Professional Documents
Culture Documents
MODDE
By Umetrics Version 9
MKS Umetrics AB Stortorget 21 SE-211 34 Malm Sweden Phone: +46 (0)40 664 2580 Email: info@umetrics.com
Welcome
Welcome to the user guide for MODDE 9. This is your guide to MODDE and its capabilities. Assumed prior knowledge We assume that you are already familiar with Windows. You should be familiar with the topics discussed in your Microsoft Windows User's Guide, including how to use the mouse select, click, shift-click, control-click, press, drag, and choose from a menu. You should also know how to search through directories to find files. Content This user guide is divided into 21 chapters. Chapter 1 gives a short introduction of how to use MODDE. Chapter 2 presents an introduction to experimental design. Chapter 3 is an overview of the program. Chapters 4 - 20 provide step-by-step procedures for creating and using experimental designs with MODDE. Chapter 21 describes how the Help function works. Appendix A, "Statistical notes" presents short explanations of statistical methods used by MODDE. Appendix B, Designs presents short descriptions of the designs available in MODDE.
Appendix C: "Optimizer" describes the optimizer feature and the properties of the different optimizer objectives. Appendix D: "Design Space" describes the design space estimation feature. References are available on the references page.
Table of Contents
How to get started with MODDE 1
Installation ..................................................................................................................... 1 Starting MODDE ........................................................................................................... 1 Experimental cycle ........................................................................................................ 1 Design phase .................................................................................................................. 2
Defining factors................................................................................................................... 2 Defining responses .............................................................................................................. 2 Defining objective ............................................................................................................... 2 Explore the data (Worksheet menu) .................................................................................... 2 Evaluate the design ............................................................................................................. 3 Fit ........................................................................................................................................ 3 Review the fit using plots and lists ...................................................................................... 3 Diagnostics .......................................................................................................................... 3 Interpret the model .............................................................................................................. 3 Refine the model ................................................................................................................. 4
General description ........................................................................................................ 5 What is modeling and experimental design? ................................................................. 5 Objectives of modeling and experimental design .......................................................... 5 Screening models and designs ....................................................................................... 5 Response surface modeling (RSM) designs................................................................... 6 Fit methods .................................................................................................................... 7
Multiple Linear Regression (MLR) ..................................................................................... 7 Partial Least Squares (PLS)................................................................................................. 7 Results ................................................................................................................................. 8 Review the model fit ........................................................................................................... 9 Assess model adequacy ....................................................................................................... 9 Limitations in investigation names.................................................................................... 10 Limitations in factor and response names ......................................................................... 10 Case sensitivity ................................................................................................................. 10 Menu reference syntax ...................................................................................................... 10 Number of factors in RSM designs ..................................................................................... 6 Number of factors in Split objective ................................................................................... 7 Number of factors in screening designs .............................................................................. 6 Number of factors with split objective ................................................................................ 6
Analysis phase ............................................................................................................... 9 Prediction - using the fitted model ................................................................................. 9 Convention................................................................................................................... 10
Overview
11
File
17
Traditional designs ............................................................................................................ 18 Advanced designs.............................................................................................................. 18 Import design from file ..................................................................................................... 21
Introduction.................................................................................................................. 17 New.............................................................................................................................. 17
Open............................................................................................................................. 21 Save and save as .......................................................................................................... 21 Delete investigation ..................................................................................................... 21 Save plot/list as ............................................................................................................ 21 Revert........................................................................................................................... 22 Send by E-Mail ............................................................................................................ 22 Generate HTML report ................................................................................................ 22 Complement design ..................................................................................................... 23
Available complementing methods ................................................................................... 24 Fold over ........................................................................................................................... 24 Estimate squares of selected factors in factorial designs ................................................... 24 D-Optimal ......................................................................................................................... 25 Complement Doehlert ....................................................................................................... 27 Screening to RSM Rechtschaffner .................................................................................... 28 Plackett Burman Super-Saturated to Plackett Burman ...................................................... 28 Encrypt/password protect an investigation ........................................................................ 29 Lock investigation ............................................................................................................. 29
Protect investigation .................................................................................................... 29 Print Setup ................................................................................................................... 30 Print Preview ............................................................................................................... 30 Print ............................................................................................................................. 30 Exit............................................................................................................................... 30
Print format for plots ......................................................................................................... 30
Edit
31
Introduction.................................................................................................................. 31 Undo ............................................................................................................................ 31 Cut, copy, paste, delete, select all ................................................................................ 31
ii
Table of Contents
Add factor, add response, add experiment, insert rows ............................................... 32 Sort............................................................................................................................... 32 Model / reference mixture............................................................................................ 34 Generators .................................................................................................................... 37 Edit the model and/or generators before creating the worksheet ................................. 38
Model ................................................................................................................................ 34 Reference mixture ............................................................................................................. 36 Sorting the candidate set ................................................................................................... 33
View
39
Standard toolbar ................................................................................................................ 39 Spreadsheet toolbar ........................................................................................................... 41 Plot toolbar ........................................................................................................................ 41 Model toolbar .................................................................................................................... 44 Window toolbar................................................................................................................. 44 Status bar........................................................................................................................... 44 Analysis advisor ................................................................................................................ 45 Audit trail .......................................................................................................................... 45 Favorites............................................................................................................................ 45 Output ............................................................................................................................... 50 Notes ................................................................................................................................. 50
Dockable windows....................................................................................................... 44
Add to Favorites .......................................................................................................... 51 Add to Report .............................................................................................................. 51 Full Screen ................................................................................................................... 51 Changing the default options using Investigation Options .......................................... 52
Plot labels .......................................................................................................................... 52 Number format .................................................................................................................. 52 Audit trail .......................................................................................................................... 53 Alpha level ........................................................................................................................ 54 Coefficients ....................................................................................................................... 55 Blocking ............................................................................................................................ 56 Confidence level ............................................................................................................... 56 List presentation ................................................................................................................ 57 R2 in plots ......................................................................................................................... 57 Residuals ........................................................................................................................... 58 Select Factor...................................................................................................................... 58 General page ..................................................................................................................... 59 List Options ....................................................................................................................... 60 Restore .............................................................................................................................. 61 Customize toolbars ............................................................................................................ 61 Customize commands ....................................................................................................... 61 Customize options ............................................................................................................. 62 Customize options ............................................................................................................. 62
Customize .................................................................................................................... 61
Design
63
Introduction.................................................................................................................. 63
iii
Factors ......................................................................................................................... 63 Responses .................................................................................................................... 63 Constraints ................................................................................................................... 63 Inclusions ..................................................................................................................... 64 Objective ...................................................................................................................... 64 D-Optimal .................................................................................................................... 64 Design wizard .............................................................................................................. 64
Factors
65
Responses
75
Linked responses ......................................................................................................... 82 Response definition spreadsheet .................................................................................. 83 Response manipulations in short ................................................................................. 84
iv
Table of Contents
Adding a response ............................................................................................................. 84 Modifying a response ........................................................................................................ 84 Deleting a response ........................................................................................................... 84 Copying and pasting a response ........................................................................................ 84
85
Constraints ................................................................................................................... 85
Specifying constraints ....................................................................................................... 85 Constraints supported ........................................................................................................ 86 Defining constraints in the spreadsheet ............................................................................. 86 Defining a constraint graphically ...................................................................................... 87 Modifying a constraint graphically ................................................................................... 88 Candidate set with a constraint.......................................................................................... 89 Constraints in qualitative or quantitative multilevel factors .............................................. 89 Inclusions vs. complement design ..................................................................................... 90 Inclusions added to the worksheet..................................................................................... 90 Inclusions as part of the design ......................................................................................... 91 Generating inclusions ........................................................................................................ 92 Modifying inclusions ........................................................................................................ 92
Inclusions ..................................................................................................................... 90
93
D-Optimal
103
What are D-Optimal designs? .................................................................................... 103 When do I use D-Optimal designs? ........................................................................... 103 D-Optimal pages in the design wizard ....................................................................... 104
Design generation criteria section ................................................................................... 104 Design alternatives section .............................................................................................. 106 Candidate set section ....................................................................................................... 107
v
Design Wizard
117
Introduction................................................................................................................ 117 Defining factors ......................................................................................................... 117 Defining responses..................................................................................................... 119 Selecting the objective ............................................................................................... 120 Selecting model and design ....................................................................................... 121 D-Optimal pages in the design wizard ....................................................................... 122
Worksheet
123
Analysis
133
Introduction................................................................................................................ 133 Organization of the Analysis menu............................................................................ 133 Evaluate ..................................................................................................................... 134 Analysis Wizard......................................................................................................... 135
Open the Analysis Wizard............................................................................................... 135 Analysis Wizard content ................................................................................................. 136 Condition number ........................................................................................................... 134 Runs, terms and degrees of freedom ............................................................................... 134
vi
Table of Contents
Prediction
165
Response prediction plot ............................................................................................ 175 Prediction list ............................................................................................................. 176 Scatter plot ................................................................................................................. 176 Sweet spot plot ........................................................................................................... 177 Optimizer ................................................................................................................... 180
Factor spreadsheet in the Optimizer ................................................................................ 181
vii
Response spreadsheet ...................................................................................................... 181 Run list ............................................................................................................................ 183 Optimizer context menu .................................................................................................. 184 Optimizer buttons ............................................................................................................ 185
Design Space
191
Introduction................................................................................................................ 191 Design Space Estimation around the selected setpoint .............................................. 191 Design Space Validation for robustness testing ......................................................... 192 Design Space window ................................................................................................ 192
Factor spreadsheet ........................................................................................................... 193 Response spreadsheet ...................................................................................................... 194 Design Space buttons ...................................................................................................... 194 Design Space Properties .................................................................................................. 195
Factor histogram ........................................................................................................ 196 Response histogram ................................................................................................... 197 Design Space Statistics list ........................................................................................ 197 Final factor adjustments ............................................................................................. 198
Show
199
Introduction................................................................................................................ 199 Objective .................................................................................................................... 199 Design matrix............................................................................................................. 200 Design region ............................................................................................................. 201 Design region for mixture designs ............................................................................. 202 Confoundings............................................................................................................. 202 Summary D-Optimal.................................................................................................. 203 Model ......................................................................................................................... 204 Reference mixture ...................................................................................................... 204
Design region properties ................................................................................................. 201
205
Introduction................................................................................................................ 205 Properties page........................................................................................................... 205 Automatic update of plots and lists ............................................................................ 205 Saving plots and lists ................................................................................................. 205 Generating multiple plots or lists ............................................................................... 206 Plot settings................................................................................................................ 206
Axis ................................................................................................................................. 207 Header and Footer ........................................................................................................... 209 Plot Area ......................................................................................................................... 210 Saving plot settings ......................................................................................................... 212 Resolution ....................................................................................................................... 213 Scale subplots equally ..................................................................................................... 213 Lock contour levels ......................................................................................................... 213 Use color ......................................................................................................................... 213 Opening property page by right-clicking......................................................................... 205 Making a change in the property page ............................................................................ 205
viii
Table of Contents
Report generator
221
Introduction................................................................................................................ 221 Starting the report generator ...................................................................................... 221 Report generator window........................................................................................... 222 Command menu bar ................................................................................................... 222
File menu ........................................................................................................................ 222 Edit menu ........................................................................................................................ 224 View menu ...................................................................................................................... 224 Insert menu...................................................................................................................... 224 Format menu ................................................................................................................... 225 Tools menu...................................................................................................................... 225 Help menu ....................................................................................................................... 226 New ................................................................................................................................. 226 Generate report specific buttons ...................................................................................... 226
Generate report toolbar .............................................................................................. 226 Format toolbar ........................................................................................................... 226 Placeholder window................................................................................................... 227 Properties window ..................................................................................................... 228 Adding plots and lists to the report ............................................................................ 228
Help
229
Introduction................................................................................................................ 229 HTML help ................................................................................................................ 229 Registration and activation ........................................................................................ 229 Manage Licenses ....................................................................................................... 229 Umetrics on the Web ................................................................................................. 230 About MODDE .......................................................................................................... 230
ix
231
Fit methods ................................................................................................................ 231 Model ......................................................................................................................... 234 Scaling ....................................................................................................................... 235 Condition number ...................................................................................................... 236
Condition number definition ........................................................................................... 236 Condition number with mixture factors........................................................................... 236 PLS and Cox reference mixture model ........................................................................... 236 MLR (regression) and the Cox model ............................................................................. 236 MLR (regression) and Scheff model ............................................................................. 237 Missing data in X ............................................................................................................ 237 Missing data in Y with Multiple Linear Regression ........................................................ 237 Missing data in Y with PLS ............................................................................................ 237 Scaling X ......................................................................................................................... 235 Scaling Y ......................................................................................................................... 235 Hierarchy......................................................................................................................... 234 Multiple Linear Regression (MLR) ................................................................................. 231 Partial Least Squares (PLS)............................................................................................. 232
N-value ...................................................................................................................... 237 Residual Standard Deviation (RSD) .......................................................................... 237 ANOVA ..................................................................................................................... 238 Measures of goodness of fit ....................................................................................... 239
Q2.................................................................................................................................... 239 R2 .................................................................................................................................... 239 Degrees of freedom ......................................................................................................... 239 Scaled and centered coefficients ..................................................................................... 240 Normalized coefficients .................................................................................................. 240 PLS orthogonal coefficients ............................................................................................ 240 Confidence intervals........................................................................................................ 240 Raw residuals .................................................................................................................. 241 Standardized residuals ..................................................................................................... 241 Deleted studentized residuals .......................................................................................... 241 Plot loadings.................................................................................................................... 242 Plot scores ....................................................................................................................... 242
Coding qualitative factors at more than 2 levels ........................................................ 241 Residuals .................................................................................................................... 241
Predictions ................................................................................................................. 242 PLS plots.................................................................................................................... 242 PLS coefficients ......................................................................................................... 242 Box-Cox plot (only MLR) ......................................................................................... 243 Mixture data in MODDE ........................................................................................... 243 Optimizer ................................................................................................................... 248
Desirability ...................................................................................................................... 248 Overall desirability .......................................................................................................... 249 Overall Distance to Target .............................................................................................. 249 Starting simplexes ........................................................................................................... 249 Sensitivity Analysis......................................................................................................... 250 Mixture factors only ........................................................................................................ 243 Process and mixture factors............................................................................................. 247
Table of Contents
Appendix B: Designs
257
Designs for process factors ........................................................................................ 257 Designs for mixture factors........................................................................................ 263
Mixture and process factors ............................................................................................ 263 Mixture factors definition................................................................................................ 263 Mixture constraint ........................................................................................................... 264 Mixture experimental region ........................................................................................... 264 Classical mixture designs ................................................................................................ 265 What are D-Optimal designs? ......................................................................................... 268 Candidate set ................................................................................................................... 268 D-Optimal algorithm ....................................................................................................... 269 Implementation of the D-Optimal algorithm in MODDE ............................................... 269 Potential terms................................................................................................................. 269 Design evaluation ............................................................................................................ 271 Inclusions and design augmentation ................................................................................ 271 Irregular region ............................................................................................................... 272 Screening onion designs .................................................................................................. 274 RSM onion designs ......................................................................................................... 274 Screening designs ............................................................................................................ 257 RSM designs ................................................................................................................... 260
Appendix C: Optimizer
275
Introduction................................................................................................................ 275 Search function .......................................................................................................... 275 Optimizer objectives .................................................................................................. 276
Limit optimization........................................................................................................... 276 Target optimization ......................................................................................................... 277 Focus optimization .......................................................................................................... 277 Response specification example in the optimizer ............................................................ 278
281
References Index
289 291
xii
Installation
You can install MODDE under Windows Vista, XP, and 2000.
Note: You must have administrative privileges to be able to install the software.
To install and activate MODDE follow the steps described below: 1. Download the password protected installation file from the Umetrics web page www.umetrics.com, click Downloads. The password is available in the delivery letter. Open the file and when prompted enter personal information as well as product information found in the delivery letter. After completing the installation, MODDE needs to be activated with a license file. Activation is done either (a) from an internal license server or (b) from Umetrics. See the delivery letter, sent to the license administrator at your company, for instructions.
2. 3.
Starting MODDE
Start MODDE by double-clicking its icon.
To read about the MODDE software look in Help (contains the same information as the user guide). To run tutorial examples, find them at www.umetrics.com (Downloads), select an example, open the investigation used in the tutorial (.mip-file included with the installation in the Investigation folder) and follow the analysis steps. To start a new investigation, on the File menu click New.
Experimental cycle
The experimental cycle consists of three phases: 1. 2. The design phase where you define your factors and within which ranges they should be varied, your responses, objective, design and model. The analysis phase where you explore your data, review the raw data and the fit, review diagnostics in plots and lists, refine and interpret the model.
3.
The prediction phase where you use the model to predict the optimum area for operability.
Design phase
Once you have clicked New on the File menu the design wizard is then opened. Enter the name and location of the new investigation in the appropriate fields. For details on the design wizard, see the Design Wizard chapter.
Defining factors
On the second page of the design wizard, enter factor names, ranges, etc. For details on defining factors see the Factors chapter.
Defining responses
On the third page of the design wizard, enter response name(s), etc. For details on defining responses see the Responses chapter.
Defining objective
On the next page of the design wizard, click your objective Screening, RSM, or Split Objective. Clicking Paste Data disables Next; clicking Finish opens the worksheet for pasting. Given your objective, the type of factors (formulation or process) and the number of factors, MODDE recommends a design, a model, and a number of center points. The total required number of runs for this design is displayed. Other choices compatible with your objective type and number of factors are listed with the required number of runs. After clicking Finish, the Worksheet is automatically generated for all classical designs. With D-Optimal designs, clicking Next opens a dialog for the selection and generation of the D-Optimal design and worksheet. After making the experiments, fill in the response values and actual factor values in the worksheet. For details on the objective pages, see the Objective, model and design chapter. For details on the worksheet, see the section Worksheet spreadsheet in the Worksheet chapter.
Analysis phase
After the response values have been entered in the worksheet you can review the raw data, fit the model, review the fitted model, interpret the model, and refine the model.
Fit
When you are ready to fit a model to your design you click Fit on the Analysis menu. MODDE automatically fits using MLR when the condition number is low and there are no mixture factors. The fit methods available are MLR, PLS and for mixtures several variants. For PLS models you can choose to extract more components than the MODDE default. To extract more components click Next Component on the Analysis menu. Next Component is unavailable when MLR is used as fit method.
Percent of the variation of the response explained by the model, R2. R2 overestimates the goodness of fit. Percent of the variation of the response predicted by the model according to cross validation, Q2, and expressed in the same units as R2. Q2 underestimates the goodness of fit. Model Validity: A Measure of the validity of the model. When the Model Validity column is larger than 0.25, there is no Lack of Fit of the model (the model error is in the same range as the pure error). Reproducibility: The variation of the response under the same conditions (pure error), often at the center points, compared to the total variation of the response.
Diagnostics
MODDE has a number of diagnostic plots:
Residual plots to find outliers, drifts, trends etc. Box Cox Plot to select the best transformation of Y. ANOVA: ANalysis Of VAriance, in particular review the Lack of Fit. The estimation of lack of fit is only available when there are replicated points as it compares the pure error and the model error.
When PLS is used for regression, scores and loadings can be plotted. These plots provide an overview of the data. On the Analysis menu, click PLS Plots to select the score or loading plot you want to display.
Note: When excluding an outlier or model term in a plot, the outlier or model terms is only excluded for the displayed response.
Remove/Add terms from/to the model that are insignificant/significant for all responses. Use Edit | Model or the interactive tool with the coefficient plot. After refining your model you should once more review the fit and diagnostics as described above.
Note: You can have a different model for each response in the same investigation.
General description
MODDE - (MODeling and DEsign) is a Windows program for the generation and evaluation of statistical experimental designs. Methods of statistical experimental designs have evolved since the pioneering work of Fisher in 1926. These methods, further refined by Box, Hunter, Scheff, Tagushi and others, provides the users with a powerful methodology for efficient experimentation. The application icon for MODDE is a pink circle with a gray M inside.
Which factors have a real influence on the responses (results)? Which factors have significant interactions (synergies or antagonism)? What are the best settings of the factors to achieve optimal conditions for best performance of a process, a system or a product? What are the predicted values of the responses (results) for given settings of the factors?
In the screening stage one uses simple models (linear or linear with interactions), and experimental designs that allows the identification of the factors with the largest effects in the fewest possible number of experimental runs. MODDE supports: Full Factorial, Fractional factorial, L-designs, Plackett Burman, Rechtschaffner, Onion, and D-optimal designs for screening experiments. With mixture factors, MODDE supports the classical axial design when the region is a simplex.
Understand in more detail HOW the factors influence the response; get a map of the system. Make predictions, optimize or find a region of operability.
MODDE supports: Three-level full factorial, central composite, (CCC and CCF), Box Behnken, Rechtschaffner, Doehlert, Onion, and D-optimal designs for RSM investigations. With mixture factors, MODDE supports the modified simplex centroid when the experimental region is a simplex. For investigations with only qualitative terms no square or cubic terms can be estimated. Creating RSM-designs for such investigations is therefore impossible.
Fit methods
The data collected by the experimental design is used to estimate the coefficients of the model. The model represents the relationship between the response Y and the factors X1, X2, etc. MODDE uses multiple linear regression (MLR) or Partial Least Squares (PLS) to estimate the coefficients of the terms in the model. MODDE recommends PLS when the investigation has a high condition number.
PLS contains the multiple regression solution as a special case, i.e. with a single response or different models, and a given number of PLS dimensions, the PLS regression coefficients are identical to those obtained by multiple regression. PLS has been extensively described in the literature and only a brief description is given here. PLS finds the relationship between a matrix Y (response variables) and a matrix X (model terms). The PLS model consists of a simultaneous projection of both the X and Y spaces on a low dimensional hyper plane with new coordinates T (summarizing X) and U (summarizing Y), and then relating U to T.
This analysis has the following two objectives: 1. 2. To well approximate the X and Y. To maximize the correlation between X and Y in the projected space (between u and t).
The dimensionality, number of significant PLS components, is determined by cross validation (CV), where PRESS (Predictive Residual Sum of Squares) is computed for each model dimension. MODDE selects automatically the number of PLS dimensions that give the smallest PRESS. PRESS is then re expressed as Q2 = (1 - PRESS/SSY), where SSY is the sum of squares of Y.
Results
Both MLR and PLS computes regression coefficients for each response. Thus Y is expressed as a function of the X's according to the selected model (i.e. linear, linear plus interactions, or quadratic).
Analysis phase
All results of model fitting, by MLR or PLS are displayed, in the same way, graphically and in lists.
Summary of the fit, R2, Q2, Model validity, and Reproducibility for every response. Coefficients and their 95% confidence intervals. ANOVA table. Effect plots for screening designs.
Normal probability plot of residuals Plot of residuals against fitted values, run order or other factors. Box-Cox plot to check for the optimal transformation of the response.
For PLS, summary of the fit by component and PLS score and loading plots are available.
Contour and rotatable 3D plots. Optimizer to find the best conditions for a desired profile of the responses. This helps in the interpretation of the results and to find a region of operability. Sweet Spot plot to draw a plot highlighting areas were the responses are within the specified ranges. Design Space to estimate the area of operability or robustness.
Convention
Limitations in investigation names
The following characters cannot be used: = \ / : * ? < > |. The length (including path) cannot be larger than 260 characters.
Case sensitivity
MODDE is case insensitive. Lower or upper case in names will be displayed as entered, but for all comparisons lower or upper case are considered the same.
Menu | Menu item. An example: File | New On the Menu menu, click Menu item. An example: On the File menu, click New. Click Menu item on the Menu menu. An example: Click New on the File menu.
4. 5. 6.
10
Overview
The symbol for MODDE is a gray cube with a pink M (above). The application icon is a pink circle with a grey M inside (below).
Investigation
Experimental plans in MODDE are organized into investigations. You can think of an investigation as a file folder containing all of the information related to a particular experiment. When you select or open a given investigation you can access, display and use all of its information. This information is organized in the following components: factors, responses, constraints, inclusions, candidate set, model, design, worksheet, analysis and predictions.
Managing investigations
Investigations are binary files saved by MODDE with the extension *.mip. When you start a new investigation you select its name and destination directory. You can open, save, save as, and delete investigations. You can generate the fold over of investigations with fractional factorial designs or Plackett Burman. You can double-click a MODDE investigation (a *.mip file), in Microsoft Explorer, to open that investigation.
11
MODDE does not save the fitted model. In order to review the results of the analysis and use make predictions, you need to fit the model by clicking Analysis | Fit or having Automatic fit turned on (default) in View | General Options tab General when the investigation is opened. After the model has been fitted, you can use the Analysis menu to create plots and list to review the model and fit and use the Prediction menu to create prediction plots and lists.
Organization
The MODDE window consists of a command menu bar and toolbars.
12
Overview
The menu bar consists of the menus File, Edit, View, Design, Worksheet, Analysis, Prediction, Show, Window, and Help.
File menu
Under the File menu the following is available:
New: Start a new investigation. Open: Open a MODDE investigation. Close: Close a MODDE investigation. Save: Saves the current investigation. Save As: Saves the current investigation under a specified path and name. Delete Investigation: Deletes the current investigation. Save Plot/List A : Saves the active plot or list. Revert: Reverts to the last saved copy of the investigation. Send by email: Sends the current Investigation by e-mail. Generate HTML report: Automatic report generator in HTML format. Complement Design: Select to complement your design by: 1. 2. 3. 4. 5. 6. Fold over: Complements screening designs of resolution III or IV by adding the fold over. Estimate square terms in a screening design: complements the design to support a quadratic model for a selection of factors. Complement the design with D-Optimal to support a specified model. Screening to RSM Rechtschaffner: Complements the Rechtschaffner screening design to the RSM Rechtschaffner. Complement the Doehlert design. Plackett Burman Super-Saturated to Plackett Burman: Complements PBSS to the regular PB.
Protect Investigation: Allows for a password protected investigation and locking of investigation.
13
Edit menu
From the Edit menu you can:
Undo Cut/Copy/Paste and Delete Select All Row insertion, addition Sort Edit the Model Edit the Reference Mixture Edit the Generators of fractional factorial designs
View menu
The following is available from the View menu:
Toolbars: View/Hide Toolbars. Dockable windows: Show or hide dockable windows. Add to Favorites: Add the current plot or list to the Favorites dockable window. Add to Report: Add the current plot or list to the current report generated by MODDE. Full screen: Maximize the plot area. Investigation Options: Change the default plot and list options. General Options: Change the default options. Customize: Customize the toolbars.
Design menu
Use the Design menu for the following:
Factors and Responses: Define and modify factors and responses Constraints and Inclusions: Define constraints and inclusions Objective: Specify and modify objective, model and design D-Optimal: Generate and evaluate D-Optimal designs, open the candidate set for viewing or editing, evaluate onion plots. Design wizard: Open the design wizard
14
Overview
Worksheet menu
Use the Worksheet menu to:
Edit the Worksheet. Set Run Order. Open the Curvature Diagnostic Plot. Create the Scatter plot. Open the Histogram plot. Open the Descriptive statistics plot and list. Open the Correlation plot or matrix. Open the Replicate plot.
Analysis menu
Use the Analysis menu to:
Evaluate the design Analyze each response in the Analysis Wizard Select Fit Method (MLR, PLS etc) Fit model by MLR or PLS Extract the Next Component: Add a PLS component
Summary of the fit PLS Summary PLS Plots Coefficients and Effects VIP Plot (PLS only) Residuals Plots Box-Cox Plot (MLR only) ANOVA
Investigate diagnostics
Prediction menu
With the fitted model you can click the Prediction menu and:
Display contour plot from the Contour plot wizard including mixture contour plots and response surface plots (3D). Create prediction plots. Make predictions in the Prediction spreadsheet.
15
Create the Sweet Spot Plot. Use the Optimizer. Estimate the Design Space.
Show menu
Use the Show menu to display the following: objective, design matrix, design region, confounding, model, D-Optimal summary, and reference mixture.
Window menu
Use the Windows menu to access the standard window commands.
Toolbars
Click Toolbars from the View menu to display or hide the following: 1. 2. 3. 4. The Standard toolbar consist of command buttons for immediate action and menu commands. The Spreadsheet toolbar displays all spreadsheets. The Plot toolbar is for zooming in and out on plots, reading plot values and coordinates, and drawing regression lines. The Model toolbar (model information) summarizes the investigation. The summary consists of the number of factors and responses, the number of runs, the Objective, Design and model and the Fit method. The Window toolbar displays buttons for arranging the open windows and Full Screen. The Status bar displays an explanation to the button that the mouse points to.
5. 6.
File
Introduction
Use the File menu to create, open, close, save, and delete investigations, save active plot or list, revert to last saved investigation, send the current investigation by e-mail, generate a HTML report, complement the design, lock or password protect the investigation, for print purposes and to close MODDE.
New
To create a new investigation click File | New. This first page allows you to:
Enter the name of the new investigation. Select the location where to save. Select the design type: Traditional Designs or Advanced Designs including RED-MUP and D-Optimal designs from an imported candidate set or scores from a SIMCA-P model. Select to import a design from file. Enable the audit trail by selecting the Enable the Audit trail check box.
17
Click Next to open the design wizard. The design wizard guides you through the design phase, from factor and response definition, objective, design and model selection, up to the generation of the worksheet. For a detail description of the design wizard see the Design Wizard chapter.
