It is proved that the scaled deviance of an inverse Gaussian sample of size n can be expressed as a sum of n - 1 independent chi-square variates. Applications to prediction intervals and squared ...