The "Uncorrelated Errors" Approach to the Problem of Causal Feedback

Political scientists increasingly have employed simultaneous equation techniques to disentangle reciprocal effects among two or more important variables. These techniques require one or more exogenous or endogenous "instrumental" variables in order to identify the system of equations properly. Thus, a tradeoff of assumptions is involved, raising the problem of finding instrumental variables that meet certain requirements, e.g., finding one exogenous variable that causes X but not Y. A solution to the dilemma exists if one is willing to assume that the error terms of X and Y are uncorrelated. This paper describes how unidentifiable equations can be solved via the uncorrelated errors assumption, and it offers an application to a substantive problem involving the reciprocal effects of party identification and issue attitudes, and using the 1976 U.S. Election Study data.