Traditional designs
When you create a new design Traditional Designs is by default selected. The traditional designs include classical designs such as factorial, Plackett Burman, Ldesigns, Rechtschaffner, and also D-Optimal and onion designs. For details, see the Design appendix.
Advanced designs
When you select Advanced Designs and click Next the Advanced Designs box is opened with the available designs:
RED-MUP. Onion or D-Optimal designs from imported candidate set. Multivariate onion or D-Optimal designs from scores.
18
File
RED-MUP
RED-MUPs are designs available for 96 (8x12), 384 (16x24), and 1536 (32x48) runs. The designs are built from sub-designs. To create a RED-MUP: 1. 2. 3. 4. 5. 6. 7. 8. 9. In the first page of the design wizard click Advanced Designs and then Next. Click RED-MUP and then Next. Define all factors for the two sub-designs and then click Next. Define the responses and then click Next. Select the objective for both the vertical and the horizontal designs: Screening or RSM. Leave the factors that should be included in the vertical design with fewer runs to the left. Move the factors that should be included in the horizontal design to the right. In this page you can also select the number of plates used, and if applicable the plate factor(s) that contain plate information. Select the desired plate size in the Plate-Size box and optionally select the Plate/Block factor interactions check box.
10. Click Next to select the vertical design then click Next again to select the horizontal design. MODDE adds center points when the selected design does not fill up the plate size.
Note: Some special RED-MUP designs, that aim to aim to make better use of the plate, are available for the 96 well plates (8 x 12).
19
5.
Note: Only regular factors are imported here. If your candidate set contains qualitative or formulation factors you have to enter the factors and settings in the Factor Definition and import the candidate set from the D-Optimal page.
The designs available here are Onion and D-Optimal designs. Onion is only available when there are enough experiments in comparison with the number of factors. For more on Onion designs, see the D-Optimal Onion designs section in the Design appendix and the Tutorial.
Multivariate designs
To create a design using the scores from a SIMCA-P project as factors: 1. 2. 3. 4. On the Advanced Designs page click Multivariate onion or D-Optimal designs from scores. Click the Browse button under Import scores (factors) from a .usp created in SIMCA-P 9 or higher to select the SIMCA-P project. SIMCA-P opens automatically. Return to MODDE without closing SIMCAP. Select the model from the Model box and click Next to import the factors from SIMCA-P. The scores are then automatically loaded from the SIMCAP .usp-file, and the candidate set for the onion design comprised of all objects (rows) in the workset of the SIMCA-P model is selected as the basis of the onion design.
Note: To be able to import scores from the SIMCA-P project, SIMCA-P 10 or later needs to be installed.
When importing the scores of a SIMCA-P project as factors in MODDE, the only designs available are the D-Optimal and the Onion designs.
Note: 'SIMCA-P' above includes both SIMCA-P and SIMCA-P+ when available. For the functionality described here the type of SIMCA-P is not relevant.
For more on Onion designs, see the D-Optimal Onion designs section in the Design appendix and the Tutorial.
20
File
Note: This dialog recognizes regular quantitative and qualitative factors. If there are formulation factors in the imported design, you have to manually alter the factor type (starting with formulation factor you want to end up last) in the factor definition page.
Open
To open a MODDE investigation, click Open on the File menu. In the Open-dialog select the directory and you get a list of all the MODDE Investigation Projects (*.mip) present in that directory. Double-click the desired investigation to open it or mark it and click Open. To open investigations created in MODDE version 4 and earlier, switch to Old MODDE Files (*. ini) in Files of type to view them, mark the .ini and click Open to convert it to the current format.
Delete investigation
To delete the current investigation click Delete Investigation on the File menu. It is not possible to delete any other investigations than the current from MODDE. For read-only investigations Delete Investigation is unavailable.
Save plot/list as
Plots can be saved as Enhanced Meta Files (*.emf), Bitmap files (*.bmp), Jpeg files (*.jpg) or PNG files (*.png). Lists can be saved as Text files (*.txt) or Web Pages (*.htm, html).
21
To save a plot, display it, make sure it is active and then click File | Save Plot As. The Save Format dialog is displayed.
In the dialog check the Keep aspect ratio box to keep the proportions of the plot as on the screen or according to a predefined plot size. In the Plot size box you can choose between Custom, 640x480(VGA), 800x600(SVGA), 1024x768, Original size.
The predefined formats are useful primarily when displaying MODDE plots on a small screen. To save a list, display it, make sure it is active and then click File | Save List As. The Save dialog is then displayed. In the Save as type box, select file type .txt or .htm/html.
Revert
To revert to the investigation as when it was last saved, click Revert on the File menu.
Send by E-Mail
To send an investigation by e-mail click Send by E-Mail on the File menu. Note that the investigation sent is the current investigation as it was last saved.
File
A placeholder tells MODDE the desired item to fill from the current investigation when you click Update Report. If you add plots and lists as placeholders and save the template, you can generate a report in the desired format, for any investigation, by selecting the saved template and clicking Update Report. For details, see the Report Generator chapter.
Complement design
To complement a design is to make a new investigation consisting of the design, or a part of the design, of the current investigation plus its complement. See also the section Inclusions vs. complement design in the chapter Constraints and Inclusions. Use Complement Design when you want to:
Estimate separately a set of terms (interactions, or main terms and interactions) that were confounded in a Resolution III or IV fractional factorial design. Complement a screening design to an RSM design supporting the full quadratic model. Complement a screening design to estimate selective curvature effects. Add additional experimental runs to improve the quality (i.e. the condition number or G-efficiency) of an existing set of experiments. Use already performed experiments in a Doehlert design to set up a new Doehlert moving the center of the design or add a factor. For more on the Doehlert designs see the Design appendix. Use already performed experiments in a Super-Saturated Plackett Burman design to add experiments resulting in a regular Plackett Burman design.
To complement the current design, click Complement Design on the File menu and the Complement Design Wizard is opened. Click the desired complementing method and click Next.
23
Fold over
When you choose to complement your design with fold over MODDE makes a new investigation consisting of the design of the current investigation plus its fold over (complement). The fold over design has as many experimental runs as the original design. Fold over designs are available for fractional factorial design of resolution III or IV and Plackett Burman designs. With the complete design (original + fold over), all main effects are clear from 2 factors interactions. With resolution III and IV designs MODDE automatically adds a block factor. You may remove the block factor from the model in the menu Edit | Model. To fold over your design: 1. 2. 3. 4. Click File | Complement Design. Click Fold over and then click Next. Enter the name and location of the new investigation. It is recommended to add an additional center point to detect a shift in the mean. Click Finish and the new investigation opens.
24
File
1. 2. 3. 4. 5.
Click Estimate square terms... and click Next. Mark the factors for which to estimate the square terms. Optionally change the star distance. Click Next. Enter the name, location of the new investigation and number of additional center points. The model has been updated with the squares of the selected factors.
D-Optimal
Complementing a design d-optimally is the most flexible way of complementing a design.
3.
25
Mixture factors
Complementing the design d-optimally with mixture factors, after giving the name and location of the new investigation, the D-Optimal wizard opens. Select the new model.
The displayed number of runs includes the original design runs. Click Next, the wizard guides you in generating the new investigation.
The displayed number of runs includes the original design runs as inclusions. Click Next; the wizard guides you in generating the new investigation.
26
File
Complement Doehlert
Doehlert designs can be complemented by expanding the design region or adding a factor. To complement a Doehlert design, click Doehlert in the Complement Design Wizard and click Next.
Leaving the default Select new center selected and selecting one of the experiments of the design as the center of a new Doehlert design. Selecting Add factor and typing the name of the new factor and typing the Center and Range in the respective boxes. The value you type in Center is the value that will be entered in the design for the already performed experiments, for the new factor.
27
Select how to complement, click Next, enter the name, location of the new investigation and number of additional center points, and then click Finish to generate the new investigation.
28
File
Protect investigation
When you click File | Protect Investigation you can select to Encrypt/Password protect or Lock the investigation.
Lock investigation
Click File | Protect Investigation | Lock to lock the investigation. Locked investigations are automatically fitted when opened. Any plot or list can be displayed, but you cannot make any changes in the investigation. The investigation becomes 'Read only' with the exception of the prediction spreadsheet. If you click Permanently, unlock on Save As, a copy of the investigation can be unlocked by using File | Save as. When Permanently is selected the investigation cannot be unlocked.
It is possible to also encrypt and password protect the investigation by selecting the Encrypt / Password Protect investigation check box. The investigation is then encrypted. This check box is optional; select it only if you want the investigation to be password protected.
Print Setup
Change the page orientation, printer, and printer settings by clicking Print Setup on the File menu. Plots are preferably printed with Orientation Landscape. The print menu is only valid for an active list or plot.
Print Preview
On File, click Print Preview and the current window can be view as it will be printed. Print Preview is only available when there is an open plot or list.
Print
Print is only available for the active plot or list. With the plot or list active, use one of the following manners to print:
On File menu click Print. Right-click the plot or list and click Print. Click the Print-button Press CTRL+P. in the Standard toolbar.
When printing plots, the plot is by default scaled to look exactly as on the screen. The lists and plots in MODDE can also be printed from another Windows application after first being copied through the clipboard to that application.
Exit
Click Exit from the File menu to close the MODDE program.
30
Edit
Introduction
Use the Edit menu to undo, cut, copy, paste, delete, select all, add experiment/insert rows/add factor/add response, sort, edit model, edit reference mixture, and edit the generators.
Undo
Click Edit | Undo or press CTRL+Z to undo changes in MODDE. Such changes can be changing values or text (factor name for instance), cutting, copying, pasting, deleting, sorting in spreadsheets, or editing the model. Undo is activated after changes in:
The spreadsheets Worksheet, Factors, Responses, Constraints, Inclusions, Prediction, and Optimizer. The model terms in the Model Properties dialog (Edit | Model) or by clicking the Exclude-button and marking model terms to exclude in a plot. Plots displaying experiments (observations) by clicking the Exclude-button and marking model terms to exclude in a plot.
Undo remembers the ten last actions in the worksheet, in the edit model dialog, and in plots displaying experiments (observations) or model terms. Undo works on the active plot or spreadsheet.
Note: After editing the values in worksheet the undo feature of plots displaying experiments stops working as manual changes in the worksheet empty the undo-memory for such plots.
31
Sort
Sorting is available for the worksheet, constraints, inclusions, candidate set, and predictions list. To sort a list, right-click and click Sort or click Edit | Sort.
In the dialog, select the column to sort the list by from the Select the column to sort box and click the Add Column button. The column appears in the list with the default sort type. Click the sort order you want under Sort selected, Ascending or Descending. If a column of the spreadsheet is marked when sort is activated it becomes, by default, the primary column to sort by. Add more columns to select secondary and tertiary etc., columns to sort by. Specify for each column the sort order. Use the Remove button to remove a column from the sort list or drag the item outside the list. The sorting starts when OK is clicked.
32
Edit
33
Model
Click Edit | Model or the Edit Model button to modify the model. The Model Properties dialog opens on the Model Terms page and you can edit the model by adding or deleting terms. Use the For Response box to specify for which response you want to alter the design, if not for all. A quadratic term is represented by factor1*factor1 and a cross term (interaction) by factor1*factor2. A cubic term is represented by factor1*factor1*factor1 and a three factor interaction is represented by factor1*factor2*factor3.
34
Edit
Add terms
To add terms in the Model Properties dialog use one of the following methods: 1. 2. Mark the factors to the left and click the relevant buttons Factors, Interactions, Squares, or Int. & Sq. (interactions and squares) Double-click a factor or mark it and click the left arrow =>; note that it is added in the New term field. Continue until the new term is fully defined and click the second arrow => adding the term to the right side. Position the cursor in the New term field and type the abbreviation of the factors to include using * to separate them. To add three factor interactions, mark the three factors, click the left arrow and then on the right =>. Or type the abbreviations in the New term field.
3. 4.
Remove terms
To remove terms in the Model Properties dialog, mark the term(s) and click the Remove button in the dialog.
Undo
The Undo button becomes active after making a change in the model. Clicking Undo only reverses the last change in the Model Properties dialog.
Reset
Click the Reset button to reset the model to the default model.
35
Reference mixture
In Model Properties, click the Reference Mixture tab to switch to the reference mixture page.
Use this dialog to change the default reference mixture of a Cox model. The reference mixture is used to impose constraints on the coefficients of the Cox mixture model. The default reference mixture is the centroid of the experimental region. If the region is a regular simplex with q mixture factors, the centroid is the point with coordinates 1/ q, 1/ q,....1/ q. If the experimental region is irregular, the centroid is the constrained centroid of the irregular constrained region. Click the Reset button to restore the reference mixture to the default values.
36
Edit
Generators
Click Generators on the Edit menu to open the Generators dialog.
A generator is a column of signs in the extended design table of the basic factors that is used to introduce additional factors in the fractional factorial designs. For example, let us assume that 5 factors are to be investigated in 8 runs. The extended design table is the table of the full factorial in three factors (basic factors), symbolically named a, b and c plus the additional columns for all the interactions. Any interaction column can be used to introduce additional factors. Let us say that to introduce the 2 additional factors, d, and e, the column of signs of a*b and a*c are selected. Then d = ab and e = ac are the generators of the fractional factorial design 25-2 (see Box, Hunter and Hunter for further information). When MODDE generates fractional factorial designs the default generators used are those published in Box, Hunter and Hunter. Editing and/or changing the default generators of a design is done in order to estimate selected interactions in a fractional factorial design of resolution III or IV instead of the default. MODDE supports the choice of positive or negative generators. To edit a generator, click in the generator column of the desired row, and enter a new generator. The confounding, in the confounding column, is updated. When you click OK, your design and worksheet are deleted and new ones are generated.
37
38
View
Introduction
Using the View menu you can show/hide the toolbars, show/hide the dockable windows Analysis Advisor, Audit trail, Favorites, Output, and Notes, and execute the commands: Add to Favorites, Add to Report, make window Full screen, change the default investigation options, change the general options, and customize toolbars, commands and options.
Toolbars
On the View menu, click Toolbars to view the available toolbars. Click the name of the toolbar to show or hide it. All toolbars can be floating or docked beneath the menu bar. Clicking a button on a toolbar will perform a certain task. As a guide to what happens when a button is clicked, a short text will appear when you hold the pointer over a button. Text will also appear in the Status bar at the bottom of the screen. You can customize all toolbars according to your wishes. For more, see Customize later in this chapter.
Standard toolbar
The Standard toolbar consists of a number of command buttons and the response box. The buttons listed below are available.
39
Edit model
Click the Edit model button to edit the model or reference mixture when applicable. Model Terms is the default page opened. To edit the reference mixture, click the Reference Mixture tab.
Response box
Use the Response box to change responses in the active plot or list.
40
View
The last response check box cannot be cleared. To get single selection when switching between responses using the response combo box, hold down the CTRL-button for single selection functionality.
Click Dockable Windows on the View menu, and then click Favorites.
Spreadsheet toolbar
The Spreadsheet toolbar contains buttons for all MODDE spreadsheets. In the order of the toolbar: the factor, response, worksheet, prediction, and optimizer spreadsheets.
Plot toolbar
The Plot toolbar holds buttons that enable you to select type of marking to be used when marking areas in plots, insert labels or text in plots, zoom in and out, read positions in graphs, do linear regression in scatter plots, get information about observations (experiments) or variables (model terms or responses), or exclude experiments and model terms. The tools are activated when you click them.
41
Arrow
The arrow button is used to select which type of marking to use. Click the small arrow to the right of the arrow and click the type of marker from the menu.
Text
Free Mark: Allows marking to take any shape. Block Mark: Marks in a rectangular shape. X Mark: Vertical bar. Y Mark: Horizontal bar. No Mark: The arrow works as a data reader displaying the number, name, and value of plotted data points when hovering near the data point symbol.
Click the T-button to insert a text-field in a plot. The field can contain text or a value and be inserted in all plots but the contour, response surface, sweet spot, and 3D scatter plots. Open the plot, click the T-button, click where you want to position the text-field, type the text or numbers, and then press ENTER or click somewhere else in the plot. MODDE will continue to open a new text-field each time you click until the T-button is pressed again. This label can be removed by clicking it again and emptying the cell.
Zoom in
Click the arrow near the Zoom in button drop down menu: , and select the type of zoom from the
Scale xy: Magnifies a rectangular region Scale x: Expands the x direction Scale y: Expands the y direction Subplot: Magnifies a subplot in a multiplot.
And then mark the desired region of the plot to zoom. Zooming in a scatter 3D plot is described in the Zoom and rotate section in the Plots and lists chapter.
Zoom out
Click the Zoom out when zooming. button to revert zoom to original scale in the steps taken
Coordinate Reader
The coordinate reader displays the coordinate values of any point of the plot area. Click the Read coordinate values button and mark the point of interest and the coordinate reader will display the coordinates at that point.
42
View
Regression line
The regression line and equation can be displayed for any 2D scatter plot in MODDE. Open a scatter plot, for instance the Observed vs. predicted plot, and click the Show/Hide regression line button .
Rotate
For rotating a scatter 3D or response surface plot see the Zoom and rotate section in the Plots and lists chapter.
In the plot mark the term(s)/experiment(s) you want to exclude from the model. Optionally click the Undo-button to reverse these exclusions.
4.
MODDE removes the terms/experiments from the model of the currently displayed response ONLY and automatically refits the model.
43
Model toolbar
The Model toolbar displays the summary of the current investigation. 1. 2. 3. 4. 5. 6. Factors: The total number of factors (the number of uncontrolled). Responses: The number of responses. Runs: The number of runs included in the design (not counting rows set as 'Excl' in the worksheet). Objective: The objective, Screening, RSM or Split Objective. Design: The selected design and model. Method of Fit: MLR, PLS etc.
Window toolbar
The Window toolbar commands are general Windows commands and work accordingly. Click the Window toolbar commands Full screen (press the F9 key to get back to regular), Cascade Windows, Tile Windows horizontally or Tile Windows vertically to arrange the windows.
Status bar
The Status bar displays the explanation to the command button that the mouse points to.
Dockable windows
MODDE has five dockable windows: analysis advisor, audit trail, favorites, output, and notes.
Open the dockable windows by clicking Dockable Windows on the View menu and then clicking the desired dockable window. When you click the Auto Hide button the window will hide away. If you then hover with the mouse over the dockable window name, it slides out and becomes visible. If you want your window to stay visible after it has slid out, just click the pin button again .
44
View
Analysis advisor
The Analysis Advisor is automatically activated after fitting an investigation. The advisor explains the analysis plots and results. Show or hide the Analysis Advisor window is available by clicking View | Dockable Windows | Analysis Advisor or by clicking the ? - button .
Audit trail
When turned on, the audit trail logs changes done to the investigation. To open or close the audit trail, click Dockable Windows on the View menu, and then click Audit Trail. For more, see the Audit trail subsection in the Changing the default options using Investigation Options section.
Favorites
To open or close the favorites window use one of the following methods:
Click Dockable Windows on the View menu, and then click Favorites.
The Favorites window by default includes a few default commands, spreadsheets, lists, and plots and can be extended and modified as desired. Execute a command, or open a plot or list by clicking it. When settings must be selected, a dialog window opens, select the settings and click OK. Plots and lists are displayed with its selected settings. Not all plots in MODDE are available to be added to the favorites as plots. Generally plots that are displayed for the first response by default, and do not require a dialog for the selection of settings, can be added to the favorites as plots. Other plots requiring the selection of settings from a dialog can preferably be added to the favorites as commands. For instance, if the Contour plot wizard is added as a command, selecting it from the favorites opens the dialog and settings can be selected.
45
Adding favorites
To add an open plot or list to favorites, right-click it and click Add to Favorites or select it from the View menu. Note that this adds the plot with the selected settings to the favorites. You can also right-click in the Favorites window and select Add Command to Favorites.
Favorites window
Right-clicking the Favorites window opens the menu below. A description of the menu items follows.
Adding favorites To add an open plot or list to favorites, right-click it and click Add to Favorites or select it from the View menu. Note that this adds the plot with the selected settings to the favorites. You can also right-click in the Favorites window and select Add Command to Favorites.
46
View
Open all items in folder Executing a folder To execute all commands in a folder, right click the folder and select Open All Items in Folder. For example with the Analysis folder:
Opening all items in the folder displays the coefficient plot and observed vs. predicted plot for the first response tiled.
Treat folder as item Click Treat Folder as Item to treat the folder as an item. The folder is then displayed as an item: Click it to open all the items. This gives the same result as Open All Items in Folder but with a single click.
47
Rename All folders, commands, plots, spreadsheets, and lists can be renamed according to your wishes by right-clicking it and clicking Rename. Delete All items in Favorites can be deleted by right-clicking the item and clicking Delete, or marking it and pressing the DELETE-key on your keyboard. Add command to favorites A number of commands can be added to favorites. The commands from the menu and toolbars are displayed on the left, under Categories, and the items available under that command are displayed to the right. For example, if you click Prediction under Categories, all the commands on the Prediction menu are displayed to the right.
Click a command, for example Prediction Plot Wizard, and click Add. The command is added to the Favorites window. You may continue adding commands, and when done, click Close.
48
View
When a command has a submenu, double-click it to view the submenu items. For example Analysis | Residuals, double-click Residuals and the submenu of four items it displayed. Select the desired one and click the Add button.
Create a new folder It is convenient to group commands in folders, and automatically execute all the commands in the folder in sequence. To create a folder in Favorites, right-click and click New Folder. The created folder is default named 'New Folder'; change the name as desired, for example to 'Residuals Plots'. To move favorites to the Residuals Plots folder, use the drag-and-drop feature.
49
Importing and exporting Favorites configuration The Favorites window configuration can be saved as an .xml-file.
Note: When importing a favorites file, the new favorites will replace the current favorites. If you want to keep you current configuration and switch back to it later, export to file before importing.
To save the current favorites configuration to file: 1. 2. 3. 4. 1. 2. 3. 4. Right-click the Favorites window. Click Export. Enter the name and location in the Save As dialog, Click Save. Right-click the Favorites window. Click Import. Browse for the file in the Open dialog. Click Open.
Restoring favorites To restore Favorites to the MODDE default, click General Options on the View menu, click the Restore tab and click the Restore button after Restore favorites.
Output
The output window is a log book of the session. All MODDE messages and actions are recorded in the Logbook. Open the output window by clicking Dockable windows on the View menu and clicking Output.
To make this window smaller, pull the top and it will remember its size. Double-click the window caption to make it floating or make it dock.
Hint: While dragging the window hold down the CTRL key. It won't dock if you are close to the frame of the main frame window.
Notes
In Notes you can record your own notes concerning the investigation. You can paste MODDE plots and lists, and embed objects such as Microsoft Excel spreadsheet, word, etc. All these commands are available by right-clicking the window. This file can then be saved as .rft (Rich Text Format) and read directly by a word processor with all plots. Open the notes window by clicking Dockable windows on the View menu and clicking Notes.
50
View
Add to Favorites
Plots and lists can be added to the Favorites dockable window. For more, see the Favorites section earlier in this chapter.
Add to Report
Plots and lists can be added to the HTML report generator. For more, see the Report Generator chapter.
Full Screen
You can use the Full screen command to maximize the plot area. Full screen can be toggled on and off using F9 and is available from the Window toolbar as well as from the View menu. When Full Screen is selected from the menu, the Window toolbar is automatically opened as a floating toolbar.
51
Plot labels
Select the plot label to display as default on plots displaying the experiments, for example residual plots, by clicking None, Experiment Number, Experiment Name, or Run Order taken from the worksheet.
Number format
Click the number of decimals 2 - 5 or the scientific (E) format in lists by clicking Auto, .00, .000, .0000, .00000, or Scientific. When exporting unscaled coefficients to use in other applications, use the scientific format to get maximum precision in the coefficients.
52
View
Audit trail
When the Audit Trail is turned on each investigation in MODDE has a separate audit trail. Each audit trail consists of one or more sessions that in turn consist of events. A new session is started and appended to the audit trail when an investigation is opened, and ends when the investigation is saved. In addition to logging events, MODDE logs information about the user, and date and time of the events. To view the audit trail, click the Audit Trail tab in the Output / Notes / Audit trail dockable window. If this tab is not shown, display it by clicking View | Dockable Windows | Audit Trail.
Alpha level
Select the significance level by clicking 1%, 5%, or 10% to be used in the Lack of fit (LoF) plot. The default is 5%.
54
View
Coefficients
In MODDE you can select the type of coefficients you want to display. The default is Scaled and centered coefficients.
PLS orthogonal
The PLS orthogonal coefficients are available only when fitting the model with PLS. The PLS orthogonal coefficients are the PLS model coefficients re-expressed to correspond to factors centered and scaled using orthogonal scaling (coded as -1 to 1). PLS orthogonal coefficients are not available in investigations with formulation factors only. With process and mixture factors, the PLS orthogonal coefficients refer to process factors scaled orthogonally, and mixture factors unscaled (original units).
Note: The PLS orthogonal coefficients are meant for comparison with MLR only. They are incorrect when the design is not balanced and the mean is not equal to the mid-range. Normalized
To make the coefficients comparable when responses (Y's) have different ranges, you can select to display the coefficients in Normalized form that is, the coefficients are divided by the standard deviation of their respective response. The normalized mode is the system default for the Coefficient Overview Plot.
55
Unscaled
The unscaled coefficients are the coefficients corresponding to unscaled, uncentered data. When exporting unscaled coefficients to use in other applications, be sure to use the E-format to get maximum precision in the coefficients.
Blocking
You can select to have the block factor treated as a fixed or random effect and the predictions computed accordingly. Select the block factor as Fixed effect when the external variability can be set at will and the primary objective for blocking is to eliminate that source of variability Select the block factor as Random effect when the external variability cannot be controlled and set at will, and the primary objective is to make prediction without specifying the block level, and taking into account the external variability. See the Statistical Appendix for more information.
Confidence level
Select the confidence level 90%, 95%, or 99% for computation of the confidence interval on the model coefficients, effects, and predictions. The system default is 95%.
56
View
List presentation
By default lists are displayed using the Regular - shows all orthogonal settings option, when there are qualitative factors. And the design matrix by default displays The current Worksheet scaled and centered.
Qualitative factors - Extended or compact format in the model list and design matrix
For qualitative factors at p levels (p > 2), MODDE generates p - 1 dummy variables numbered from 2 to p. Click Extended shows all settings to display all the p levels of the qualitative factor or Regular shows all orthogonal settings to display only p - 1 levels of the qualitative factor.
R2 in plots
Click R2 or R2 adjusted to display in the Summary plot and PLS Summary plot. The default is to display R2.
57
Residuals
Select the type of residuals to be used in the residual plots.
Raw residuals
The raw residual is the difference between the observed and the fitted (predicted) value.
Standardized residuals
The standardized residual is the raw residual divided by the residual standard deviation (RSD).
Default
With MLR and 2 or more degrees of freedom, deleted studentized residuals are the MODDE default when plotting residuals. With PLS and models with less than 2 degrees of freedom, MODDE uses as default the Standardized residuals.
Select Factor
Select the factor to display as default in the Residuals | vs. Variable plot and Effects | Main Effect plot.
General Options
The General Options has three pages with settings that are used for all investigations.
General with general settings such as automatic fit. List Options like coloring and grid styles. Restore with buttons to restore plot settings, favorites, and dont show again-messages.
Click General Options on the View menu to access the options. Click the Show Details button to display a short description of the marked option.
58
View
General page
The General page is divided into the parts General, Investigation Settings, Program Limits, and Audit Trail.
General
Under General you can change the following options:
With Automatic fit = Yes, each investigation opened is automatically fitted. This is MODDE's default. If you set Automatic Fit = No MODDE will not fit when you open, nor when you make changes in the worksheet, but will still fit when you make changes using the interactive exclude tool. With Automatically display log = Yes, the Output window is automatically opened when opening an investigation. This is MODDE's default. In Most recent file list you can enter the number of investigations you want to see in the recent file list under the menu File. MODDE's default is 6. When Show expanded design factors in RED-MUP worksheet = No, the expanded design factors will be hidden. This is MODDE's default. For more see the section on RED-MUP in the Design appendix. In Theme you can change the look of MODDE. The default is Office 2003 Classic Theme. Available are also Office 2000, Office XP Classic Theme, Windows XP, and Whidbey. OpenGL plots = No is the default and recommended for the Contour and Sweet Spot Plots. In 3D scatter plots, OpenGL is always used.
59
Investigation settings
The available investigation settings are:
Max number of layers in Onion Design to limit how many layers that will be possible to make. MODDE's default is 10. Replicate tolerance is the number used when MODDE decides whether experiments can be considered to be replicates or not. MODDE's default is 0.1, that is, 10% of the range for each factor. Show units in worksheet = Yes results displaying the specified factor and response units in parenthesis after the factor resp. response names.
Program limits
Under Program Limits you can set the Maximum candidate set size. The size of the candidate set in MODDE is by default limited to 512 000 rows when MODDE creates the candidate set for you. You can change this limit. The maximum size of the candidate set that you can create and generate a design from is limited by the RAM in your computer.
Audit trail
Prompt for extended user info, found under Audit Trail, is by default No. If you want MODDE to prompt the user to enter extended user information, change this setting to Yes.
List Options
Use this page to change the default colors on lists. Other options available in tab list options are:
Default number format Print lists in color Headers in bold Threshold for the correlation matrix
60
View
Restore
Use the Restore page to restore Umetrics defaults for Plot Settings, Favorites, and 'dont show again'-messages.
