fitting to Get Prior Parameter Estimates Suppose you have already collected experimental data and placed it in a JMP data table. and a column that contains a formula showing the functional relationship between the factor(s) and the response. Select Analyze > Modeling > Nonlinear. The first example in this section describes how to approach creating a nonlinear design when you have prior data. then click OK.jmp (Figure 9. 1. one column for the response. Chemical Kinetics.1) contains a column (Model (x))whose values are formed by a formula with parameter values that represent a poor initial guess. when they are descriptive of the underlying process. 4.

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

Use these limits when creating a nonlinear design in JMP. Click the Save Estimates button to have JMP place the fitted parameters in Chemical Kinetics.9 Nonlinear Designs—Example of Creating a Nonlinear Design 171 9 Nonlinear Figure 9. named Model (x). 1. 6. 2. .3 Results After Clicking Go These ranges are the intervals for Vmax and k. Identify the responses by highlighting the column that contains the response values and click OK.jmp’s column formula. use Velocity (y). Click the Confidence Limits button to produce confidence intervals. Select DOE > Nonlinear. The formula looks like: Now create a nonlinear design for fitting the model’s nonlinear parameters. For this example. 7. They are asymptotically normal.

Figure 9. 3. .6 Changing the Distribution 5. as shown in Figure 9. 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 9.7 Changing the Values Now you have described your current level of uncertainty of the two parameters.5). those values appear as the high and low values in the Factors control panel.3 to 7 (Figure 9. Type the desired number of runs into the text box and click Make Design to preview the design (Figure 9. the range of data for Concentration goes from 0.5 Changing the Factor Values 4. Figure 9. Change the values for Vmax and k to correspond to those limits. Therefore.4 Choosing Response Columns Note that in Chemical Kinetics.7.jmp (Figure 9.8).25. Change the range to a broader interval: from 0.1).417 to 6. 6. Look back at the analysis report you created above (Figure 9.3) and locate in the Solution table the upper and lower confidence limits for Vmax and k.

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

) For this example.10 Choosing Response Columns 4.3. Select DOE > Nonlinear Design. 5. Change the two factors’ values to be a reasonable range of values. Open Sample Data > Design Experiment> Reaction Kinetics Start. Figure 9. as described above. Open Reaction Kinetics Start. . Identify the response by highlighting the column that contains the response (predictor) values and click OK (Figure 9.174 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. To follow along with this example. 2.10). use the values for Reaction Temperature of 520 and 540. 3.jmp. (This may be an educated guess. but you have a guess for the unknown parameters. The formula looks like: This model was taken from Box and Draper (1987).11). Use values for Reaction Time of 0.1 and 0. 1. It is as a function of time and temperature. use observed yield. For this example. open the Sample Data folder that was installed when you installed JMP. The formula arises from the fractional yield of the intermediate product in a consecutive chemical reaction. Leave each parameter’s distribution as at the default Uniform.jmp. 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.

12 Design Table You have now completed the design without data. Click Make Design. then Make Table.9 Nonlinear Designs—Example of Creating a Nonlinear Design 175 9 Nonlinear Figure 9. Figure 9.11 Changing the Number of Runs 6. Your results should look similar to those in Figure 9.jmp Now examine the design region. Figure 9.13 Reactor Kinetics. .12.

Highlight yield model and click the X. and highlight Reaction Time and click the X button (Figure 9. 11.Click the Go button on the Nonlinear control panel. 12. Predictor Formula button. Create an overlay plot by selecting Graph > Overlay Plot.69% at a reaction temperature of 540 degrees Kelvin and a reaction time of 0. 8.Select Analyze > Modeling > Nonlinear. Response button. 13. Maximize Desirability. Highlight Reaction Temperature and click the Y button. there are no points at low temperature and high time—the lower right corner of the graph. 10. 9. Click OK. 14. In particular.Highlight observed yield and click the Y. . then click OK.14 Create an Overlay Plot Notice that the points are not at the corners of the design region.Now you can choose Profilers >Profiler from the red triangle menu.1126 minutes. The maximum yield is 60.176 9 Nonlinear Designs—Example of Creating a Nonlinear Design 7.14). Figure 9.

9 Nonlinear Designs—Example of Creating a Nonlinear Design 177 9 Nonlinear Figure 9.15 Time and Temperature Settings for Maximum Yield .

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

Click to enter or change factor values. Lognormal.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. Selecting the Number of Runs The Design Generation panel shows the minimum number of runs needed to perform the experiment. Click to edit the distribution: Uniform. Type the desired number of runs into the text box and click Make Design. . Normal. or Exponential.18 Example of Setting Up Factors and Parameters Double-click to edit the factor or parameter name. Figure 9.

180 9 Nonlinear Designs—Creating a Nonlinear Design Figure 9.20 Example Preview Design Making or Augment the Table After previewing the design. click either Make Table or Augment Table. Clicking Make Table creates a new table with all runs included.19 Selecting the Number of Runs Viewing the Design Before creating the design table. . Clicking Augment Table adds the new runs to the existing table. Figure 9. 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.

.

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

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

= – 10 log ⎜ 1 ∑ ----. This becomes the response for a fit across the inner design runs.table found in the Design Experiment folder of the Sample Data. Or. The signal and noise factors for this example appear in the table below.jmp.= – 10 log ⎜ 1 ∑Y 2⎟ n i i ⎠ N ⎝ 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). An inner design constructed over the control factors finds optimum settings. generate the original design table on your own using DOE > Taguchi Arrays. A performance statistic is calculated across the outer runs for each inner run. 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 . To follow along with this example. The experiment is performed on all combinations of the inner and outer design runs.⎟ -⎜ n i Y 2⎟ N i ⎠ ⎝ ⎛ ⎞ S---. open the Sample Data folder that was installed when you first installed JMP. Open the Byrne Taguchi Data. The table below lists the recommended performance statistics.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.= 10 log ⎜ Y -⎟ ----N ⎝ s2 ⎠ ⎛ ⎞ S1--. Goal S/N Ratio Formula ⎛ 2⎞ 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. An outer design over the noise factors looks at how the response behaves for a wide range of noise conditions.

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. Click Continue. Select DOE > Taguchi. 5. The factors panel then shows the four three-level control (signal) factors and three noise factors. 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. Click Make Table to create the design table shown in Figure 10. . as shown in Figure 10. 7. Open the factors table by opening the Sample Data folder that was installed when you first installed JMP. Highlight L9-Taguchi to give the L9 orthogonal array for the inner design. Highlight L8 to give an eight-run outer array design. 2. 6. and Signal and Noise Factors for the Byrne-Taguchi Example 4. The outer design has three two-level factors. Open Design Experiment > Byrne Taguchi Factors. However.jmp. 3. Figure 10.1. Click the red triangle icon on the Taguchi Design title bar and select Load Factors.2. A full factorial in eight runs is generated.1 Response.

