Precision for relative potency estimates is well described if the standard statistical methodology is used for parameter estimation. Currently I am confronted with a special case. It is a parallel line assay with clear heteroscedasticy. As such I need to perform a weighed regression. This gives my the parameter estimates (intercept 1 and 2; slope), their variances and their covariances. It should be possible to use this information to calculate the precion on the relative potency estimate: (intercept1-intercept2)/slope. But how?