that the ordinary least squares coefficient corresponding to the variable with the error will be biassed downward (in absolute value), and it is also seen that the direction and extent of bias in the other coefficients will depend directly on the covariances between those coefficient estimates and that of the offending variable. In particular, it may be useful to note that the sign of the bias is given by: sgn ( gj 8 *j) = ( l)(sgn cov ,2j,, ) (sgn j,