2 Taguchi Design Before Data Entry Now suppose the pull-off adhesive force measures are collected and entered into the columns containing missing data.is the column of measurements taken with the three noise factors set at their low levels. which is –10 times the common logarithm of the average squared reciprocal: –10Log10 Mean 1 1 1 1 1 1 1 1 .10 Taguchi Designs—Taguchi Design Example 187 10 T aguchi Figure 10. . Y--. it is the “larger–the–better” (LTB) formula. 1. .3 Complete Taguchi Design Table (Byrne Taguchi Data. To see the completed experiment and continue following this example. 8. Figure 10.+ +2 y+ .2 y .jmp. . The goal of the analysis is to find factor settings that maximize both the mean and the signal-to-noise ratio.. For example. Analyze the Data The data in Byrne Taguchi Data.2 y .2 y +++2 This expression is large when all of the individual y values are small.+2 y ++.2 y+ ... y . Click the red triangle icon next to the Model script and select Run Script as shown in Figure 10. .3 by selecting Help (View on the Macintosh) > Sample Data Directory > Design Experiment > Byrne Taguchi Data.3. The missing data column names are appended with the letter Y before the levels (+ or –) of the noise factors for that run..4.+ 2 y .jmp are now ready to analyze. open the table shown in Figure 10. .+ . . In this case.jmp) The column named SN Ratio Y is the performance statistic computed with the formula shown below. as shown in Figure 10. . .

The two responses are the mean and signal-to-noise ratio over the outer array.5) indicate that different settings of the first three factors would increase SN Ratio Y. The prediction profiler is a quick way to find settings that give the highest signal-to-noise ratio for this experiment. It includes the main effects of the four signal factors.5 The Prediction Profiler .4 Fitting a Prediction Model 2. Figure 10. Click Run Model. Figure 10.188 10 Taguchi Designs—Taguchi Design Example The window that appears is automatically set up with the appropriate effects. The profile traces (Figure 10. 3. Open the Prediction Profiler by clicking the red triangle on the Response Mean Y title bar and selecting Factor Profiling > Profiler.

6).” p.6 Best Factor Settings for Byrne Taguchi Data Creating a Taguchi Design To start a Taguchi design. follow the steps below: • “Detailing the Response and Add Factors.” p.9075. Figure 10.” p. The Depth setting changed from L1 to L3.0253 to 26. 192 . click the red triangle on the Prediction Profiler title bar and select Desirability Functions. These new settings increased the signal-to-noise ratio from 24. which is what you want in this experiment. select DOE > Taguchi Arrays. The default desirability functions are set to larger-is-better. or click the Taguchi Arrays button on the JMP Starter DOE page. 191 • “Making the Table.10 Taguchi Designs—Creating a Taguchi Design 189 10 T aguchi 4. 5. See the chapter “Standard Least Squares: Perspectives on the Estimates” of JMP Statistics and Graphics Guide for more details about the prediction profiler.” p. Then. In this example. There was no change in Adhesive. the settings for Interfer and Wall changed from L1 to L2 (Figure 10. This adds the row of traces and column of function settings to the profiler. 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. 190 • “Choosing Inner and Outer Array Designs. To find optimal settings. 191 • “Displaying and Modify the Design.

Figure 10. 2 Click to change the response goal: Larger Is Better. 3 1 Double-click to edit the response name. Nominal is Best. or Click to add a noise. click and then type the new value. 3 2 Double-click to edit the factor name. or None. For information on importance weights and lower and upper limits. The steps for setting up the details of this response are outlined in Figure 10.” p.190 10 Taguchi Designs—Creating a Taguchi Design Detailing the Response and Add Factors The Responses panel has a single default response. 58. Smaller is Better. then select a signal type: 2 Level. . or 3 Level. see “Understanding Response Importance Weights.8. The steps for setting up the factors are outlined in Figure 10.7.7 Setting Up the Response Click to enter lower and upper limits and importance weights. 1 To change a signal or noise’s value. Figure 10.8 Entering Factors Click to add a signal.

Figure 10. see “Aliasing of Effects.9 shows your choices when you have eight factors. Choosing Inner and Outer Array Designs Depending on the number of factors you have.” p. Figure 10. see “Understanding the Coded Design. For more details on the coded design. . • Coded Design—Shows the pattern of high and low values for the factors in each run. Figure 10.10).10 Display and Modification Options • Aliasing of Effects—Shows the confounding pattern for fractional factorial designs. 95 and “Changing the Coded Design. JMP will ask you to select an inner and outer array design. 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. click Continue. For more details on aliasing of effects.10 Taguchi Designs—Creating a Taguchi Design 191 10 T aguchi When you finish adding factors. Highlight the design(s) you would like to use and click Continue. 96.” p. 93.9 Selecting a Design Using Eight Factors Displaying and Modify the Design After you select a design type.” p.

see “Understanding the Coded Design. For details on codes and how to change them. and “Changing the Coded Design. 95. each row represents a run. In the data table. plus signs designate high levels and minus signs represent low levels. 96.11 Taguchi Design Table . Figure 10. In the values for the Pattern variable.11 appears.” p.192 10 Taguchi Designs—Creating a Taguchi Design Making the Table When you click Make Table.” p. a table similar to that shown in Figure 10.

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. you can master the temptation to assume that one successful screening experiment has optimized your process.11 Augmented Designs 11 Augmented It is best to treat experimentation as an iterative process. That way. . The augment designer modifies an existing design data table. It also presents a case study of design augmentation. You can also avoid disappointment if a screening experiment leaves behind some ambiguities. supporting your iterative process.

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

To begin. In this study. the Augment Design dialog shown in Figure 11. the purpose of this example is to estimate all two-factor interactions in 16 runs. However. which can’t be done when there is the additional blocking factor in the model. However. 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. demonstrates how to use JMP’s augment designer to resolve ambiguities left by a screening design. open the Sample Data folder that was installed when you installed JMP. After you identify the factors and response and click OK. This model is the result of the model stored with the data table when it was created by the Custom designer. Figure 11. Note: You can check Group New Runs into Separate Blocks to add a blocking factor to any design. a chemical engineer investigates the effects of five factors on the percent reaction of a chemical process.1 Augment Design Dialog for the Reactor Example Now click Augment on the Augment Design dialog to see the display in Figure 11. et al. (1996).2.1 appears. adapted from Meyer. the augmented design is to have 16 runs in order to estimate all two-factor interactions.jmp. . Then select Augment Design from the DOE menu.

