Statgraphics Centurion 19, released in 2020, is the latest version of our flagship Windows desktop product. It is a major upgrade, adding many new features for statistical analysis, data visualization and predictive analytics. Highlights include a new user interface, important enhancements to the Design of Experiments Wizard, new machine learning procedures, and methods for exchanging data and scripts with Python.
To discover what Statgraphics 19 has to offer: click the download button above, watch the video link below, check out the statistical capabilities listed in the left sidebar, or simply scroll through this page.
Contents - a complete list of all statistical tests and procedures available in Statgraphics 19.
Version 19 Enhancements
Statgraphics 19 is a major upgrade that contains many new features, including:
A new graphical user interface featuring a ribbon bar which makes it easy to find the analysis you want to use and the controls to select options.
An interface to Python, with facilities for exchanging data, executing scripts, and accessing libraries such as the K-means clustering procedure.
Over a dozen new statistical procedures, including equivalence analysis and noninferiority testing for variances, quantile regression, piecewise linear regression, zero-inflated Poisson and negative binomial regression, fitting mixtures of univariate and bivariate normal distributions, machine learning procedures such as decision forests, waterfall plots, and Venn and Euler diagrams.
Newalias optimal designsin the DOE Wizard, together with a feature for selecting optimal runs to enhance an existing design.
A dashboard that indicates the status of selected procedures such as control charts and capability indices.
A new Dashboard has been added to the set of StatFolio windows that can display tables and graphs from different analyses side-by-side. For procedures such as control charts, capability analyses, regressions, stock charts and measurement studies, the background of a table or graph can be colored green, yellow or red to indicate the status of selected indices, large changes or unusual residuals. Watch Video
The Distribution Fitting (Univariate Mixture Models) procedure fits a distribution to continuous numeric data that consists of a mixture of 2 or more univariate normal distributions. The components of the mixture may represent different groups in the sample used to fit the overall distribution, or the mixture model may approximate some distribution with a complicated shape. The procedure fits the distribution, creates graphs, and calculates tail areas and critical values. Tools are provided for determining how many components are needed to represent a data sample. Watch Video
The Distribution Fitting (Bivariate Mixture Distributions) procedure fits a distribution to continuous numeric data that consists of a mixture of 2 or more bivariate normal distributions. The components of the mixture may represent different groups in the sample used to fit the overall distribution, or the mixture model may approximate some distribution with a complicated shape. The procedure fits the distribution and creates graphs of the fitted model. Tools are also provided for determining how many components are needed to represent a data sample. Watch Video
The Piecewise Linear Regression procedure is designed to fit a regression model where the relationship between the dependent variable Y and the independent variable X is a continuous function consisting of 2 or more linear segments. The function is estimated using nonlinear least squares. The user specifies the number of segments and initial estimates of the locations where the segments join. The procedure then estimates the slopes, slope changes, and the locations at which the slope changes occur. Watch Video
Stability studies are commonly used by pharmaceutical companies to estimate the rate of drug degradation and to establish shelf life. Measurements are typically made on samples from multiple batches at different points in time. Of primary interest is estimating the time at which the lower prediction limit from the degradation model crosses the lower specification limit for the drug. Depending on the structure of the data, batches may be treated as either a fixed or random factor. Watch Video
The Quantile Regression procedure fits linear models to describe the relationship between selected quantiles of a dependent variable Y and one or more independent variables. The independent variables may be either quantitative or categorical. Unlike standard multiple regression procedures in which the model is used to predict mean response, quantile regression models may be used to predict any percentile. Median regression is a special case where the quantile to be predicted is the 50th percentile. Watch Video
New Alias-Optimal designs generated by the DOE Wizard consider not only the precision in the estimated model coefficients but also potential bias in those estimates caused by active effects that are not in the assumed model. Criteria such as D-optimality do not take into account aliasing caused by omitted effects. Sometimes, alternative D-optimal designs may be subject to considerably different amounts of aliasing. At other times, a small reduction in the efficiency of the selected design may result in a large reduction in potential bias. Watch Video
A new feature added to the DOE Wizard is the ability to add runs to an existing experiment so as to maximize a selected optimality criterion. The user first selects the number of runs to be added and then completes the dialog box shown below. Watch Video
New procedures have been added for demonstrating the equivalence or noninferiority of population variances. One procedure compares the variance of a single sample to a target value, while the other compares the variances of samples taken from 2 different populations. In the second case, the samples are considered to be “equivalent” if the ratio of their respective standard deviations falls within some specified interval surrounding 1. Watch Video
The GLM Method estimates the repeatability and reproducibility of a measurement system based on a study in which m appraisers measure n items r times. It also estimates important quantities such as the total variation, the precision-to-tolerance ratio, the standard deviation of the measurement error, and the percent of study contribution from various error components. In addition to variation introduced by appraisers and parts, additional factors may also be included. The additional factors may be treated as having either fixed or random effects. Note: This procedure will handle unbalanced data. Watch Video
The Decision Forests procedure implements a machine-learning process to predict observations from data. It creates models of 2 forms:
Classification models that divide observations into groups based on their observed characteristics.