Customize
Use Customize to customize toolbars and menus as in Office 2003 and later. Click Customize on the View menu to open the dialog. With the Customize dialog open you can interactively customize all available toolbars and menus by dragging an item to a new location.
Customize toolbars
On the Toolbars page you can:
Select which toolbars to display by clearing or selecting the available toolbars. Create new toolbars by clicking New. Rename toolbars you have created Delete toolbars you have created. Reset MODDE toolbars to their default content and position.
Customize commands
On the Commands page you can customize available menus and create new menus.
61
Customize options
On the Options page you can select:
To Always show full menus. To display Large icons. To Show ScreenTips on toolbars. To Show shortcut keys in ScreenTips. Menu animations among: (System default), Random, Unfold, Slide, Fade, None.
Customize keyboard
On the Keyboard page you can select to assign and reassign shortcut keys to new and old commands.
62
Design
Introduction
When you create or modify a design, some or all items on the Design menu are used. The available items are:
Factors and Responses: Define and modify factors and responses. Constraints and Inclusions: Define constraints and inclusions. Objective: Specify and modify objective, model and design D-Optimal: Generate and evaluate D-optimal designs, open the candidate set for viewing or editing, view onion plots. Design wizard: Open the design wizard.
Factors
Click Factors on the Design menu to open the factor spreadsheet. To modify one of the factors listed, mark it and press the ENTER key on your keyboard or right-click the factor and click Edit. To add a new factor, double-click the last line or right-click the spreadsheet and click Add Factor. For details, see the Factors chapter.
Responses
Click Responses on the Design menu to open the response spreadsheet. To modify one of the responses listed, mark it and press the ENTER key on your keyboard or right-click the response and click Edit. To add a new response, double-click the last line or right-click the spreadsheet and click Add Response. For details, see the Responses chapter.
Constraints
Click Constraints on the Design menu to open the Constraints dialog. To add a constraint, use the spreadsheet part or the graphical part of the dialog. For details, see the Constraints section in the Constraints and inclusions chapter.
63
Inclusions
Click Inclusions on the Design menu to open the inclusions dialog. In this dialog you can do the following:
Import runs from a text file or another investigation to inclusions. Import the current worksheet. Add the current inclusion runs to the current worksheet. Open the D-Optimal page in the design wizard.
For details, see the Inclusions section in the Constraints and inclusions chapter.
Objective
Click Objective on the Design menu to open the objective pages. In the objective pages you define the purpose of the design by selecting Screening, Response Surface Modeling (RSM), Split Objective, or Paste Data. Click Next to select which design you want MODDE to create. When selecting Paste Data, clicking Finish opens the worksheet allowing you to paste the data. For details, see the Objective, model and design chapter.
D-Optimal
When you have created a D-Optimal design a number of items are available under DOptimal on the Design menu. For all D-Optimal designs you can click:
Generate to regenerate the design. Candidate set to view the current candidate set and which design runs that are included in the current design. Evaluate to view the D-Optimal results and select another one of the generated designs. Design Plot to create the D-Optimal design plot.
For Onion designs you can additionally create onion plots by clicking Onion plot and Onion 3D Scatter. These plots display the current candidate set colored according to layer and the experiments currently included colored black. For details, see the D-Optimal chapter.
Design wizard
The design wizard is automatically opened when starting a new investigation from File | New. It can also be opened after creating an investigation by clicking Design Wizard on the Design menu or by pressing CTRL+W. The design wizard includes pages for creating/modifying factors, creating/modifying responses, defining/modifying a constraint numerically, objective and design selection, and when applicable the D-Optimal pages with the possibility to define inclusions. For details, see the Design Wizard chapter.
64
Factors
Introduction
Factors are variables that can be varied, or vary, during an experiment. A typical example of a factor is amount of raw material or temperature. Open the Factor spreadsheet by clicking Design | Factors or File | New, and then Next. In the Factors spreadsheet (window), you define (enter), modify, and delete factors. MODDE supports quantitative, qualitative, and mixture factors. Quantitative factors may be used in a transformed metric. When factors are transformed, the design is made in the transformed units, but the worksheet is expressed in original units.
Double-clicking the last line of the spreadsheet or Right-clicking the spreadsheet and clicking Add Factor.
65
The Factor Definition dialog is divided in an upper part and a lower part. The upper part displays Factor name, Abbreviation, and Units and is available independently of what is displayed in the lower part of the dialog. The lower part has two pages: General, which is the default page when opening the dialog, and Advanced.
Factor name
Enter the Factor name with up to 50 alphanumeric characters.
Abbreviation
The Abbreviation is automatically filled with the first 3 characters of the factor name. You can change the abbreviation as desired using up to 5 characters. The abbreviation is used as plot label in plots, in the Model and Confoundings lists found under the Show menu, in Worksheet | Correlation | Matrix, and in Edit | Generators.
Units
Enter the unit of the factor (optional). The units are displayed in the factor spreadsheet and can optionally be displayed in the worksheet, see the General page section in the View chapter for more.
General page
On the General page of the factor definition dialog you select which type of factor you are defining in Type of factor, the factor settings in Low and High or Settings, and how it is used in Use.
66
Factors
Note: MODDE is limited in the precision of the factor values. Factors ranging over a larger range than that, for instance. low at 0.0001 and high at 10000, cannot be treated as a factor.
For quantitative multilevel and qualitative factors the levels planned to be used should be entered. For a qualitative factor each entry in Settings has to be text, optionally including numbers, while for quantitative multilevel each entry has to be a number.
Note: In the same experiment you can have both mixture factors and regular process factors defined as quantitative or qualitative. Up to 12 factors are allowed when both process and formulation factors are defined in the design.
Quantitative (default) Quantitative factors are continuous factors defined at two levels, Low and High. To define more than two levels, see Quantitative multilevel. Up to 32 factors are allowed for screening designs and up to 20 for RSM designs. Quantitative multilevel To specify more than two levels for a quantitative factor click Quantitative multilevel under Type of Factor. MODDE supports up to 24 levels for quantitative multilevel factors. Constraints are not allowed with this type of factor, and the available designs are D-Optimal, Mixed full factorial, and three level designs when applicable. Up to 32 factors are allowed for screening designs and up to 20 for RSM designs. Qualitative To specify a qualitative factor, click Qualitative under Type of Factor. Qualitative factors are discrete. For a qualitative factor, the levels should not stand in relation to each other. If the levels are a range although discrete then the factors should be defined as quantitative multilevel. MODDE supports up to 24 levels for qualitative factors. RSM designs cannot be created with only qualitative terms. With one or more quantitative terms present up to 20 extended qualitative terms are allowed for RSM designs.
67
Formulation To specify a mixture factor, click Formulation under Type of Factor. Define a mixture factor as Formulation, when it is not an inert filler. Define the experimental range of the mixture formulation factor by entering its Low and High values. Up to 20 formulation factors are allowed in screening designs and up to 16 in RSM designs. Filler Specify a mixture factor as filler, by clicking Filler under Type of Factor, when:
It is always present in the mixture. That is, the sum of the High of the other mixture factors does not exceed 1. It accounts for a large percentage of the mixture. There are no restrictions on its range. Rather that factor is added at the end to bring the mixture total to the desired amount, 1 (100%). You are NOT interested in estimating the effect of the filler per se.
For a filler factor, the experimental range Low and High values are grayed out as it will be calculated as 1 minus the sum of the other mixture factors.
If the conditions are not met, MODDE issues a message, and changes the filler factor to formulation. See the Statistical appendix for more details on mixture factors.
Use
Under Use you can select how the factor will be used in the design. A factor can be Controlled, Uncontrolled, or Constant. Controlled (default) When you can control the settings of the factor, leave the default under Use (Controlled). These factors can be regular process factors (i.e. pH, Temp, etc.) defined as quantitative, quantitative multilevel or qualitative, or mixture factors, defined as formulation or filler. A Filler factor can only be defined as Controlled. Uncontrolled and Constant are unavailable for a filler factor.
68
Factors
Uncontrolled Define a factor as Uncontrolled (under Use) if you cannot control it but want to measure and record its value. Examples of such factors are ambient temperature or humidity. Mixture factors (defined as Formulation or Filler) cannot be uncontrolled and the option Uncontrolled is therefore unavailable. Constant Define a factor as Constant (under Use) when you want the worksheet to display a setting of a factor that is not changed. Quantitative, qualitative, and formulation factors can be defined as Constant factors. Quantitative and qualitative constant factors are displayed in the worksheet and counted in the model toolbar. When mixture factors are constant, the mixture total T for the controlled mixture factors is equal to: T = 1 - [constant mixture factors]. Multilevel quantitative and filler factors cannot be defined as Constant.
Advanced
For quantitative and quantitative multilevel factors the Advanced tab offers the possibility to transform and change the MLR scaling. For regular quantitative factors it is also possible to define number of decimals to be used for the factor from this page.
69
Transformation
To transform a factor, click the Advanced tab, and in the Transform box click the transformation of your choice. When you transform a factor, the design is created in the transformed units, but the worksheet is expressed in original units. Hence transformation of a factor will change the center point and the star point values in the worksheet. All transformed factors are displayed with a ~ (tilde) near the name in lists and plots. The following transformations are available:
Transformation
None Lin Log Negative Log Exp Logit Power
Description
Default C1 * Y + C2 Log10(C1 * Y + C2) -Log10(C2 - C1 * Y) e(C1 * Y + C2) Log10((Y - C1)/(C2 - Y)) (C1 * Y + C2)C3 where C3 can be any value from -2 to 2.
When a transformation is selected (except None), the relevant constant fields are displayed. The field C3 is only displayed for the power transformation.
70
Factors
Mixture factors When fitting the model with MLR, the mixture factors are not scaled. You can select to fit the model in pseudo components. This is recommended when the mixture region is regular. For investigations containing both process and mixture factors, by default process factors are orthogonally scaled and the mixture factors are unscaled. The coefficients displayed as scaled and centered correspond to this default scaling of the variables i.e. mixture unscaled and process orthogonally scaled. If you select to display the unscaled coefficients, they correspond to all factors unscaled, including the process factors.
Note: You should select the same scaling for all the factors, the system default is recommended. PLS scaling
When fitting the model with PLS, all factors including mixture factors are always scaled and centered to unit variance. For mixture factors, when you select pseudo components, the mixture factors are first transformed to pseudo components and then scaled to unit variance (pseudo components can be switched on/off).
Number of decimals
In the Advanced page of the factor definition dialog you can select Number of decimals. The values for number of decimals are: Free, 0, 1, 2, 3, 4 and represent the number of digits displayed after the decimal point. The Number of decimals-value should correspond to the precision with which the factor can be set in your equipment. It is important that it is not set too low since after setting this value all values for that factor will be rounded accordingly in the worksheet. If you do not know the precision of the instrument, leave No. of decimals 'Free'. 'Free' means that no rounding off of the results and values corresponding to this factor will take place. For example, the settings of a factor in a CCC design, is its high value (in orthogonal scaled units) multiplied by the 4th root of the number of runs in the factorial part of the design. If the precision of this factor is set to 0 all decimal digits are removed from this factor setting in the worksheet and computation will take place using the values in the worksheet.
71
Factors can be added by double-clicking the last row of the factor definition spreadsheet or right-clicking the spreadsheet and clicking Add Factor. To copy factors, mark the factors to copy, click Edit | Copy or press CTRL+C, and click Edit | Paste or press CTRL+V. MODDE copies the factors and adds a digit after the name when pasting to make it unique.
72
Factors
Adding a factor
To add a factor through the factor definition dialog use one of the following methods:
Double-click the last line of the factor definition spreadsheet. Right-click the spreadsheet and click Add Factor. Open View | Design Wizard and click New in the Define factors page.
Quantitative factors can be added by typing on the last (empty) row starting with entering the factor name, see the example:
Modifying a factor
It is possible to edit the fields Name, Abbr., Units, and Settings directly in the spreadsheet. To modify any of the other fields, double-click that factor in the factors spreadsheet. The Factor Definition dialog opens with the attributes of the factor to modify. The Factor Definition dialog can also be opened by clicking View | Design Wizard and clicking Edit in the Define factors page.
Copying a factor
Factors can be copied and pasted in the factor spreadsheet by clicking Edit | Copy and then Edit | Paste or pressing CTRL+C and then CTRL+V.
73
Deleting a factor
To delete a factor, mark it in the factor definition spreadsheet and use one of the following methods:
Press the DELETE key on the keyboard. Right-click and click Delete. Click Edit | Delete.
You are warned of the possible consequences. Click Yes/No in the confirmation box.
74
Responses
Introduction
A response is the result from an experiment. A typical example of a response is yield. Open the response definition spreadsheet from Design | Responses or File | New, Next, Next. In the response definition spreadsheet, you define (enter), modify, delete, copy, print, and list responses. MODDE supports only quantitative responses. Responses may be transformed, and MODDE supports several transformations. For transformed responses, predictions, contour plots and 3D plots, are back transformed to original units.
Clicking Design | Responses and double-clicking the last line of the spreadsheet or Right-clicking the spreadsheet and clicking Add Response.
75
Response name
Enter the response name with up to 50 alphanumeric characters in the Response name box.
Abbreviation
The Abbreviation is automatically filled with the first 3 characters of the response name. You can change the abbreviation as desired using up to 5 characters. The abbreviation is used as plot label in plots and in Worksheet | Correlation | Matrix.
Units
Optionally enter the unit of the response in the Units box. The units are displayed in the response definition spreadsheet and can optionally be displayed in the worksheet, see the General page subsection in the View chapter for more.
Limits
Fill in the Min, Target, and Max fields when that information is available to you. These values are then automatically used in the Design Space, Sweet Spot, and Optimizer windows.
76
Responses
Regular responses
Regular responses are the standard responses measured and fitted in the current investigation. Regular responses can be transformed and it is also possible to change the modifier for PLS scaling.
Transformation
The following transformations are available:
Transformation
None Lin Log Negative Log Exp Logit Power
Description
Default C1 * Y + C2 Log10(C1 * Y + C2) -Log10(C2 - C1 * Y) e(C1 * Y + C2) Log10((Y - C1)/(C2 - Y)) (C1 * Y + C2)C3 where C3 can be any value from -2 to 2.
When a transformation is selected (except None), the constants in the formula are entered in the fields displayed after selecting a transformation. The C3 field is only displayed for the power transformation. Specifying a transformation for a response is done to get the best mathematical fit of the estimated function.
Note: You can specify or modify the current transformation by right-clicking the Histogram plot and clicking Transform.
MLR scaling
When fitting the model with MLR no scaling of responses is available.
PLS scaling
When fitting the model with PLS it is possible to scale to unit variance with or without a modifier. With the default scaling option, the responses are centered and scaled to unit variance when fitting.
77
Autoscale modifier
In the PLS Scaling box, click Autoscale Modifier to change the modifier. Leaving the modifier at the default=1 gives the same result as when selecting Unit Variance. Enter a different value of the modifier and the response will be scaled to unit variance multiplied by the value of the modifier.
Note: To keep a response out of the analysis set its autoscale modifier to zero (i.e., enter 0 in the edit field).
Derived responses
A Derived response is a computed response as function of the factors and/or fitted regular responses. When you add a derived response, you enter its formula. Derived responses can be edited and deleted. Derived responses are displayed in the responses spreadsheet. The values of the derived responses are entered automatically in the worksheet when the model is fitted. The derived responses are also available for all plots and list under the Worksheet and Prediction menus and can be used as regular responses under these menus.
Note: Derived response values are only available after fitting (clicking Fit on the Analysis menu) the model. When responses are included in the formula, MODDE uses the fitted (predicted by the model) values of the responses in the computation.
78
Responses
When clicking the Edit button, the Derived Response Wizard opens. The first page contains information about derived responses. Select the Dont show this page again check box if you do not want to see this information again.
79
Clicking Next opens the next page in which you Enter the formula for the derived response.
When you click Finish, MODDE parses the formula for correctness, and only computes and displays the derived response in the worksheet, when you fit the model. The derived response is added to the responses spreadsheet and the worksheet.
Note: Derived responses are deleted when regular factors are deleted, or changed and when responses that are part of the derived response are deleted.
Responses
Integer, Variables (factors or responses), List of Integers and Sets of Variables Integer constant and floating points Operator : denotes a sequence, i.e., from: to, for example 6:8 means 6,7,8 List of integers, such as 1,3,5:8 is the same as 1,3,5,6,7,8 Variables (factors or responses) are denoted by vint, where int refers to the variable number in the worksheet (i.e. v5, v15, for variables 5 and 15) A set of variables (matrix) is denoted by v[int1,int2, int3:int4,int6] with square brackets. Int refers to the variable number in the worksheet. For example v[1,5,7:10] refers to the set of variables v1, v5, v7, v8, v9, and v10.
Note: To denote a set of variables you have to use the square brackets [ ], and not regular parenthesis ( )
Functions
The functions available are Log10, Ln (natural log), and Exp (exponential).
A set of variables with a constant A set of variables with a single variable A set of variables with another set of variables of the same size (they are added pair wise). A single variable with a constant or a single variable
Note: Power, multiplication and division cannot be applied to a set of variables and another set of variables. The first operand can be a constant, a variable or a set of variables, but the second and following operands must be a single variable or a constant.
81
Additional operators
The following additional operators apply to variables or sets of variables:
Avg(v[int1:int2]) Stdev(v[int1:int2]) Sum(v[int1:int2]) Average of variables vint1... to vint2 Standard deviation of variables vint1... to vint2 Sum of variables vint1... to vint2
Note: The parser is not case sensitive (t and T mean the same thing).
Linked responses
A Linked response is a response available in one investigation but fitted in another. Linked responses are no longer available in MODDE. Investigations containing linked responses are converted in MODDE 9 to instead hold different models and/or worksheets. MODDE supports as many models in one investigation as there are responses.
82
Responses
83
Adding a response
To add responses, double-click the last entry in the Responses spreadsheet window. Enter the responses name, abbreviation etc, in the Response Definition dialog.
Modifying a response
The fields Name and Abbr. can be edited directly in the Responses spreadsheet. To modify other fields for regular or derived responses, mark the desired response in the responses spreadsheet and:
The Response Definition dialog opens with the attributes of the response to modify.
Deleting a response
To delete one or more responses, mark the response(s) in the responses spreadsheet and press the DELETE key on the keyboard or click Edit | Delete. A dialog is displayed to confirm the deletion.
84
Constraints
A common problem is that experimentation may not be possible in some region of the experimental space. For example it may not be possible, in an experiment, to have high temperature and simultaneously low pH, and you want to cut-off the corner High temp, Low pH. In MODDE this is solved by adding a Constraint using Design | Constraint. A linear constraint is a function of the factors that specify a part of the experimental region to be included or excluded. The resulting experimental region is an irregular polyhedron. The corners of this region are called the extreme vertices; they constitute part of the candidate set, i.e. a discrete set of potentially good runs. D-optimal designs are the only designs available when the experimental region is constrained to an irregular polyhedron. Constraints can be defined for quantitative or formulation factors.
Specifying constraints
Enter your constraints in the Constraints spreadsheet. The Constraints window is opened by clicking Constraints on the Design menu. In the upper part, the spreadsheet, you define each constraint (one per row) as a mathematical relation. In the lower part, the graphical view, you can define constraints, to be added to the upper part, geometrically. Such constraints may include two factors only and are shown in the upper part after clicking the Add button.
85
Constraints supported
MODDE supports linear constraints, specified as exclusions, for quantitative process factors or mixture factors Xk of the form
The first constraint specifies to exclude experimental runs where the sum of X3, X4, and X5 is < 0.6 The second constraint specifies to exclude experimental runs where the sum of X4 and X5 < 0.3
86
3. 4.
87
Note: Click a row in the spreadsheet defining a constraint in two factors and MODDE displays the graphical constraint.
88
2. 3. 4.
Note: To sort the candidate set in order to find undesirable factor combinations you need to create a D-Optimal design, click Design | DOptimal | Candidate Set, and then click Edit | Sort.
89
Inclusions
In MODDE designs can be augmented either using Complement Design (see the section Complement Design in chapter File menu) or Inclusions. Inclusions are extra runs that will be part of the worksheet. You can include a set of experimental runs (inclusions), either at the end of the worksheet or to be part of a DOptimal design. The Inclusions spreadsheet can be opened by clicking Design Inclusions or by clicking the import or Edit buttons in the Change D-Optimal settings page in the Design Wizard.
The extra experiments to include are found in another investigation or in a text-file as complement design can only complement the current investigation. The experiments were not saved in a MODDE investigation or a text file. In Inclusions such experiments can be entered manually or pasted. When adding experiments after the design has already been created. That is, when the inclusions should not be part of the design generation. The desired design should be a classical design. Use Fold over or Estimate square terms in a screening design. When using inclusions D-optimal designs is the only available choice. The desired design should include star points. Use Estimate square terms in a screening design and change the Star distance.
Note: The inclusions are added to the worksheet only when you click the Add to Worksheet button
90
Note: When generating D-Optimal designs, and the Include in design check box is selected, the inclusions are a part of the design and included in the number of runs.
91
Generating inclusions
Click Design | Inclusions to open the Inclusions spreadsheet, or click the Edit button in the Inclusions area in the D-Optimal page in the design wizard. In the inclusions spreadsheet you can add experiments by typing or pasting, by importing from file, and by importing the current worksheet. You can do several imports.
Importing worksheet
Click the Import Worksheet button to import the current worksheet to use as inclusions.
Modifying inclusions
If the inclusions already exist, clicking Design | Inclusions or the Edit button in the DOptimal page in the design wizard, opens the spreadsheet with the inclusions for editing. To delete rows, mark them, press the DELETE key or right-click the spreadsheet and click Delete.
92
Introduction
The Objective is the purpose for creating the design. MODDE recognizes two objectives: Screening (first stage of an investigation when little is known) and Response Surface Modeling (RSM) (optimization with the important factors.). The Split Objective supports both screening and RSM. Paste Data defaults to the screening objective with a linear model. After defining your factors and responses, clicking Design Objective opens the design wizard which guides you through the selection of objective, design, and model of the investigation. The following sections describe the definition of the objective and selection of model and design. See the Design Appendix chapter for details concerning the available designs. There are two dialogs associated with the menu item Objective:
First: the Select Objective dialog to select the purpose. Second: the Select Model and Design dialog to select the type of design and model.
93
Screening objective
Select the Screening objective when:
You are starting an investigation and know little about the effects of the factors on the response, the behavior of the response in the experimental region, or the true size of that region. The goal is to reduce the number of factors to those with the largest effect on the response.
This objective is available for all types of factors and factor combinations.
RSM objective
Select the RSM objective when:
A lot is known about the investigation i.e. important factors, the size of the region etc. The goal is to approximate the response by a mathematical model for the purpose of prediction, optimization or finding a region of operability.
Split Objective
Select the Split Objective when the investigation holds both process and mixture factors AND you want to specify separate models for each. If you want to specify one model for both mixture and process factors, select Screening or RSM as objective. The split objective is only available when there are both process and mixture factors available.
94
Paste Data
Select Paste Data when you have the design and want to paste it instead of MODDE creating one for you. After selecting Paste Data and clicking Finish, the MODDE worksheet will expand dynamically to fit the size of the pasted data. Click Analysis | Evaluate to view the condition number of the current worksheet and model. See also the Import design from file section in the File chapter.
for the current objective and number of factors, with the number of runs and the associated models.
Designs in MODDE
The design is the protocol for varying the factors in each experiment. Thus the design is a set of experimental runs spanning the experimental region. See the Design appendix for more details concerning the designs available in MODDE.
95
Recommended designs
According to the selected objective and number of factors, MODDE recommends the two most likely designs. Continue with the recommended design or select another one by pointing and clicking or pressing the UP and DOWN arrow keys on the keyboard. MODDE recommends, whenever possible, classical designs. The recommendations are marked First and Second in the Recommendation column.
Note: By clearing the Show extended list of designs check box you can display all supported designs in MODDE available for the defined factors and selected objective.
Runs in design
In the Runs column the number of runs in the design is displayed. When there is a '+' and/or a '-' sign after the number that means that the number of runs can be changed for that particular design.
Model
MODDE supports polynomial models, such as linear, interaction, and quadratic. Third order terms such as cubic or three factor interactions may be added to the model in Design Edit Model after the design generation. The model for each design is listed in the Model column.
Screening models
Linear and Interaction models are appropriate for the screening objective. When the model you select is:
Linear MODDE generates the linear model. You may edit the model and enter selected interactions. Interaction MODDE generates the full interaction model, i. e. all the two factor interactions are included.
RSM models
Quadratic models are used for the RSM objective. For classical mixture designs cubic designs are also available. When the model you select is:
Quadratic MODDE generates the full quadratic model holding all two-factor interactions and all the square terms of all the factors. Special cubic or cubic MODDE generates models accordingly. Such models are only supported with mixture factors. Such models include all two-factor interactions, all square terms, and some or all cubic terms.
Split models
When selecting the Split Objective, the model for the process factors and the mixture factors can be specified independently of each other by clicking the Settings button.
96
Design runs
The Design runs box displays the number of runs for the selected design.
When there exists 2 or more fractional factorial designs of the same resolution, with different number of runs, the number of runs, in the Runs column, is marked with a +. MODDE defaults to the design with smallest number of runs. Use the Design runs arrow to select the larger design. For example, with the screening objective for 7 or 8 factors there exist two-resolution IV design, one with 16 runs and the other with 32. MODDE selects the one with 16 runs. To select the design with 32 runs, click the Design runs arrow. With D-Optimal designs the number of runs, in the Runs column, is marked with a + and - indicating that there exists smaller and larger designs. When augmenting a design D-Optimally, the number of runs includes the number of inclusions.
97
Center points
The Center points box displays the number of center points. MODDE always recommends 3 center points. To change the number of center points:
Type the desired number Click the Center points arrows and click a number. Mark the Center points box and press the up or down arrow keys on the keyboard.
Replicates
The Replicates box displays the number of times to replicate the whole design including center points. The default is '0', meaning that the design is not replicated. Enter '1' here to replicate the design once. To change the number of replicates:
Type the desired number Click the Replicates arrows and click a number. Mark the Replicates box and press the up or down arrow keys on the keyboard.
Total runs
After Total runs, the total number of runs included in the worksheet is listed and includes: runs in the design plus center points and replicates.
98
Settings
For some designs the Settings-button is active. Click the Settings button to:
Edit the generators and/or model for fractional factorial designs of resolution III, V, and V. Change the star distance for a CCC design. Specify the model when you have selected the split objective.
99
Note: The model and generators can be edited outside the design wizard by clicking Edit | Model and Edit | Generators.
For more details see, the sections Model / reference mixture and Generators in the Edit chapter.
100
Select the desired models and select the Add interaction between linear process and mixture factors check box to add interactions between the process and mixture factors. Click the Edit model button in the D-Optimal page to edit the models further.
Note: It is not possible to have interactions between the mixture factors as these are part of the quadratic model. Scheff Models are not supported for investigations with both process and mixture factors.
Description
Click the Description button, positioned below the Settings button, to display a short description of the selected design. To hide the text, click the button again.
101
Blocks
MODDE supports Orthogonal Blocking for the 2 levels Factorial, Fractional factorial, Plackett Burman, CCC, Box Behnken, and D-Optimal designs. The maximum number of blocks supported by MODDE is 8, with a minimum block size of 4. Select the number of blocks to include in your design from the Blocks box.
Orthogonal blocking
The method of dividing experiments into blocks, so that the block effect is uncorrelated with the main factor effects is called orthogonal blocking. Orthogonal blocking is a way to deal with extraneous sources of variability that are not included in the model. For example if one is making 32 experiments and the batches of raw material are sufficient for 8 experiments, one would like to run the experiments in blocks of 8 such as the variation between batches of raw material does not affect the estimate of the main factor effects. See also the Orthogonal blocking section in the Statistical appendix.
Block interaction
An interaction between a main effect and a block effect is called a block interaction. When the design supports the interactions between the block effects and the main effects, the Block interactions check box, in the Select the model and design page is active. You can select the check box if you want to add the block interactions to your model. For details see the Orthogonal blocking section in the Statistical Appendix.
102
D-Optimal
2. 3.
4. 5. 6.
103
This Change D-Optimal settings page consists of three sections relating to: 1. 2. 3. Design Generation Criteria Design Alternatives Candidate set
Design runs
Design runs is the number of runs the D-Optimal algorithm will generate, not including the center points. You can change this number as desired. The smallest number of runs accepted is the number of terms currently included in the model.
Model terms
The number of terms currently in the model is listed after Model terms. This number is updated after changes in Edit Model.
104
D-Optimal
Note: When the investigation contains only mixture factors, the Edit Model button is unavailable. The D-Optimal design is always generated from the full model specified in the design page.
With investigations containing both mixture and process factors, you can only edit the process factor terms and the interactions between mixture and process factors.
Potential terms
By default MODDE includes a set of potential terms, i.e. additional terms not included in your model that might be important. The objective is to select a D-Optimal design rich enough to guard for the potential terms. If you want your design to be just optimal for your specified model, clear the Use potential terms box.
Inclusions
To use runs available from file as inclusions, click the Import button. To edit the available inclusions, or paste/type runs to use as inclusions, click the Edit button. If you have specified runs as inclusions in the Design | Inclusion window prior to entering the design wizard, the Include in design check box found under Inclusions, is by default selected and the inclusions will automatically be part of the D-Optimal design. Clear this check box if you do not want the inclusions to be part of the DOptimal design (but rather manually added at the end of the worksheet). To add the inclusions after generating the worksheet, open the Inclusions under the Design menu, and then click Add to worksheet.
