package mvtnorm: sigma parameter in pmvnorm() (PR#2478)

Full_Name: Jerome Asselin
Version: 1.6.2
OS: RedHat Linux 7.2
Submission from: (NULL) (142.103.173.179)



pmvnorm() may fail for a univariate distribution when
its parameter "sigma" is defined as a matrix. It will
fail if sigma < 1.

library(mvtnorm)

#THIS WORKS

pmvnorm(lower=-Inf,upper=2,mean=0,sigma=matrix(1.5))

[1] 0.9487648
attr(,"error")
[1] 0
attr(,"msg")
[1] "univariate: using pnorm"

#THIS FAILS

pmvnorm(lower=-Inf,upper=2,mean=0,sigma=matrix(.5))

Error in checkmvArgs(lower = lower, upper = upper, mean = mean, corr = corr,  :
        diag(sigma) and lower are of different length

#THIS WORKS

pmvnorm(lower=-Inf,upper=2,mean=0,sigma=.5)

[1] 0.9976611
attr(,"error")
[1] 0
attr(,"msg")
[1] "univariate: using pnorm"

The bug is in checkmvArgs():

  if (!is.null(sigma)) {
          if (!is.matrix(sigma)) {
             if (length(sigma) == 1)
                 UNI <- TRUE
             if (length(sigma) != length(lower))
                stop("diag(sigma) and lower are of different length")
          } else {
+           if (length(diag(sigma)) != length(lower))
+              stop("diag(sigma) and lower are of different length")
             if (length(sigma) == 1) {
                 UNI <- TRUE
                 sigma <- sigma[1,1]
             }
-           if (length(diag(sigma)) != length(lower))
-              stop("diag(sigma) and lower are of different length")

          }
     }


The fix is to change the position of the validation, i.e. before the 
matrix is coerced to a vector of length 1.

Uwe Ligges