Diagram

The Diagram tab is shown below:

The Diagram tab contains the following elements:

•

2D plot of the Partial Correlation Diagram.

This plot is the principal result of this process. Each variable is represented by a node point in the plot, and points are connected with line segments. The default position of each node is decided by eigenvalue decomposition of a partial correlation matrix. The aspect ratio (the ratio of y-axis length to x-axis length) is adjusted heuristically. The width of the lines is proportional to the strength of the partial correlations. Red and blue colors represent positive and negative partial correlations, respectively. You can click and move nodes or click and select lines joining them.

Partial correlations are correlations adjusted for all other variables in the plot. Strong partial correlations can reveal potential causal relationships amongst variables while controlling for confounding due to the other variables specified. The strength of the partial correlation between two variables can be estimated by the distance between their respective nodes. Assuming that the partial correlation matrix is approximated well in these two dimensions, the closer the nodes are to each other, the stronger the correlation.

Note: These drawing methods are not directly related to the estimation theory of covariance selection models. Even if you change the positions of nodes, the lines (edges) between nodes are not changed.

This tab contains the following Drill-down options:

•

Delete or Add Lines. Click the small triangle next to the Delete or Add Lines outline box to show available options. A typical analysis would proceed by deleting all lines that are not statistically significantly different from zero. You can perform this deletion automatically by setting the appropriate stepwise criterion and clicking the Go button. You can optionally perform it one step at a time with the Step button. Alternatively, you can select lines with the mouse and delete or add them by hand. You can also freeze lines so that they are not deleted or release them after they have been frozen.

•

Graphics Options. Click the small triangle next to the Graphics Options outline box to show available options.

Node Orientation controls the layout of the points. The default layout uses eigenvectors, but you can optionally use a circular or sorted circular arrangement. The former is based on the eigenvectors and projects them out onto a circle, whereas the latter uses the order in which you specified the variables. You can also rotate the diagram clockwise or counterclockwise.

Line Display Mode controls how the lines are colored and sized. The default lines use the partial correlations themselves, but you can optionally use p-values for all of the lines or just those that have been retained or deleted.

Line Properties controls the color, width, and transparency of the lines. Use the pull down menu and sliders to change these properties.

Text Properties controls the text on the diagram. You can change the font sizes used and also add the partial correlations themselves to each line.

•

Additional Graphs and Statistics. Click the small triangle next to the Additional Graphs and Statistics outline box to show available options.

Additional Graphs include ellipse plots, scatterplots, a 3D spinning plot, and the scatterplot matrix from JMP's Multivariate platform. These views can provide additional insight into partial correlation patterns.

Available Statistics include tests for all lines and the estimated correlation and partial correlation matrices.

Save Results options enable you to copy results to a JMP journal or save results as a GraphViz DOT file.