Degrees of freedom
Number of Degrees of freedom of the residuals is calculated as: Number of design runs Model terms +1 (when you have center points) The number of degrees of freedom recommended for D-Optimal designs in MODDE is at least five.
105
Repetitions
In the Repetitions box, select the number of designs you want to generate with the same number of runs, N. This will give a set of designs for each value of N.
If you want MODDE to only select balanced designs, select the Use balanced only check box. To be able to get a balanced design, the selected number of design runs must be a multiple of the number of levels of the qualitative factor. The number of design runs may be updated, if necessary, to be a multiple of the number of levels of the qualitative factor.
Note: It is not always possible to generate balanced designs. When MODDE does not succeed in generating a balanced design, it issues a message. In this case, to generate a design, you must clear the Use balanced only check box.
106
D-Optimal
The candidate set can also be opened for editing by clicking Design | D-Optimal | Candidate set.
107
D-Optimal results
When you click Next on the Change D-Optimal settings page, MODDE generates the D-Optimal designs and displays them in the D-Optimal results page.
By default, the best design according to G-Efficiency is selected. Use the Auto-select design by box to instead select the best design according to Determinant or Condition number. Or select another design manually by marking the design. Click any column header to sort the list. To see the D-Optimal results as a plot, select the Display as plot check box.
The columns with the grid pattern represent the currently selected design.
108
D-Optimal
Right-click the plot and click Properties to open the D-Optimal Design Plot dialog allowing:
Selection of which of the criteria to display, G-efficiency, Determinant, or Condition number under Show. Sorting on Type instead of Design (default) under Sort by. Selection of which layers to display when D-Optimal onion designs have been generated in the Show Layer box.
Click Finish to generate the worksheet. Any already existing design and worksheet will be deleted. Click Design | D-Optimal | Generate to regenerate the D-Optimal designs as many times as needed.
Note: The menu Design | D-Optimal | Generate is not available for onion designs. To generate a new onion design, select Design | Objective.
Generate
Click Design | D-Optimal | Generate to open the Change D-Optimal Settings page to re-enter the D-Optimal pages and generate a new set of D-Optimal designs. See the section D-Optimal pages in the design wizard previously for more.
Candidate set
Click Design | D-Optimal | Candidate Set to open the Candidate set for viewing or editing. See the section Candidate set previously for more.
109
Evaluate
Click Design | D-Optimal | Evaluate to open the D-Optimal Results page to view the other generated designs and their properties. Here you can also select one of the other designs as your worksheet by marking and clicking Finish. See the section D-Optimal Results previously for more.
Design plot
Select Design | D-Optimal | Design Plot to display the same information as Evaluate in a column plot. MODDE displays G-efficiency, Log of the determinant of X'X, and the condition number of X, for the D-Optimal designs. When you generate several D-Optimal designs with different N (number of runs), you can plot any of the D-Optimal criteria (G-efficiency, Condition No, Log of the determinant of X'X) as a function of N. This plot can also be created from the Design | D-Optimal | Evaluate and selecting the Display as plot box. See the Design appendix for more information.
Note: All of the statistics available in the D-Optimal Design Plot are computed from the runs selected D-optimally and do not include the possible center points added to the worksheet. With mixture factors, the condition number refers to the slack variable model with all mixture components scaled orthogonal.
110
D-Optimal
Candidate set
The D-Optimal onion design in MODDE is created from a candidate set. The candidate set can be created by MODDE, imported from one of the supported file formats, or imported from SIMCA-P.
111
The following is listed: 1. 2. 3. The number of the Layer (starting from inside). The number of Candidate Runs in the layer for imported candidate sets. The span of the layer defined by its % From (Percentile) and % To distance to the center of the multivariate space. You can change the span of a layer as long the number of runs in the candidate set remains one and half times larger than the number of design runs in that layer. Overlapping span between layers is not allowed. When you change the span of the layers, MODDE updates the number of candidate set runs in each layer. If the span of the layers overlap, or the number of runs in the candidate set is not large enough, the layer is colored in red and a message indicating the problem is displayed. You must fix the problem before clicking Finish. The number of Design Runs in each layer. You can change the number of design runs. The number of desired runs must be at least equal to the number of terms in the model. The recommended number of runs includes 3 degrees of freedom for the outer layer and 1 degree of freedom for the rest of the layers. Select the number of D-Optimal designs, in each layer, you want to generate with the same number of runs. The default number for Repetition is 1. The Model for each layer. You can change the model in each layer. Click the model you want to change and select another model or mark the layer and click Edit Model to customize the model. After editing a model, MODDE writes Model edited in the Comment column. MODDE updates the design runs according to the new selected model.
4.
5. 6.
112
D-Optimal
When you click Finish, MODDE generates several D-Optimal designs in each layer varying the number of runs by plus and minus 2, and displays the Onion D-Optimal results.
The table on this page displays, for every layer, the generated designs statistics. By default, in this table, the designs with the highest G-efficiency are selected. You can select a different design in a given layer by marking it in the list or using the Auto-select design by box and selecting a different criterion.
113
Candidate set
Click Design | D-Optimal | Candidate set to display the candidate set. The columns in the spreadsheet are as follows when the candidate set was imported: 1. 2. 3. 4. 5. Design Run number. Corresponds with the experiment number in the worksheet. The Exp Name (experiment name) when available. The Layer number (the innermost layer = 1). The distance to the center of the multivariate space in Percent. The design variables.
Runs in the candidate set that are not used to generate the D-Optimal design (not included in the selected percentile ranges), are colored light gray. When the candidate set is generated by MODDE the additional columns Layer and Percent are unavailable. The layer belonging is displayed in the onion plots and when listed (right-click and click Create List) the list is organized according to layer.
Evaluate
Click Design | D-Optimal | Evaluate to display the Onion D-Optimal results page. You can use the table to select different designs, in selected layers. See the D-Optimal Results section previously for more.
114
D-Optimal
Onion plots
There are two onion plots to visualize the candidate set and the selected D-Optimal design, the Onion Plot and the Onion 3D scatter. For the onion scatter plots, the property page has two tabs: Select Factors and Plot Labels. Use Select Factors to select which factors to display on the X, Y, and Z-axes. Use Plot Labels to select which labels to display in the plot.
Onion plot
To create the 2D onion plot, click D-Optimal on the Design menu, and then click Onion Plot. The onion plot is a 2D scatter plot of the candidate set. The candidate set runs are colored by layer, and the selected design runs are black.
115
Onion 3D Scatter
To create the 3D Onion plot click D-Optimal on the Design menu, and then click Onion 3D Scatter. The Onion 3D scatter plot displays the candidate set colored by layer with the selected design runs in black. You can customize the plot using the plot settings. For more, see the Plots and List chapter.
116
Design Wizard
Introduction
The Design Wizard opens when you click File | New, to create a new investigation, and guides you from the start of the investigation to the generation of the worksheet. The design wizard can also be opened by clicking Design | Design Wizard. Exit the wizard at any time by clicking Finish. The accelerator for the Design Wizard is CTRL+W. Following is a description of the pages in the Design wizard.
Defining factors
The first page of the design wizard is the Define Factors page. On this page you can define (enter new factors), modify, and/or delete factors.
117
All functionality available in the Factor spreadsheet described in the Factor chapter is available in this page of the design wizard. Specific for the design wizard factor page:
The New and Edit-button open the Factor Definition dialog. The Delete-button deletes the factor(s) currently marked. Selecting the Place constraints on the experimental region box and clicking Next opens the Constraints page of the design wizard. The Constraints spreadsheet, including graphical constraints, is described in the chapter Constraints and inclusions.
Note: The graphical constraint using the interface to define the constraint is not available from the design wizard. If you want to use the graphical constraint, click Design | Constraints
Clicking Next opens the Define responses page.
118
Design Wizard
Defining responses
In the Define responses page you can define (enter new response), modify, or delete responses.
All functionality available in the Responses spreadsheet described in the Response chapter is available in the design wizard. Specific for the design wizard response page:
The New and Edit-buttons opens the Response Definition dialog. The Delete-button deletes the response(s) currently marked.
119
Clicking Next opens the Select the model and design page.
120
Design Wizard
Note: You can sort the Design list on any column by clicking on its header.
121
The selected design is highlighted and marked with the worksheet icon.
Note: Sort the list according to any column by clicking its header.
122
Worksheet
Introduction
After you have selected the model and design, MODDE generates the worksheet. The worksheet is a spreadsheet used for storing the data (factors and response values) and is where you enter the experimental results. The worksheet is by default sorted in standard order. Whenever you change the design (by editing the generators or selecting another design) both the design and the worksheet are deleted and new ones are generated. MODDE issues a message whenever the worksheet is regenerated. The data in the worksheet are used when selecting to plot or list from the Worksheet menu.
Worksheet menu
After the worksheet has been created and you have entered results, there are a number of plots and lists available under the Worksheet menu. All responses, regular and derived, are available in the plots and lists of the Worksheet menu.
123
To re-randomize the run order after the worksheet has been created but before entering results, on the Worksheet menu, click Set Run Order. The run order can only be randomized for experiments with no results in the worksheet.
Fully randomized
With RSM designs and Screening designs at more than 2 levels, Fully randomized is the order in which you should perform the experiments and this is also the only available order for these cases.
After selecting To detect curvature and creating the worksheet you will find the three points above at run order 1-3 and the rest of the runs randomized.
124
Worksheet
Strong curvature
If the Curvature Diagnostic Plot exhibits strong curvature as is shown above, you should first re-measure the center point and re-do the plot. If the plot still exhibits strong curvature, reduce the ranges of the factors by 2 / 3 and restart the project.
No curvature
If the Curvature Diagnostic Plot does not exhibit curvature, as the one below, continue performing the rest of the experiments.
125
Click the Delete button in the dialog to remove factors or responses from the Yaxis or from Series. To change what to display on the X-axis, mark and click the arrow to add the new variable. Click the Color by Variable tab to color by a factor or a response. With the scatter plot created from the Worksheet menu you can also color by run order and experiment number.
You can customize the 3D plot using the property page and plot settings. See the Plots and List chapter.
126
Worksheet
Histogram
On the Worksheet menu click Histogram to display the histogram of the currently selected response. Select the desired response or add more responses by right-clicking the plot and clicking Properties, or by making the selection in the Response box.
Transform in Histogram
You can transform a response by right-clicking the Histogram plot and clicking Transform.
127
Descriptive statistics
Descriptive statistics can be displayed both as plot and in a list for all available responses.
128
Worksheet
Correlation
The linear correlation coefficients R between all the terms in the model and all the responses are displayed in the Correlation Matrix and Correlation Plot. Process factors are transformed, scaled, and centered as specified in the factor definition for MLR (default = orthogonal scaling). Responses are transformed as specified in the response definition. Formulation factors are always scaled orthogonally. The value of the correlation coefficient R represents the extent of the linear association between two terms. The value of R ranges from -1 to 1. When R is near zero there is no linear relationship between the terms.
Correlation matrix
On the Worksheet menu, click Correlation, and then click Matrix. Correlation coefficients above the threshold, between a term in the model and the responses are colored green and those between terms of the model are colored red.
To change the threshold, colors, or number format, right-click and click Properties.
Correlation plot
On the Worksheet menu, click Correlation, and then click Plot. The default plot displays the 10 largest correlation coefficients. To change number of correlations to display or limit the number of correlations according to a threshold: 1. 2. Right-click the plot and click Properties. Make the change, for instance click Show absolute correlations above threshold and enter a value. With '0' all correlations are displayed
129
Replicate plot
In the replicate plot the values of the response are plotted vs. experimental runs displaying the variation in the response for replicated experiments. On the Worksheet menu click Replicate Plot to display the plot. Select the desired response or add more responses by right-clicking and clicking Properties, or by making the selection in the Response box.
Note: When the response has been transformed the Replicate plot displays the back transformed values. To display the plot in the transformed metric, select the Show transformed values check box in the Options tab in Properties.
Replicated experiments
MODDE checks the rows of all the factors (both included and excluded) in the worksheet for replicates. Rows in the worksheet with the same factor values plus or minus a tolerance are considered replicates. The default Replicate tolerance is 0.1 (10%). You can change the Replicate tolerance used in General Options on the View menu, tab General. The replicates are used for the computation of the pure error and displayed on the same Replicate Index in the Replicate Plot.
130
Worksheet
Worksheet spreadsheet
Description of the worksheet
Experiment number: Exp No start with number one and are assigned sequentially. They cannot be edited. Experiment name: MODDE assigns a default experiment name, Exp Name, on the form Nxx, where xx is the experiment number. You may edit the name and enter your own identification. The Plot Label named Experiment Name displays the text entered under Exp Name when plotted. Run order: The Run Order is the order in which the experiments should be performed. MODDE suggests a randomized run order. Sort the worksheet according to run order before performing the experiments. Include or exclude: The fourth column labeled Incl/Excl indicates if the experiment is included or excluded from the analysis. When the worksheet is generated all experiments are marked Incl and are included in the analysis. To exclude an experiment from the analysis select Excl in the worksheet.
Note: Excluded rows are excluded from the analysis for all responses. To exclude the response value for only one response, right-click the cell and click Exclude value(s).
Factors: In the Factor columns the factors are listed in original units. Blocking: When you have selected Blocks in the Select the model and design page, the column $BlockV displays which block each experiment has been assigned to. For details on blocking, see the Orthogonal blocking section in the Statistical appendix. Responses: In the columns to the far right all responses are found. The response values are listed in original units.
131
Open the worksheet window, then click Edit | Add Experiments. Open the worksheet window, right-click and click Add Experiment. On the Design menu, click Inclusions. Type values or import the experimental runs you want added to the worksheet. Then click the Add to Worksheet button. The inclusions will be added at the end of the worksheet.
The first four columns of the worksheet are automatically filled in when adding experiments.
Right-click the desired column and then click Sort. Mark the worksheet, then click Edit | Sort.
132
Analysis
Introduction
After entering the response values in the worksheet, analyze the resulting data in the menu Analysis. Use the Analysis Wizard to guide you through from raw data analysis to interpretation and diagnostics. You can also manually fit the model to the data by using either MLR (Multiple Linear Regression) or PLS (Projection to Latent Structure) and clicking Fit. To review the fit, display as lists or plots, the summary of the fit, the coefficients, the effects, and the analysis of variance (ANOVA). Residual plots and the Box-Cox plot (only for MLR) are available for diagnostic purposes.
133
Evaluate
Click Evaluate on the Analysis menu to display an evaluation of the current worksheet and model. The table includes condition number, (number of) worksheet runs, (number of) model terms, DF (degrees of freedom) residual, DF lack of fit, and DF pure error individually for each response. These values differ between the responses when they have different models or excluded values.
Condition number
The Condition Number is the ratio of the largest and the smallest singular values of X (eigenvalues of X'X) where X is the extended design matrix. This condition number represents a measure of the sphericity of the design (orthogonality). All factorial designs, without center points, have a condition number of 1 and the design points are situated on the surface of a sphere. For more see the Statistical appendix. With factors orthogonally scaled all classical screening designs have a condition number =1 without center points and <3 with center points. The condition number of classical RSM designs varies according to the number of factors but remains <10.
Worksheet runs: the number of runs included in the Worksheet spreadsheet. Model terms: the number of terms included in the current model. DF residuals: The degrees of freedom of the residuals. DF lack of fit: the degrees of freedom of the Lack of Fit DF pure error (repl. runs): degrees of freedom of the pure error calculated using the replicated runs (experiments). If there are no replicated runs the pure error cannot be calculated.
134
Analysis
Analysis Wizard
The Analysis Wizard guides you through the main steps from review of the raw data and the fit, to the diagnostics and refining of the model. When opening the Analysis Wizard, the model is automatically fitted using the default fit method. In the Analysis Wizard each response is handled separately in a step wise manner where you start looking at plots, pruning the model etc. for one response, and when done you step over to the next response. The following features are available as buttons in the Analysis Wizard:
Display or hide Min, Target, and Max limits when available in the response definition. Transform the response in the Histogram Plot Exclude model terms or severe outliers for the current response. Undo after exclude. Edit model in the Coefficient Plot for the current response. Display regression line in the Residual Normal Probability Plot and Observed vs. Predicted Plot.
Note: All changes done in the Analysis Wizard take effect immediately. Clicking Close closes the Analysis Wizard but all changes remain.
135
136
Analysis
Note: If your X matrix has a condition number > 3000, MODDE will only fit the model with PLS and the condition number selecting MLR is displayed as infinite. Scheff MLR
Select Scheff MLR to fit the mixture data with a Scheff type model. When you select this fit method, the model is restored to its default specification. When your investigation contains both mixture and process factors Scheff MLR is unavailable as such models are only available in MODDE for experiments with mixture factors only.
137
Note: The default method of fit with the Cox reference mixture model is PLS. When the model obeys mixture hierarchy you can if you want fit the model with MLR. When fitting the model with PLS, the condition number refers to the X matrix, with unit variance coding. MODDE Analysis Advisor automatically opened
The advisor opens when you fit the model and guides through the analysis. To turn on or off the advisor:
On the View menu, click Dockable Windows, and then click Analysis Advisor.
138
Analysis
Note: R2 and Q2 provide the best summary of the fit of the model. R2 is an overestimated and Q2 an underestimated measure of the goodness of fit of the model Saturated models
R2 and R2 Adjusted are not available for saturated models, i.e. models with DF (degrees of freedom) equal to zero (see statistical appendices). However, with PLS Q2 can still be computed for saturated models and is hence displayed in plots and lists.
Statistics
In the plot footer MODDE displays: N = Number of experimental runs. DF = N-p (the degrees of freedom of the residuals). Cond. no. = Condition number of the extended design matrix coded as selected in the factor definition box: MLR scaling, or for PLS, unit variance.
The first column in the summary plot is R2 and is the fraction of the variation of the response explained by the model:
R 2 = SS REG /SS SSREG = the sum of squares of Y corrected for the mean, explained by the model. SS = the total sum of squares of Y corrected for the mean. R2 overestimates the goodness of fit. The R2 value is always between 0 and 1. Values close to 1 for both R2 and Q2 indicate very good model with excellent predictive power. For details see the Statistical appendix. You may select to plot R2 Adjusted instead of R2 by clicking it in the Properties page or set it to be default displayed by clicking Investigation Options on the View menu, tab R2. R2 Adjusted: The fraction of variation of the response explained by the model adjusted for degrees of freedom.
139
Q2
The second column in the summary plot is Q2 and is the fraction of the variation of the response predicted by the model according to cross validation and expressed in the same units as R2. Q 2 = 1 - PRESS/SS PRESS = the prediction residual sum of squares. SS = the total sum of squares of Y corrected for the mean. Q2 underestimates the goodness of fit. The Q2 is usually between 0 and 1. Q2 can be negative for very poor models. With PLS negative Q2 are truncated to zero for computational purposes. Values close to 1 for both R2 and Q2 indicate a very good model with excellent predictive power. For details see the Statistical appendix.
Model validity
The third column in the summary plot is the Model Validity and is a measure of the validity of the model. When the model validity column is larger than 0.25, there is no lack of fit of the model. This means that the model error is in the same range as the pure error. When the model validity is less than 0.25 you have a significant lack of fit and the model error is significantly larger than the pure error (reproducibility). A model validity value of 1 represents a perfect model. Validity = 1 + 0.57647*log(plof) where plof = p for lack of fit.
Reproducibility
The forth column in the summary plot is the Reproducibility which is the variation of the response under the same conditions (pure error), often at the center points, compared to the total variation of the response. Reproducibility = 1 - (MS(Pure error)/MS(total SS corrected)). MS = Mean squares, or Variance. A reproducibility value of 1 represents perfect reproducibility.
140
Analysis
For every fitted response the plot displays R2 and Q2. The definition for R2 and Q2 is the same as in the Summary of Fit Plot. You may select to plot R2 Adjusted instead of R2 by selecting it in the Properties page or setting it as default in View | Investigation Options, tab R2. The condition number is calculated for the extended design matrix with the factors scaled to unit variance.
141
This plot displays the R2 and Q2 per PLS component for the selected response. You can change the selected response using the Response box. You may select to display R2 adjusted instead of R2 in the Property page and in the Investigation Options menu.
Saturated models
R2 and R2 Adjusted are not available for saturated models, i.e. models with DF (degrees of freedom) equal to zero (see statistical appendices). However, with PLS Q2 can still be computed for saturated models and is hence displayed in plots and lists.
142
Analysis
Investigating diagnostics
Use the diagnostic plots and lists to find outliers, needed transformations etc.
Normality of the residuals. If the residuals are normally distributed, the points on the probability plot follow close to a straight line Outliers. These are points deviating from the normal probability line, and having large absolute values of studentized residuals i.e. larger than 4 standard deviation indicated by red lines on the plot.
143
Residuals
Under Analysis | Residuals three residual plots and one list are available.
144
Analysis
Right-click the plot and click Properties to select which variable to display on the xaxis.
Residual list
To display the Residuals List, on the Analysis menu, click Residuals, and then click List. The number under the response name, is the experiment number. Observed: The value of the response as listed in the worksheet. Predicted: The predicted value for that observation. Observed - Predicted: The residual for that observation. Confidence Interval: The 95% confidence interval on the predicted value. You can change the confidence level in the Property page and in Investigation Options.
145
146
Analysis
The Box Cox plot displays the maximum likelihood as a function of the power of the transformation by plotting values of lambda, , vs. the maximum likelihood. If the response values vary more than a magnitude of ten in the experimental domain, a transformation is often recommended. The maximum point on the Box Cox plot gives the value of (lambda, ) for the response transformation Y that gives the best fit of the model. This is the maximum likelihood estimator for . For more, see the Statistical appendix.
max = -1, use the Power transformation and C3 = -1. max = 0, use the Logarithmic (10Log) transformation. max = 0.5, use the Power transformation and C3 = 0.5.
147
The observed vs. predicted plot displays the plot of the observed values vs. the fitted or predicted values. Plots with the points close to straight line indicate good models. A regression line can be fitted by clicking the regression button on the Plot toolbar.
148
Analysis
The lack of fit plot compares the Lack of Fit (LoF) component to the pure error component and displays a graph with 3 bars.
SD-LoF: Shows the variation of the response due to the lack of fit of the model (i.e. the model error) adjusted for degrees of freedom and in the same units as Y. This is the square root of MS (mean square) lack of fit. SD-pe (Pure error): Shows the variation due to the replicated experiments (observations) adjusted for degrees of freedom and in the same units as Y. This is the square root of MS (mean square) pure error. SD-pe*sqrt(F(crit)): Shows SD pure error (second bar) multiplied by the square root of the critical F.
The critical F is the value of the F-distribution over which SD LoF is statistically significant at the 95% confidence level. Hence, when the third bar is smaller than the first, the lack of fit is significant at the 5% level (see statistical appendices).
149
ANOVA plot
To display the ANOVA plot, on the Analysis menu, click ANOVA, and then click Anova Plot.
In the ANOVA plot the regression component is compared with the residual component and 3 bars are displayed.
SD Regression: Shows the variation of the response explained by the model, adjusted for degrees of freedom and in the same units as Y. This is the square root of MS (mean square) regression. RSD: Shows the variation of the response not explained by the model, adjusted for degrees of freedom and in the same units as Y. This is the residuals standard deviation. RSD*sqrt(F(crit)): Shows RSD (second bar) multiplied by the square root of the critical F.
The critical F is the value of the F-distribution over which SD regression is statistically significant at the 95% confidence level. Hence, when the third bar is smaller than the first, the model is significant at the 5% level. For more see the Statistical appendix.
150
Analysis
ANOVA table
To open the ANOVA table, on the Analysis menu, click ANOVA, and then click Anova Table.
The analysis of variance (ANOVA) table is displayed for the selected response. If there are replicated observations, the residual sum of squares is further partitioned into PURE ERROR and LACK OF FIT. A goodness of fit test is performed by comparing the MS (mean square) lack of fit to the MS (mean square) pure error. See the Statistical appendix for more information.
Replicated observations
MODDE checks the rows of the factors in the worksheet for replicates. Rows in the worksheet with the same factor values plus or minus a 10% tolerance interval are considered replicates and used for the computation of the pure error. The replicate tolerance can be changed in General Options.
Note: The red coloring of the p-values always refer to the 95% resp. 5% levels. Footer of the ANOVA plots and tables
At the bottom of plots and tables the following is displayed: N = Number of Runs. DF = N-p degrees of freedom of the residuals. R2, R2 adjusted, Q2 and RSD (residual standard deviation).
151
Coefficient plot
To create the coefficient plot, on the Analysis menu click Coefficients, and then click Plot.
The coefficient plot displays the coefficients, when changing from 0 to high, for the selected response with the confidence interval as error bars. By default, the coefficients refer to the data scaled and centered. You can select which type of coefficient to display from Properties. Select Scaled and centered (default), Normalized, Unscaled, or PLS orthogonal (only available for PLS). When you have confounded terms in your investigation these terms are marked with a bracket #. Hoover over the column in the plot and you will get information about which term(s) it is confounded with. This information is available also in the coefficient overview plot.
152
Analysis
153
Coefficient list
To create the coefficient list, on the Analysis menu click Coefficients, and then click List.
P-values signaling non significant coefficients at the selected confidence interval are colored in red. For the selected response(s), this list includes: Terms: Under the response name the name of the terms included in the model are listed. To display the list for another response, or more than one response, use the Response bar or Properties. Coefficients: Value of the coefficient. Standard error: Standard error of the coefficient P Value: Probability to get the displayed value for the coefficient if its true value was zero. Confidence interval: The 95% confidence interval on the coefficient value. To select a different level for the confidence interval open Properties or Investigation Options.
154
Analysis
The Effects List displays the effects and their confidence intervals.
Note: When an expanded list of the qualitative levels is desirable, use the coefficients plots and lists to display or list the coefficients for every level of a qualitative variable instead of the effects plots and lists. Note that the effects for linear and interaction models are twice the corresponding coefficients. Sorted effects plot
To open the effects plot, on the Analysis menu, click Effects, and then click Plot. For process factors the values of the effects (computed as twice the MLR coefficients) are plotted sorted (in absolute value) in descending order. The 95% confidence interval is shown as error bars.
For mixture factors this plot displays the adjusted Cox effects (unavailable for Scheff models). This effect represents the change in the response values when component k varies over its range, all other mixture factors kept in the same proportion as in the reference mixture. For details on how the mixture effects are calculated, see Statistical appendix, section Mixture data in MODDE.
155
This plot, proposed by Daniel's in 1954, is based on the fact that if all estimated effects were noise, they would have a normal distribution and when plotted on a normal cumulative plot, would fall on a straight line. Hence effects significantly different from zero (noise) will fall outside the normal line. Note that this plot assumes independent effects, and that all estimable effects are plotted. Hence, it is only relevant for screening designs with saturated models DF = 0. Also for this plot to be meaningful, one needs models with at least 10 effects. If these conditions are not met MODDE will warn you that this plot may not be statistically correct.
Note: The normal probability of effects plot is not available with mixture factors
156
Analysis
Main effect plot for mixture factors When you selected to display a mixture factor Xk, this plot displays the predicted change in the response when Xk varies from its low to its high level, the relative amounts of all other mixture factors are kept in the same proportion as in the standard reference mixture. For details, see the Statistical appendix, section Mixture data in MODDE. With formulation factors you can adjust according to reference or to range using Properties.
157
Interaction plot
To open the interaction plot, on the Analysis menu, click Effects, and then click Interaction Plot. To switch to another interaction and/or switch the factor displayed on the x-axis: 1. 2. Right-click the plot and click Properties. In tab Interaction Effects select the term in the Interaction term box and/or select the factor in the Factor on X-axis box.
Note: This plot is only available for process factor. No interaction plots are available for mixture factors.
When you select a 2 factor interaction, the predicted values of the response, when one factor varies from its low to its high level, are plotted for both levels of the other factor, all remaining factors in the design being set on their average. If you have mixture factors in the model, these are all set at the standard reference mixture.
158
Analysis
3 factor interaction plot When you select a 3-factor interaction, the predicted values of the response, when a factor varies from its low to its high level, are plotted for all the combinations of levels of the other two factors.
159
Effects list
To open the effects list, on the Analysis menu, click Effects, and then click List. The values of the effects (twice the coefficients) are listed with their 95% confidence interval sorted (in absolute value) in descending order.
160
Analysis
The VIP plot displays the VIP values as a column plot sorted in descending order.
The VIP list displays the sorted VIP values and the PLS coefficients for all responses in the investigation.
161
Score Scatter Plot displays the score vectors t and/or u in a scatter plot. Use the score plots to reveal groups, trends, outliers, and similarities. The plot marks represent your experiments. Score Column Plot displays the score vectors t or u in a column plot. Loading Scatter Plot displays the loading vectors p, c, w and/or wc in a scatter plot. Use the loading plots to investigate the correlation between terms in your model. The plot marks represent the terms currently in the model and/or the response-variables. The response variables are displayed when plotting the c-vectors. Loading Column Plot displays the loading vectors p, c, w or wc in a column plot.
The score plots reveal groups, trends, outliers, and similarities. You can create three different score scatter plots: 1. 2. 3. T scores (for ex.: t1 vs. t2): t scores are windows in the X space displaying the objects as situated on the projection plane or hyper plane. U scores (for ex.: u1 vs. u2): u scores are windows in the Y space, displaying the objects as situated on the projection plane or hyper plane. T vs. U scores (for ex.: t1 vs. u1): Displays the objects in the projected X (T) and Y (U) space, and how well the Y space coordinate (u) correlates to the X space coordinate (t).
