A very flexible approach to multivariate data analysis and design of experiments
Unscrambler statistical software is geared to two of the most useful areas of industrial R&D, namely multivariate analysis and experimental design. The latest version (10.3) of this useful niche software has a number of additions and upgrades, including regression and classification methods, exploratory data analysis tools, predictive modeling, extensive pre-processing options, and descriptive statistics with tests. Originally targeted to chemical engineers, it is now useful for a number of industrial disciplines and includes transformation and validation routines for analytic instrumentation and process control equipment. Technical Specifications/requirements are shown in Table 1.
For the cut-and-paste crowd, as well as the drag and drop group, Unscrambler makes data importation ludicrously simple. You can use Excel to highlight the data cells, then merely drag it to the Unscrambler screen and drop it. It immediately appears in the project spreadsheet. Alternatively, you may use File/Import Data, and there are a number of file types that may be imported. The next step is to classify/categorize/format the data. This is my personal gripe with many software packages and Unscrambler is, unfortunately, no exception.
Rather than intelligently categorizing the data (which to some extent it does) and easily letting you specify the predictor and predicted variables, this software demands several steps and filling out a dialog box, which I didn’t find as intuitive as I would like. However, once this hurdle is over and the method becomes more automatic to the new user, the software will amaze with numerous plots and calculations. The interactive nature of the graphics allows drill-down to individual data points, zoom and rotate the graphics, as well as label important areas with arrows and text. As is the custom with much modern software, the program comes with a Navigator pane and project-based workflows to ensure rapid access to each area.
The main panels, taskbar and spreadsheet will be familiar to most Excel users. The data shown in Figure 1 was previewed in the Excel Preview box, which comes up automatically upon importing the data (Figure 2). Plots of simple data means and deviations are readily constructed when Descriptive Statistics are requested (Figure 3). Of course, the more detailed 3-D plots (scatter, line, or matrix) may be easily and quickly produced (Figure 4).
Specific statistical tests each have their own sets of diagnostic graphics, as well as the summary tables (as for Principal Components analysis) — shown in Figure 5.
Both the Navigation pane and the bottom tabs allow for rapid movement among graphics, warnings, and data.
There are even helpful dialog boxes with outlier warnings and consequences. Once data formatting is conquered, the rest is downhill, as there are a number of helps for the novice user. The tutorials are mercifully detailed and there are many of them, including full case studies.
This software also contains numerous transformations for spectroscopic data. Also, the workflow smoothness lends itself to enhanced business value. The combination of experimental design, widely used in most manufacturing sectors, coupled with multivariate analysis, including prediction and classification, extends the knowledge base to research usage and more detailed knowledge of how a system works and performs.
Lastly, the addition of specialized plots, calculations of process limits, and inclusion of audit trails greatly enhance the quality assurance work that is a staple of modern process control. Interested readers are urged to try the 30-day free trial.
• $6,994 (single seat, industrial)
One Woodbridge Center, Suite 319, Woodbridge, NJ 07095
1-732 726 9200; 1-973 556 1229
John Wass is a statistician based in Chicago, IL. He may be reached at editor@ScientificComputing.com.