The SAS applications calculate multivariate t-probabilities by running executable files during the course of the application. The source code of these exe-files can be downloaded on the website of Alan Genz. The numerical methods used for the calculations can be found in Genz and Bretz (2002). The SAS applications provide examples of how the exe-files can be called dureing the SAS session. The first part of the code provide an example, which runs the exe-files from PROC IML. The second part of the code provide an example, of how the exe-file can be run from an arbitrary data step, without IML. The exe-files must be saved under the pathname "c:\temp". There are two exe-files, corresponding to two different approaches of calculating multivariate t-probabilities.
 
 

The file mvtprob.exe uses randomised lattice rules (Koroboy type), which is of stochastic nature. It requires an existing input file called "tempin.txt". It writes the results to the output file "tempout.txt" (which need not to exist before running the exe-file). The input file should contain the input parameters (delimiter is a blank) organised as follows:

N NU MAXPTS ABSEPS
LOWER
UPPER
INFIN
DELTA
CORREL

The parameters are explained here. The only difference is that CORREL is a N(N-1)/2 row vector, which contains the elements of the lower triangular part of the correlation matrix. The output file contains the integral value, an error estimation and a flag variable. The pathname of both I/O files are given interactively in the SAS application. Note that the given SAS applications open and read the output file at runtime, such that the results are available in an ordinary SAS data set.
 
 

The file mvtprob2.exe uses a subregion adaptive method, which is of deterministic nature. It requires an existing input file called "tempin2.txt". It writes the results to the output file "tempout2.txt" (which need not to exist before running the exe-file). The input file should contain the input parameters (delimiter is a blank) organised as follows:

N NU MAXPTS ABSEPS
LOWER
UPPER
INFIN
CORREL

The parameters are explained here. The only difference is that CORREL is a N(N-1)/2 row vector, which contains the elements of the lower triangular part of the correlation matrix. The output file contains the integral value, an error estimation and a flag variable. The pathname of both I/O files are given interactively in the SAS application. Note that the given SAS applications open and read the output file at runtime, such that the results are available in an ordinary SAS data set.