Score column plots can only be created for one vector at a time.
162
Analysis
The loading plots display the correlation between the X variables T(X) and the Y variables U(Y).
163
You can create four different loading plots: 1. 2. p loadings (for ex.: p1 vs. p2): These plots show the importance of the X variables in the approximation of the X matrix. w loadings (for ex.: w1 vs. w2): The w's are the weights that combine the X variables (first dimension) or the residuals of the X variables (subsequent dimensions) to form the scores t. These weights are selected so as to maximize the correlation between T and U, thereby indirectly Y. X variables with large w's (positive or negative) are highly correlated with U (Y). Variables with large w's are situated far away from the origin (on the positive or negative side) on the plot. c loadings (for ex.: c1 vs. c2): These plots display the correlation between the Y variables and the X scores T (X). The c's are the weights that combine the Y variables with the scores u; so as to maximize their correlation with X. Y variables with large c's are highly correlated with T (X). wc loadings (for ex.: wc1 vs. wc2): These plots show both the X-weights (w) and Y-weights (c), and thereby the correlation structure between X and Y. One sees how the X and Y variables combine in the projections, and how the X variables relate to Y.
3.
4.
Loading column plots can only be created for one vector at a time. For details see the Statistical appendix.
164
Prediction
Introduction
To make prediction using the model means:
Using contour plots for interpretation. Making predictions in the prediction plots and list. Optimizing for selection of best conditions. Investigating robustness to disturbances using the Design Space. Contour Plot Wizard Prediction Plot Wizard Response Prediction Plot Prediction List Scatter Plot Sweet Spot Plot Optimizer Design Space Validation - only briefly described here but more thoroughly in the Design Space chapter and Appendix D: Design Space.
Note: The Prediction menu applies to all responses, regular and derived.
165
Selecting responses
You can select to display all responses for the 2D contour plots but with more than 9 the plots become very small. The contours are not overlaid, but displayed next to each other. Select which response(s) to display by marking in Available responses and clicking the => and/or marking in Selected responses and clicking <=.
Using constraints
Select the Use constraints box if you want the available constraints to be displayed in the plot.
166
Prediction
Plot options
To select the resolution of the plot, that is, the grid calculated to create the contour plot, click Plot Options. Note that you can type a resolution value here or click one of the predefined resolutions. You can also select to scale the subplots equally, to lock the contour levels, to produce the plot with/without color and with/without contour level labels. For more, see the Customizing contour plots sections in the Plots and lists chapter.
2D contour
The 2D contour plot displays the predicted response values, spanned by two factors, in a response surface contour plot. For mixture the plot is spanned by three factors. On the first page of the contour wizard, under Plot type, click Contour. Under Inner plot type, if you have both process and mixture factors in the model, click Process or Mixture to indicate which factors you want to display on the inner axes. Select which responses to display and click Next.
167
To hide the constants, right-click the window, and clear View Constants.
Zooming subplot
To zoom in on a subplot, click Subplot in the Zoom plot menu from the Plot toolbar.
168
Prediction
4D contour
The 4D Contour plot displays the predicted response values, spanned by two factors, in 9 response surface contour plots in a 3x3 grid spanned by another two factors. For mixture the plots are spanned by three factors. On the first page of the contour wizard, under Plot type, click 4D Contour.
Note: The 4D plot is available for investigations with only process factors, and with both mixture and process factors, but not with mixture factors only.
Under Inner plot type, if you have both process and mixture factors in the model, click Process or Mixture to indicate which factors you want to display on the inner axes. Select which response to display and click Next.
169
Inner axes factors Select two process factors for the axes under Factors at the plot axes, or three mixture factors Outer axes factors Select two process factors for the outer axes. The response contours are plotted for the low, middle, and high levels of these factors. You can also select qualitative factors for the outer axes and select for which settings to display the contour plots. When there are more than 4 factors in the model, the remaining factors are held constant
170
Prediction
By default all plots are equally scaled, that is, the color coding is the same for all plots.
171
Response Surface
The Response Surface plot displays the predicted response values, spanned by two factors, in a response surface plot. This plot is only available for models with two or more process factors and can only display one response. On the first page of the contour wizard, under Plot type, click Surface. Select the responses to display and click Next.
To hide the constants, right-click the plot and clear View Constants.
172
Prediction
Plot type
To select to create a prediction plot, under Plot type click Prediction or Overlay prediction on the first page of the prediction plot wizard.
Factor type
Under Factors at the X-axes, click the type of factor to vary: Process or Mixture. You cannot vary simultaneously both process and mixture factors.
Selecting responses
You can select to display up to 3 responses for the prediction plot and 9 for the overlay prediction plot. Select which response(s) to display by marking in Available responses and clicking the => and/or marking in Selected responses and clicking <=.
173
Range or reference
You can select to keep the relative amount of the other mixture factors in the same proportion as their ranges, rather than the reference mixture (default) by clicking Adjusted to range.
174
Prediction
Select the response(s) on the Select Responses page. You can also use the Response box. Select which factor to display on the Select Factor page. For mixture factors you can also select to Adjust proportional to reference mixture or Adjust proportional to ranges Select confidence level on the Confidence Level page.
Factor type
The response prediction plot can be displayed for process and mixture factors although only one at a time.
Process factor
When the selected factor is a process factor, all other factors are kept constant on their averages.
Mixture Factor
When the selected factor is a mixture factor, the relative amounts of all other mixture factors are kept in the same proportion as in the standard reference mixture. If no standard reference mixture is specified, the centroid of the constrained region is used as default. Response prediction plot example
175
Prediction list
In the prediction list you can type or paste factor settings to make predictions. The lower and upper level 95% confidence intervals for these predictions are also calculated. To open the prediction list, on the Prediction menu, click Prediction List. To insert rows, right-click the list and click Insert Rows or press the DOWN arrow key when on the last row. If the model has been fitted the predictions are calculated automatically after entering settings for all factors. If you do not want the predictions to be automatically updated, right-click the list and clear Auto Update. To make predictions with auto update cleared, right-click the list and click Update Predictions or press the F5 key. To change the confidence level of the confidence intervals, right-click the list and click Properties.
Note: With PLS and an X matrix with large condition number, the standard error of predictions is computed and displayed instead of the confidence interval.
Scatter plot
In the prediction scatter plot you can view your factors and predicted responses as 2D and 3D scatter plots.
Note: The points displayed in the scatter plots are the points in the current prediction list. The scatter plot is updated with new points when you enter them in the prediction list.
To create a prediction scatter plot, on the Prediction menu, click Scatter Plot. For 2D plots, select the variable to be plotted on the X-axis by marking it and clicking the arrow =>. Mark the other variable(s) and click the arrow to Series. For 3D plots, select the variables to be plotted on the X, Y, and Series axes. in the dialog to remove factors or responses from the YClick the Delete button axis or from Series. To change what to display on the X-axis, use the arrow to add the new variable. Click the Color by Variable tab to color by a factor or a response. You can customize the 3D plot using the property page and plot settings. See the Plot settings for 3D scatter and Onion 3D plots section in the Plots and Lists chapter.
176
Prediction
Plot type
To select to create the 2D sweet spot plot click Sweet Spot and to select the 4D with up to four factors (5 with mixture) click 4D Sweet Spot. Click Next.
Response Selection
In the Response Selection page, enter the settings for the relevant responses.
Note: If you have entered Min and/or Max values in the response definition, this page is automatically filled with those values. To update with updated values from the response definition, click the Get Limits button.
For each response you have to select:
To Include or Exclude each response under Incl/Excl. Which type of values, Value or Percent, you are entering under Min and Max. Type the values that are of interest under Min and Max. MODDE has entered the smallest and the largest values found in the worksheet. Change them according to your desired sweet spot. You can also select or clear the Use Constraints check box and change the resolution in the Resolution box.
177
Plot axes
Click Next to open the Axes and Constants page, or for 4D sweet spot, the 4D Axes page. Mixture factors can only be selected as inner factors in a 4D plot.
Select two process factors for the axes under Factors at the plot axes, or three mixture factors For the 4D, select two process factors for the outer axes. The sweet spot contours are plotted for the low, middle, and high levels of these factors.
178
Prediction
Green for the sweet spot, that is the areas where all responses are within the selected range. Blue for areas where one of the responses is within its selected range. White for areas where none of the responses are within their selected ranges. Other colors for areas with more than one response within its range but not all.
The above are the default settings and you can define specific colors in the plot settings dialog.
179
Optimizer
You can use the optimizer to find an experimental area limited both by factor settings and response criteria. To open the optimizer, on the Prediction menu click Optimizer. MODDE opens a dialog box with 3 spreadsheets.
The optimizer can be set up for different objectives: 1. 2. Limit optimization where the objective is to reach a solution where all responses are within the specification limits. This is default in MODDE. Target optimization where the objective is to reach a solution where all responses are as close to target as possible. For the target optimization it is necessary that all responses can be optimized close to or to reach the target. Otherwise you may end up with an unacceptable solution. Focus optimization where you want to favor one or several responses over the others using individual weights.
3.
For more information about the optimizer objectives see Appendix C: Optimizer and the Optimizer section in the Statistical appendix.
180
Prediction
Response spreadsheet
In the response spreadsheet all the responses used in your model are available. Before starting the optimization you must select the Criteria for your response(s), weights and limits. If you have specified Min, Target, and/or Max in the response definition, these specifications are copied to the Optimizer response section. If no Min, Target, or Max settings are defined in the response definition, the response will be by default Predicted.
Note: You can fetch updated limits from the response definition by rightclicking and clicking Update limits from response definition.
181
Minimize the response. Type the highest value you can accept under Max and your target value under Target. Maximize the response. Type the smallest value you can accept under Min and your target value under Target. Target the response. Type the smallest, largest, resp. desired value under Min, Max, resp. Target. Predicted response. The response will not participate in the optimization but the prediction will be displayed. Excluded response. The response will not be a part of the optimization nor displayed in the run list.
When you have not entered Min, Target, or Max values and you choose to Minimize or Maximize, MODDE makes an educated guess of your limits. It is imperative that the limits are well chosen and reflect the data at hand. When the difference between the minimum/maximum and target values is too small the optimizer will not run. When this happens you should increase the range.
Weight
In the column Weight you can enter a number between 0.1 and 1 reflecting the importance of the response. Default is 1 indicating that all responses are of equal importance. The weight will change the optimization objective described in the first section of this chapter as follows:
Weight = 1 for all responses results in a limit optimization (default). Weight = 0.2 for all responses results in a strict target optimization. Individual weights for the responses results in an optimization where the responses with higher weights are favored in the quest for all responses to reach inside the limits.
182
Prediction
Run list
The run list includes the factors, responses, iterations, Log(D), and DPMO. The D in Log(D) is a normalized distance to the target. See Appendix C: Optimizer for more information.
In the run list above, row 6 has a DPMO = 0 meaning that with disturbances + 5% on the factor settings will give a solution inside the specifications. Row 1 has approximately the same Log(D) but a DPMO = 5600 indicating that a small disturbance in the factor settings from this point will result in some hits outside the specifications. Therefore row 6 is preferable.
183
Sensitivity analysis
The Analyze sensitivity check box is by default selected to display complementary information about the proposed optimal factor settings. The sensitivity is expressed as DPMO (Defects Per Million Opportunities) outside specifications, based on Monte Carlo simulations and including prediction error. Note that DPMO is an estimate of the result after the number of simulation performed; 10 000 simulations will give a fair estimate.
Optimizer Properties
In the Options tab optimizer settings are specified.
Feature
Use absolute response limits
Description
When selected, only the runs where all responses are predicted inside the specified limits are displayed in the run list.
Default
Not selected.
Calculate DPMO
When selected, the DPMO is calculated for runs where all responses are predicted inside the specified limits. The Sensitivity Range for the factors specifies the range used in the DPMO calculations. Selected. This is a sensitivity analysis that indicates if a solution is sensitive or insensitive to small changes. When selected, the sensitivity analysis (DPMO) includes the model error in addition to the sensitivity range. Selected.
The number of simulations in the calculation of the DPMO. The result 10 000 is scaled up to the unit 'dots per million operations'. When selected, the weight specified in the response spreadsheet is included when calculating Log(D). Not selected.
184
Prediction
Optimizer buttons
There are seven optimizer buttons: Generate start runs, run optimizer, generate start points from selected, contour plot wizard, copy to predictions, sweet spot plot wizard, and analyze design space. The Design Space, opened by clicking the Analyze Design Space - button is described in the Design Space chapter.
MODDE generates start runs from the corners and center of the experimental region, plus the three best runs. For details on how the start runs are selected see the Statistical appendix, section Optimizer.
185
Final factor settings. Predicted responses values. Number of iterations. Log(D), i.e., the logarithm of the overall distance to Target. The smaller the Log(D) the better is the result. At Log(D) = -10 all predicted responses are on target. DPMO, estimated number of hits outside specification on one million simulations. A smaller DPMO means that the solution is less sensitive to disturbances.
Restarting the optimizer After convergence, you can always click the Run Optimizer (play) button to restart the optimizer. It restarts from the displayed resulting runs of the previous search. If you do not want to continue with the resulting runs click the Generate new starting points or Generate starting points around selected button to specify other start runs.
In the dialog box displayed the run you marked is the default run to optimize around. To switch to another run, enter another run in the Center around optimizer run box. Enter the percent of the factor range in the Factor range box. The percentage entered here is used to calculate the new high and low limits for the start runs.
186
Prediction
Center around the optimum found by the optimizer is by default marked, and the run to use in the center is the one you marked. If you want to select another run enter the new run number. Selecting Factor definitions results in displaying same factor settings as when you enter the contour plot wizard from the Prediction menu. Selecting Optimizer factor setup results in displaying the current factor settings in the Optimizer.
Center around the optimum found by the optimizer With Center around the optimum... selected, clicking OK opens the contour plot wizard dialog box. If you create a 2D contour plot the first two factors (for process factors; first three factors for mixture factors) are varied around their optimum settings (20% range) and the other factors are set constant to their optimal values. If you create a 4D contour plot the first two factors (for process factors; first three factors for mixture factors) are varied around their optimum settings (20% range), the 3rd and 4th are set at their low, high, and center values using the 20% optimum range, and other factors are set constant to their optimal values. The resulting plot displays the area outside the factor range shaded.
187
Contour plot displayed from optimizer In the created contour plot the predicted parts that are outside the original factor settings are displayed shaded when the plot is extrapolated. The selected run is displayed in the plot as lines from the axes with arrows pointing toward the position of the selected run.
188
Prediction
189
190
Design Space
Introduction
The calculation of the design space is a search function that expands the possible factor ranges from a setpoint (optimum) to the largest possible range where all response predictions are still within the specifications. Predictions in the design space are done with Monte Carlo simulations. The resulting distribution of predictions simulates a real situation with a random combination of factor setting disturbances within a given range. There are two Design Space features in MODDE:
Predictive Design Space Estimation for optimization. Design Space Validation for robustness testing.
191
192
Design Space
Factor spreadsheet
In the factor spreadsheet the following functionality is available for each factor:
Item Description Setting
For DS Validation: by default the low, center, and high values from the factor definition. The acceptable factor range For Predictive DS Estimate: by default the Optimum is the value for the selected run in the run list. The Low and High values are the 95%* confidence with normal distribution. Calculated.
St.Dev
The standard deviation of the factor Free means that the largest possible range of variation for the specific factor is used, with respect to the settings for other factors and response specifications. Locked means that the range settings of the factor are locked. The Distribution of each factor range. Uniform, Normal, Triangular or Target.
Role
Distribution
Uniform means that all factor settings within the specified range have the same probability to appear. For quantitative factors, by default Normal means that the simulations are 'Normal'. normally distributed within the factor For formulation factors, by default range. 'Triangular'. Triangular means that the distribution has the shape of a triangle; a good way to get a skewed distribution for a factor. Target means that the factor will be set to a fixed value, the "Optimum" value.
The left side of the T bar in the Estimated acceptable range is the Estimated Min estimated minimum factor value for which the predictions still fall inside the specifications. The right side of the T bar in the Estimated acceptable range is the Estimated Max estimated maximum factor value for which the predictions still fall inside the specifications.
Calculated.
Calculated.
The yellow lines are the factor settings of the selected optimal run in the optimizer. The red lines are the Low and High factor settings of the experimental region. Estimated acceptable range The black T bar represents the region of acceptable variability valid for that factor when all other factors are locked at the optimum. Valid means that all predictions of the responses are within the specifications. No model error is considered in this search. The blue region represents the 95%* part of random factor variability with normal distribution where all predictions are within the specifications.
*95% is the default. In the Design Space Properties you can change to 99%.
193
Response spreadsheet
In the response spreadsheet the following functionality is available for each response:
Item
Min, Target, Max
Description
The optimization range
Default
Min, Target, and Max as specified in the Optimizer. 'Minimize, Maximize, Target, and Predicted as specified in the optimizer.
Criterion
When the search does not reach inside the limits for a response, the Criterion is set to 'Not Met'
Cpk
A Process Capability Index, Cpk, which originates from the SixSigma statistics and is estimated in this Estimated by MODDE when simulation. Cpk =1 means that approximately 0.13%* selected. of the predictions will fall outside the specifications. Defects Per Million Opportunities is an estimate of how many predictions will be outside the specification per Estimated by MODDE. one million predictions in the selected DS with the selected distribution. The yellow line represents Target value for the responses as specified in the Optimizer. The red lines are the specification limits for each response as specified in Optimizer. The faded green region represents the probability of a prediction for a random distribution of factor settings in the given ranges (low-optimum-high), the DS.
DPMO
* See table in the DPMO and Cpk section in Appendix A: Statistical notes.
Resample reruns the Design Space calculations. Create Histogram opens the response histogram plot. Individual response analysis displays a detailed DS estimate for the individual responses. This window displays an overview of the acceptable factor ranges which makes it easier to understand which the limiting responses are. In the window below the response Soot has the smallest acceptable interval and will therefore be the most limiting response, affecting the complete region of operability for each
factor.
Properties opens the Design Space Properties dialog. For more see the Design Space Properties sub section.
194
Design Space
Parameters section
In the Parameters section the settings used in the design space estimation are displayed and can be adjusted.
Option
Acceptable range simulations Response profile simulations DPMO target
Description
Number of simulations done in each step of the search for the acceptable range for the factors. Number of simulations for the final predictions of the response profile. Defects Per Million Opportunities outside the specification target. The first response that exceeds this target will stop the expansion of the accepted factor range. Stop criterion will be the total number of predictions outside the target for all responses. When this check box is not selected the DPMO target is for each response. Includes model error in the predictions of the response distribution. Any change in the settings results in a recalculation of the Design Space.
Default
20 000 100 000
1 000
Automatic update
Selected. As this calculation includes many computational operations you may want to inactivate it and do the calculations when all settings have been adjusted instead of after each adjustment. Defines the Low and High values in such a way that 95%, in the default case, of the Monte Carlo simulations are inside the limits. 95% Selecting 99% widens the included area by lowering the Low limit and increasing the High limit in the factor spreadsheet.
195
Show section
In the Show section what to display and how can be adjusted.
Option Description Default
Columns shown in the factor spreadsheet of the DS. The available fields are: By default all the columns but Factors (factor names), Low, Optimum, High, St.Dev, Possible Min and Possible Max are displayed. Role, Distribution, Possible Min, Possible Max and Estimated acceptable range. By default Responses, Min, Target, Max, Criterion, DPMO, and Predicted response profile are displayed. Blue for factors, green for responses.
Factor fields
Columns shown in the response part of the DS. The Response available fields are: Responses (response names), Min, fields Target, Max, Criterion, Cpk, DPMO, Average, St.Dev, %Outside, Median, 1st Quartile, 3rd Quartile, and Predicted response profile. Color range Displays the range of colors of the distribution bars.
Factor histogram
When double-clicking the estimated acceptable range for the factors, the Factor Histogram opens. This histogram displays the distribution of the factor settings used for the simulation.
196
Design Space
Response histogram
The result of the Monte Carlo simulations can be presented as frequency histograms by:
197
198
Show
Introduction
On the Show menu the following items are available: Objective list, Design Matrix, Design Region plot, Confoundings list, D-Optimal Summary table, Model list, and Reference Mixture list.
Objective
To open the Objective list, on the Show menu click Objective. The Objective list displays the selections made in the Objective pages of the design wizard
The following information is listed: Maximum Runs: the maximum number of runs that MODDE can include in a design created by MODDE. Objective: the selected objective. Process Model: the type of model created for the process factors. Mixture Model: the type of model created for the mixture factors. Design: the selected design. Runs in Design: the selected number of runs in the design created by MODDE. If you have added runs after creating the design they will not be included here. Center points: the number of center points selected. Replicates: the number of times the entire design has been replicated. N = Actual Runs: the number of runs created by MODDE.
199
Design matrix
To display the Design Matrix, on the Show menu, click Design Matrix. The Design Matrix displays the experimental plan in coded unit for quantitative factors, as in the worksheet for qualitative factors and in pseudo components for formulation factors. If you have qualitative factors at more than two levels, you can select to display the design with the qualitative factors extended. To display the design matrix with the qualitative factors extended: 1. 2. 3. Right-click the matrix and click Properties. Click Extended shows all settings and click OK. Optionally switch back by clicking Regular shows all orthogonal settings.
By default the Design Matrix is derived from the worksheet, and reflects any changes (excluded runs, changed values, additional runs etc.,) done to the factor part of the worksheet after its generation. To display the original design matrix generated by MODDE: 1. 2. 3. Right-click the matrix and click Properties. Click The design as generated by MODDE and click OK. Optionally switch back by clicking The current Worksheet scaled and centered.
200
Show
Design region
Open the Design Region plot to get an overview of your experimental plan. To open the Design Region plot, on the Show menu, click Design Region. Use the Response box or right-click the plot and click Properties to switch to another response.
Is displayed as one or more cubes. Can be displayed for classical designs with up to 5 process factors. Displays the response values at the star points on the face of the cubes for CCC designs. Displays the points in the design region plot color coded according to the response values entered in the worksheet. The response range is parted equally resulting in three groups named High, Medium, and Low. Displays points where the factor settings are not at the original value, at the closes intersection for low, center, and high factor settings. In this plot, the replicate tolerance is therefore NOT used to identify replicates. Displays a black point, Conflicting colors when the response values to be displayed have been assigned to different color groups (High, Medium, and Low).
Hint: When you want to view 3 factors in 3D, use the Scatter Plot found under the Worksheet and Prediction menus.
201
Confoundings
Open the Confoundings list to see which terms that are mathematically identical in the current design. For instance, in the example below the term M*TH is included in the model, but the effect of this term is confounded with the effect of H*S. This means that using this design there is no way of telling whether the coefficient displayed for M*TH reflects M*TH, H*S, or a mixture of both. To list the confoundings click Confoundings on the Show menu. For factorial designs resolution III or IV, the Confoundings list displays the confounding pattern for the complete interaction model. In the Term column, the background of the terms included in the model is colored. In the coefficient plots and lists, confounded terms are marked with a bracket #-sign.
202
Show
Summary D-Optimal
To displays a list summarizing the generation of a D-Optimal design click D-Optimal Summary on the Show menu.
Under Candidate set the properties of the used candidate set are listed in the form of Extreme Vertices, Edge points, Centroid of high dim. Surfaces, and Total Runs. Under D-Optimal the properties of the selected design are listed in the form of Objective, Model Type, Potential Terms, Number of Inclusions, Constraints, Design Runs, Selected Design Number, G-Efficiency, log(Det. of X' X), Norm. log(Det. of X' X), and Condition number.
203
Model
On the Show menu click Model to display the selected model. If you have qualitative factors at more than two levels and want to display the model with the qualitative factors extended: 1. 2. 3. Right-click the Model list. Click Properties. Click Extended Show all settings.
Reference mixture
To display the reference mixture used in the analysis of the design click Reference Mixture on the Show menu. To change the reference mixture, click Model / Reference Mixture on the Edit menu and click the Reference Mixture tab.
204
Introduction
You can customize the plots in MODDE using the Plot Settings page. You can change the properties of plots and lists from the property page accessed by right-clicking the plot or list and clicking Properties. Most plots can be transformed into lists by right-clicking the plot and selecting Create List.
Properties page
Opening property page by right-clicking
To open the property page of a plot or list, right-click the plot or list and click Properties.
With the plot or list active, on the File menu, click Save Plot/List As. Right-click the plot or list and click Save Plot/List As.
For details, see the section Save plot/list as in the File chapter.
205
Note: When displaying multiple plots, opening the Properties page displays the properties for the first displayed response.
Plot settings
Use Plot Settings to customize the most common attributes of the axes, of the plot area and of the header and footer. There are two different types of plot settings dialogs in MODDE:
Plot settings for 3D scatter plot and contour, response surface, and sweet spot plot where all options are in the same dialog. Plot settings for other plots are divided in Axis, Header and Footer, and Plot Area pages.
Open Plot Settings by double-clicking the plot or right-click and click Plot Settings, then when applicable click Axis, Header and Footer, or Plot Area. After customization, most features can be saved as default plot settings in two ways:
Click the Save Settings button in the Plot Settings dialog. Right-click the plot and click Save as Default Plot Settings.
206
Axis
To change the settings of the plot axes, right-click the plot, click Plot Settings, and then click Axis. This dialog is also opened if you double-click one of the axes. All of the properties apply to the selected axis: X, Y, or Y2 (the second Y-axis). In this dialog you can change general settings, such as scale, gridlines, titles, and fonts by clicking the appropriate tab.
Scale
To change the scale, enter the Minimum and Maximum data values and the value to Increment by. You can specify the spacing of the minor tick marks and the number of Decimal places to display on the tick label. Click the Recalculate Scale button to restore the Minimum and Maximum values.
Miscellaneous
By default the axes are in normal scale. You can change them to Logarithmic scale or Values in reverse order by selecting the corresponding check box. The axes are default displayed. To hide an axis, clear the Show axis check box. The Always recalculate scales check box is by default selected. If you do not want MODDE to recalculate the scale you should clear the check box.
207
Gridlines
In the Gridlines tab, select the Show gridlines check box to display the grid lines of the selected axis. The grid lines are placed on the tick marks. Select the Pattern, Color, and Width of the gridline.
Title
In the Title tab, select to show or hide the axis title, change the text and its orientation.
208
Font
In the Font tab, select the font for both the axis titles and the tick mark labels.
For how to customize the axes of the 3D scatter plots, see the section later in this chapter.
For how to customize the header and footer of the 3D scatter plots, see the section later in this chapter.
209
Plot Area
To change the settings of the interior of the plot, right-click the plot, click Plot Settings, and then click Plot Area. This dialog is also opened if you double-click the plot area. In this dialog you can change symbol style, line style, data labels, and font by clicking the appropriate tab.
Symbol style
Select the shape, color and size of all the symbols in a series.
Line style
Click the Line Style tab to change the properties of lines in line plots.
Data labels
Click the Data Labels tab to change the attributes of all the labels on data points inside the plot area. Color, alignment, or orientation (Rotation) applies to all the labels in a given series.
Font
Click the Font tab to select the font for all the data labels.
210
Background
Click Background, found under Series to the left in the dialog, to change the borders of the window, plot, and chart areas.
Legend
Click Legend, found under Background to the left in the dialog, and select or clear the Show Legend check box to show or hide the legend. Here you can also change text placement, color, background and border styles of the legend.
211
Column style
When you have a column plot you can change the pattern, color, column width, and overlap for each or all series in the Format Plot Area dialog.
For how to customize the plot area of the 3D scatter plots, see the section later in this chapter.
Click the Save Settings button in the Plot Settings dialog. Right-click the plot and click Save as default Plot Settings.
You can always restore Umetrics default Plot settings in View | General Options | Restore.
212
Resolution
Resolution is the precision used when creating the contour plot. Selecting a higher resolution requires more calculations and is therefore more time consuming. To select the resolution of the plot, click one of the predefined resolutions or type a resolution value here.
Use color
By clearing the Use color check box you can choose to display the contour and surface plots without colors.
213
Axis
Click Axis to change settings of the axes. For more, see the Plot settings for 3D scatter and Onion 3D plots section later in this chapter.
Header
Click Header or Footer to customize the positioning and coloring. For more, see the Plot settings for 3D scatter and Onion 3D plots section later in this chapter.
Regions
Use the Region style page to customize how the area outside the factor setup is visualized. By default, this area is shaded white.
Labels
Click Labels to:
Clear the Show label check box to hide the plot labels. Select the position of the plot labels in the Anchor box. Select the Connect with attach point check box when the label is separated from the line (using Offset) and you want a line between the label and the point. Customize the Text and Background color and Transparency of the plot label. Select the Is visible check box under Border to display a border around each label. Enter custom values in the Margin and Width fields. Click the Font tab to customize the font.
Contour
In Contour Level Colors in the Plot Settings page you can customize levels and colors according to the below. The default plot displays a number of contour levels. To display fewer or more contour levels, type a new value in the Number of Levels box. In Individual level colors the new number of levels will be displayed.
Note: To for instance display the contour levels for 10, 20, 30 100 you need to type Min=0 and Max=110 in Color level range, and in the Number of Levels box type 10.
