We consider a linear normal model Y = X theta + e with theta verifying
a linear restriction and the standard estimators <(theta)over cap> (u
nrestricted MLE) and theta (restricted MLE). We prove that theta* is
preferable to <(theta)over cap> using a new and strong criterion which
implies the domination under other usual criteria; in particular it i
s proven that the standard simultaneous confidence intervals centered
at theta have more confidence than those centered at <(theta)over cap
>. (C) 1997 Academic Press.