Regression models that predict the value of a dependent variable.
The models are constructed by creating a large number of decision trees and averaging the predictions made from those trees. Many trees are constructed using a procedure similar to that of Classification and Regression Trees, with randomized node optimization and “bagging”. Watch Video
The Zero Inflated Count Regression procedure is designed to fit a regression model in which the dependent variable Y consists of counts. The fitted regression model relates Y to one or more predictor variables X, which may be either quantitative or categorical. It is similar to the procedures for Poisson Regression and Negative Binomial Regression except that it contains an additional component that represents the occurrence of more zeroes that would be expected in those models. Data containing excess zeroes is very common, including such diverse examples as the number of days a student is absent from school, the number of insurance claims within a population where not everyone has insurance, the number of defects in a manufactured item, and wild animal counts. Watch Video
The Venn and Euler Diagrams procedure creates diagrams that display the relative frequency of occurrence of discrete events. They consist of circular regions that represent the frequency of specific events, where the overlap of the circles indicates the simultaneous occurrence of more than one event. Watch Video
3 types of waterfall plots have been added to version 19: an ordered plot, a sequential plot, and a 3-dimensional plot. Ordered Waterfall Plots are used to illustrate how a variable of interest increases or decreases amongst a sample of individuals. Data values are sorted and plotted as a barchart, usually with respect to a baseline equal to 0. A reference line may be added to the plot to display a target value. Watch Video
Sequential Waterfall Plots are used to illustrate the cumulative effect of positive and negative contributions to a total value. Bars are drawn representing each contribution as well as totals and subtotals. Applications include financial analysis, inventory analysis, performance analysis, hiring, and demographics.
Three Dimensional Waterfall Plots are used to display multiple columns of data versus a common variable. One frequently encountered example is a Cumulative Spectral Decay plot, in which spectra are plotted at multiple times to illustrate changes in amplitude as a function of both frequency and time. In general, the plots are used to show changes in a quantitative variable versus both time and some other factor.
Version 19 adds an interface to the Python programming language that is similar to the interface to R that was added in Version 18. Procedures have been added that make it easy to pass data between Statgraphics and Python. Python scripts may also be written and executed from within Statgraphics. Watch Video
The K-Means Clustering procedure implements a machine-learning process to create groups or clusters of multivariate quantitative variables. Clusters are created by grouping observations that are close together in the space of the input variables.The calculations are performed by the “Scikit-learn” module in Python. Watch Video
Conformance analysis has been added to the procedures for determining capability based on attribute data. Conformance analysis is used to determine how well a process conforms to specifications stated in terms of the number of nonconforming items per package. Watch Video
Stepwise variable selection has been added to the GLM procedure for both quantitative and categorical factors. In addition, entry of interactions and other high-order terms has been simplified. Watch Video