Statistica

Statistica is a data mining and analysis program sold by StatSoft, Inc., of Tulsa, Oklahoma. The program is broad in scope with separate modules for advanced or specialty features oriented to data-intensive industries such as power production, pharmaceuticals, insurance, finance and banking, health care, and electronics. The program is available for individual computers, enterprise solutions for multiple users, and as a Web-based program. This entry describes features that are most apt to be of interest to readers of this encyclopedia, such as typical analyses, user interface, graphics, and documentation.

The basic module contains standard frequentist analysis techniques such as descriptive statistics; analysis of variance (ANOVA)/multivariate analysis of variance (MANOVA), including post hoc tests and planned comparisons; multiple regression; correlations; cross-tabulation analyses; basic nonparametric analyses; and distribution fitting statistics. An advanced module adds numerous additional analysis methods. A partial list of these methods includes generalized linear models, time-series forecasting, survival and failure time analysis, structural equation modeling, log linear analysis, and power analysis. There are also quality control statistics such as those associated with the Six Sigma movement and root cause analysis. It also contains multivariate exploratory methods such as principal components and factor analysis, cluster analysis, discriminant analysis, multidimensional scaling, and reliability/item analysis. There are other modules that permit neural network analyses and data mining.

Although Statistica is a highly sophisticated analysis system, it does not offer any current options for dealing with missing data other than casewise or listwise deletion and using means to impute missing values. There are some additional alternative missing data imputation methods in the time-series analysis routine.

For many psychological investigations, it is necessary to calculate effect sizes, both for interpretation and to allow later inclusion of the study's results in a meta-analysis. Effect sizes for mean differences obtained through ANOVA/MANOVA are partial eta-squared. These rely on the amount of variance in the dependent variable that was explained by the independent variable. There are several variations of effect sizes (the g and d family of effect sizes) that express the results as difference between means in standard deviation units. These are often used in meta-analyses. Statistica does not produce these effect sizes, although they can be relatively easily calculated from the statistics reported by Statistica.

Bayesian analyses are limited to certain data mining techniques and neuro network analyses, which are inherently Bayesian in nature. Other analyses do not permit the introduction of prior distributions or probabilities, at least not through the graphical user interface. Bayesian methods are becoming increasingly common in statistics, but other programs will need to be used to perform them.

Statistica contains an extensive array of graphic techniques. Because the program is used in many industries, it has the capability of creating realtime graphs. For typical social science research, the usual graph types (line, histogram, scatterplot) are available, plus graphs useful for descriptive purposes such as variations on pie charts and similar depictions. A potentially useful variation on the scatterplot that is not available in some other software is the bag plot, which is an extension of the Tukey box plot to two dimensions, illustrating central location, skewness, correlation, and allowing for quick identification of outliers.

...