Optimal Design Window
The Optimal Design window is used to select test runs from a design table in order to minimize the estimation uncertainty of the coefficients in a given regression model. It takes into account a) the specific effects that are under investigation, b) limitations on the number runs that can be performed and c) constraints on the factor level combinations that can be used during the experiment. The ReliaWiki resource portal provides more information on optimal custom designs at: http://www.reliawiki.org/index.php/Optimal_Custom_Designs.
To access the Optimal Design window, build a design and then choose Data > Design > Optimal Design.
The window includes two tabs discussed next.
Design Settings Tab
This tab is used to specify the method that will be used to create the design. Follow the steps outlined below to enter the settings.
- If you wish to optimize the design for a regression
model, select the Regression
Model-Based Optimal Design option at the top of the
window.
- In the Desired Model Terms area, select which terms will be included in the regression model.
- In the Number of Runs in the Optimal Design area, specify the number of runs that will be used in the optimal design. The minimum and recommended numbers of runs are calculated automatically by the software. If you want to use a different number of runs, select the Custom number of runs option and then enter a number in the input field.
- If desired, you can modify the method for creating
the design in the Algorithm
Settings area.
The first section in this area concerns the random elements that are used in the generation of the design.- Select the Use Seed check box if you would like to set a consistent starting point from which the random numbers will be generated. Using the same seed value and keeping all other settings the same will allow you to replicate the resulting design.
- Specify the Number of random initial runs that will be used. The software uses each random initial run to generate a design (using the algorithm you specify next), and then returns the best design that was found. Using more initial runs may produce a better design, but it will also increase the time required to generate the final design.
The second section is used to select the algorithm that will be used to generate the design. Three algorithms are available.
|
How likely to produce the optimal design? |
How fast? |
Federov's Method |
Most likely |
Slowest |
Modified Federov's Method |
Somewhat likely |
Somewhat fast |
K-Exchange Method |
Least likely |
Fastest |
- If you want the design points (i.e., runs) in the design
to cover as large of a design space as possible, select the
Distance-Based Optimal Design
option.
- Then use the Number of Runs in the Optimal Design area to specify the number of runs that will be used in the optimal design. The minimum and recommended numbers of runs are calculated automatically by the software. If you want to use a different number of runs, select the Custom number of runs option and then enter a number in the input field.
Candidate Runs Tab
This tab is used to select which runs will be considered in the final design. By default, the table on the Candidate Runs tab shows all the runs used in the current design. If desired, you can modify the factor levels in any of the runs.
Follow the steps outlined next to choose the runs to consider and, if desired, add/remove runs.
- In the Included in
optimal design? column, make sure the appropriate option
is chosen for each run:
- Choose Consider if you are not sure whether the combination should be included in the optimal design. The software will use the settings you specified on the Design Settings tab to determine whether the combination will be included.
- Choose Include if want that combination to be used in the optimal design (e.g., you might be interested in examining the response at a particular combination).
- Choose Exclude if you do not want to include the combination (e.g., there might be safety concerns with setting all factor levels to the highest level).
- If you want to add a new run to the table, click the Add button at the bottom of the window. The software will create a new row in the table with center points for all the factor values. If you do not wish to use center points, you can enter new factor values. To remove an added row, select it and click the Delete button.