Opening JMP

When you first open JMP, an initial splash screen showing the details of your version displays and then disappears. After this, two windows appear. The one in the foreground is the Tip of the Day window, which provides helpful tips and features about JMP (Exhibit 3.1 shows Tip 2). In Exhibit 3.1 these windows are tiled so that you can see the initial contents of each.

We recommend that you use the tips until you become familiar with JMP. Unchecking Show tips at startup at the bottom left of the window disables the display of the Tip of the Day window.

Behind the Tip of the Day window, you will find the JMP Home Window. The JMP Home Window has four panels, each providing links to its contents. The top right pane lists all open JMP windows. The content of the other three panes is self-explanatory, and their state persists between JMP sessions. You can resize the panes in the JMP Home Window, and hide or unhide these panes using the menu commands under View > Home Window Panes.

The Window List pane in the JMP Home Window, which you can also obtain by selecting View > Window List, gives a tree-based view of all other windows currently open in your JMP session. The tree structure shows the name of a currently open table, and the names of any open tables or report windows that depend on it. Tooltips show the contents of a window in miniature, and clicking on the name brings that window forward. Using a right mouse click gives other options for managing that window in relation to others currently open.

By default, the JMP Home Window is designated as the Initial JMP Window. You can change this by selecting File > Preferences > General and selecting from the Initial JMP Window dropdown list. The possible choices are Home Window, JMP Starter, or Window List. If you make a change, click OK to save your new preference.

The JMP Starter Window can also be made visible by selecting View > JMP Starter. It provides an alternative way to access commands that are available in the main menu and toolbars. The commands are organized in groupings that may be more intuitive to some users. You may prefer using the JMP Starter to using the menus. However, to standardize our presentation in this book, we will illustrate features using the main menu bar. At this point, if they are open, please close the JMP Starter and Tip of the Day windows.

Data Tables

Data structures in JMP are called data tables. Exhibit 3.2 shows a portion of a data table that is used in the case study in Chapter 8 and which can be opened via the link in the book's Journal under the “Chapter 8—Informing Pharmaceuticals Sales and Marketing” outline. This data table, PharmaSales.jmp, contains monthly records on the performance of pharmaceutical sales representatives. The observations are keyed to 11,833 physicians of interest. Each month, information is recorded concerning the sales representative assigned to each physician and the related activity. The data table covers an eight-month period. (A thorough description of the data is given in Chapter 8.)

A data table consists of a data grid and data table panels. The data grid consists of row numbers and data columns. Each row number corresponds to an observation, and the columns correspond to the variables. For observation 17, for example, the Practice Name is Clapton Park, Greater London, Visits has a value of 0, and Prescriptions has a value of 5. Note that for row 17, the entry in Visits with Samples is a dot, an indicator that the value for this numeric variable is missing for this row. For a character variable, a blank is used to represent a missing value.

There are three data table panels, shown to the left of the data grid. The Table panel is at the top left. In our example, it is labeled PharmaSales. Below this panel, you see the Columns panel and the Rows panel.

The Table panel shows a listing of scripts. These are pieces of JMP code that produce analyses. You can run a script by clicking on the red triangle to the left of the script name and choosing Run Script from the dropdown menu. As you will see, you will often want to save scripts to the data table in order to reproduce analyses. This is very easy to do.

The Columns panel, located below the Table panel, lists the columns, or variables, that are represented in the data grid. In Exhibit 3.2, we see that the data table has 16 columns. The columns list contains a grouping of variables, called IDs, which groups four variables together. Clicking on the gray disclosure icon next to IDs reveals the four columns that are grouped.

In the Columns panel, note the small icons to the left of the column names. Each icon represents the modeling type of the data in the corresponding column. JMP indicates the modeling type for each variable as shown in Exhibit 3.3.

Specification of these modeling types tells JMP which graphs and analyses are appropriate for these variables. For example, JMP will construct a histogram for a variable with a continuous modeling type, but it will construct a bar graph and frequency table for a variable with a nominal modeling type. Note that our PharmaSales.jmp data table contains variables representing all three modeling types.