2 Initial Augmented Model To continue with the augmented reactor design. as shown in the Design Generation text edit box.3. This adds all the two-factor interactions to the model.196 11 Augmented Designs—The Reactor Example: Using D-Optimal Augmentation Figure 11. . choose 2nd from the Interactions popup menu as shown in Figure 11. The Minimum number of runs given for the specified model is 16.

Figure 11. similar to the design shown in Figure 11.4 D-Optimally Augmented Factor Settings Note: The resulting design is a function of an initial random number seed. JMP computes D-optimally augmented factor settings.4.11 Augmented Designs—The Reactor Example: Using D-Optimal Augmentation 197 11 Augmented Figure 11. To reproduce the exact factor settings table in Figure 11.3 Augmented Model with All Two-Factor Interactions When you click Make Design.4. . A dialog shows the most recently used random number. Click OK to use that number again.5 shows the random number (12834729) used to generate the runs in Figure 11. or Cancel to generate a design based on a new random number. (or the most recent design you generated).4. choose Set Random Seed from the popup menu on the Augment Design title bar. The dialog in Figure 11.

suppose you augmented that data table . This table property contains the JSL commands that display the Fit Model dialog with all main effects and two-factor interactions as effects. Figure 11.7.jmp sample data file. Analyze the Augmented Design Suppose you have already run the experiment and recorded results in the Percent Reacted column of the data table.5 Specifying a Random Number Click Make Table. To see these results. Also. These folders were installed when you installed JMP.6 shows the Reactor Augment Data. Figure 11. . run the Model script stored as a table property with the data table by right-clicking the script name and selecting Run. Change the fitting personality from Standard Least Squares to Stepwise.198 11 Augmented Designs—The Reactor Example: Using D-Optimal Augmentation Figure 11.6 Completed Augmented Experiment (Reactor Augment Data. open the Reactor 8 Runs.jmp. found in the Design Experiment folder of the Sample Data . as shown in Figure 11.jmp) To start the analysis.

8. You should see the dialog shown in Figure 11. Then. .11 Augmented Designs—The Reactor Example: Using D-Optimal Augmentation 199 11 Augmented Figure 11.050 and Prob to Leave is 0.7 Fit Model Dialog for Stepwise Regression on Generated Model When you click Run Model. the stepwise regression control panel appears. choose Mixed from the Direction menu and make sure Prob to Enter is 0.100. Click the check boxes for all the main effect terms.

Note that Feed Rate is out of the model while the Catalyst*Temperature.9. Figure 11. 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. shows the result of this example analysis.200 11 Augmented Designs—The Reactor Example: Using D-Optimal Augmentation Figure 11. .

and find the optimal factor settings. Figure 11. click Make Model on the Stepwise control panel to generate this reduced model.11 Augmented Designs—The Reactor Example: Using D-Optimal Augmentation 201 11 Augmented Figure 11.10 New Prediction Model Dialog . as shown in Figure 11.9 Completed Stepwise Model After Stepwise is finished. make predictions.10. You can now fit the reduced model to do additional diagnostic work.

shows that Catalyst has the largest main effect. and Temperature and the low level of Concentration.11.202 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. When you drag the prediction traces for each factor to their extreme settings. indicate a highly significant regression model with no evidence of Lack of Fit. Stir Rate. the significant two-factor interactions are of the same order of magnitude as the main effects.12 Prediction Model Estimates Plot It is desirable to maximize the percent reaction. Figure 11.1 to 81. shows that maximum occurs at the high levels of Catalyst. However.8. 81. had ambiguous results.12. . Figure 11. This is the reason that the initial screening experiment.” p. The prediction profile plot in Figure 11. shown in the chapter “Screening Designs. the estimate of Percent Reacted increases from 65.13.

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

• Add Centerpoints—adds centerpoints. 207. 2. It gives the following five choices: • Replicate—replicates the design a specified number of times. Figure 11. Select DOE > Augment Design.15). See “Adding Axial Points. 208.204 11 Augmented Designs—Creating an Augmented Design Creating an Augmented Design The augment designer modifies an existing design data table. Replicating a Design Replication provides a direct check on the assumption that the error variance is constant. which can have more terms than the original model.jmp. It also reduces the variability of the regression coefficients in the presence of large process or measurement variability.” p.” p. Identify the responses you want to use for the augmented design by highlighting the column name(s) and clicking OK (Figure 11. This example uses Reactor 8 Runs. • Add Axial—adds axial points together with center points to transform a screening design to a response surface design.” p. 210.14) and clicking OK. See “Replicating a Design. 209.” p. which is located in the Sample Data folder’s Design Experiment folder. See “Adding Centerpoints. These folders were installed when you installed JMP.” p. Open a data table that contains a design you’d like to augment.14 Identify Factors 4. Identify the factors you want to use for the augmented design by highlighting the column names (Figure 11. See “Creating a Foldover Design. • Augment—adds runs to the design (augment) using a model. 204. To replicate the design a specified number of times: 1. • Fold Over—creates a foldover design. 3. See “Adding New Runs and Terms. .

11 Augmented Designs—Creating an Augmented Design 205 11 Augmented Figure 11. Figure 11.16).17 Enter the Number of Runs 8.16 Choose an Augmentation Type 7. Click the Replicate button.18.17). Figure 11. entering 2 specifies that you want each run to appear twice in the resulting design. If you want the original runs and the resulting augmented runs to be identified by a blocking factor. as show in Figure 11. For example.15 Identify Responses 5. View the design. . check the box beside Group New Runs into Separate Block (Figure 11. This is the same as one replicate (Figure 11. 6. Enter the number of times you want JMP to perform each run and click OK.

Click Make Table.18 Reactor Data Design Augmented With Two Replicates 9. If desired.19 Prediction Profiler and Surface Plot 10. The design table appears (Figure 11. view the prediction variance profile and the prediction variance surface plots (Figure 11.206 11 Augmented Designs—Creating an Augmented Design Figure 11.19) by clicking the disclosure icon to the left of the associated title bars.20). . Figure 11.

which can be used in a lack-of-fit test. 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. for example. Open a data table that contains a design you’d like to augment. The table below shows two center points appended to the end of the design. check the box beside Group New Runs into Separate Block (Figure 11. Select DOE > Augment Design. If you want the original runs and the resulting augmented runs to be identified by a blocking factor. 5. . 3. 8. Identify the factors you want to use for the augmented design by highlighting the column names (Figure 11.16). Click Make Table. Identify the responses you want to use for the augmented design by highlighting the column name(s) and clicking OK (Figure 11. 2.jmp. Enter the number of centerpoints you want and click OK. To add centerpoints: 1. 4. These folders were installed when you installed JMP. 9.11 Augmented Designs—Creating an Augmented Design 207 11 Augmented Figure 11.15). 7. 6. which is located in the Sample Data folder’s Design Experiment folder. Click the Add Centerpoints button and add two centerpoints.14) and clicking OK. This example uses Reactor 8 Runs.