214
After Color range on the contour levels the color buttons, defining the color scale used in the plot, are positioned. To change from the default colors, click the respective buttons, and click the color to use. To change the color of one of the available levels under Individual level colors, mark the level and click the Color button, and then click the new color. To remove one of the available levels, mark it and click the Remove button. You can also add levels by typing a value in the field under the Add button, and then clicking the Add button. Click the Contour Line Style tab to customize the contour line color, width, and pattern.
Axis
Click Axis to change settings of the axes. For more, see the Plot settings for 3D scatter and Onion 3D plots section later in this chapter.
Header
Click Header or Footer to customize the positioning and coloring. For more, see the Plot settings for 3D scatter and Onion 3D plots section later in this chapter.
Contour
In Contour Level Colors in the Plot Settings page you can customize the colors according to the below.
215
The default plot displays up to 5 colors in the legend. These levels are by default colored from blue to green where green represents the sweet spot. To change the color of one of the available levels: 1. 2. Under Individual level colors, mark the level and click the color button. Then click the new color.
To change the range used, click the respective color buttons and click the new color.
Cube
Click Cube to:
Clear the Color cube sides check box to make the cube sides transparent. When the Color cube sides check box is selected the sides are colored in the color specified in Side color. Customize the Side color. Select/clear the Show gridlines and Gridlines on minor ticks check boxes to show/hide the gridlines. Change the Line width, Pattern, and Line color to specify the properties of the gridlines.
216
Axis
Click Axis to change settings of the axes. For each axis you can select:
To display the annotation of the axis. The minimum and maximum values of the axis and the step size. To autoscale the axis. The axis title. Customize the axes color and size. Customize the annotations color and distance from the axis. Select/clear the Show axis arrows check box to display arrows/regular axes. Select the axis arrow color. Select the level of Transparency for the axes and annotation respectively. Adjust autoscaling properties. Customize the tick's direction and size for normal and minor ticks. Clear the Show ticks and Show minor ticks check boxes to hide the ticks.
In the Font tab you can select the font and size of the annotation and titles of the axes.
217
Header
Click Header to:
Select/clear the Is visible check box under Header to show/hide the header. Select the position of the header in the Anchor box. Customize the Border and text color of the header. Customize the Background color of the header. Select/clear the Is visible check box under Border to show/hide the border. Customize the Margin and Width of the border.
In the Font tab you can select the font and size of the header.
Legend
Click Legend to:
Select/clear the Show legend check box to show/hide the legend. Select the position of the legend in the Anchor box. Select the orientation of the legend in the Orientation box. Select the text adjustment of the legend in the Text adjust box. Customize Border and Text color and Background color of the legend. Select/clear the Is visible check box under Border to show/hide the border. Customize the Margin and Width of the border.
Footer
Click Footer to:
Select/clear the Is visible check box under Footer to show/hide the footer. Select the position of the footer in the Anchor box. Customize the Border and text color of the footer. Customize the Background color of the footer. Select/clear the Is visible check box under Border to show/hide the border. Customize the Margin and Width of the border.
In the Font tab you can select the font and size of the footer.
218
Symbols
Click Symbols and the individual series to change the properties of the symbols. When you mark the individual series you change the properties of that series only. When you mark Symbols you change the properties of all series. In these pages you can
Select the Shape of the symbol. Customize the Color of the symbol. Select the Size of the symbols. Select the Draw style of the symbol, where the choices are Fill, Wireframe, and Point. Fill is the default. Select the Line style when the Draw style is Wireframe.
Labels
Click Labels to:
Select/clear the Show label check box to show/hide the plot labels. Select the position of the plot labels in the Anchor box. Select the Connect with point check box when the labels are separated from their position (using Offset) and you want a line between the label and the point. Customize the Offset in X, Y, and Z space of the plot labels. Customize the Text color of the plot label.
Colors
Click Colors to:
Change the Highlight Style displayed when hovering over a point. For more about the content on this page, see the Symbols topic above. Change Marking Color for the marked points.
219
220
Report generator
Introduction
MODDE has an automatic report generator. You start the report by clicking File | Generate HTML Report. A dialog opens where you can select the MODDE default template, or any other report as a template. In the report generator, basic formatting functionality is available for writing text. Plots, lists, and model results of MODDE can be added to the report at any time. These items are added to the report as placeholders. A placeholder stands in the place of contents which MODDE will provide, let it be a plot, list or any text or number. The placeholders enable you to use the same report, as a template, in different investigations. You can then edit the text, and just click the Update Report button to update all MODDE results from the current investigation.
If you have saved a report with the same name as the investigation, MODDE default suggests opening that report. When your investigation has not been fitted MODDE will ask to fit the investigation so it can automatically fill all the placeholders. You can prevent this question if you select under View | General Options, to automatically fit investigations when opened.
221
A command menu bar. A standard toolbar, named Generate Report, with commonly used commands. A format report toolbar with commonly used formatting commands. A main window showing the report template. A placeholder window with a list of built-in placeholders. A Report Generator FAQ window with a short introduction how to use the report generator. A Properties window where plot size and placeholder properties can be edited.
File menu
On the File menu you find the general Windows commands New, Open, Save As, Print Preview, Print and Exit. Additionally you find Continue Edit Report With, Templates, and View in Browser.
222
Report generator
Templates
You can save templates, restore templates and add or remove custom templates by clicking Templates on the File menu. Click:
Save as Default Template when you have changed/created a template according to your wishes and want to use it as the default template next time you generate a report. Restore Default Templates if you have made changes to the default templates and want to remove those changes. Save as Custom Template when you have changed/created a template according to your wishes and want to save it to be able to use it again. Custom templates will be listed in the Select template or open existing dialog and when you click Insert | Template. Add / Remove Custom Templates when you want to add an already created template or remove one of your custom templates.
View in browser
To view the current report in your default internet browser, on the File menu click View in Browser. This works with Internet Explorer and Mozilla Firefox.
223
Edit menu
On the Edit menu you find the standard commands Undo (CTRL+Z), Redo, Copy (CTRL+C), Paste (CTRL+V), Paste Unformatted (CTRL+SHIFT+V), Clear (press the Delete key), Select All (CTRL+A), and Find (CTRL+F).
View menu
On the View menu you can select to hide or show the following:
Generate Report toolbar opened/closed by clicking View | Toolbar | Generate Report. For more, see the Generate report toolbar section later in this chapter. Format toolbar opened/closed by clicking View | Toolbar | Format. For more, see the Format toolbar section later in this chapter. Status bar which displays an explanation when you point to a button. To show/hide it, click View | Status Bar. Placeholders window which displays the placeholders that can be added to the report. For more see the Placeholders window section. Properties window. The properties of plots and images can be customized in the Properties window. To display the window click Properties. For more see the Properties window section. Click Customize to customize toolbars and menus as in Office 2003 and later. Find a short introduction in the Customize section in the View chapter.
Insert menu Use the Insert menu to insert a Hyperlink, Image, File, or Template in the current report. Here you also can toggle the Grab Plot or List mode explained below.
Hyperlink
To insert a hyperlink, select text in the report and then click View | Hyperlink, also available by right-clicking the report and clicking Insert Hyperlink. In the dialog that opens, enter the address of your hyperlink the URL field.
Image
Click View | Image to insert a picture in the report. Also available by right-clicking the report and clicking Insert Image. Click the Browse button to find your file.
File
Select View | File to insert a Web page file (*.htm, *.html), a Text file (*.txt), or a picture file (*.jpg, *.png, *.gif, *.bmp.)
Template
Clicking View | Template opens a dialog where you can select a custom template to insert in the report. Templates must first have been added using File | Templates | Save as Custom Template.
224
Report generator
Format menu
On the Format menu you can customize Font and Styles and Formatting.
Tools menu
On the Tools menu, find the commands Update Report, Update Placeholder, Show All Placeholders, Show Placeholder, Remove All Placeholders, and Remove Placeholder.
Update Report
Click Update Report to update all placeholders with the plots and lists of the current investigation.
Update Placeholder
Click Update Placeholder to update the marked placeholder.
Show Placeholder
Click Show Placeholder to show the underlying placeholder of marked plot, list, or item.
Remove Placeholder
Click Remove Placeholder when you do not want a certain item, plot, or list to be updated. Remove Placeholder is also available as a button on the Generate Report toolbar.
225
Help menu
To access the FAQ of the report generator, click Welcome Page and FAQ on the Help menu.
New
Click the arrow next to the New button and the following commands are displayed.
Click New on the menu displays the generate report dialog from which you can select which report/template to use or to open an existing report/template. Click New Blank Report to start a new report with no text. Click New From Default Template if you have created a template that you have saved as default template. Save a report as a default template by clicking Templates on the File menu and then clicking Save as Default Template. Umetrics default template is used if no other template has been specified.
For more on the Placeholder buttons see the Tools menu section. For more on the Grab Plot or List button see the Insert menu section. For more on the Continue Edit With button see the File menu section.
Format toolbar
The Format toolbar is the standard toolbar for formatting text with three additional and Insert Image , and View in Browser buttons: Insert Hyperlink For more, see the Insert menu and File menu sections in this chapter. .
226
Report generator
Placeholder window
Open the Placeholder dockable window by clicking View | Placeholders.
The combo box lists different categories of placeholders. To insert a placeholder in the report, mark the placeholder and click Insert. To update the inserted placeholder, click the placeholder then on the Tools menu click Update Placeholder. You can also click the Update Report button on the Generate report toolbar to update all placeholders in the report. Click Show Placeholders to show the underlying placeholders.
227
Properties window
Open the Properties dockable window by clicking View | Properties.
In the Properties window you can change the default plot size and to save a plot as .png, .bmp, or .jpg. Properties for placeholders are displayed when you click the placeholder. You can change the properties of the placeholder in the Data field in the properties window. Click the Data field to view a description on how you can change the properties of the current placeholder.
228
Help
Introduction
MODDEs help is based on this user guide. The user guide documents are transferred in to a compiled HTML file. To read the Help file Internet Explorer version 4.0 or higher must be installed but does not need to be your default browser.
HTML help
The HTML help file is installed to include interactive help and stand alone where the program is installed. Open the help by:
Clicking Contents and Index from the Help menu. Clicking the Help-button in one of the dialogs or wizards.
Clicking MODDE Help from the Program menu. Pressing F1 opens the context sensitive help.
Use the Contents, Index, or Search tabs to find what you are looking for. Additionally, the Analysis Advisor is available to guide you through the analysis. For more on the Analysis Advisor see that section in the View chapter.
Manage Licenses
If your company has a license server that handles you licenses, follow the instructions delivered with the delivery letter. If you choose to activate later, click Manage Licenses on the Help menu and then follow the instructions delivered with the delivery letter.
229
About MODDE
To find the version number of MODDE, on the Help menu, click About MODDE.
230
Fit methods
MODDE supports Multiple Linear Regression (MLR) and PLS (Projection to Latent Structures) for fitting the model to the data.
231
Note: When the models for the responses are different, PLS fits each response separately.
PLS has been extensively described in the literature and only a brief description is given here. PLS finds the relationship between a matrix Y (response variables) and a matrix X (predictor or factor variables) expressed as: Y = XB+E The matrix Y refers to the characteristics of interest (responses). The matrix X refers to the predictor variables and to their square or/and cross terms if these have been added to the model. PLS creates new variables (ta) called X-scores as weighted combinations of the original X-variables: ta = Xwa, where wa is the combinations weights. These X-scores are few, often just two or three, and orthogonal (independent). The X-scores are then used to model the responses. With several responses, the Y-variables are similarly combined to a few Y-scores (ua) using weights ca, ua = Yca. The PLS estimations are done in such a way that it maximizes the correlation, in each model dimension, between ta and ua, One PLS component (number a) consists of one vector of X-scores (ta), and one of Y-scores (ua), together with the X and Y-weights (wa and ca). Hence the PLS model consists of a simultaneous projection of both the X and Y spaces on a low dimensional hyper plane with new coordinates T (summarizing X) and U (summarizing Y), and then relating U to T. This analysis has the following two objectives:
To well approximate the X and Y spaces by the hyper-planes To maximize the correlation between X and Y (u and t).
232
Geometrically, we can see the matrices X and Y as n points in two spaces, (see figure), the X-space with p axes, and the Y-space with m axes, p and m being the number of columns in X (terms in the model) and in Y (responses).
X3 t2 X2 t1 u 1
Y 3
u 2 Y 2
X1 t1
Y1
u 1
The model dimensionally, (number of significant PLS components), is determined by cross validation (CV), where PRESS (see below) is computed for each model dimension. One selects the number of PLS dimensions that give the smallest PRESS.
233
First fit the model. MODDE extracts only the significant PLS components. This is the PLS solution. Then click Next Component menu and continue extracting PLS components until no correlation between X and Y remains. This is the MLR solution.
Model
You may edit the model and add or delete terms for individual responses. You may add up to third order terms (cubic terms, or 3 factors interaction). If your design is singular with respect to your model, MODDE will fit the model with PLS, and MLR will not be available.
Hierarchy
MODDE enforces hierarchy of the model terms. You cannot delete the constant term. You can only delete a linear term if no higher order term containing the factor is still in the model.
234
Scaling
Scaling X
When fitting the model with multiple linear regression, the design matrix X is scaled and centered as specified in the factor definition box: MLR scaling. If the choice is not orthogonal, the condition number will differ from the one displayed when clicking Analysis | Evaluate. When fitting the model with PLS the X matrix is always scaled to unit variance. If warranted, the scaled X matrix is extended with squares and / or cross terms according to the selected model. The choices of scaling are: (x denotes the original factor value and z the scaled one) Orthogonal scaling:
zi = (xi - M)/R
Where M = Mid-range, R = Range/2. Mid-range scaling:
zi = (xi - M)
Unit variance scaling:
zi = (xi - M)/s
Where m = average, s = standard deviation computed from the worksheet. Note that Orthogonal and Mid-range scaling are only available with MLR. MODDE default scaling for MLR is the orthogonal scaling.
Scaling Y
The matrix of responses Y, when fitting the model with PLS, is by default scaled to unit variance. You can modify the unit variance scaling weight by using the PLS scaling box in the factor definition. With MLR the Y's are not scaled.
235
Condition number
The condition number of the orthogonally scaled and centered extended design matrix using the SVD (Singular value decomposition) is computed when clicking Evaluate on the Analysis menu. The X matrix is taken from the worksheet. The calculation depends on fit method (MLR, PLS) and which factors are involved.
2.
If you select Mid-range scaling, and your factors have different ranges, the condition number of the worksheet becomes very large. This is only a numerical artifact, but due to the fact that MODDE uses the SVD with MLR, the model should be fitted with PLS.
Note: If you selected a different scaling than orthogonal, the condition number will be different than the one computed in Analysis | Evaluate. In particular if you select Mid-range scaling and your factors have different ranges, the condition number of the worksheet becomes very large. When fitting the model with PLS, the condition number refers to the X matrix, with unit variance coding.
236
Note: With formulation factors and fitting with PLS, the condition number is computed by excluding the factor with the largest range, and scaling the remaining ones orthogonal. When fitting with MLR the condition number is computed without centering and scaling the factors.
Missing data
Missing data in X
Missing data in X are not allowed, and will disable the fit. This does not apply to uncontrolled X-variables. For MLR experiments with missing in an uncontrolled factor results in excluding that row in the calculations while PLS can handle this.
N-value
The N-value used in ANOVA, and for the computation of R2 adjusted, is the actual number of non-missing observations (experiments) for each response-column. This Nvalue and DF = N-p are displayed at the bottom of the ANOVA plots and lists, and on all residual plots, including observed vs. predicted Y.
237
ANOVA
The analysis of variance (ANOVA) partitions the total variation of a selected response SS (Sum of Squares corrected for the mean) into a part due to the regression model and a part due to the residuals.
SS = SSregr + SSresid
If there are replicated observations (experiments), the residual sum of squares is further partitioned into pure error SSpe and Lack of fit SSlof.
A goodness of fit test is performed by comparing the MS (mean square) of lack of fit to the MS of pure error: Two ANOVA plots are displayed: 1. 2. The regression goodness of fit test The LoF goodness of fit test
238
Q2
Q2 = (SS - PRESS)/SS with
$ (Yi Yi )2 PRESS = 2 i (1 hi )
where hi is the ith diagonal element of the hat matrix: X(X'X)-1X'
R2
R2 = (SS - SSresid)/SS R2 adj = (MS - MSresid)/MS MS = SS / (n - 1) MSresid = SSresid / (n - p) RSD = Residual Standard Deviation = MSEresid
Degrees of freedom
MODDE always computes the real degrees of freedom RDF of the residuals: RDF = n - p - ( ni - 1 ) where the loops over i n = number of experimental runs ni = number of replicates in the ith set p = number of terms in the model, including the constant
Saturated models
When RDF = 0 the model is saturated, and MODDE does not compute or display R2, R2 Adjusted or Q2 when fitting the model with MLR. With PLS, only Q2 is computed and displayed.
Singular models
Singular models (condition number > 3000) are only fitted with PLS. If p > n - ( n i - 1 ) , the degrees of freedom of the residuals are computed as: DFresid = 0, with no replicates in the design DFresid = (n i - 1), with replicates in the design
239
Coefficients
Scaled and centered coefficients
The regression coefficients computed and displayed by MODDE refer to the centered and scaled data. You may also select to display the unscaled and uncentered coefficients.
Normalized coefficients
In the overview plot, to make the coefficients comparable between responses, the centered and scaled coefficients are normalized with respect to the variation of Y. That is, they are divided by the standard deviation of their respective Y's.
For matrices with condition number > 3000, PLS does not compute confidence intervals on the coefficients.
Confidence intervals
Confidence intervals of coefficients and predictions are computed using the total number of observations, regardless of missing values for PLS models with identical models for all responses. For MLR and PLS with different model for the responses, the total number of observations is the number of non-missing values per response. This total number of observations is displayed as N at the bottom of all other plots and lists. The approximation for PLS with identical models is possible because the confidence intervals computed with the regression formulas are somewhat too large because the PLS solution is a shrunk estimator with smaller prediction errors than those of regression. Hence a small number of missing elements in Y does not make the PLS confidence intervals larger than those computed with the regression formulas and the total number of observations.
240
C(2)
-1 1 0 0
C(3)
-1 0 1 0
C(4)
-1 0 0 1
The coefficients of these expanded terms are given as the coefficients for level 2 (b), 3 (c), and 4 (d) of C, while the coefficient for level 1 (a) is computed as the negative sum of the three others. MODDE displays all the four coefficients in the coefficient table but notes that they are associated with only three degrees of freedom.
Residuals
Raw residuals
The raw residual is the difference between the observed and the fitted (predicted) value ei = Y i i The raw residuals are displayed in the residual lists.
Standardized residuals
The standardized residual is the raw residual divided by the residual standard deviation ei / s (s = RSD) These are MODDE default for PLS Residual plots.
241
Predictions
For X matrices with condition number < 3000, both MLR and PLS computes a confidence interval for the average predicted y: Yi + hi * RSD * t( / 2, DFresid) hi is the ith diagonal element of the hat matrix: X(X'X)-1X' For X matrices with condition number > 3000 and for all Cox mixture models, PLS computes only the standard error of the average predicted Y: SE(Y) = [(1 / N) + t'0 * (T'T)-1* t'0] * RSD
PLS plots
Both scores and loading plots are available:
Plot loadings
WC plots (PLS)
Plots of the X- and Y-weights (w and c) of one PLS dimension against another, say, no.'s 1 and 2, show how the X-variables influence the Y-variables, and the correlation structure between X's and Y's. In particular one better understands how the responses vary their relation to each other and which ones provide similar information.
Plot scores
TT, UU, and TU plots (PLS)
The tt and uu plots, of the X- and Y-scores of, say, dimensions 1 and 2 (i.e. t1 vs. t2, and u1 vs. u2), can be interpreted as windows into the X- and Y-spaces, respectively, showing how the design points (experimental conditions, X) and responses profile (Y) are situated with respect to each other. These plots show the possible presence of outliers, groups, and other patterns in the data. The tu plots (t1 vs. u1, t2 vs. u2, etc.) show the relation between X and Y, and display the degree of fit (good fit corresponds to small scatter around the straight line), indications of curvature, and outliers.
PLS coefficients
PLS computes regression coefficients (Bm) for each response Ym expressed as a function of the X's according to the assumed model (i.e. linear, linear plus interactions or quadratic,). These coefficients are (columns of B) computed as: B = W(P'W)-1 C' W and C are (p*A) and (m*A) matrices whose columns are the vectors wa and ca. p = number of terms in the model m = number of responses A = Number of PLS components
242
A simple and parsimonious model An approximately constant model error variance An approximately normal model error distribution.
If the response values vary more than a magnitude of ten in the experimental domain, a transformation is often recommended. The maximum point on the Box-Cox plot gives the value of (lambda, ) for the response transformation Y that gives the best fit of the model. This is the maximum likelihood estimator for . For more information see Draper and Smith Applied Regression Analysis, Second Edition Wiley, New York or Box and Draper Response Surface Modeling. The Box-Cox plot is not available for PLS.
243
Cox reference model When all mixture factors are formulation factors, MODDE generates, by default, the Cox reference model, and i.e. the complete polynomial model linear or quadratic. MODDE supports also a special cubic and a full cubic model (see section Objective, model and design). Scheff model You may select to fit a Scheff model, by selecting Scheff MLR as fit method on the Analysis | Select Fit Method menu. MODDE expresses the mixture model in the Scheff form. The full cubic model is not supported as a Scheff model.
Changing the Standard Reference Mixture. Click Model / Reference Mixture on the Edit menu to change the coordinates sk of the standard reference mixture. By default MODDE selects as reference mixture the centroid of the constrained region. Mixture Hierarchy with the Cox reference model. By default all Cox reference models, linear and quadratic, obey mixture hierarchy. That is the group of terms constrained by: bksk = 0 ckjbkjsk = 0 for k = 1,,,,q (1) and for j = 1,,,,q (ckj = 1 when j k and ckj = 2 when k = j.) are treated as a unit, and terms cannot be removed individually.
If you want to remove terms individually, as with regular process models, clear the Enforce the mixture model hierarchy check box in the Edit | Model / Reference Mixture dialog. When the mixture hierarchy is not enforced (this includes cubic models), the Cox reference model can only be fitted by PLS. The coefficients are the regular PLS coefficients computed from the projection and not re-expressed relative to a stated standard reference mixture. Note that in all cases model hierarchy is enforced a term cannot be removed, if a higher order term containing that factor is still in the model.
244
ANOVA with the Cox model In the ANOVA table, the degrees of freedom for regression are the real degrees of freedom, taking into account the mixture constraint. These are the same as for the equivalent slack variable model. Screening plots When the objective is to find the component effects on the response, the coefficients of the Cox reference linear model are directly proportional to the Cox effects. The Cox effect is the change in the response when component k varies from 0 to 1 along the Cox axis. That is the axis joining the reference point to the kth vertex. Effect plot The effect plot displays the adjusted Cox effects. The adjusted effect of component k is: k = rk*tk rk = Uk - Lk tk = bk/(T-sk) where: rk is the range of factor k tk is the total Cox effect T is the mixture total. In most cases T=1. bk is the unscaled uncentered coefficient sk is the value of the factor at the reference mixture The Effect Plot is only available for screening designs using the Cox model. Main effect plot For a selected mixture factor Xk, this plot displays the predicted change in the response when Xk varies from its low to its high level, adjusted for all other mixture factors, that is, by default, the relative amounts of all other mixture factors are kept in the same proportion as in the standard reference mixture (MODDE does not check if the other mixture factors are kept within their ranges). For example, if the Main effect of the mixture factor X1 is being displayed, when X1 takes the value x1, the other mixture factors are assigned the values: xj = (T - x1)* (sj / T - s1). Sk are the coordinates of the standard reference mixture. The standard reference mixture is the one used in the model. You can change this default and select to have all other mixture factors kept in the same proportion as their ranges (this ensures no extrapolation). Interaction Plot Interaction plot is not available when you only have mixture factors.
Linear models bksk = 0 (1) Quadratic models bksk = 0 (1) ckjbkjsk = 0 for k = 1,,,,q (1) and for j = 1,,,,q (2) Here ckj = 1 when j k and ckj = 2 when k = j. and sk are the coordinates of the standard reference mixture.
bksk = 0 (1)
Quadratic models
bksk = 0 (1) ckjbkjsk = 0 for k = 1,,,,q (1) and for j = 1,,,,q (2)
Here ckj = 1 when j k and ckj = 2 when k = j. The scaled and centered coefficients are recomputed afterwards.
Note: In MODDE, with linear and quadratic models obeying the mixture hierarchy, (i.e. terms constrained by (1) or (2) can only be removed as a group, and not individually), by default the PLS coefficients are always expressed relative to a stated standard reference mixture.
With models containing terms of the third order (cubic), or disobeying mixture hierarchy, no constraints are imposed on the PLS coefficients. The coefficients are in this case, the regular PLS coefficients and the reference mixture is implicit and results from the projection.
246
Quadratic Scheff bk = Cox (PLS) b0 + bk + bkk Scheff bkj = Cox (PLS) bkj - bkk - bjj
Scheff models
The Scheff models are only fitted with MLR and only the main effect plot is available. Scheff models are only available for investigation with all mixture factors. ANOVA As described by Snee in Test Statistics for Mixture Models (Technometrics, Nov. 1974), the degrees of freedom in the ANOVA table are computed in the same way as with the slack variable model.
247
Note: When the model contains terms of order 3, or contains qualitative and formulation factors, the PLS coefficients are not adjusted relative to a stated standard mixture. MODDE plots
All MODDE plots are available when you have both mixture and process factors. For both the Main Effect and prediction plots, when you select to vary a process factor, all of the mixture factors are set to the values of the standard reference mixture. When you select to vary a mixture factor, process factors are set on their average and the other mixture factors are kept in the same proportion as in the standard reference mixture or their ranges.
Optimizer
The Optimizer uses a Nelder Mead simplex method with the fitted response functions, to optimize an overall desirability function combining the individual desirability of each response. For details on how to use the Optimizer, see the Appendix C: Optimizer chapter.
Desirability
For every response y, the desirability function is computed as follows: f(g(y)) = 100*(e^(*g(y)) - 1) where g(y) = 100*((y -P)/(T - P)) T, L and P are defined as follows: T = User desired Target L = User defined worst acceptable response value(s) P = Worst response value computed from the starting simplex. P is never closer to the Target than the L(s). When the response is to be maximized, L is the smallest acceptable value, when the response is to be minimized L is the largest acceptable value. When the response is to be on Target, the user gives the smallest and largest acceptable values. When the response is to be minimized, we must have T < L and when the response is to be maximized we must have T > L. For responses to be on target the user must supply lower and upper limits such as L(lower) < T < L(upper) L is generated internally when not supplied by the user.
( L P) (T P)
248
Where Limit = 90 + 80*Log10(w), and w = the weight (importance) assigned to each response by the user. The weight w is a number between 0.1 and 1 (default value). W effects the shape of the desirability function f(g(y)). For more information about the effects see Appendix C: Optimizer. This definition of makes f(g(y)) = - (Limit) when y = L and gives the exponential function f(g(y)) the theoretical range: 0 to -100 (this latter limit can only be reached asymptotically when y gets close to Target). When the user wants the response to be on target, L(upper) is used in the calculation of when y > T and L(lower) is used in the calculation when y < T.
Overall desirability
The overall desirability f(ds) is a weighted average of the individual desirability function. The weights, denoted w, are the user entered weights between 0.1 and 1, reflecting the relative importance of the responses.
M = number of responses. D = -10 when all responses have reached T. D is not used in the optimization but is displayed as Log(D) in the run list.
Starting simplexes
The optimizer starts 8 simplexes from 8 starting runs selected as follows:
The first four are from the corners of a 23-1 design in the factors with the largest coefficients for the first response. The fifth is the overall center point. The last three are the 3 best runs from the worksheet considering only the first response and the predicted values.
Duplicated start points will not be shown. The user can modify these runs or add own. Each simplex is generated from the starting run by adding an additional run for each factor with an offset of 20% of the distance from the center to the maximum value, the other factors being kept at the same values. A check is made that all runs are within the defined experimental region.
249
Sensitivity Analysis
Sensitivity Analysis is an estimate of how sensitive the proposed factor settings are to small disturbances. The sensitivity analysis is done with Monte Carlo simulations (MC) and displayed as DPMO (Defects Per Million Opportunities outside specifications). Normally distributed random disturbances of + 5% of the range of each factor are applied and result in an estimate of the total number of DPMO for all active responses. The user can modify the sensitivity range and the number of MCs. DPMO=0 means that the proposed factor settings represent a robust point of operations for factor disturbances of + 5%.
Orthogonal blocking
When you cannot perform all of the experiments in a homogeneous way, randomizing the run order of the experiments may not be sufficient to deal with the extraneous sources of high variability. You may want to run the experiments in homogeneous groups, i.e., blocks, in such a way that the external source of variability does not influence the effects of the factors. For example, suppose you are running a full factorial with 5 factors and 32 runs, and the batch size of raw material allow you to perform only 8 runs per batch. You may want to run your experiment in 4 blocks, each composed of 8 runs using homogeneous material. The method of dividing 32 runs in 4 blocks of 8 runs, each such as the difference between the blocks (the raw material) does not affect the estimate of the factors, is called Orthogonal Blocking. MODDE supports orthogonal blocking for the 2 level factorial, fractional factorials, Plackett Burman, CCC, and Box Behnken designs. MODDE can also block D-Optimal designs. These designs are more flexible with respect to the number of blocks and the block size, but the blocks in D-Optimal design are not usually orthogonal to the main factors. The only restriction with D-Optimal designs is that the number of runs must be a multiple of the block size.
