With the addition of Design of Experiments (DOE) functionality to the 2013 release, Quantum XL obtains an unprecedented state in the market. In one application, you can find statistical analysis, DOE, Monte Carlo Simulation, and companion tools (QFD, FMEA, Pugh).
Before we implemented the DOE functionality, we spoke with hundreds of users about what they wanted. Almost universally, the first words they uttered were “Easy to Use”. Of course, “easy” varied greatly based on the user. One particularly vocal customer voiced frustration with a different software application’s menus. Calculating the regression with all Quantitative inputs, some categorical inputs, and logistic outputs required clicking on many different menu items. Further still, if the data was historical (not from a Designed Experiment) then the regression results were displayed differently and some graphing and optimization functionality was lost. “Why,” he asked, “would you do this to a customer?”
With Quantum XL, we’ve packed an incredible amount of power into a tiny menu. You don’t have to remember which type of regression to use with quantitative outputs vs. logistic, just click on Analyze Design and Quantum XL runs the correct regression for you. Design reduction, prediction, optimization, and graphing for quantitative outputs works exactly the same as for logistic outputs.
Design of Experiments (DOE) Functionality
Quantum XL 2013 includes multiple design creators, such as a 2/3/Mixed level factorial builder, Taguchi, Plackett-Burman, CCD, Box-Behnken, custom, and even D-Optimal Designs. Quantitative, binary, and nominal outputs are handled seamlessly with Mixed Response Modeling. Support for blocking, folding, and randomization is fully implemented.
Most users will probably rely on the Design Wizard. The wizard steps you through the process of selecting the best design based upon the number of factors, type inputs, type outputs, and other design needs.
Advanced users can directly choose a design from the many design builders.
D-Optimal design creation has been simplified and helps prevent the selection of poor designs. Any design, including full/fractional, historical, or custom designs, can be used as the candidate runs. Quantum XL can generate a D-Optimal design with the target number of runs +/- N runs. For example, the results below are for 36 +/- 2 runs. For each of the D-Optimal designs, the Average VIF and Maximum VIF are displayed so that you may make an informed decision as to which design to choose.
Quantum XL seamlessly analyzes Quantitative and Categorical inputs and outputs.
When you click on “Run Regression”…
Quantum XL uses the correct type of regression automatically, without making you remember when to use Binary Logistic vs. Nominal Logistic vs. Ordinary Least Squares.
The regression results are written to a new worksheet …
… along with a prediction area.
Quantum XL includes an advanced optimization algorithm that allows constraints in addition to the objective. The constraints and the objectives can be based on quantitative, binary, and nominal. For more information, see the Mixed Response Modeling article.
Quantum XL supports main effect, surface, interaction, and residual plots. You can also block in replicates or via aliasing. We’ve included a regression advisor, which checks your models for problems such as poor orthogonality and high residual values.
While the bulk of the new features have been around DOE, several new non-DOE features have been added as well. In particular, we’ve added time series, correlation coefficient, and distribution calculators.
Quantum XL includes Trend Analysis, Moving Average, Single and Double Exponentially Weighted Moving Average, Cross Correlation, Auto Correlation, and Partial Auto Correlation.
Correlation and Covariance
Pearson’s and Spearman’s Correlation Coefficient as well as co-variance calculations have also made their way into the 2013 release.
And now you can find Binomial, Normal, and Weibull calculators with graphs in Quantum XL.