The Rows panel appears beneath the Columns panel. Here we learn that the data table consists of 95,224 observations. The numbers of rows that are Selected, Excluded, Hidden, and Labelled are also given in this panel. These four properties, called row states, reflect attributes that can be associated with a given row and that define how JMP utilizes that row. For example, Selected rows are highlighted in the data table and graphs, and can easily be turned into subsets. In Exhibit 3.2, row 17 is selected and a 1 appears to the right of Selected in the Rows panel. Excluded rows are not included in calculations. Hidden rows are not shown in graphs. Labelled rows are assigned persistent labels with values taken from chosen columns that can be viewed in most plots. Excluded, hidden, and labeled rows are also appropriately flagged next to the row number in the data grid.

Visual Displays and Reports

Commands can be chosen using the menu bar, the icons on the toolbar, or, as mentioned earlier, the JMP Starter window. We will use the menu bar in our discussion. Exhibit 3.4 shows the menu bar and default toolbars for a table in JMP. Note that the name of the active data table is shown as part of the window's title. Also shown are the File, Edit, and Data Table toolbars. Toolbars can be customized by selecting View > Customize > Menus and Toolbars.

Reports can be obtained using the Analyze and Graph menus. JMP reports obtained from commands under the Analyze menu provide visual displays (graphs) along with numerical results (in text tables). Commands under the Graph menu primarily provide visual displays, some of which also provide analytical information.

The Analyze and Graph menus are shown in Exhibits 3.5 and 3.6. These high-level menus lead to submenus containing a wide array of visual and analytical tools, also displayed in Exhibits 3.5 and 3.6. Some of these tools appear later in this chapter and in the case studies.

The menu choices under Analyze and Graph launch platforms. A platform dialog allows you to make choices about variable roles, plots, and analyses. A platform generates a report consisting of a set of related plots and tables, with which you can interact so that the relevant features of the chosen variables reveal themselves clearly. We will illustrate this idea with the Distribution platform.

Whenever you have a new data table, you should use the Distribution platform to understand each variable. At this point, please open the data table PharmaSales.jmp. With this data table active, select Analyze > Distribution to open the dialog window shown in Exhibit 3.7.

Suppose that you want to see distribution reports for Region Name, Visits, and Prescriptions. Select all three variable names in the Select Columns list by holding down the Control key while selecting them. Enter them in the box called Y, Columns; you can do this either by dragging and dropping the column names from the Select Columns box to the Y, Columns box, or by selecting them and clicking Y, Columns. In Exhibit 3.8, these variables have been entered and given the Y, Columns role. Note that the modeling types of the three variables are shown in the Y, Columns box.

Click OK to obtain the output shown in Exhibit 3.9. The graphs are given in a vertical layout to facilitate dynamic visualization of numerous variables at one time. This is the JMP default. However, under File > Preferences > Platforms > Distribution (use the JMP menu on the Mac) you can set a preference for a stacked horizontal view by checking the Stack command. You can also make this choice directly in the report: Click the red triangle next to Distribution and select Stack.

By default, menus and toolbars for reports are set to auto-hide based on the size of the report window. You can change this preference via the Windows Specific Preference Group in File > Preferences. If menus and toolbars are hidden, you can reveal them by moving your mouse to the top of the window or by clicking the ALT key on your keyboard. Once menus and toolbars are exposed, a right mouse click will show additional toolbar selections. (These are also available via View > Toolbars.) Selecting these options allows you to control which toolbars are shown.

Note that, by default, the toolbars associated with a report are different from those associated with a table (compare the toolbars in Exhibits 3.9 and 3.4).

The report provides a visual display, as well as supporting analytic information, for the chosen variables. Note that the Distribution reports for the different modeling types are tailored to the modeling type. For Region Name, a nominal variable representing unordered categories, JMP provides a bar graph as well as frequency counts and proportions (using the default alphanumeric ordering). Visits, which is an ordinal variable, is displayed using an ordered bar graph, accompanied by a frequency tabulation of the ordered values. Finally, the graph for Prescriptions, which has a continuous modeling type, is a histogram accompanied by a box plot. The analytic results consist of sample statistics, such as quantiles, sample mean, standard deviation, and so on.