Note: Blocking introduces extra factors in the design, hence reduces the degree of freedom of the residuals, and the resolution of the design. You should only block when the extraneous source of variability is high and cannot be dealt with by randomizing the run order.
Block interaction
An interaction between a main effect and a block effect is called a block interaction. When the design supports the interactions between the block effects and the main effects, the Block interactions check box, in the Select model and design dialog in the design wizard is active. You can select the check box if you want to add the block interactions to your model.
250
$B2
+ + Block 1 Block 2 Block 3 Block 4
When you select Show | Design Matrix, the design matrix is displayed in coded units including the blocking factors. When the worksheet is generated, the blocking factors are recoded and the model is reparameterized. Rather than keeping the d blocking factors, such as 2d = k (the number of blocks), MODDE generates one qualitative variable called $BlockV, with k levels called B1, B2 ...Bk.
251
MODDE blocks the D-Optimal design by generating a qualitative factor, $BlockV, with as many levels as the selected number of blocks. By default, it then selects only balanced designs with respect to the blocking factor.
252
It is not always possible to generate a balanced D-Optimal design with respect to the blocking factor. In this case you may want to change the model, the number of blocks, or generate an unbalanced design.
253
Normal
Triangular
254
The low and high factor settings are the distribution boundaries. For a normal distribution 95% of the distribution is found within the boundaries, by default. The automatic search procedure expands the distribution for each factor until one or more response limits are exceeded according to the specified DPMO target. The automated search procedure is symmetrical around the setpoint but random for the factors.
= predicted average.
= estimated standard deviation for predictions.
255
DPMO
115070 35930 8198 1350 483 159 48 13 3.40 0.79 0.17 0.03 0.0060 0.0010
%Outside
11.51 3.59 0.82 0.13 0.05 0.02 0.0048 0.0013 0.0003 7.93328E-05 1.69827E-05 3.33204E-06 5.99037E-07 9.86588E-08
For more information about Cpk (Process Capability Index) search in Wikipedia.
The above is true for PLS models too but PLS models with condition number > 3000 and non hierarchical mixture models display the standard error of prediction instead of confidence intervals.
256
Appendix B: Designs
Resolution III designs where main effects are confounded with 2 factor interactions. Resolution IV designs where two factor interactions are confounded with each other. Resolution V designs where main effects and all two-factor interactions are clear of each other (unconfounded). MODDE supports resolution V designs.
With both resolution III and IV designs, you can only select the linear model. You may edit the model and enter selected interactions. In that case, you may have to edit the generators of the design. With resolution V designs, MODDE generates the interaction model.
257
The default generators used by MODDE for fractional factorial designs are those recommended by Box, Hunter and Hunter (page 410). You may edit and change the generators in the Generator dialog available on the Edit menu. When you update the confounding, MODDE will warn you if some of the effects in your model are confounded with each other, i.e. if your model is singular.
L9: design with up to 4 factors at three levels. L27: design with up to 13 factors at three levels. L36: design with up to 13 factors at three levels. L18 is called mixed as it has one factor at 2 levels and up to 7 factors at three levels.
With these 3 levels designs MODDE (objective = screening) lets you select only the linear model, because these designs do not support interactions. In Edit | Model you may edit the model and include square terms.
D-Optimal designs
D-Optimal designs are computer generated designs that maximize the determinant of the X'X matrix, X being the extended design matrix. D-Optimal designs are available for all objectives. For more see the D-Optimal designs section later in this chapter and see also the DOptimal chapter.
Onion designs
Like regular D-Optimal designs, D-Optimal Onion designs can be used both in screening and in RSM with quadratic models. The Onion designs comprise layers of designs, usually D-Optimal, where the outermost layer determines which type of model (screening or RSM) that the Onion design supports. For more see the D-Optimal onion designs section later.
258
Appendix B: Designs
Rechtschaffner designs
Rechtschaffner designs are orthogonal, saturated fractions of resolution V of the 2n and 3n factorial designs. They allow the estimation of all main effects and all first order interactions without confounding. They are saturated designs, with no degrees of freedom remaining for the estimation of residuals and diagnostics. The 2n Rechtschaffner designs are well suited when the objective is screening, with 6 or more factors, and little knowledge about the importance of each individual first order interaction. In this case it is of interest to estimate all first order interactions, unconfounded, and then eliminate the insignificant (small) ones, hence recovering some degrees of freedom for diagnostics and residual analysis. The required number of runs N for the 2n Rechtschaffner designs with k factors is: N =1 + k + k(k - 1)/2 It is recommended to add 3 to 4 center points to these designs.
RED-MUP designs
The RED-MUP designs are custom designs developed for the use with 96 well plates (see figure) and larger (384, 1536, etc.). These are widely used platforms for experimentation in biochemistry, microbiology, pharmaceutical development, etc., with some special properties (buildable and extendable to other factor intervals). The RED-MUP designs consist of two sub-designs corresponding to the vertical and horizontal directions of the plates, i.e., 8 and 12, respectively, for 96-well plates. The total design is made by multiplying the two sub-designs together. Hence, this total design supports a model with all interactions between the factors in the sub-designs, plus from each sub-design, the main effects, and when these sub-design so support, interactions, and quadratic effects. Below, we use n1 and n2 for the number of rows and columns in the plate, i.e., 8 and 12 in a 96 hole plate. A 96-well plate can handle from 5 full RSM factors up to 18 factors for a stretched screening situation. The layout of a 96 well plate has 8 rows and 12 columns. Hence, the vertical direction has n1=8, and the horizontal direction has n2=12.
If both sub-designs support only main effects, using for example Plackett Burman subdesigns, up to n1+n2-2 factors can be investigated, i.e., up to 18 factors in a 96 well plate. Such a sparse design without center points is not recommended. More reliable designs with center points in the larger sub-design would allow n1-3 + n2-5 = n1+n2-8 factors, i.e., 12 factors for a 96 well plate.
259
When specifying the RED-MUP design, it is important to distribute the factors over the two sub-designs (vertical and horizontal) so that (a) the actual experimental protocol remains simple and doable, (b) the sub-designs and final design make chemical/biological/engineering sense, and (c) a-priori interesting interactions and higher order terms can be estimated. Note that all the interactions between each of the factors in the vertical design and each of the factors in the horizontal designs can always be estimated. Hence, factor pairs for which interactions are expected should be split into the two sub-designs. Then their interaction can always be estimated, regardless of choice of sub-design. Special designs When selecting to create a RED-MUP design, there are special designs for the 96 well plates (8 x 12) which aim to make better use of the plate.
RSM designs
RSM designs are used in later stages of an investigation to develop more elaborate models (quadratic) in the few important factors, usually not more than 5 or 6. MODDE supports the following RSM designs:
MODDE also supports a reduced CCC and CCF for four factors, with the fractional part of the design reduced from 16 to 12 runs. Note that with the CCC designs you may edit the model and include cubic terms, if you wish.
D-Optimal designs
D-Optimal designs are computer generated designs that maximize the determinant of the X'X matrix, X being the extended design matrix. D-Optimal designs are available for all objectives. For more see the D-Optimal designs section later in this chapter and see also the DOptimal chapter.
260
Appendix B: Designs
Onion designs
Like regular D-Optimal designs, D-Optimal Onion designs can be used both in screening and in RSM with quadratic models. The Onion designs comprise layers of designs, usually D-Optimal, where the outermost layer determines which type of model (screening or RSM) that the Onion design supports. For more see the D-Optimal onion designs section later.
Rechtschaffner designs
Rechtschaffner designs are orthogonal, saturated fractions of resolution V of the 2n and 3n factorial designs. They allow the estimation of all main effects and all first order interactions without confounding. They are saturated designs, with no degrees of freedom remaining for the estimation of residuals and diagnostics. The 3n Rechtschaffner designs are well suited for the RSM objective with 6 or more factors as they require fewer runs than the classical CCC or CCF non saturated designs. The intent with these designs is to estimate quadratic terms but performing fewer runs than with CCC or CCF. Eliminating insignificant terms, after performing the experiments, results in recovering some degrees of freedom. The required number of runs N for the 3n Rechtschaffner designs with k factors is: N = 1 + 2k + k(k - 1)/2 It is recommended to add 3 to 4 center points to these designs.
RED-MUP designs
The RED-MUP designs are custom designs developed for the use with 96 well plates (see figure below) and larger (384, 1536, etc.). These are widely used platforms for experimentation in biochemistry, microbiology, pharmaceutical development, etc., with some special properties (buildable and extendable to other factor intervals). The RED-MUP designs consist of two sub-designs corresponding to the vertical and horizontal directions of the plates, i.e., 8 and 12, respectively, for 96-well plates. The total design is made by multiplying the two sub-designs together. Hence, this total design supports a model with all interactions between the factors in the sub-designs, plus from each sub-design, the main effects, and when these sub-design so support, interactions, and quadratic effects. The RED-MUP designs are well suited for the RSM objective with up to 5 or 6 factors. The intent with these designs is to get a precise model that can be used for optimization and for detailed understanding. The maximum number of RSM factors depends on the sizes of the sub-designs. An 8 run sub-design, e.g., a Doehlert design with 2 center points, supports 2 RSM factors (1 constant, 2 linear, two quadratic, and one interaction terms), and a 12 run subdesign, e.g., a three level Rechtschaffner design with 2 center points, supports 3 RSM factors (1 constant, 3 linear, 3 quadratic, and 3 interaction terms) for a total of 5 RSM factors for a 96-well plate. Mixed objective Since the RED-MUP designs are constructed from two sub-designs, one of these can be an RSM design and the other a screening design. In such a case the objective is said to be mixed.
261
Special designs When selecting to create a RED-MUP design, there are special designs for the 96 well plates (8 x 12) which aim to fill up the plate.
Doehlert designs
The Doehlert designs are quadratic RSM designs with some special properties (buildable and extendable to other factor intervals). They allow the estimation of all main effects, all first order interactions, and all quadratic effects without confounding. They are saturated designs with similar properties to the CCF and CCC designs. Geometrically they are polyhedrons based on hyper-triangles (simplexes), with a hexagon in the simplest two-factor case. Doehlert design in 2 factors with 6 runs + center points, can be extended to a new design by adding 3 experiments. Usually also one or two new center points are added in the new design (i.e., in the figure the right-most point in the old design).
The Doehlert designs are well suited for the RSM objective with up to 5 or 6 factors (respectively 33 and 45 runs with 3 center points). The intent with these RSM designs is to get a precise model that can be used for optimization and for detailed understanding. The required number of runs N, except for replicated center points, for the quadratic Doehlert designs with k factors is: N = 1 + k + k2 It is recommended to add 3 to 4 center points to these designs.
262
Appendix B: Designs
Formulation factor
Formulation factors are the usual mixture factors used in formulations with specifically defined experimental ranges. Most mixture experiments have only formulation factors.
Filler factor
The presence of filler is typical of certain types of simple mixture experiments. For example in a synthesis the solvent is typical filler, as is water in a juice punch. A filler is a mixture component, usually of little interest, making up a large percentage of the mixture, and added at the end of a formulation to bring the mixture total to the desired amount. It is recommended to define a mixture factor as filler when all three conditions below are fulfilled:
The factor is always present in the mixture. The factor accounts for a large percentage of the mixture and there is no restriction on its range. It is added at the end to bring up the mixture total to the desired amount (usually 1 when no mixture factors are kept constant). You are not interested in the effect of the filler per se.
When you specify a filler factor, MODDE checks that these conditions are met and defaults to a slack variable model, with the filler factor omitted from the model.
Use
All mixture factors are controlled or constant. The Uncontrolled option is unavailable for both formulation and filler factors. Formulation factors can be defined as Constant when you want to keep them constant in the experiment.
263
When mixture factors are constant, the mixture total T = 1 - Sum (constant mixture factors). When no formulation factors are defined as constant, the mixture total has to be equal to 1. MODDE issues an error message and stops whenever the mixture total is not equal to T or 1.
Note: When the mixture region is regular, mixture factors are first transformed to pseudo components, and then scaled with PLS models.
Mixture constraint
In a mixture experiment the mixture total (i.e. the sum of all the mixture factors in the experiment) is equal to a constant T. The mixture Total T is generally equal to 1 when no mixture factor is kept constant. This mixture constraint implies that the mixture factors are not independent, and this collinearity has implications on the mixture experimental region, the mixture designs, and the mixture model formulation.
264
Appendix B: Designs
Note: You can select to have the analysis done on the mixture factors expressed in original unit under the Analysis menu. Region is the L simplex
When the mixture region is the L simplex, the L pseudo component transformation is defined as: Pi =(Xi - Li)/(RL) The transformed mixture factors Pi vary from 0 to 1.
Screening designs
MODDE provides three variants of the axial design. Axial designs locate all the experimental points on the axis of the simplex and are recommended for screening, see Snee (references). Standard Axial (AXN) The standard axial design includes the following 2*q +m runs (q = number of mixture factors, m centroid points as specified by user). 1. 2. 3. All the q vertex points. The coordinates of the ith Vertex point is xi = (0, 0, 0..1, 0, 0..). All q interior points of the simplex. The coordinates of the ith Interior point is xi = (1/2q, 1/2q, 1/2q,..(q+1)/2q, 1/2q, 1/2q..). The overall centroid of the simplex with coordinates x = (1/q, 1/q,....., 1/q..) replicated (m-1) times.
265
Extended Axial (AXE) The extended axial design includes the following 3*q +m runs (q = number of mixture factors, m specified by user). 1. 2. 3. 4. All the q vertex points. The coordinates of the ith Vertex point is xi = (0, 0, 0..1, 0, 0..). All q interior points of the simplex. The coordinates of the ith Interior point is xi = (1/2q, 1/2q, 1/2q,..(q+1)/2q, 1/2q, 1/2q..). All the q End points. The coordinates of the ith End point is xi = (1/(q-1), 1/(q-1), 1/(q-1), 0, 1/(q-1), 1/(q-1)..). The overall centroid of the simplex with coordinates x = (1/q, 1/q,....., 1/q..) replicated (m-1) times.
Reduced Axial (AXR) The reduced axial design includes the following (q+m) (specified by user) points: 1. 2. 3. All the q vertex points. A subset or none (specified by user) selected from the q interior points. The overall centroid replicated as desired.
RSM
MODDE provides 2 variants of the quadratic model designs, one special cubic and one cubic. The simplex centroid design has all the experimental points on the vertices, and on the center of the faces of consecutive dimensions. Modified simplex centroid (SimM) The modified simplex centroid design supports a quadratic model and includes the following: 1. 2. 3. 4. The q vertex points. The coordinates of the ith Vertex point is xi = (0, 0, 0..1, 0, 0..). The (q (q-1))/2 Edge centers. The coordinates of the ijth edge point is xij = (0, 0, 1/2, 1/2 0, 0..). The q Interior check points. The coordinates of the ith interior point is xi = (1/2q, 1/2q, (q+1)/2q, 1/2q, 1/2q..). The overall centroid with coordinates x = (1/q, 1/q,...1/q), replicated as desired.
266
Appendix B: Designs
Modified simplex centroid Face center (SimF) The modified simplex centroid face center design supports a quadratic model and includes the following: 1. 2. 3. 4. 5. The q vertex points. The coordinates of the ith Vertex point is xi = (0, 0, 0..1, 0, 0..). The (q (q-1))/2 Edge centers. The coordinates of the ijth edge point is xij = (0, 0, 1/2, 1/2 0, 0..). The q Face centers of dimension (q-1). The coordinates of the ith face center is: (1/q-1, 1/q-1,..,0, 1/q-1..1/q-1). The q Interior check points. The coordinates of the ith interior point is xi = (1/2q, 1/2q, (q+1)/2q, 1/2q, 1/2q..). The overall centroid with coordinates x = (1/q, 1/q,...1/q), replicated as desired.
Simplex centroid Special Cubic (SimSC) The simplex centroid special cubic design supports a special cubic model and includes the following: 1. 2. 3. 4. 5. The q vertex points. The coordinates of the ith Vertex point is xi = (0, 0, 0..1, 0, 0..). The (q (q-1)) 1/3, 2/3 Edge points. The coordinates of the ijth edge point is xij = (0, 0, 1/3, 2/3, 0, 0..), xji = (0, 0, 2/3, 1/3, 0, 0..). The q(q-1)(q-2)/6 Face centers of dimension 2. The coordinates of the ith face center is (0, 0, 0, 1/3, 1/3, 1/3..0, 0, ). The q Interior check points. The coordinates of the ith interior point is xi = (1/2q, 1/2q, (q+1)/2q, 1/2q, 1/2q..). The overall centroid with coordinates x = (1/q, 1/q,...1/q), replicated as desired.
Simplex Centroid Cubic (SimC) The simplex centroid cubic design supports a cubic model and includes the following: 1. 2. 3. 4. 5. 6. The q vertex points. The coordinates of the ith Vertex point is xi = (0, 0, 0..1, 0, 0..). The (q (q-1)) 1/3, 2/3 Edge points. The coordinates of the ijth edge point is xij = (0, 0, 1/3, 2/3, 0, 0..), xji = (0, 0, 2/3, 1/3, 0, 0..). The q(q-1)/2 Edge centers. The coordinates of the ith edge center is xi = (0, 0, 0, 1/2, 1/2, 0...0). The q(q-1)(q-2)/6 Face centers of dimension 2. The coordinates of the ith face center is (0, 0, 0, 1/3, 1/3, 1/3..0, 0, ). The q Interior check points. The coordinates of the ith interior point is xi = (1/2q, 1/2q, (q+1)/2q, 1/2q, 1/2q..). The overall centroid with coordinates x = (1/q, 1/q,...1/q), replicated as desired.
267
D-Optimal designs
What are D-Optimal designs?
D-Optimal designs are computer generated designs, tailor made for a specific problem. They allow great flexibility in the specifications of your problem. They are particularly useful when you want to constrain the region and no classical design exists. D-Optimal means that these designs maximize the information in the selected set of experimental runs with respect to a stated model. For a specified regression model Y = X* + where: Y is a (N x 1) vector of observed responses, X is a (N x p) extended design matrix, i.e. the n experimental runs extended with additional columns to correspond to the p terms of the model (i.e., the added columns are for the constant term, interaction terms, square terms, etc..) (beta) is a (p x 1) vector of unknown coefficients to be determined by fitting the model to the observed responses. (epsilon) is a (N x 1) vector of residuals (the differences between the observed and predicted values of the response y). They are assumed to be independent of each other, normally distributed and with constant variance 2 The D-Optimal design maximizes the determinant of the X'X matrix, which is an overall measure of the information in X. Geometrically; this corresponds to maximizing the volume of X in a p dimensional space.
Candidate set
D-Optimal designs are constructed by selecting N runs from a candidate set. This candidate set is the discrete set of all potential good runs. MODDE generates the candidate set as follows: I) For a regular process region, the candidate set consists of one or more of the following sets of points (depending on your model and the number of factors):
The full factorial for up to 10 factors, reduced factorial for up to 32 factors. Centers of edges between hyper-cube corners Centers of the faces of the hyper-cube. Overall centroid
II) For constrained regions of mixture or/and process factors, the candidate set consists of one or more of the following set of points:
The extreme vertices of the constrained region The centers of the edges. If these exceed 200, the center of the 200 longest edges The centers of the various high dimensional faces The overall centroid.
MODDE has implemented an algorithm to compute the extreme vertices, center of edges, center of faces etc. as described by Piepel (1988).
268
Appendix B: Designs
D-Optimal algorithm
D-Optimal designs have been criticized for being too dependent on an assumed model. To reduce the dependence on an assumed model, MODDE has implemented a Bayesian Modification of the K-Exchange algorithm of Johnson and Nachtsheim (1983), as described by W. DuMouchel and B. Jones in A Simple Bayesian Modification of D-Optimal designs to reduce dependence on an Assumed Model, Technometrics (1994). With this algorithm one can add to the primary terms i.e. the terms in the model, potential terms, i.e. additional terms that might be important. The objective is to select a D-Optimal design, rich enough to guard for potential terms, and enable the analysis to detect possibly active ones. In order not to increase the number of runs N, and to avoid a singular estimation, one assumes that the coefficients of the potential terms are likely to have a mean of 0 and a finite variance (tau, )2.
Potential terms
Potential terms are higher order terms not included in the model but taken into account during the creation of the candidate set. Potential terms are default added but can be removed by clearing the Use potential terms box. Depending on the number of factors, the objective and the model, MODDE adds the following potential terms:
Model
Linear Linear + interactions
Potential terms
All interactions All squares
RSM
Factors
2-8
Model
Quadratic
Potential terms
All cubes
269
Model
Linear Linear All interactions
Potential terms
All interactions Interactions between the first 20 factors All squares
RSM
Factors
2-6 7 - 12
Model
Quadratic Quadratic
Potential terms
All cubes None
Model
Linear
Potential terms
All squares + interactions
RSM
Factors
2 - 12
Model
Quadratic
Potential terms
All cubes
Note: No potential terms are added for investigations with all factors defined as qualitative.
270
Appendix B: Designs
Design evaluation
To evaluate and compare D-Optimal designs, MODDE computes the following criteria:
LogDetNorm
The log of the determinant of X'X normalized for number of terms in the model p, and number of runs N. This is the criterion used, by default, to select the best design. MODDE selects the design with the largest value (closest to 0) of LogDetNorm. LogDetNorm = Log10 [ Det(X'X)1/p / N] The maximum value of LogDetNorm, for an orthogonal design, is 0.
LogDet
The Log of the determinant of the X'X matrix
Condition No
The condition number of the X design matrix coded orthogonal, and extended according to the model.
G efficiency
G efficiency is a lower bound on D efficiency, which compares the efficiency of a DOptimal design to a fractional factorial. G efficiency is defined as: Geff = (100*p)/(n*d) Where p = number of terms in the model n = number of runs in the design d = Maximum relative prediction variance v over the candidate set, where the prediction variance v = x(X'X)-1x'
Note: All of these statistics are computed from the runs selected D-optimally and do not include the possible center points added to the worksheet.
271
Irregular region
Screening
When the mixture region is an irregular polyhedron, MODDE computes the extreme vertices (corners) delimiting the region. These extreme vertices constitute the candidate set and the centers of the high dimensional faces are added to support potential terms. The design is a D-optimal selection of N (specified by user) runs from the candidate set.
RSM
MODDE computes the extreme vertices, 1/3, 2/3 centers of edges, centers of faces of dimension (q-1) and the overall centroid of the experimental region. When there are too many extreme vertices, only the center of the 25% longest edges is computed. These experimental points constitute the candidate set. The design is a D-Optimal selection of N runs (specified by the user) from the Candidate set.
Note: All mixture designs are displayed in pseudo components. Mixture models
Because of the mixture constraint, (the mixture factors are not independent) the analysis of mixture data with multiple regression requires a special model form. The traditional approaches have been:
Defining the model omitting one mixture factor, hence making the others independent. This is the Slack Variable approach. Omitting some terms from the model, so that the terms remaining in the model are independent. This is Scheff model, with the constant term removed from the linear model and the quadratic terms removed from the quadratic model. Using the complete model including all the mixture terms, but putting constraints on the coefficients to make them estimable. This is the Cox reference model, and the constraints on the coefficients are defined with respect to a standard reference mixture. This standard reference mixture serves the same function as the centering constant with process variables models.
272
Appendix B: Designs
Note: When the model contains terms of order 3, or contains qualitative and formulation factors, the PLS coefficients are not adjusted relative to a stated standard mixture.
273
274
Appendix C: Optimizer
Introduction
The optimizer works according to a given set of specifications. The specification of the factors and responses are selected according to the desired result. A drawback with this strategy can be that the best possible solution might not be reached. With unrealistic response specifications it will be impossible for the optimizer to do a good job. With a good strategy and by using complementary tools such as contour plots, DPMO estimates, sweet spot plots and Design Space estimates, a good understanding for the possibilities can be obtained. The optimizer is used to find an experimental setpoint that fulfills various criteria. The optimizer uses a search function to find the best possible solution to an equation system given a number of operating criteria. The optimizer starts with a number of criteria set in the optimizer window. This appendix describes the possibilities and limitations of the optimizer function. The first part is a description of how the optimizer works and the second part discusses how different objectives can be reached by selecting different start criteria for the optimization.
Search function
The optimizer works with a desirability function (f(ds)) that searches for the best possible combination of factor settings that predicts a result inside the response specifications and as close as possible to the target(s). When searching for one solution with many criteria, the result will be a compromise. This compromise is based on a summary function that is a measure of the distance to target for all results. It is expressed as f(ds). How f(ds) works depends on the weight for each response and the limit and target specifications: the optimizer objective.
275
Optimizer objectives
The optimizer can be set up for different objectives: 1. 2. Limit optimization where the objective is to reach a solution where all responses are within the specification limits. This is default in MODDE. Target optimization where the objective is to reach a solution where all responses are as close to target as possible. For the target optimization it is necessary that all responses can be optimized close to or to reach the target. Otherwise you may end up with an unacceptable solution. Focus optimization where the objective is to favor one or several responses over the others using individual weights.
3.
To control the optimization criteria the weight function has a key role, as well as reasonable limits and targets for the responses. The optimizer works with a desirability function, f(ds), and will strive to reach the lowest possible value. The shape of the function is controlled by the weight and the settings of criteria (Min, Target, Max) for each response. In the following pictures two desirability functions are shown, the first with weight=1 and the second with weight=0.2. With the weight 1 the lowest possible f(ds) is -100 and with a weight of 0.2, the lowest possible f(ds) is -20.
Limit optimization
With weight=1 the desirability function decreases rapidly close to the limit and then flattens out. As consequence it will be easier to reach a compromise where all responses are inside the specification limits but maybe not as close to target as possible.
Weight = 1. The desirability function reached the lowest possible value, -100, just inside the specification limit. This will work well to find a compromise when many responses strive to get inside the specification limits. Note that in this plot the 'weight scaled limit' and the 'limit' are displayed on top of each other.
276
Appendix C: Optimizer
Target optimization
If all weights are set to 0.2, the optimizer will search for a target solution for all responses if that is possible. If that is not possible, the optimizer might find a solution that will predict some responses very close to target and some outside the specification limits.
Weight = 0.2. The desirability function will reach the lowest possible value, -20, close to the target. If the weight is set to 0.2 for all responses it will go for a target solution due to the slope of the function. This solution might be unacceptable if a common target solution isnt present.
Focus optimization
If the weights are set differently for different responses, responses with higher weights take priority in the search for a solution inside the specifications. The overall optimization criterion is to reach the lowest sum of f(ds). With the weight 1 the lowest possible f(ds) is -100 and with a weight of 0.2, the lowest possible f(ds) is -20.
277
Replicate Plot: All values for NOx can be found between 10 and 35.
The response Soot is minimized while the response Fuel is set to Target for illustrational purposes. Generally we also want to minimize the Fuel consumption.
278
Appendix C: Optimizer
Optimizer result
The optimizer will search for a solution to the specifications from 8 different starting points using a Nelder-Mead Simplex algorithm trying to minimize the desirability value f(ds). The result is expressed as a normalized distance to target Log(D) and DPMO (Defects Per Million Opportunities outside specifications). For more information about the calculations, see the Optimizer section in the Statistical appendix. After running the optimizer the best proposal (lowest Log(D)) is selected. A Log(D) < -1 means that all results should be safely within specification limits. The optimal value of Log(D) is -10; then all response predictions are on target. DPMO gives information about robustness to small disturbances introduced by the Sensitivity Range specified for the factors.
In the run list above, row 6 has a DPMO = 0 meaning that with disturbances + 5% on the factor settings will give a solution inside the specifications. Row 1 has approximately the same Log(D) but a DPMO = 7000 indicating that a small disturbance in the factor settings from this point will result in some hits outside the specifications. Therefore row 6 is preferable. The result from the optimizer can be evaluated with three main tools:
Contour Plot Wizard generates a plot around the factor settings of the selected row, showing the dynamics around the selected point, e.g. if it is a flat region or very sensitive to small changes. The selected run is displayed in the plot as lines from the axes with arrows pointing toward the position of the selected run. Sweet Spot Plots Wizard generates an overlay of contour plots where the common region within specifications is shown in green. The selected run is displayed in the plot as lines from the axes with arrows pointing toward the position of the selected run. Analyze Design Space displays how the factor settings can be varied around the selected point (optimum) and still fulfill the response criteria. We can make an estimation of a safe DS region with Monte Carlo simulations on the factor settings.
279
Introduction
The calculation of the design space is a search function that expands the possible factor ranges from a setpoint (optimum) to the largest possible range where all response predictions are still within the specifications. Predictions in the design space are done with Monte Carlo simulations. The resulting distribution of predictions simulates a real situation with a random combination of factor setting disturbances within a given range. There are two Design Space features in MODDE:
Predictive Design Space Estimation for optimization. Design Space Validation for robustness testing.
A good overview of where the selected point is located in the design region. An estimate of the largest possible factor variation that still results in predictions within specification.
The International Conference on Harmonisation (ICH); Draft Guidance: Q8(R1) Pharmaceutical Development Revision 1 (http://www.fda.gov/cber/ich/ichguid.htm) has outlined quality by design (QbD) principles for pharmaceutical development which introduced the concept of Design Space (DS). ICH Q8 defines DS as the multidimensional combination and interaction of input variables (e.g. material attributes) that have been demonstrated to provide assurance of quality.
