Consistent Sets of Estimates for Restricted Regressions with Errors in All Variables.

In the errors-in-variables model, positive semidefinite conditions on covariance matrices, together with sufficiently small prior limits on the severity of measurement error, imply that the set of feasible values for population coefficients is bounded. The authors show how maintained restrictions can be used to reduce further the set of feasible values for the coefficients. If the restrictions are incorrect, the restricted feasible set will be empty for sufficiently small limits on the severity of measurement error. Hence the restrictions are testable even though the coefficients are not identified. Copyright 1987 by Economics Department of the University of Pennsylvania and the Osaka University Institute of Social and Economic Research Association.