This example uses Reactor 8 Runs.14) and clicking OK. check the box beside Group New Runs into Separate Block (Figure 11. 5.16). The default. Identify the responses you want to use for the augmented design by highlighting the column name(s) and clicking OK (Figure 11. To create a foldover design: 1.22 folds on all five factors and includes a blocking factor. Click the Fold Over button. 4. if you choose no factors. 6.jmp. If you choose a subset of factors to fold over. Note: Adding centerpoints or replicating the design also generates an additional Block column in the JMP design table. Choose which factors to fold. Identify the factors you want to use for the augmented design by highlighting the column names (Figure 11. 7. which is located in the Sample Data folder’s Design Experiment folder. If you choose a subset of factors to fold over. the remaining factors are replicates of the original runs. To identify the original runs and the resulting augmented runs by a blocking factor. Open a data table that contains a design you’d like to augment. the remaining factors are replicates of the original runs.15). The design data table that results lists the original set of runs as block 1 and the new (foldover) runs are block 2. 8. Select DOE > Augment Design. . The example in Figure 11. These folders were installed when you installed JMP. 2. Click Make Table.208 11 Augmented Designs—Creating an Augmented Design Figure 11. 3. is to fold on all design factors. This is especially useful as a follow-up to saturated or near-saturated fractional factorial or Plackett-Burman designs.21 Design with Two Center Points Added Creating a Foldover Design A foldover design removes the confounding of two-factor interactions and main effects.

7. Open a data table that contains a design you’d like to augment. 6. 2. To do this: 1. . Enter the axial values in units of the factors scaled from –1 to +1. then enter the number of centerpoints you want. Click Add Axial. 3.15). Select DOE > Augment Design.14) and clicking OK.jmp.11 Augmented Designs—Creating an Augmented Design 209 11 Augmented Figure 11. therefore transforming a screening design to a response surface design. 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. Identify the factors you want to use for the augmented design by highlighting the column names (Figure 11.22 Listing of a Foldover Design On All Factors Adding Axial Points You can add axial points together with centerpoints.16). This example uses Reactor 8 Runs. Identify the responses you want to use for the augmented design by highlighting the column name(s) and clicking OK (Figure 11. 5. If you want the original runs and the resulting augmented runs to be identified by a blocking factor. These folders were installed when you installed JMP. the augmented design includes the number of centerpoints specified and constructs two axial points for each variable in the original design. When you click OK.

Figure 11.24 Design Augmented With Two Center and Ten Axial Points center points axial points Adding New Runs and Terms Using the augment designer. suppose you start with a two-factor. D-optimal augmentation is a powerful tool for sequential design. Click Make Table. which optimally blocks the second set with respect to the first. Adding runs to a design is powerful. To add new runs and terms to the original model: . Figure 11.24 shows a table augmented with two center points and two axial points for five variables. you can add runs using a model that can have more terms than the original model. If you add quadratic terms to the model and five new points. You can also group the two sets of experimental runs into separate blocks. For example.210 11 Augmented Designs—Creating an Augmented Design Figure 11.23 Entering Axial Values 8. four-run design. The design table appears. two-level. Using this feature you can add terms to the original model and find optimal new test runs with respect to this expanded model. 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. JMP generates the 3 by 3 full factorial as the optimal augmented design.

jmp. Select DOE > Augment Design. 2.11 Augmented Designs—Creating an Augmented Design 211 11 Augmented 1. 5. 6. 3. The original number of runs (Figure 11. Enter the number of total runs you want this design to contain.25 Viewing the Existing Design 7. check the box beside Group New Runs into Separate Block.25) appear in the Factor Design panel. Click the Augment button. Figure 11. which is located in the Sample Data folder’s Design Experiment folder. This number is the original number . 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. If you want the original runs and the resulting augmented runs to be identified by a blocking factor. 4. Open a data table that contains a design you’d like to augment. These folders were installed when you installed JMP. Identify the factors you want to use for the augmented design by highlighting the column names and clicking OK.

10. Figure 11. view the prediction variance profile and the prediction variance surface. The augmented design table (Figure 11. .27) appears with the additional runs. If desired.Click the Make Table button.26) appear in the Design panel. Click the Make Design button. 8.26 24 Total Runs 9. The new number of runs (Figure 11.212 11 Augmented Designs—Creating an Augmented Design of runs plus the additional runs.

The following sections describe commands found in this menu that are specific to augment designs. they are available in all designs and more information is in “Special Custom Design Commands. 66.” p. the window in Figure 11. Click the red triangle icon on the Augment Design title bar to reveal a list of commands.11 Augmented Designs—Special Augment Design Commands 213 11 Augmented Figure 11.27 The Augmented Design Table with New Runs additional runs Special Augment Design Commands After you select DOE > Augment Design and identify factors and responses.28 appears. . Most of these commands are for saving and loading information about variables.

Optimality) .or I. Modifying the Design Criterion (D. The script is saved as a table property called Design Matrix.28) and select Save X Matrix.214 11 Augmented Designs—Special Augment Design Commands Figure 11.28 Click the Red Triangle Icon to Reveal Commands Viewing Diagnostics Reports To display a table with relative D. When you run this script. click the red triangle icon in the Augment Design title bar and select Show Diagnostics before clicking the Make Design button. If you do not have the log visible. select View > Log. 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 (Figure 11. G and A efficiencies in augmented designs. JMP creates a global matrix called X and displays its number of rows in the log.

