Specification error in causal models

A typology of model specification errors, initially devised for single-equation multiple-regression models, is extended to include systems of equations embodied in recursive causal models. Analytical results for the error forms identified are presented. On the basis of these results a number of inadequacies of traditional techniques of causal inference are identified. In place of the traditional techniques of causal inference an alternative strategy for constructing recursive causal models, based on specification error knowledge, is offered. An illustrative application is also included.