Some observations on the use of conjugate initial distributions for exponential families.