” p. see “Why Change the Number of Starts?. Enter the appropriate value and click OK. You can disallow any combination of levels of categorical factors if you have not already defined linear inequality constraints. 71. then choose Make D-Optimal Design or Make I-Optimal Design. see “Accounting for Factor Level Restrictions.0 with any positive number. To edit the sphere radius for the design in units of the coded factors (–1. you can define general factor constraints on the factors. the window in Figure 11. For information on how to do this.” p.29 Changing the Number of Starts For additional information on the number of starts. When you select this command. The number you enter overrides the default number of starts. click the red triangle icon in the Augment Design title bar (Figure 11. Specifying the Sphere Radius Value Augment designs can be constrained to a hypersphere. click the red triangle icon in the Augment Design title bar (Figure 11. submit the following command before you build a custom design: DOE Sphere Radius = 1. which varies depending on the design. .0. In this statement you can replace 1. 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.29 appears with an edit box for you to enter the number of random starts for the design you want to build.28) and select Sphere Radius. using JSL. 73.28) and select Number of Starts. Selecting the Number of Random Starts To override the default number of random starts. Figure 11.28) and select Optimality Criterion. click the red triangle icon in the Augment Design title bar (Figure 11. 1).11 Augmented Designs—Special Augment Design Commands 215 11 Augmented To modify the design criterion. Disallowing Factor Combinations In addition to linear inequality constraints on continuous factors and constraining a design to a hypersphere. Or.

216 11 Augmented Designs—Special Augment Design Commands .

given my proposed sample size. JMP computes the third. “Will I detect the group differences I am looking for. the result is a plot of the other two. given that the true effect size is at least a certain size. 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). two-sample.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. and power. sample size. and k-sample situations. . Using the Sample Size and Power command when doing a prospective analysis helps answer the question. power. The calculations assume that there are equal numbers of units in each group. You can apply this platform to more general experimental designs. and the magnitude of the effect. 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. These depend on the significance level—alpha—of the hypothesis test for the effect and the standard deviation of the noise in the response. if they are balanced. and a number-of-parameters adjustment is specified. You can supply either one or two of the three values. difference to detect. If you supply only one of these values. It requires that you enter any two of three quantities. 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. This capability is available for the single sample. and alpha level?” In this type of analysis. If you supply two values.

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

or samples) are involved in the experiment.1 Sample Size and Power Choices . the panel shown in Figure 12. • Effect Size Effect size is how different the means are from each other or from the hypothesized value. power expected. 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. The following sections describe each of these selections and explains how to enter estimated parameter values and the desired computation. a two sample comparison. When you select DOE > Sample Size and Power. • Error Standard Deviation Error Standard Deviation is the unexplained random variation around the 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. or when comparing k sample means. • Sample Size Sample size is how many experimental units (runs. • Power Power is the probability of declaring a significant result.1 appears with button selections for experimental situations. Figure 12.

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

as shown in the left-hand dialog in Figure 12. which you need to adjust to real units.3. 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 0.90 or 0. Sample size is not the number per group. Power is equal to alpha when the specified effect size is zero. The dialog on the right in Figure 12.99998. See the section. but the total over all groups. or samples). when there is one. • Power Power is the probability of getting a statistic that will be declared statistically significant. If an experiment requires considerable effort. rounding to 1. and three parameters for the interactions. 223. For single sample problems this is the difference between the hypothesized value and the true value. • Sample Size Sample size is the total number of observations (runs. Bigger power is better. • Back Goes back to the previous dialog. To calculate the power when the sample size is 10. but the cost is higher in sample size.12 Prospective Power and Sample Size—One-Sample and Two-Sample Means 221 12 Power five—two parameters for the extra main effects.9. • Difference to Detect Difference to detect is the smallest detectable difference (how small a difference you want to be able to declare statistically significant). experimental units. .95 if you can afford it. • Continue Evaluates at the entered values.” p. • Animation Script The Animation Script button runs a JSL script that displays an interactive plot showing power or sample size. This is usually done by increasing the estimated sample size to the smallest number evenly divisible by the number of groups. leave Power missing in the dialog and click Continue.3. for an illustration of this animation script. You should go for powers of at least 0. 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. Computed sample size numbers can have fractional values. In practice. “Power and Sample Size Animation for a Single Sample. shows the power is calculated to be 0. plan so that the experimental design has the power to detect a sizable effect.

222 12 Prospective Power and Sample Size—One-Sample and Two-Sample Means Figure 12.4 A One-Sample Example Plot When only Sample Size is specified (Figure 12.95. Change the range of the curve by changing the range of the horizontal axis. For example. a plot of power by difference appears. the plot on the right in Figure 12. has the horizontal axis scaled from 1 to 8. Figure 12.4. The left-hand graph in Figure 12. . shows a range of sample sizes for which the power varies from about 0. which gives a more typical looking power curve. leave both Sample Size and Power empty and click Continue.4.3 A One-Sample Example To see a plot of the relationship of power and sample size.5) and Difference to Detect and Power are empty. Double click on the horizontal axis to get any desired scale.2 to 0.

often represented as β in literature. Alpha is 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.05. .6.4. and the Difference to Detect is set to 0. You can drag the handles over the curves drag to see how their positions affect power. In the example shown in Figure 12.5 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. The animation begins showing a normal curve positioned with mean at zero (representing the estimated mean and the true mean).12 Prospective Power and Sample Size—One-Sample and Two-Sample Means 223 12 Power Figure 12.4 (the difference to be detected). The probability of committing a Type II error (not detecting a difference when there is a difference). and another with mean at 0. Sample Size is 10. Also.

To find out how large. 0. Suppose the error standard deviation is 0.6 Example of Animation Script to Illustrate Power Two-Sample Means The dialogs work similarly for two samples. the desired detectable difference is 1.7. as shown in the dialog on the left in Figure 12. shown on the right in Figure 12.9.224 12 Prospective Power and Sample Size—One-Sample and Two-Sample Means Figure 12. This is considerably lower than in the single sample because each mean has only half the sample size.5433.7. . The crosshair tool estimates that a sample size of about 35 is needed to obtain a power of 0. the Difference to Detect is the difference between two means. and the sample size is 16. To increase the power requires a larger sample. Leave Power blank and click Continue to see the power calculation. leave both Sample Size and Power blank and examine the resulting plot. The comparison is between two random samples instead of one.9 (as before).

and the Error Std Dev is 0.12 Prospective Power and Sample Size—k-Sample Means 225 12 Power Figure 12. the power and sample size calculations produce the power curve shown on the right in Figure 12. In this case it is the square root of (–1.9. The next example considers a situation where 4 levels of means are expected to be about 10 to 13.5)2.7 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. as shown in the dialog on the left in Figure 12.5)2+ (–0.8. If both Sample Size and Power are left blank.5)2+(0.236. calculated as square root of the sum of squared deviations from the grand mean. When a sample size of 16 is entered the power calculation is 0. This confirms that a sample size of 16 looks acceptable. which is the square root of 5. .5)2+(1.8.95. Notice that the difference in means is 2.

