New Modeling Capabilities

Logistic regression for binary responses

When results are expressed as binomial data (ie – 0/1), ANOVA doesn’t work, and logistic regression becomes the analysis method of choice. It’s a predictive analysis that gives an estimate of the probability of success based on input settings. Perfect for indicating outcomes in scenarios like pass/fail.

Kowalski-Cornell-Vining (KCV) models 

Experiments that combine mixture components with process factors can have a large number of runs involved to gather the necessary data. By eliminating the high order model terms, the KCV models can greatly reduce the number of required runs, saving you time and money.

New Interface Improvements

New graphic options

Histogram and 3D scatterplots are now available. Visualize your data the way you want! Legends can now be customized in many different ways, including location and font size.

Side-by-side view of factorial effects plots

View the half-normal and Pareto plots of effects next to each other for better analysis.

Multi-graph notebook view

Have multiple graphs on one screen. You pick the graphs and order. Choose the results and views that help you visualize your modeled responses.

Many more features in DX12

Logistic Regression

• Responses with binary data (every cell is a 0 or 1) will have a Logistic Regression option on the Transform tab.
• Chi-squared tests for logistic regression model term significance.
• McFadden, Adj. McFadden, and Tjur pseudo-R-squared statistics are available for logistic regression.
• Binary responses can be optimized using criteria based on the probability of success or failure.
• When analyzing with logistic regression, you may now change which level is considered a “success” on the Transform tab.
• The Confirmation node now shows a mean and intervals for confirmation data entered for a binary response.
• Logistic regression can now be performed on any response that consists of two values, not just responses with only 0s and 1s.

• Kowalski-Cornell-Vining (KCV) models are available as a Model Order choice.
• KCV models provide a more efficient alternative to a full crossed model when building.
• KCV models can be selected during the analysis as well.

Interface

• The font size in the simulation editor, via the Dialog Control font preference.
• Icons have been added to the new graph menus.
• The factors tool now appears in optimization graphs for single factor designs.
• The 3d Surface graph title is now editable.
• Model Graphs have been moved to a notebook interface.
• You can create multiple graphs simultaneously in side by side views.
• All graphs now have a dockable legend that can be moved independently of the graph.
• The legend can also be docked to the right or bottom of the graph via the right-click menu.
• Elements in the legend can be toggled individually via the right-click menu.
• The font size of the legend can be increased or decreased by hovering over it, holding control, and using the mouse scroll wheel.
• A histogram of any column of data can be generated in the Graph Columns node.
• You can set a third axis on the Graph Columns scatterplot to make it into a 3d scatterplot.
• Effects graphs have been moved to a notebook interface and can be compared side by side.
• The Cube plot can be toggled between observed and predicted values.
• Pairwise comparisons when clicking on a point in Interaction or One Factor plots are now available in a separate toolbox.

Categoric Coding

• Categoric variables can now use Helmert contrasts, which compares each level to the mean of the remaining levels.
• Treatment contrasts are also available, which compares each level to a user-selected control, or reference level.
• Ordinal contrasts no longer need to be numeric. If they are not numeric they are assumed to be evenly-spaced (e.g. “Low”, “Medium”, “High”).
• The choice of contrast types for categoric factors is shown in the ANOVA.