Correction to "Scoring with Constraints"

This paper has a serious error in that the key result on the expectation of the log likelihood gradient is not applicable in this context.

This leads to the incorrect conclusion that the asymptotic limits of the Lagrange multipliers associated with a finite set of  constraints is zero.

A correct version is given in the "Notes on least squares and maximum likelihood" available on these web pages.

The Lagrange multiplier results in given in chapter 3, The method of maximum likelihood, in section 3.4 on equality constrained problems.

The numerical example is reworked in chapter 4, Computational methods for maximum likelihood, in section 4.5 on constrained likelihood problems.