Because accounting for the constraints on model parameters lead to more efficient and interpretable estimators. Therefore, constrained parameter problems arise in a variety of statistical applications. For example, if θi is the toxic of a certain medicine at dose i(i = 1,2,… ,k) , then it is well known that the toxic is increasing as the dose increasing, it is reasonable to assume the simple ordering, that is θ1 ≤ θ2 ≤ … ≤ θk. Another example, if μi denote the efficiency of some medicine at dose i(i = 1,2, … , k), we know that the efficiency is usually increasing up to some dose m, and then is decreasing. Thus an umbrella ordering with change-point m seems more reasonable. Let μ1 ≤ … ≤ μm ≥ … ≥ μk . Furthermore, in life insurance , let pi denote the true mortality rate at age ti, it is reasonable toassume that {pi} are increasing convex, that is reasonable to assume {pi} to beThis paper concerns the method for estimating the restricted parameters in multivariate normal via the EM-type algorithms. The model is yl = Xθ + el, el Nm(0, ∑), l = 1,2, … ,p. where yl is an m-vector of responses for individual l, Xm*q is a known matrix, regression coefficients θ = (θ1, … , θq) are restricted to the simple ordering θ1 ≤ … ≤ θq, the umbrella ordering θ1 ≤ … ≤ θh ≥ θh+1 ≥ … ≥θq, and the increasing convex ordering 0where {dk} are known and d1 < … < dq .
|