A modified Newton method for constrained estimation in covariance structure analysis

Abstract

This paper presents a modified Newton algorithm for estimation of parameters in covariance structure analysis, subject to general equality and inequality constraints. The method requires solving a quadratic programming problem at each iteration. An algorithm based on an active set method for solving the quadratic programming problem is proposed. A number of applications of equality constraints are discussed and the modified Newton method, proposed here, is compared to the multiplier method. Our examples indicate that the modified Newton method is superior to the multiplier method both in terms of speed of computation and robustness to poor starting values.