Looking at the report for Region Name, we see that some regions have relatively few observations. Northern England is the region associated with roughly 43 percent of the rows. For Visits, we see that the most frequent value is one, and that at most five visits occur in any given month. Finally, we see that Prescriptions, the number of prescriptions written by a physician in a given month, has a highly right-skewed distribution, which is to be expected.

Additional analysis and graphical options are available in the menus obtained by clicking on the red triangle icons in JMP. If we click on the red triangle next to Distributions in the report shown in Exhibit 3.9, a list of commands appears from which we can choose Stack, for example (see Exhibit 3.10). This gives us the report in a stacked horizontal layout as shown in Exhibit 3.11. (Although the red triangle icons will not appear red in this text, we will continue to refer to them in this fashion, as it helps identify them when you are working directly in JMP.)

Click the red triangle next to the variable name to see a menu containing commands that are specific to the modeling type of each variable being studied. Exhibit 3.12 shows the commands that are revealed by clicking on the red triangle next to Prescriptions. These commands are specific to a variable with a continuous modeling type.

The red triangles support the unfolding style of analysis required for EDA. They put context sensitive commands right where you need them, allowing you to look at your data in a graphical format before deciding which analysis might best be used to describe, further investigate, or model these data.

You also see gray triangles, called disclosure icons, in the report. These serve to conceal or reveal certain portions of the output in order to make viewing more manageable. In Exhibit 3.12, to better focus on Prescriptions, the disclosure icons for Region Name and Visits are closed to conceal their report contents. Note that the orientation of the disclosure icon changes depending on whether it is revealing or concealing contents.

When you launch a platform, the initial state of the report can be controlled with the Platforms option under File > Preferences. Here, you can specify which plots and tables you want to see, and control the form of these plots and tables. For the analyses in this book, we use the default settings unless we explicitly mention otherwise. (If you have already set platform preferences, you may need to reset your preferences to exactly reproduce the reports shown above and in the remainder of the book.)

Finally, notice the three icons and arrow in the right area on the lower border of the report window shown in Exhibit 3.12. Clicking on the first icon brings the JMP Home Window to the front. The home icon appears in all report and data table windows. Clicking on the second icon brings the data table used to create the report or table to the front. The data table icon only appears in reports or tables that are based on a parent data table (a similar icon appears on the top right border on the Mac). You will find these two icons incredibly useful when you have multiple data tables and reports open. If you have more than one report window open, selecting the check box and using the arrow widget on the far right allows you to arrange or combine the selected windows.

Dynamic Linking to Data Table

JMP dynamically links all graphs and plots that are based on the same data table. This is arguably its most valuable capability for EDA. To see what this means, consider the PharmaSales.jmp data table (Exhibit 3.2). Using Analyze > Distribution, select Salesrep Name and Region Name as Y, Columns, and click OK. In the resulting report, click the red triangle icon next to Distribution to select Stack. Also, close the disclosure icon for Frequencies next to Salesrep Name. This gives the plots shown in Exhibit 3.13.

Now, in the report for Region Name, click in the bar representing Scotland. This selects the rows where Region Name = Scotland, as shown in Exhibit 3.14. Simultaneously, areas in the Salesrep Name bar graph corresponding to rows with the Region Name of Scotland are also highlighted. In other words, we have identified the sales representatives who work in Scotland. Moreover, in the Salesrep Name graph, no bars are partially highlighted, indicating that for each of the sales representatives identified, all of their activity is in Scotland. The proportion of a bar that is highlighted corresponds to the proportion of rows where the selected variable value (in this case, Scotland) is represented.

Click the data table icon in the bottom right of your Distribution report to bring the data table to the front. What has happened behind the scenes is that the rows in the data table corresponding to observations having Scotland as the value for Region Name have been selected in the data grid. Exhibit 3.15 shows part of the data table. Note that the Scotland rows are highlighted. Also, note that the Rows panel indicates that 11,048 rows have been selected.

Since these rows are selected in the data table, points and areas on plots and graphs corresponding to these rows will be highlighted, as appropriate. This is why the bars of the graph for Salesrep Name that correspond to representatives working in Scotland are highlighted. The sales representatives whose bars are highlighted have worked in Scotland, and because no bar is partially highlighted, you can also conclude that they have not worked in any other region.