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.nist. entered as the Baseline Variance. The computations then show whether the true variance is larger or smaller than its hypothesized value. 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.1) determines sample size for detection of a change in variance. In the dialog.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. you enter two of the items and the Power and Sample Size calculations determines the third.226 12 Prospective Power and Sample Size—One-Sample Variance Figure 12.gov/div898/handbook. select either Larger or Smaller from the Guarding a change menu. To indicate direction of change. alpha level. . and direction of change you want to detect. You can access the NIST manual examples at http://www. As with previous dialogs.itl. The examples throughout the rest of this chapter use engineering examples from the online manual of The National Institute of Standards and Technology (NIST). specify a baseline variance.

as indicated on the dialogs (Figure 12. enter a negative amount in the Difference to Detect box.10).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. When there are two samples. Note: Remember to enter the variance in the Baseline Variance box.9. but the computations to determine sample size and test proportions use a normal approximation. When you click Continue.99. The dialogs are the same except you enter Baseline Proportion and also specify either a one-sided or two-sided test. Enter these items as shown in Figure 12. Figure 12. If you want to detect a change to a small variance. the Baseline Proportion is the average of a known baseline proportion and the single sample proportion. the computed result shows that you need a sample size of 170. For a one-sample situation.05 and power of 0. The sampling distribution for proportions is actually binomial. the Baseline Proportion you enter is the average of the two sample proportions. with an alpha of 0. .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. not the standard deviation.

and the computed sample size of approximately 77. leave both Difference to Detect and Sample Size blank. or average of two-sample proportions. Use the grabber tool (hand) to move the x-axis and show a specific range of differences and sample sizes. 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. Figure 12.The manager does not want to stop the process unless it has degenerated greater than 16% defects (6% above the 10% known baseline defective).10 Default Power and Sample Dialog for One-Sample and Two-Sample Proportions Enter average of baseline and one-sample proportions. Figure 12. which is the average of the baseline (10%) and and the proportion above the baseline (6%).11.228 12 Prospective Power and Sample Size—One-Sample and Two-Sample Proportions 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. The Baseline Proportion in this example is 0. 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. Enter 1 or 2 to indicate the type of test (one. where proportion of defects is part of process control monitoring.11 Dialog To Compare One Proportion to a Baseline and Sample Size Plot .08. To see the plot on the right in Figure 12.or two-sided) Testing proportions is useful in production lines.

Enter the Difference to Detect in terms of the baseline count per unit (defects per unit).128 (Figure 12. The computed sample size is expressed in those units. A unit can be an area and the counts can be fractions or large numbers. 1. The test is for one-sided (one-tailed) change. alpha is sometimes called the producer’s risk and beta is called the consumer’s risk. The process meets the target if there are less than 48 defects (6 defects per wafer in a sample of 8 wafers). Then enter two of the remaining fields to see the calculation of the third. 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.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. 2.12). As an example. Figure 12.9. In this kind of situation. which is the chance of detecting a change larger than 2 (6 defects per wafer). Click the Counts per Unit button. Although the number of defects observed in an area of a given size is often assumed to have a Poisson distribution. Enter an alpha of 0. Select DOE > Sample Size and Power. 3. 5. the area and count are assumed to be large enough to support a normal approximation. Enter a power of 0.1 to be the chance of failing the test if the new process is as good as the target. 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. Click Continue to see the computed sample size of 8.12 Dialog For Counts Per Unit Example . 4.

000. Enter 50 for the number of defects and 1. enter 6 as the Sigma Quality Level and leave the Number of Defects blank (window to the left in Figure 12.000 opportunities.13. are a Sigma quality level of 5. 4.4. Click Continue. use the Sample Size and Power feature to compute the Sigma quality level for 50 defects in 1. The results. by entering any two of the following three quantities: • number of defects • number of opportunities • Sigma quality level When you click Continue. the sigma quality calculator computes the missing quantity using the formula Sigma Quality Level = NormalQuantile(1 – defects/opportunities) + 1. as shown in the window on the right in Figure 12.230 12 Prospective Power and Sample Size—Sigma Quality Level Sigma Quality Level Use the Sigma Quality Level feature. Figure 12. .13 Sigma Quality Level Example 1 If you want to know how many defects reduce the Sigma Quality Level to “six-sigma” for 1.14) shows that the Number of Defects cannot be more than approximately 3. Click the Sigma Quality Level button.000 as the number of opportunities. 2.5. 3.14).000.000 opportunities: 1. accessed by selecting DOE > Sample Size and Power.13.000. As an example. as shown in the window to the left in Figure 12.3. Select DOE > Sample Size and Power. The computation (window to the right in Figure 12.

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

.

p131-145. (1990).X. (1980). Box. W. K. Box. WI: American Society of Quality Control.R. Inc. “An Analysis of Unreplicated Fractional Factorials. Revised Edition. (1978). (1960). “On the Experimental Attainment of Optimum Conditions. Hunter. Bose.W.. Inc. Lucas. (1957). G. . R.P. 1.E. 12:4.M. 1980. J. (1947) “Mathematical Theory of the Symmetrical Factorial Design” Sankhya: The Indian Journal of Statistics. G. Empirical Model–Building and Response Surfaces. 214–219. G.M. Series B. Fitting Equations to Data.” International Statistical Review. “Letter to the Editor: Response to James M. and Transformations. F. G. and Taguchi. 13. Box. 168–177.G. G. D. and Hunter.S. Oct.” Technometrics 32. New York: John Wiley and Sons. Daniel C. C.P. R.P. Performance Criteria. Derringer.” Journal of the Royal Statistical Society. “Some New Three-Level Designs for the Study of Quantitative Variables.E..P. and Nachtsheim. C. and Wood. G. and Behnken. 11–18. and Donev.C. 455–475.148. “Use of Half–normal Plots in Interpreting Factorial Two–level Experiments. “Simultaneous Optimization of Several Response Variables. R. N. D. 363-364. and Suich. (1988). J. Cornell. New York: John Wiley and Sons.. D. 1-45. and Cox. Sun. C. D. New York: John Wiley and Sons. A. and Meyer.J.P. Experimental Designs. International Statistical Institute. G. “Signal–to–Noise Ratio. pp. (1986).E. Second Edition New York: John Wiley & Sons.A. 1–40. Box. and Wu.” Journal of Quality Technology.P. (1986). Byrne. Vol. Optimum Experimental Designs Clarendon Press.F.G. (1951). (1990). 1..D. New York: John Wiley and Sons.. Second Edition. “A Catalogue of Two-level and Three-Level Fractional Factorial Designs with Small Runs. Chen. (1993). Statistics for Experimenters. Cook.” Technometrics. Cochran. Box. Daniel. C. J. Oxford (1992) p.B. 311–314.” Technometrics 2. and Draper. Part 2. G. 8.E. (1987). Milwaukee. R. 107-166. (1980).E.E.” Technometrics 28.D.” Technometrics 30. (1959). A.W.References References Atkinson. N.J. ASQC 40th Anniversary Quality Control Congress Transactions. Experiments with Mixtures. Box. 61. and Wilson.

