JMP is the SAS institute's package for exploratory data analysis. JMP offers a plethora of features for distribution analysis (including contingency tables, outlier analysis, and nonparametric statistics), and graphs (including contour, profile and ternary charts), time series modeling, regression, {m}anova, PCA, logit analysis, cluster analysis, survival, design of experiments (added in version 4.0), and quality control. In addition, JMP makes it extremely quick and easy to perform "hands-on" data manipulation such as: creating derived and aggregated variables and tables, subsetting, stacking, and joining multiple tables together. And Version 4 adds facilities to import a wider variety of data formats, extract information from databases, and preview text file input. (Unlike SAS, however, the size of JMP tables is limited by the amount of memory available on your machine.)
Exploratory data analysis with JMP is extraordinarily easy and powerful, because of three design decisions that distinguish it from all of its competitors: (1) let the user describe the data, (2) organize the program my task rather than by statistical model, (3) and make everything displayed in the program "live."
Unlike most statistical packages, which will only keep track of the "data type" of each variable. You can tell JMP almost anything you know about a variable, including its data types (integer, floating point, string), its modeling type (nominal, ordinal, continuous), its role in your model (dependent, independent, weight, frequency, time axis), and range or set constraints that apply to it. You can also attach notes (of arbitrary length, or formulas relating one variable to others. This makes it easy to build documentation into your research, or course, but JMP uses this information intelligently to guide analysis and visualization.
Functions in JMP are organized by task, not statistical model. The broad task categories are "design of experiments", "analysis", and "graphing", which are further divided by purpose.
For example, the "analysis" category comprises "distribution", "fit Y by X", "matched pairs" "fit model", "nonlinear fit", "multivariate", clusters" ,"survival," and "time series." When a user chooses an
analysis, JMP usually "does the right thing" based upon the roles of the variables selected. For example, running "fit Y by Y" will produce contingency tables if both variables are nominal or ordinal but different analyses if either variable is continuous (see Fig 1). As befitting an EDA
package, most analysis and graphing operations are performed quite quickly, although when an operation does take longer than expected one wishes that the designers had included a button to interrupt processing. Also, occasionally the arrangement of models can seem idiosyncratic -- e.g., scatter plots are
performed through "Analyze…Distributions" while principle components analyses can be performed either through "Analyze…Multivariate" or "Graph..Spinning Plots".
In contrast to such competitors as Stata, SPSS, and Minitab, tables, scripts, formulas, and graphs are very well integrated, and the connections among these are "live." For example if you select data in a summary table, the corresponding entries are selected in the original table, and all graphs based on those tables instantly highlight points or regions to reflect the selection. (One sometimes wishes, however that summary statistics were updateable along with the graphs.) Graphs can be directly rotated, rescaled, recolored, or otherwise manipulated in a "live" fashion. At a higher level, formulas and scripts are used to further link data and analysis: the user can attach formulas to variables, with the result that these variables are automatically generated from other data, and automatically change when the related data changes. In version 4, scripts can be attached to tables to perform similar automatic linking between tables and analyses. Scripting is made very easy, because the system allows you to run a set of analyses first, and to save the resulting analysis set as a scripts or in a journal for later use. (See Fig. 2)