Suppose that you are interested in identifying the region where a specific sales representative works. Look at the second sales representative in the Salesrep Name graph, Adrienne Stoyanov, who has a large number of rows. You can simply click on the bar corresponding to Adrienne in the Salesrep Name bar graph to highlight it, as shown in Exhibit 3.16.

This has the effect of selecting the 2,440 records corresponding to Adrienne in the data table (check the Rows panel). Note that the bar corresponding to Northern England in the Region Name plot is partially highlighted. This indicates that Adrienne works in Northern England, but that she is only a small part of the Northern England sales force.

To view a data table consisting only of Adrienne's 2,440 rows, simply double-click on her bar in the Salesrep Name bar graph. A table (Exhibit 3.17) appears that contains only these 2,440 rows—note that all of these rows are selected, since the rows correspond to the selected bar of the histogram. This table is also assigned a descriptive name. If you have previously selected columns in the main data table, only those columns will appear in the new data table. Otherwise all columns will display in the new table, as shown in Exhibit 3.17.

To deselect the rows in this data table, click in the blank space in the lower triangular region located in the upper left of the data grid (Exhibit 3.18). Click in the upper right triangular region to deselect columns.

Note that JMP also tries to link between data tables when it is useful. For example, when using the Tables > Summary menu command (described later), reports generated from summary data can be linked to reports based on the underlying data.

By way of review, in this section you have seen the ability of JMP to dynamically link data among visual displays and to the underlying data table. You have done this in a very basic setting, but this capability carries over to many other exciting visual tools. The flexibility to identify and work with observations based on visual identification is central to the software's ability to support Visual Six Sigma. The six case studies in Chapters 5 through 10 elaborate on this theme.

Window Management

In exploring your data, you will often find yourself with many data tables and reports simultaneously open. Of course, these can be closed individually. However, it is sometimes useful to close all windows, to close only data tables, or to close only report windows. You can do this and more from the Window List pane in the JMP Home Window.

To see the current state of your JMP session, click the home window icon in the bottom right of your top-most window to bring the JMP Home Window to the front (Exhibit 3.19). To open a similar home window on the Mac, select JMP Home from the Windows menu. In the Window List panel, you see that you have five open windows—two data tables, two reports, and the journal file. The list of open windows also appears in the bottom panel of the Window menu. When running analyses on data, it is important to make sure that the appropriate data table is active—when a launch dialog is executed, commands are run on the active data table.

At this point, you could select Window > Close All. However, you want to continue your work with PharmaSales.jmp. Right-click on PharmaSales in the Window List and select Close All But This. Then click Save None in the window that appears. Alternatively, you can close the other three windows individually selecting them in the Window List, right-clicking, and selecting Close. Or you can close them by navigating to each window separately and clicking on the close button (X) in the top right corner of each window.

Graph

Techniques that have visual displays as their primary goal are available from the Graph menu shown in Exhibit 3.6. Many of these displays allow you to view patterns and anomalies in your data in one, two, three or more dimensions. In the remaining chapters, you will see examples of the following:

• Graph Builder allows highly interactive simultaneous visualization of multiple variables using a wide variety of graphical elements.
• 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).
• Scatterplot Matrix gives a matrix of scatterplots.
• Scatterplot 3D gives a three-dimensional data view.
• Surface Plot creates three-dimensional, rotatable displays.
• Profiler shows traces of a function, often a prediction model, and is used in optimization and simulation.
• Contour Profiler gives a multidimensional view of a function for use in optimization.
• Treemap is a two-dimensional version of a bar graph that allows better visualization of variables with many levels.

Analyze

Techniques that combine analytic results with supporting visual displays are found in the Analyze menu, shown in Exhibit 3.5. Analyze commands often include 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 command, which addresses the relationship between two variables. If both are continuous, then the Bivariate platform presents a scatterplot, allows you to fit a line or curve, and provides regression results. If Y is continuous and X is nominal, then Fit Y by X 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 output 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.

