The visual displays and analyses that are featured in the case studies represent many different menu items in JMP. In this section, we will give a quick overview of the more basic items and where they can be found.
Techniques that have visual displays as their primary goal are available from the Graph menu shown earlier in Exhibit 3.6. Many of these displays allow you to view patterns in the data in three or more dimensions. The Bubble Plot is animated, giving one way for you to see behavior over time. In the case studies, we will see examples of the following:
Graph Builder allows highly interactive simultaneous visualization of multiple variables using a wide variety of graphical elements.
Scatterplot 3D gives a three-dimensional data view.
Bubble Plot is a dynamic extension of a scatterplot that is capable of showing up to five dimensions (using x position, y position, size, color, and time).
Tree Map is a two-dimensional version of a bar graph that allows visualization of variables with many levels.
Scatterplot Matrix gives a matrix of scatterplots.
Diagram is used to create cause-and-effect diagrams.
Control Chart constructs a variety of control charts that, in particular, allow for phases.
Variability/Gauge Chart is useful for measurement system analysis and for displaying data across the levels of multiple categorical variables, especially when the focus is on displaying the variation within and between groups.
Pareto Plot gives a bar chart ordered by decreasing frequency of occurrence.
Capability provides a goal plot as well as box plots to assess the performance of numerous responses.
Profiler provides an interactive display that is used in optimization and simulation.
Surface Plot creates three-dimensional, rotatable displays.
Techniques that combine analytic results with supporting visual displays are found in the Analyze menu, as shown earlier in Exhibit 3.5. Analyze commands often produce displays similar to those found under the Graph menu. For example, a Fit Model analysis allows access to the profiler, which can also be accessed under Graph. As another example, under Multivariate Methods > Multivariate, a scatterplot matrix is presented. The selection Graph > Scatterplot Matrix also produces a scatterplot matrix. However, Multivariate Methods > Multivariate allows you to choose analyses not directly accessible from Graph > Scatterplot Matrix, such as pairwise correlations.
Each platform under Analyze performs analyses that are consistent with the modeling types of the variables involved. Consider the Fit Y by X platform, which addresses the relationship between two variables. If both are continuous, then Fit Y by X presents a scatterplot, allows you to fit a line or curve, and provides regression results. If Y is continuous and X is nominal, then the platform produces a plot of the data with comparison box plots, and allows you to choose an analysis of variance (ANOVA) report. If both X and Y are nominal, then a mosaic plot and contingency table are presented. If X is continuous and Y is nominal, a logistic regression plot and the corresponding analytic results are given. If one or both variables are ordinal, then, again, an appropriate report is presented.
This philosophy carries over to other platforms. The Fit Model platform is used to model the relationship between one or more responses and one or more predictors. In particular, this platform performs multiple linear regression analysis. The Modeling menu includes various modeling techniques, including neural nets and partitioning, which are usually associated with data mining.
Our case studies will take us to the following parts of the Analyze menu:
Distribution provides histograms and bar graphs, distributional fits, and capability analysis.
Fit Y by X gives scatterplots, linear fits, comparison box plots, mosaic plots, and contingency tables.
Fit Model fits a large variety of models and gives access to a prediction profiler that is linked to the fitted model.
Modeling > Screening fits models with numerous effects.
Modeling > Neural Net fits flexible nonlinear models using hidden layers.
Modeling > Partition provides recursive partitioning fits, similar to classification and regression trees.
Multivariate Methods > Multivariate gives scatterplot matrices and various correlations.
Another menu that will be used extensively in our case studies is the Tables menu (Exhibit 3.20). This menu contains options that deal with data tables. The Summary platform is used to obtain summary information, such as means and standard deviations, for variables in a data table. Subset creates a new data table from selections of rows and columns in the current data table. Sort sorts the rows according to the values of a column or columns. Stack, Split, and Transpose create new data tables from the data in the current data table. Join and Update operate on two data tables, while Concatenate operates on two or more data tables.
Tabulate is an interactive way to build tables of descriptive statistics. Missing Data Pattern produces tables that help you determine if there are patterns or relationships in the structure of missing data in the current data table. Recall that for a numeric variable JMP uses "." to denote a missing value, while JMP uses an empty string for character data, which appears as a blank cell in a data table.
Our case studies will take us to the following platforms:
Summary summarizes columns from the current data table in various ways.
Sort sorts a data table by the values of designated columns.
Concatenate combines rows from two or more data tables.
Tabulate constructs tables of descriptive statistics in an interactive and flexible way.
Missing Data Pattern creates a table describing where data are missing across a set of columns.
We will also use features that are found under the Rows menu. The Rows menu is shown in Exhibit 3.21, with the options under Row Selection expanded. The Rows menu allows you to exclude, hide, and label observations. You can assign colors and markers to the points representing observations. Row Selection allows you to specify various criteria for selecting rows, as shown.
A row state is a property that is assigned to a row. A row state consists of information on whether a specific row is selected, excluded from analysis, hidden so that it does not appear in plots, labeled, colored, or has a marker assigned to it. You often need to assign row states to rows interactively based on visual displays. The Clear Row States command removes any row states that you have assigned.
Data Filter will be used extensively in our case studies. This command provides a very flexible way for you to query the data interactively to identify meaningful, and possibly complex, subsets. It also provides a way of animating many of the visual displays in JMP. Depending on the analysis goal, subsets that you define using Data Filter can be easily selected and placed into their own data tables for subsequent analysis or excluded from the current analysis.
The Cols menu, shown in Exhibit 3.22, provides options dealing with column properties and roles, information stored along with columns, formulas that define column values, recoding of values, and more.
The Column Info command opens a dialog that allows you to define properties that are saved as part of the column information. To access Column Info for a column, right-click in the column header area and choose Column Info. Exhibit 3.23 shows the Column Info dialog for Visits. For example, you can add a note describing the column (this has already been done), save specification limits, control limits, and so forth. When defining a new column, you can select Formula to define that column using a formula.
In Column Info note that you can specify the data type and modeling type for your data. You can specify a format here as well.
In Chapter 2, we introduced the idea of experimental data, which arise when we deliberately manipulate the Xs. An experimental design is a list of trials or runs defined by specific settings for the Xs, with an order in which these runs should be performed. This design is used by an experimenter in order to obtain experimental data. JMP provides comprehensive support for generating experimental designs and modeling the results.
The DOE menu, shown in Exhibit 3.24, generates settings for a designed experiment, based on your choices of experimental design type, responses, and factors. There are nine major design groupings (Custom Design to Taguchi Arrays in the menu listing). The Augment Design platform provides options for adding runs to an existing design. The Sample Size and Power platform computes power, sample size, or effect size for a variety of situations, based on values that you set.
The Custom Design platform allows great flexibility in design choices. In particular, Custom Design accommodates both continuous and categorical factors, provides designs that estimate user-specified interactions and polynomial terms, allows for situations with hard-to-change and easy-to-change factors (split-plot designs), and permits you to specify inequality constraints on the factors. The Custom Design platform is featured in one of our case studies (Chapter 6), while the Full Factorial Design platform is used in another (Chapter 8).
18.224.54.255