281
The predictive DS estimate shows the factor settings of the selected run in the optimizer and the accepted variability around these settings that still result in predictions in the accepted response region.
We can make an estimation of a safe DS region with Monte Carlo simulations on the factor settings. MODDE will perform a search to identify the largest possible range for each factor (faded region) that can be used and still meet all response requirements. The default target is 1000 (0.1%) hits outside the limits (DPMO) for one response. In this case the limiting response will be Soot, the final predictions will give 0.17% of the predictions outside the upper limit.
random factor settings according to the selected distribution, around their optimum value but within the Low and High limits, followed by predictions of the responses. In this case 100 000 predictions are performed. The distribution as well as the number of simulations and the range can be changed by the user.
The resulting distributions can be presented as a histogram, one for each response.
282
These settings give results within specifications but with a combination of factor settings that is more extreme and close to the experimental limits. Another consequence is that the accepted region of variability for the proposed factor settings is narrower than the previous proposal starting from row 6.
283
284
Evaluation
The Prediction | Design Space Validation tests the robustness by making a large number of random disturbances (Monte Carlo Simulation) in the specified region. In this example the specified region is the Experimental Region. In the DS window (shown below) the factor part shows the original investigation settings with a specific selection for the qualitative factor Column. ColA was manually selected as this column gave the worst results (See Tutorial example for more information about the specific example). The result is shown as a distribution of random samples including model prediction errors and it is well within the specification limits for some responses. The result can be expressed in general statistics as well as capability indexes Cpk or DPMO = Defects Per Million Opportunities outside specifications.
A description of the details of this window is found in the Design Space window section in the Design Space chapter.
Factor spreadsheet
All factors are varied within the design limits with Monte Carlo simulations according to a Normal distribution. These are the default settings.
Response spreadsheet
The result for response k1 is optional; there are no specific demands for this response. The result for response k2 is partially outside the specification limits. The result for response Res1 is above the low specification limit. The result for response PlateN(2) is above the low specification limit From the above we conclude that this system is robust against disturbances in the factors for Res1 and PlateN(2). k2 is not robust against disturbances in the factors.
285
Final adjustments
Design Space can be used to estimate the maximum accepted variability in factors that still predict all results within the specifications. The problem in the described example is response k2. The requirement for k2 is that less than 0.1%, corresponding to DPMO = 1000, of the predictions may be outside the specification limits. There are constraints when handling this type of situation;
Which factors affect the result? How can we adjust the factor limits without causing too much problems in the normal use of the procedure?
First we have to check the model to understand which factors are the most influential. The model has to be significant for an adjustment in factor ranges to have affect on the result distribution. In this example the model for k2 is very significant and the most important factor is Acetonitrile (ACN).
286
Assuming that the factor Temperature is easy to control with a narrower range we start with this factor by adjusting temperature to +/- 0.5 C. At the same time we can open the Role for ACN to 'Free'. This instruction together with the specification limit for k2 (DPMO = 1000) will give an estimate of a range for ACN where we can predict that the system is robust according to the specifications. The picture below displays the result of the settings.
The proposed settings for ACN are now 25.51 to 26.49 and the estimated distribution for k2 is 1410 hits outside the specification limits. A final step might be to make an adjustment of the factor settings to some practical new specification within the range for ACN, for instance 25.5 to 26.5. The result shown below implies that the critical response k2 will have 0.16% of future predictions outside the specifications.
287
To view statistics, right-click the Design Space window, and then click Design Space Statistics.
A more detailed description of this example is found in the tutorial named Robustness Testing.
288
References
1.
R.A. Fisher, Statistical methods, experimental design, and scientific inference, A re-issue (J. H. Bennett, Ed.), Oxford University Press, Oxford, England, (1990). G.E.P.Box, W. G. Hunter, and J. S. Hunter, Statistics for experimenters, Wiley, New York, (1978). G.E.P. Box, The collected works, Vol 1. (G.C. Tiao, Ed.), Wadsworth Advanced Books and Software, Belmont, CA, 1985. Morgan, Chemometrics: Experimental Design, ACOL, London, and Wiley, New York, (1991). Wold, Soft modeling, The basic design and some extensions, In Vol. II of K-G. Jreskog and H. Wold, Ed.s. Systems under indirect observation, Vol.s I and II, North-Holland, Amsterdam, (1982). Wold, A. Ruhe, H. Wold and W. J. Dunn III, The Collinearity Problem in Linear Regression. The Partial Least Squares Approach to Generalized Inverses, SIAM J. Sci. Stat. Comput. 5, 735-743, (1984). Hskuldsson, PLS Regression Methods, J.Chemometrics, 2, 211-228, (1988). Wold. Cross validatory estimation of the number of components in factor and principal components models, Technometrics 20, 397, (1978). Draper and Smith, Applied Regression Analysis, Second Edition, Wiley, New York.
2. 3. 4. 5.
6.
7. 8. 9.
10. Cornell, Experiments with Mixtures, New York: Wiley, (1990). 11. Cox, A Note on Polynomial Response Functions for Mixtures, Biometrica, 58, 155-159, (1971). 12. Crosier, Mixture Experiments: geometry and Pseudo components, Technometrics, 26, 209-216, (1984). 13. Kettaneh-Wold, Analysis of mixture data with partial least squares, Chemometrics and Intelligent laboratory Systems, 14, 57-69, (1992). 14. Rechtschaffner R.L., Saturated fractions of 2n and 3n factorial designs, Technometrics, 1967, Vol.9, N4, 569-575. 15. Ing-Marie Olsson, Erik Johansson, Martin Berntsson, Lennart Eriksson, Johan Gottfries, and Svante Wold, Rational DOE-protocols for 96 well plates, Chemometrics and Intelligent Laboratory Systems, 2006. 16. Doehlert, D.H., Uniform shell designs, Journal of the Royal Statistical Society, 1970, Serie C, N19, 231-239. 17. Snee, Test Statistics for Mixture Models, Technometrics, Nov. 1974.
289
18. Nordfelth, R., et al., 2005 Small-Molecule Inhibitors Specifically Targeting Type III Secretion, Infection and Immunity, 73, 3104-3114. 19. Design space; ICH Q8(R1) Pharmaceutical Development Revision 1, http://www.ich.org/cache/compo/276-254-1.html. 20. Quality by Design; http://en.wikipedia.org/wiki/Quality_by_Design
290
Index
Analysis Fit............................................ 136, 137 Menu ......................................... 15, 133 Of variance ............................. 150, 238 Phase ............................................... 2, 9 Wizard ............................................ 135 Analysis Advisor ................................... 45 Analysis menu Organization ................................... 133 Select Fit Method ............................ 136 Analysis of Variance ................... 150, 238 Analysis phase ......................................... 2 Analysis Wizard .................................. 135 Arrow .................................................... 41 Audit trail .............................................. 53 Augmenting designs .............................. 23 Auto update predictions....................... 176 Auto, MLR, PLS, Scheff ................... 137 Automatic fit ......................................... 59 Automatic update .......................... 59, 205 Autoscale modifier ................................ 77 Axes ............................................ 174, 207 Axial designs ....................................... 265 AXN .................................................... 265 AXR .................................................... 265
2
2D contour ................................... 167, 187
3
3D Labels..............................................216 Onion plots......................................115 Rotation response surface .................41 Scatter ............................. 115, 126, 176 Surface ............................................172 Zoom and rotate ................ 41, 214, 220
4
4D Contour ...........................................169 Sweet spot .......................................177
A
Abbreviation .................................... 65, 75 Absolute Limits ...................................181 Accelerators ......... 30, 31, 45, 64, 221, 224 Activation ............................................229 Add Command to Favorites ......................46 Experiments ....................................131 Factor .......................................... 27, 65 Inclusions ..........................................90 Response ..................................... 32, 75 Row ...................................................32 Squares and interactions.............. 34, 38 Terms ................................................34 To Favorites ......................................51 To Report .................................. 51, 228 Add to Favorites ....................................51 Add to Report ........................................51 Adjust according reference .......... 174, 243 Alpha level ............................................54
B
Balanced ...................................... 106, 257 Block Blocking............................................ 56 Blocks ....................................... 95, 102 BlockV ............................................ 131 D-Optimal designs .......................... 252 Interactions ............................. 102, 250 Mark ................................................. 41 Orthogonal ...................................... 250 Random..................................... 56, 253 RSM designs ................................... 252 Screening designs ........................... 251 Box Behnken designs .......................... 260 Box Whisker........................................ 128 Box-Cox plot ............................... 147, 243
291
Buttons Command menu bar ........................222 Format bar .......................................226 Generate report bar .........................226 Optimizer buttons............................185 Plot bar ..............................................41 Spreadsheet bar .................................41 Standard bar ......................................39 Status bar ..........................................44 Window bar.......................................44
PBSS to PB ..................................... 258 Screening to RSM Rechtschaffner .... 28 Condition Number ............................... 236 Confidence interval for predictions ..... 242 Confidence level Changing ........................................... 56 Used in .... 145, 149, 150, 173, 175, 176 Confoundings .............................. 152, 202 Constraint Candidate set..................................... 89 Constraints supported ....................... 86 Defining ................................ 85, 86, 87 Mixture ........................................... 264 Modifying graphically ...................... 88 Qualitative or multilevel ................... 89 Why .................................................. 85 Continue Edit Report With .......... 222, 226 Contour Customize ............................... 213, 214 From Optimizer .............................. 187 Insert text .......................................... 41 Labels ............................................. 214 Levels ............................................. 214 Options............................................ 214 Plot .................................. 167, 169, 187 Wizard .................................... 166, 187 Contour levels ..................................... 214 Controlled.............................................. 68 Convention ............................................ 10 Coordinate Reader ................................. 41 Copy .................................................... 188 Correlation................... 129, 162, 163, 232 Correlation matrix ......................... 60, 129 Correlation plot ................................... 129 Cross-validation rules .......................... 234 Cubic centroid ................................. 6, 265 Cubic model .................................. 34, 260 Cubic terms ......................................... 234 Curvature ............................................. 125 Customize Contour plot ............................ 213, 214 Header and footer............................ 209 Properties of the axes ...................... 207 Properties of the plot area ............... 210 Scatter plot ...................................... 216
C
Candidate runs .....................................112 Candidate set Constraints in qualitative and multilevel ..........................................89 Created by MODDE .......................268 Imported ..........................................111 Max size ............................................59 Candidate set section ...........................107 Case sensitivity ......................................10 CCC Design .............................................260 Settings .............................................99 CCF .....................................................260 Central Composite design Circumscribed Design .............................................260 Settings .............................................99 Centroid ................................. 34, 265, 272 Classical mixture designs ....................265 Coding qualitative factors....................241 Coefficients ................... 55, 152, 240, 241 Collinearity ..........................................264 Color Scatter plot ......................................176 Sweet spot plot ................................177 Worksheet .......................................131 Color by variable .................................176 Command menu bar ..............................13 Compatibility .......................................111 Complement design Doehlert ............................................27 D-Optimal .........................................25 Estimate square term .........................24 General..............................................23 Inclusions .................................. 90, 271
292
Index
D
Decimal ......................................... 71, 207 Default options ......................................52 Default plot settings.............................206 Default template .......................... 222, 226 Degrees of freedom .............................239 Delete investigation ...............................21 Deleted Studentized Residuals .... 146, 241 Derived responses Defining ............................................78 Operators...........................................81 Qualitative factors .............................82 Sets of variables ................................80 Syntax ...............................................81 Description button ...............................101 Descriptive Statistics ...........................128 Design Advanced designs .............................18 Import design from file .....................21 Traditional designs ............................18 Design augmentation ...........................271 Design Region ............................. 201, 202 Design Runs .......................... 97, 104, 112 Design Runs Span ...............................106 Design Space Buttons ............................................194 Estimation ............................... 190, 191 Statistics .................................. 197, 254 Validation .......................................284 Window...........................................192 Design Wizard ............................... 64, 117 Desirability ..........................................248 Detect curvature .......................... 124, 125 Determinant Criteria .................................... 106, 271 D-Optimal design maximizes ..........268 Log .......................................... 109, 271 Normalized Log ..............................109 DF residuals.........................................134 DFresid ................................................239 Diagnostic part ....................................152 Diagnostics .................................. 125, 143
Distance to model .......................................... 146 to target ........................................... 186 DModY ............................................... 146 Dockable Windows ............................... 44 Doehlert ......................................... 27, 262 D-Optimal Algorithm........................................ 269 Complement...................................... 25 Onion design ........................... 113, 274 Pages ....................................... 108, 122 Results ............................................ 108 Summary ......................................... 203 What ....................................... 103, 268 When....................................... 103, 272
E
Edge ............................................ 203, 265 Edit menu .............................................. 31 Edit model ............................... 34, 99, 104 Effects ................................................. 155 E-format ................................................ 52 Eigenvalues ................................. 134, 236 Email ..................................................... 22 Encrypt .................................................. 29 Estimate squares .................................... 23 Evaluate ............................... 109, 134, 271 Evaluation plot .................................... 184 Exclude button ...................................... 41 Exclude tool .......................................... 41 Execute folder ....................................... 46 Exit ........................................................ 30 Experiment Name .......................... 52, 131 Experiment Number ...................... 52, 131 Experimental cycle .................................. 1 Experimental design .......................... 5, 10 Exporting Favorites configuration ......... 50 Extended................................ 57, 200, 204 Extended Axial .................................... 265 External variability ...................... 124, 250 Extreme Vertices ................................. 203
293
F
Factor Add ...................................................32 Advanced page ..................................69 Definition ..........................................65 Max number ......................................67 Mixture and process ........................ 263 Modifying .........................................73 Name .................................................10 Qualitative................................. 82, 241 Scaling ...................................... 70, 235 Setting ...............................................67 Spreadsheet .......................................72 Transformation..................................70 Types ................................................67 Use ....................................................68 Factorial designs .......................... 257, 260 Favorites Add command ...................................46 Add to Favorites................................51 Button ...............................................39 Create a new folder ...........................46 Restore ..............................................61 Window....................................... 45, 46 F-distribution ............................... 149, 150 Filler ................................................ 67, 68 Fit Goodness of ....................................239 Lack of ............................ 140, 149, 150 Model ...................................... 136, 137 Multiple Linear Regression (MLR).231 Partial Least Squares (PLS) ............232 Summary .........................................139 Fit methods .............................. 7, 231, 232 Fold over ......................................... 23, 24 Folder .............................................. 45, 46 Footer .......................................... 209, 214 Format Axes ................................................207 Bar in report ....................................226 Header/Footer .................................209 Menu in report ................................225 Plot area ..........................................210 Formulation ............................. 67, 68, 263 Fractional factorial Complement ......................................23 Design .............................................257 Free Mark ..............................................41
294
Full factorial design ..................... 257, 260 Full Factorial Mixed ............................ 257 Full Screen ...................................... 44, 51
G
G-efficiency......................... 108, 112, 203 General Options General page ..................................... 59 List presentation................................ 57 Restore .............................................. 61 General page.................................... 59, 66 Generate D-Optimal button ................. 107 Generate new ............................... 107, 186 Generate Report bar ............................ 226 Generate Report window ..................... 222 Generators ................................. 37, 38, 99 Grab Plot ..................................... 224, 226 Graeco-Latin square ............................ 258 Gridlines ...................................... 207, 212
H
Hat matrix ........................... 239, 241, 242 Header ......................................... 209, 214 Help ............................................. 226, 229 Help menu ........................................... 229 Help-button ......................................... 229 Hierarchy ............................................. 234 High Limit ........................................... 181 Histogram ............................................ 127 Hypercube ........................................... 260 Hyperlink ............................................ 224 Hyper-triangles .................................... 262
I
Image................................................... 224 Import Candidate set............................. 18, 111 Design from file ................................ 21 Favorites configuration ..................... 50 Scores ....................................... 18, 111 Worksheet to inclusions .................... 92
Index
Inclusions Added to the worksheet.....................90 Design augmentation ......................271 Editing inclusions..............................92 Generating.........................................92 Part of the design ..............................91 vs. complement design ......................90 Increment.............................................207 Insert Image .........................................226 Insert Rows.............................. 32, 92, 176 Installation ...............................................1 Interactions .............................. 34, 99, 158 Interior points ......................................265 Interpreting the model .........................152 Investigation Options.............................52 Irregular region ....................................272
M
Main Effect.................................... 58, 157 Manage licenses .................................. 229 Maximum runs .................................... 199 Menu item ............................................. 10 Mid-range ...................... 70, 235, 236, 240 Minimum ..................................... 207, 248 Mip-file ................................................. 11 Missing ................................................ 237 Mixture and process factors .................. 247, 263 Constraint........................................ 264 Contour plot wizard ........................ 166 Data................................................. 243 Designs ........................................... 263 Experimental region ........................ 264 Factor definition.............................. 263 Hierarchy ........................................ 243 Irregular region ............................... 272 Models ............................................ 243 Prediction plot wizard ..................... 173 MLR scaling .................... 70, 77, 235, 236 MLR solution from PLS ...................... 234 MODDE .................................... 1, 11, 230 Model Distance .......................................... 146 Fit............................................ 136, 137 Hierarchy ........................................ 234 Individual ........................................ 134 Interpret .......................................... 152 List .................................................. 204 Saturated ......................................... 142 Select .......................................... 38, 95 What is ................................................ 5 Model bar .............................................. 44 Model predictive power....................... 233 Model validity ..................................... 140 Multilevel factor .................................... 67 Multiple Linear Regression ............. 7, 231 Multiplots ...................................... 39, 206 Multivariate ................................... 18, 232
L
Labels ..................................................176 Lack of Fit Alpha level ........................................54 ANOVA ..........................................238 DF ...................................................134 Model validity .................................140 Plot ..................................................149 Latent structures ..................................232 Layers .................... 59, 112, 258, 261, 274 Layers overlap .....................................112 L-designs ................................... 5, 18, 258 Legend ......................................... 210, 214 Line style .............................................210 Linked responses ...................................82 List presentation ....................................57 Loading plot ................................ 162, 163 Lock Contour levels .................................213 Investigation......................................29 Log Determinant ............................ 109, 271 In audit trail.......................................53 LogDet.................................................271 LogdetNorm ................................ 109, 271 Low Limit............................................181
N
Network installation ............................ 229 New ......................................... 11, 17, 226
295
New From Default Template ...............226 New worksheet from file .......................21 Next Component Analysis menu......................... 137, 138 Extract .............................................138 MLR solution from PLS .................234 No Mark ................................................41 Non-English ........................................214 Normal probability ...................... 143, 156 Normalized coefficients................. 55, 152 Normalized Log Determinant ..............109 Number format ................................ 52, 60 Number of decimals ...................... 71, 207 N-value ................................................237
Options Customize ......................................... 61 Default .............................................. 52 General.............................................. 58 List .................................................... 60 Scaling .............................................. 70 Options button ..................................... 169 Organization .................................. 12, 133 Orthogonal Blocking.................................. 102, 250 Correlation ...................................... 129 Scaling .............................. 70, 235, 236 Scaling by MODDE ................ 111, 264 Settings ........................................... 200 Outliers ........................ 143, 146, 162, 232 Output.............................................. 50, 59 Overall distance to target ..................... 184 Overlay prediction ....................... 173, 174
O
Objective ................................... 5, 94, 199 Observed vs Predicted ................... 41, 148 Office 2003.............................. 59, 61, 224 Older MODDE versions ........................11 Onion Candidate set ...................................111 Design ............................. 258, 261, 274 Generate .................................... 18, 113 Layers max........................................59 Plots ................................................115 Onion D-Optimal ......................... 112, 113 Open All Items ......................................46 Open investigation .................................21 Operators ...............................................81 Optimization criteria............................181 Optimizer Copy to prediction list .....................188 Definition ........................................248 Generate new ..................................186 Generating the start runs .................185 Lists ........................................ 181, 183 Log ..................................................183 Play button ......................................186 Running the Optimizer ....................186 Optimizer buttons ................................185 Optimizer runs .....................................186
P
Partial Least Squares ....................... 7, 232 Password protect ................................... 29 Paste Design .......................................... 94 Paste unformatted ................................ 224 PBSS ................................................... 258 Percent......................................... 113, 177 Placeholders ........................ 224, 226, 227 Plackett Burman .................................. 258 Plackett Burman Super-Saturated........ 258 Plate-Size ...................................... 18, 259 Plot area....................................... 210, 214 Plot bar .................................................. 41 Plot Labels............................................. 52 Plot Loadings .............................. 163, 242 Plot options.................................. 166, 213 Plot Settings 3D scatter ........................................ 216 Axis................................................. 207 Header and footer............................ 209 Plot area .......................................... 210 Plots....................................... 16, 205, 228 PLS plots ............................. 162, 163, 242 PLS Scaling ............................. 71, 77, 235
296
Index
PLS summary list ................................142 PLS Summary plots ..................... 139, 141 Potential terms ............................. 203, 269 Prediction Copy from optimizer .......................188 Derived responses .............................78 Formula ...........................................242 Linked responses ...............................82 Plot wizard ......................................173 Response prediction plot .................175 Scatter plot ......................................176 Sweet spot plot ................................177 Prediction List .....................................176 Prediction plot wizard..........................173 Predictivity ..............................................4 PRESS/SSY .............................................7 Print .......................................................30 Print format dialog.................................30 Print preview ................................... 30, 39 Print setup..............................................30 Print-button............................................30 Process and mixture factors .................263 Product ID ...........................................229 Program limits ............................... 59, 107 Properties window in report ................228 Property page............................... 205, 213 Protect investigation ..............................29 Pseudo components ..................... 137, 200 Pseudo-resolution ................................250 Pure error ..................................... 134, 150
Range Contour level range ......................... 214 Curvature ................................ 124, 125 Factor ........................................ 67, 181 Recalculate Scale button ..................... 207 Rechtschaffner design ................. 259, 261 Recommended designs .................. 95, 274 RED-MUP ....................... 18, 59, 259, 261 Reduced Axial ..................................... 265 Reduced CCC ...................................... 260 Reduced CFF ....................................... 260 Reference mixture ......................... 34, 204 Regions................................................ 214 Register ............................................... 229 Regression ........................................... 231 Regular presentation ...................... 57, 200 Regular responses .................................. 77 Remove All Placeholders .............................. 225 Button ................................. 32, 34, 214 Encryption ........................................ 29 or add header/footer ........................ 209 Placeholder ............................. 225, 226 Terms ................................................ 34 Rename...................................... 46, 61, 82 Repetitions................................... 106, 112 Replicate plot ...................................... 130 Replicate tolerance ................ 59, 130, 201 Replicated experiments . 97, 130, 134, 150 Report Generator ........................... 22, 221 Report Generator FAQ window .......... 226 Report/template ................................... 226 Reproducibility ............................ 140, 141 Reset button........................................... 34 Residual Standard Deviation ............... 237 Residual types ..................................... 146 Residuals ....................... 58, 144, 145, 241 Residuals plots .................................... 144 Resolution III ................................ 23, 257 Resolution IV designs ......................... 257
Q
Q2 ................................................ 140, 239 Qualitative factors ................... 67, 82, 241 Quantitative ............................. 67, 69, 263 Quantitative multilevel .............. 67, 68, 69
R
R2 .................................. 57, 139, 142, 239 Random ................................... 56, 61, 253
297
Response Add ........................................... 32, 119 Box............................................ 39, 206 Definition .................................... 75, 83 Exclude using modifier .....................77 Manipulations ...................................84 Name .................................................10 Regular..............................................77 Spreadsheet .......................................83 Sweet spot .......................................177 Response box................................. 39, 206 Response Surface Modeling ...... 6, 94, 260 Response value label .............................41 Restore...................................................61 Revert ....................................................22 Rich Text Format...................................50 Robustness ................................... 192, 284 Role .....................................................181 Rotate ............................................ 41, 220 Rotation ................................. 41, 207, 220 RSD ............................................. 237, 239 RSD*sqrt .............................................150 RSM Objective ................................ 94, 95 Run list ................................................183 Run Optimizer ............................. 185, 186 Run order ....................... 52, 124, 131, 144 Runs....................................... 97, 186, 268
Save Settings button ............................ 206 Scale Contour plot ............................ 213, 214 Options.............................................. 70 Scale XY ............................................... 41 Scaled coefficients............................... 240 Scatter plot 3D or 2D ......................... 115, 126, 176 Loading ................................... 162, 163 Onion .............................................. 115 Plot settings..................................... 216 Score ............................................... 162 Scheff ........................................ 137, 236 Scientific ....................................... 52, 207 Score ............................. 18, 111, 162, 274 Score column plot................................ 162 Score vectors ....................................... 162 Screening ....................................... 94, 257 Screening complemented ...................... 23 ScreenTips ............................................. 61 SD ............................................... 149, 150 SD Regression ..................................... 150 SD-LoF................................................ 149 SD-pe .................................................. 149 SD-pe*sqrt........................................... 149 SDY..................................................... 141 Select all ................................................ 31 Select fit method.................................. 137 Send by e-mail....................................... 22 Set Run Order ................................ 15, 124 Settings Button ............................................... 38 Copy factor settings ........................ 188 D-Optimal ....................................... 122 Factor ................................................ 67 Investigation ..................................... 59 Plot .................. 206, 207, 209, 210, 214 Restore .............................................. 61 Saving ..................................... 206, 212 Show ............................................... 200 Shaded ................................................. 214 Shortcut menu ..................................... 205
S
Safe Region ................................. 190, 191 Saturated Models .................................142 Save Audit trail ..........................................53 Button ....................................... 39, 226 Inclusions ..........................................90 Investigation......................................21 Plot or list .................................. 21, 205 Plot settings ............................. 206, 212 Save As.......................................... 21, 205 Save Format dialog ................................21 Save List As .................................. 21, 205 Save Plot As .................................. 21, 205 Save Settings ............................... 206, 212
298
Index
Show All Placeholders ..............................225 Details button ....................................58 Menu .................................................16 Placeholder...................... 225, 226, 227 ScreenTips on toolbars ......................61 SimC ....................................................265 SIMCA-P............................... 18, 111, 274 SimF ....................................................265 SimM ...................................................265 Simplex ...............................................264 SimSC .................................................265 Slack variable ......................................272 Slide.......................................................61 Sort ................................................ 32, 155 Special Cubic Model ...........................265 Split objective .................................. 94, 99 Spreadsheet................................ 41, 72, 83 Square terms ..........................................95 SS ........................................ 139, 140, 238 SS explained ........................................232 Standard bar...........................................39 Standardized residuals ................. 146, 241 Star Distance .........................................99 Start runs .............................................185 Status bar ....................................... 44, 224 Stdev................................................ 80, 81 Styles ...................................................225 Summary D-Optimal ...........................203 Summary List .............................. 141, 142 Summary plot .............................. 139, 141 Surface......................................... 172, 214 Sweet Spot Plot ........................... 177, 189 Symbol Style .......................................210 Syntax for derived responses .................81 System default .......................................61
Templates ............................ 222, 224, 226 Text file ................... 21, 92, 107, 111, 224 Theme.................................................... 59 Tile Windows ........................................ 44 Time .................................................... 207 Time stamp .......................................... 209 Title ............................................. 207, 209 Toolbars Command menu in report ............... 222 Customize ......................................... 61 Format bar in report ........................ 226 Generate report bar in report ........... 226 Model ................................................ 44 Plot .................................................... 41 Reset ................................................. 61 Spreadsheet ....................................... 41 Standard ............................................ 39 Status ................................................ 44 Window ............................................ 44 Tools menu .......................................... 225 Traditional designs ................................ 18 Transformations .............. 70, 77, 147, 264 Treat folder as item ............................... 46 Txt ........................... 21, 92, 107, 111, 224
U
Uncentered .......................................... 236 Uncentered coefficient ........................ 240 Unconfound ........................................... 99 Uncontrolled .................................. 68, 237 Undo .................................. 31, 34, 39, 226 Unfold ................................................... 61 Unit Variance .......................... 70, 71, 235 Unit variance coding ................... 137, 236 Unit variance scaling ..................... 70, 235 Units ................................................ 65, 75 Update Constraint-button .............................. 88 Placeholder button .......................... 227 Placeholders .................... 225, 226, 227 Predictions ...................................... 176 Report ....................................... 22, 225 URL..................................................... 224 Use True Type ..................................... 214
299
T
Tagushi ....................................................5 Target in Optimizer ..................... 181, 248 T-button ................................. 41, 167, 169
V
Variable importance ............................161 View menu .................................... 14, 224 View/Hide Toolbars ..............................14
W
WC plots...................................... 163, 242 Weight ......................................... 163, 181 Window Bar ....................................................44 Dockable ...........................................44 Menu .................................................16 Toolbar..............................................44 Wizard Analysis Wizard ..............................135 Contour plot wizard ........................166 Design wizard ...................................64 Prediction plot wizard .....................173 Workset ....................................... 111, 274
Worksheet Add experiment ................................ 32 Copy ................................................. 31 Description...................................... 131 Design matrix.................................. 200 Menu ............................................... 123 Missing data .................................... 237 Open ................................... 39, 41, 123 Sort ................................................... 32
X
X Mark .................................................. 41 X'X Condition number ........................... 236 Eigenvalues ............................. 134, 236 Matrix ..................... 258, 260, 268, 271
Y
Y Mark .................................................. 41 Y-miss ................................................. 237
Z
Zoom ............................................. 41, 220
300