The remaining chapters will take you 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, contingency tables, and associated statistical tests.
• Tabulate is an interactive approach to constructing tables of descriptive statistics and other summary tables.
• Fit Model fits a large variety of models and gives access to a prediction profiler that is linked to the fitted model.
• Modeling > Partition provides recursive partitioning models, similar to classification and regression trees.
• Modeling > Neural Net fits flexible nonlinear models using hidden layers.
• Modeling > Model Comparison (JMP Pro only) compares models relative to various performance measures.
• Multivariate Methods > Multivariate gives scatterplot matrices and various correlations.
• Multivariate Methods > Cluster attempts to arrange rows into groups where the variation within a group is less than the variation between groups.
• Multivariate Methods > Principal Components exploits correlations between variables to represent most of the variability in a space of reduced dimensionality.
• Quality and Process > Control Chart Builder provides the interactivity of Graph Builder and constructs a variety of control charts.
• Quality and Process > Measurement Systems Analysis addresses measurement system variability and encompasses both a traditional approach and Wheeler's Evaluate the Measurement Process (EMP) approach.
• Quality and Process > Variability/Attribute 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.
• Quality and Process > Process Capability provides a goal plot and box plots to assess the performance of numerous responses, as well as related capability measures.
• Quality and Process > Pareto Plot gives a bar chart ordered by decreasing frequency of occurrence.
• Quality and Process > Diagram is used to create cause-and-effect diagrams.

Tables

Another menu that is used extensively in the remaining chapters is the Tables menu (Exhibit 3.20). This menu contains commands that perform operations on data tables. In the remaining chapters you will see examples of the following:

• Summary provides 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.
• Transpose creates new data tables from the data in the current data table.
• Join operates on two data tables, joining them by adding columns of data.
• Concatenate operates on two or more data tables, joining them by adding rows of data.
• Missing Data Pattern produces a table that helps you determine if there are patterns or relationships in the structure of missing data in the active data table.

Rows

You will also use many features that are found under the Rows menu. The Rows menu is shown in Exhibit 3.21, with the commands 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.

Recall that a row state is a property that is associated with 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 change row states of rows interactively based on visual displays. The Clear Row States command removes any row states that are currently in effect.

Data Filter and the Local Data Filter, which filters a specific report, are used extensively in the case studies. These options provide a flexible and interactive way to change the row states in order to identify meaningful, and possibly complex, subsets of rows. They also provide a way of animating many of the visual displays in JMP. Depending on the analysis goal, subsets that you define using Data Filter can easily be selected and placed into their own data tables for subsequent analysis or excluded from the current analysis.

Columns

The Cols menu, shown in Exhibit 3.22, provides commands dealing with column properties and roles, information stored along with columns, formulas that define column values, recoding of values, and more.

Note that some of the commands in Exhibit 3.22 are active because columns were selected in the table.

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.

A column can have column properties associated with it. You can add a note describing the column using the Note property. (The Note column property has already been added for each column in PharmaSales.jmp.) You can also have specification limits using the Spec Limits property, control limits using the Control Limit property, and so forth. To define a column using a formula, select the Formula column property to define that column.

Note that you can specify the Data Type and Modeling Type for your data in the Column Info window. The Numeric and Character data types are the most common, but you can also define a column to have a data type of Row State or Expression. The Row State data type allows you to construct a column that contains row states, providing a permanent record of row states. The Expression data type allows you, among other things, to store images in table cells. These can be displayed on certain plots, which can be very useful for providing extra context and meaning. You can specify an appropriate data format in the Column Info window as well. For more information refer to the JMP documentation.

DOE

Chapter 2 introduced the idea of experimental data, which arise when you 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, factors, number of runs, and other settings. There are five major groupings.

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, and allows for situations with hard-to-change and easy-to-change factors and covariates (split-plot, split-split-plot, and strip-plot designs). You can specify inequality constraints on the factors or disallowed combinations of factor settings. The Custom Design platform is featured in one of the case studies (Chapter 7) while the Full Factorial Design platform is used in another (Chapter 9).

The Evaluate Design platform allows you to assess the effectiveness of a design prior to actually running the experiment. If you construct your design using JMP's Custom Design platform, design evaluation results are presented before you construct your design table. 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.³