E. E. Learner (Amer. Econom. Rev. 75, 308-313, 1985; Econometrica 50, 725-736, 1982) and M. McAleer, A. R. Pagan, and P. A. Volker, (Amer. Econom. Rev. 75, 293-307, 1985) have shown how to assess omitted variable bias by computing extreme bounds of coefficient estimates. Far less attention has been focused on the potential bias from inclusion of variables that are mismeasured. S. Garber and S. Klepper (Econometrica 48, 1541-1546, 1980) derive the bias of the OLS estimator for correctly measured variables due to a single mismeasured variable. The evaluation of this bias is straightforward. However, a more reasonable presumption in many environmental economics problems is that most explanatory variables are mismeasured. In this case, S. Klepper and E. E. Learner (Econometrica 52, 163-183, 1984) show that measurement error bias is now a complex function of the sample covariances of all variables and the covariances of all measurement errors. Biases are harder to assess and may be greater than measured previously. Using the Bayesian diagnostics of Klepper and Learner, we find that bias due to omitted variables appears to be minimal for the OLS coefficient estimator of a frequently used work-days lost data set. However, this estimator is much more susceptible to bias from measurement error than indicated by the simple Garber-Mepper formula. Even small degrees of measurement error are troublesome due to high collinearity between environmental and personal attribute variables, and this problem likely plagues many other similar studies. © 1992.
