Version 8 Enhancements to SAS/INSIGHT Software
SAS/INSIGHT software provides an intuitive point-and-click interface environment for exploratory data analysis. You can use it to visualize and understand your data quickly through dynamic graphics and advanced statistical techniques. With Version 8, SAS/INSIGHT software is updated with several new graphical and statistical features.
New graphical enhancements include 3D surface plots, contour plots, 3D response surfaces, mean comparison circles for multiple comparisons, and color blending of up to five colors. Multivariate techniques include advanced principal components with biplots, canonical correlation analysis, maximum redundancy analysis, and canonical discriminant analysis. Also included are new robust measures of scale and tests of normality for univariate data as well as tests for differences across groups.
Graphics
Surface and contour plots are now available in SAS/INSIGHT software. You can produce a surface plot using the Rotating Plot task, fit a response surface in the Fit task, and select the new Contour Plot task from the Analyze pull-down menu. The software provides two methods for fitting surface and contour plots: linear interpolation and thin-plate smoothing spline. Although linear interpolation is faster and in many cases sufficient, the thin-plate smoothing spline produces a smoother surface.
Figure 1: Contour Plot using Thin-Plate Smoothing Spline
From the Fit dialog, you can request either the thin-plate spline or kernal method to fit a response surface to your model. Through this approach, you have access to the fitted surface R-square, mean squared error, parameter estimates, and dynamic bandwidth control. Using a slider control, you can interactively modify the value of the bandwith to see how the shape of the surface changes.
Figure 2: Surface Plot using a Thin-Plate Smoothing Spline
The color palette in the Tools window has been enhanced to support five-color blending. You can now drop a color in the mixing strip in one of five positions and thus create a multiple color mixture to display more information.
With Version 8, SAS/INSIGHT software adds extensive principal component analysis capabilities, including standardized and raw regression and scoring coefficients, biplots based on standardized or raw interval variables, and component rotations. Biplots simultaneously display observations as points and interval variables as vectors in plots of principal components. They are useful for examining data patterns and outliers. The software also provides the equamax, orthomax, parsimax, quartimax, and varimax rotation methods to orthogonally transform principal components to obtain factors that are more easily interpretable.
Figure 3: Biplots in Principal Components Analysis
Also included are canonical correlations for analyzing two sets of interval variables to reduce their high-dimensional relationship into a few canonical variables. This analysis finds successive linear combinations such that the canonical correlation between two variables is maximized. You can test series of hypotheses that each canonical correlation and all smaller correlations are zero, request two- and three-dimensional plots of canonical variables, and produce canonical correlation component biplots.
Maximum redundancy analysis is another new capability provided in Version 8. Given two sets of variables, maximum redundancy analysis finds a linear combination from one set of variables that best predicts the variables in the opposite set. SAS/INSIGHT software uses either raw or standardized variances, provides raw and standardized scoring coefficients, and produces two- and three-dimensional maximum redundancy plots and biplots.
SAS/INSIGHT also includes canonical discriminant analysis, which is a dimension-reduction technique for deriving canonical variables from a set of interval variables to summarize between-class variation in a categorical variable. The software provides computations for pooled within-class or total-sample variances equal to one, likelihood ratio tests for signficance of canonical correlations, and biplots for standardized or raw canonical discriminant components.
Figure 4: Comparison Circles
Several methods are provided for investigating differences across groups of a categorical variable. By choosing Comparison Circles from the pop-up menu of a box plot, you can request a test for multiple comparison of means at a given confidence level. Multiple comparison methods include student’s paired t-test, the Tukey-Kramer method, Bonferroni’s paired t-test, Dunnett’s t-test, and Hsu’s multiple comparisons with the best and worst.
For more detail, refer to the paper New Features in SAS/INSIGHT in Version 7 by Marc-david Cohen et al.
Statistics and Operations Research Home Page | What's New in Data Analysis