C. Jones. 33. Meyer. G. (1976). M. 8. New York: Marcel Dekker. and Burman. (1990)..J.D. “Letter to the Editor: Comments on Cook and Nachtsheim (1989).V. New York: Marcel Dekker. . The Coordinate Exchange Algorithm for Constructing Exact Optimal Designs.H. (1983). 16:2. Response Surface Methodology. 37–47. R..M. Johnson. R. 363–364. R. 469–473. pp. St. “Quick and Easy Analysis of Unreplicated Fractional Factorials. J. and Cornell. (1994).A. “Some Guidelines for Constructing Exact D–Optimal Designs on Convex Design Spaces.C.D.” Technometrics.G. 28-31. B. New York: Springer Haaland. W. Lenth. and Box.D. pp. Vol 37. Meyer.” Technometrics.” Technometrics.A. Inc.K. (1995).” The Indian Journal of Statistics. “Bayesian Design and Analysis of Computer Experiments: Use of Derivatives in Surface Prediction. P. A. 305–325.C. “A New Class of Supersaturated Design. Cornell. John. P. and Nachtsheim. “An Interactive Graph For Exploring Multidimensional Response Surfaces. J. Mahalanobis. “A Simple Bayesian Modification of D-Optimal Designs to Reduce Dependence on an Assumed Model. (1989). 38:4. D.F. Mitchell. K.” 1991 Joint Statistical Meetings. I. (1947). 60-69.J. N.L. (1987) Response Surfaces: Design and Analysis. and Jones. 36. 35. 17 pp 15-23.” Technometrics. (1976) Response Surface Methodology. M.W. J. P. John. R. G. Virginia Polytechnic and State University. J.M. Follow-up Designs to Resolve Confounding in Multifactor Experiments. (1993). Myers. and Nachtsheim. Boston: Allyn and Bacon.. 271–277. (1993). Part 2. Meeker.(1996). New York: Macmillan Publishing Company. “Programs for Generating Extreme Vertices and Centroids of Linearly Constrained Experimental Regions. Technometrics. (1988).” Technometrics 44: 72-79. “An algorithm for the Construction of D-Optimal Experimental Designs.” Technometrics 25. Steinberg. Hahn.” Journal of Quality Technology 20:2. D. Lucas. and Vining.I. “Sankhya. #3. Myers. Piepel. G.P. 31.” Journal of Quality Technology.” Technometrics.H. Inc. 140-157. Bradley (1991). Experimental Design in Biotechnology. (1989).” Technometrics.” Technometrics 35:2. Statistical Design and Analysis of Experiments. (1988). p307. 125-139.. Plackett. (2002) “Split Plot Designs and Estimation Methods for Mixture Experiments with Process Variables. April. P. (1972).” Technometrics.R.. pp. R. Goos. and Ylvisaker.. Mitchell. S. Kowalski.Q. (1947). R. T. Vol. J. (1974). 32. (2002) The Optimal Design of Blocked and Split-Plot Experiments.234 References DuMouched. Georgia Khuri.M. Vol 8.J.. Morris. and Feder.. “D-Optimality for Regression Designs: A Review.203-210. C.J. Lin. and Draper. R. Atlanta. G. “The Design of Optimum Multifactorial Experiments.E..M. W. P. T. “The Evaluation and Comparison of Experimental Designs for Fitting Regression Relationships.. D.” Biometrika. J. (1975). 243-255.

303–326. “Extreme Vertices Designs for Linear Mixture Models. “Computer-Aided Design of Experiments for Response Estimation.” Technometrics. 4 October 1985 p.” Technometrics.D. Japan: Central Japan Quality Control Association. Ronald D. JRSS B 20. (1975). Taguchi. Snee. D.J. “An Introduction to Quality Control. (1979).D. Snee. 16. “Experimental Designs for Quadratic Models in Constrained Mixture Spaces. 344-360. H. (1958) Experiments with Mixtures. 391–408. Snee. G. 26.” Technometrics.W. Vol 17. Snee.231. Statistics.Some Practical Experiences. (1984). “Experimental Designs for Mixture Systems with Multicomponent Constraints. R. (1974). R.” Nagoya.D. Welch. .” Commun. R. A8(4).References 235 References Sheffé. (1985)Computer Aided Design of Experiments . W. No. (1976). Journal of Quality Technology. 17:2. 149–159. 217–224. and Marquardt.

.

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

90 contour plots 143 control factors 183. 88 desirability values. 215 E effect aliasing 94 eigenvalue 105 eigenvector 105 . defining 57 how they work 80 models. 214 resolutions 91 roles 67 Design of Experiments. 111 central composite 101 foldover 208 fractional factorials 90 full factorial 81. 62 examples 15.jmp 42 D-optimal augmentation 210 designs 29. describing 61 screening 17 D defaults number of random starts 71 defects 229 D-efficiencies 41 describing models 61. 52–53 optimality criteria 70. 215 entering 68 linear 138 saving 68 continuous factors 60. 103.238 Index adding 135 disallowing combinations 73. 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. 117. as column properties 77 determinant formula 72 diagnostics for custom designs 69. 130 design matrix table properties 70. 214 Difference to Detect option 220–221. 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. 215 Cubic Model. 189 maximizing 106 traces 106 values 58. 53 Box-Behnken 101. See DOE 76 design roles 78 designers augment 204 designs ABCD 139 aberration 91 augment 204 balanced 62 Bayesian D-optimal 29. 26 factors. 224 disallowed combinations 73.jsl 35 custom designs data tables 65 Design Generation panel 7. 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. 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. 97. 215 F factor column 79–80 for mixture experiments 76 factor combinations. 90 continuous 60. G. 123. 113. and A 69. 117. 131. 90 control factors 183. 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. 107. 126. 59 categorical 60. 185 Inscribe option 112 interactions 92 high-order 91 intercepts hidden 142 I-optimal designs 52 optimality criteria 70. 117. disallowing 73 Factor Profiling option 105. 188 factorial designs fractionals 90 full 81. 90. 185 key factors 81 nonmixture 42 tables. entering 77 importance weight 67 inner arrays. inner designs 183. 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. 190 importance values. 135. 88 goals matching targets 58. 139 k-Sample Means (power and sample size) 225 L L18 Chakravarty 92 L18 Hunter 92 L18 John 92 L36 92 .Index 239 size 219 sparsity 81. 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. 88 goals for DOE (Design of Experiments) response variables 77 Group New Runs into Separates Block 207 Index H hidden intercepts 142 hyperspheres 72. 145. 125 three level 92 factors adding 39. 215 I identifying key factors 81 importance of responses 57. 92 effects nonestimable 91 orthogonal 112. 90. 89 minimizing and maximizing 58. 125 Design Generation panel 62 examples 119 functions desirability 106. 183 efficiencies D.

