We propose a data-based extremum formulation that extends theempirical-likelihood and information-theoretic methods of estimation andinference. It is demonstrated how this method may be used in a general linearmodel context to mitigate the problem of an ill-conditioned design matrix. Adual loss...