Processes | Pattern Discovery | Partial Correlation Diagram

Partial Correlation Diagram process helps you infer and potential causal relationships between a set of . This process fits so-called selection (also known as graphical Gaussian models ), in which partial correlations ( the correlation between two variables adjusted for all other variables) are estimated, and then plots each variable as a node. The nodes are then connected with line segments, whose size and color are determined by the partial correlations. Additional graphs are also available, along with options for controlling them.
: Although a strong partial correlation is not the same as a causal relationship, it can suggest one.
Let the partial correlation matrix be R . The decomposition is applied to R .
where E is the matrix, and M is the diagonal matrix of eigenvalues.
The X and Y position of each node is determined by the following:
X = Sqrt(Abs(M[1]))*E[,1]
Y = Sqrt(Abs(M[2]))*E[,2]
and M[2] are the 1st and 2nd eigenvalue, respectively. E[,1] and E[,2] are the 1st and 2nd eigenvector, respectively.
is required to run the Partial Correlation Diagram process. Because the process calculates the distance between the (rows), a wide -formatted data set (in which the rows comprise each of the variables) is normally used as the input data set. If you are working with a tall data set and you want to compute distance between columns, first run the process.
adsl_diit.sas7bdat data set, shown below, was generated by transposition of the adsl_dii.sas7bdat data set included with JMP Clinical (see ). Patients are listed in columns, are listed in rows. There are 911 columns for 906 patients and 350 rows listing events.
The output generated by this process is summarized in a Tabbed report. Refer to the output documentation for detailed descriptions and guides to interpreting your results.