214 matrix. 114 larger–the–better formulas (LTB) 187 Latin Hypercube space filling design 149. 97. 88 maximizing desirability 106 goals 58. 215 M matching targets 77 matching target goals 57–58. 139 orthogonal array designs 92. 114 models custom design 61 describing 61. 133 with nonmixture factors 42 mixture. 126 patterns confounding 94 performance statistics 185 Plackett-Burman designs 91 Plastifactors. 138 local optimum 71 no-intercept model 76 noise factors 183. adding 59 N Responses button 57. 88 minimizing goals 58. 88–89 matrix 70. extra 220 Pattern column 84.240 Index L9-Taguchi 186 Label column 98. 135. 88 means one and two sample 220 methods CONAEV 139 minimize responses 57. 113. column properties 76 Model script Model Specification dialog 98. 215 Optimality Criterion 214 optimality criterion 70 order of runs 65. 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. outer designs 183. 104. 185 nonestimable effects 91 nonlinear design 178 nonmixture factors 42 number of runs 15. 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. alias 64 maximize responses 57. 131. 126.jmp 143 plots Actual-by-Predicted 99 contour 143 spinning 109 ternary 127. 62 screening designs 91 number of starts 71. 185 P parameters. 154–157 limits. 98. 183 orthogonal designs screening designs 90 screening experiments 127 surface designs 112 Orthogonal option 112 outer arrays. 140 points axial 101 center See center points centroid 135 Poisson-distributed counts 229 potential terms (DOE) 29 power N N factors. 88 . responses 77 linear constraints 135. 114.

88 goals 57. 123 robust engineering 183 roles design roles 78 roles. 113. 88 saving 67 upper limits 58. 139 randomizing runs 65. 126 rescaling designs 112 resolution numbers 91 resolutions of designs 91 response limits.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. 97. design 78 Rotatable option 112 RSM (Response Surface Methodology) 110 runs additional 65. 110 desirability values 58.jmp 198 Reactor Factors. 97. 139 requesting additional 65. 37 properties. 136. 135. 131. 135. desirability functions 106 lower limits 58. 97. 113. 131. 215 random starts 71 Randomize within Blocks 65. 88 goals. 37 traces 106 variances 33. column property 77 response surface designs examples 103. 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.jmp 119–120 Reactor Response. 126. 126 order they appear in table 65. 97. multiple 77 RMSE 99. 88. 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. 97. 97. 113.jmp 195 Reactor Augment Data. 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- . 113. columns 75 proportions (power and sample size) 227 pseudocomponent (mixture column property) 77 pure error 207 Q quadratic model 32–64 R radius. 113. 139 range constraints 135 Reaction Kinetics Start. 113.jmp 174 Reactor 32 Runs. 126. 88 responses. sphere 72.jmp 119 Reactor 8 Runs. 131.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.

92 sphere packing Borehole problem 161 methods 151 space filling designs 149 sphere radius 72. 139 Sort Right to Left 65. 126. number of 71. effect 81. 113. 215 spinning plots Box-Behnken designs 109 Split plot designs 60 split plot designs 65 standard deviation. creating 68 making in custom designs 65 Taguchi designs 183–189 description 183 methods 183 target value 77 target values 58. 89 ternary plots 127. 133 single-sample means (power and sample sizes) 221 solutions equivalent 80 Sort Left to Right 65. 135. random 71 statistics. 113. 224 two-sample proportion (power and sample size) 227 U Uniform (space filling design) 149. 97. 157 uniform precision designs 112 User Defined option 112 V values high and low for columns (in DOE) 76 target 58. error 219 star points 101 starts. 139 space-filling designs 149–165 simplex lattice 133 simplex centroids 130 sparsity. 215 starts. 97. 89 variance error and prediction 33 . 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. 140 traces desirability 106 trade-off in screening designs 91 tutorial examples augment designs 195–203 custom designs 15. performance 185 subsets. 136. 131. 26 DOE 83 full factorial designs 119 mixture designs 143–145 response surface designs 103. finding 136 supersaturated designs 22 surface designs See response surface designs T tables factors table. 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. 131.242 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. 126.

190 weight.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. saving 70. 88. importance 57.

- Nitnem in Gurmukhi (Uni)
- Nitnem in Gurmukhi (Uni).pdf
- 2013 - General Income Tax and Benefit Guide
- MY 64 Multimeter Manual
- MY 64 Multimeter Manual
- SukhaasanSeva
- ’07Camry_specs
- By.giani.sant.Singh.maskin
- 52533824 Okuma Lathe Manual
- Sant Isher Singh Ji Maharaj, Sant Kishan Singh Ji Maharaj - Bhai Daya Singh Ji's Online Library - Www.ik13
- Rahitnama Bhai Chaupa Singh Ji - With Commentary by Amrit Pal Singh Amrit
- Rahitnama Bhai Chaupa Singh Ji - With Commentary by Amrit Pal Singh Amrit
- Soft Skills
- ifsc code
- Powerful English Speaking
- 268 Sinumerik 810D 840D 840Di ISO Dialect
- Tata Photon Plus Prepaid

- NCSS 3
- Discovering Jmp
- Stata - Outreg command
- Design of Experiments
- excel-hm3
- 9783642395116-c1
- Design of Experiments
- MEDINA OLIVA_Résumé afis
- Book
- Hlm Comparison-1 Copy
- SAS JMP Modeling Multivariate
- Ali
- Design Experiments Applications i to 13
- pert & cpm
- Design Of Experiments
- Design of Experiments via Taguchi Methods Orthogonal Arrays
- Design of Experiments Tutorial
- IntroToArena.3 (1)
- Pukelsheim Optimal DoE
- SimulationToolsGuide
- Design of Experiments With MINITAB Escrito Por Paul G. Mathews
- Network Analysis [Compatibility Mode]
- Introduction to Design of Experiments
- 10.1.1.66.3176
- Design of Experiments1
- REGRESSION ANALYSIS
- Design of Experiments Guide
- ES - 92 Project Management
- PracticalData AnalysiswithJMP

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