The ggobi application provides a graphical interface for each collection of datasets being viewed together. The interface has a main control panel to display the different datasets and their variables and to allow one to select different combinations of these variables to display in different displays and their plots. The different displays are created by selecting the appropriate menu entry on this dataset and variable panel, and each appears in its own window.
Control Panel
The ggobi control panel, including variable selection area, and a single scatterplot.
A ggobi instance with two datasets, as you can see by the named tabs above the variable selection area. For this data, both datasets are stored within a single xml file, and the rules for linking between them are defined there as well.
For a parallel coordinates plot, the control panel looks something like this, with each plotted variable selected.
Plot Types
The first image above includes a scatterplot; ggobi has other plot types as well, including a parallel coordinates plot. (This corresponds to the control panel configuration just above.)
And ggobi can also show a scatterplot matrix. The plots on the diagonal are one-dimensional ASHes (Average Shifted Histograms).
Brushing, Identifying & Linking Plots
This auxiliary control panel is called the color & glyph chooser; it is used to select the symbol to be used in brushing.
Now we can interactively brush points. Since views of the same data are linked, we can change the characteristics of several displays at the same time.
 |
 |
|---|
Auxillary Control Panels
Other auxiliary control panels are used to view variable characteristics, clone variables, transform or jitter variables, subset the data, select a new colorscheme, support the exploratory analysis of missing data, view and control the groups formed by brushing, and to keep track of displays and their individual plots.
The variable manipulation tool reports summary statistics about each variable, and includes buttons that allow variable ranges to be set, or new variables to be added.
Using the colorscheme tool, we can choose from a variety of colorschemes. If we like, we can then automatically assign colors to the points according the values of a selected variable. The result might look like the plot on the right.
 |
 |
|---|
After brushing, we can work with the groups directly using the color & glyph groups tool. We can use the H button next to each symbol to hide all the cases drawn with that symbol, or double-click on the symbol itself to change the symbol for those points.
Missing Data
ggobi datasets can include missing values, which are usually specified by embedding the strings "." or "na" in the data. It is useful to explore the missingness directly by creating a new dataset composed of 0's and 1's, and then linking the missingness displays to plots of the original data. In this example, the original dataset is named "sleep" and the new one is "sleep (missing)".
This pair of linked plots shows an example of how we can explore missingness. We're brushing all the cases for which the variable NonD has missing values, and looking at the distribution of these missings in a scatterplot of Span vs Sleep.
 |
 |
|---|
Extensible Programmed Layouts with R/SPlus-ggobi Interface
One can extend the different types of plots that one can create from the ggobi graphical interface by creating new ones with different combinations of plots. Using the S-language interface to ggobi, one can create new plots individually and then arrange them in arbitrary cells of a grid to present them in a single window. More details here.