It has been a challenge to avoid double counting when economic values (EV) of traits are derived for breeding goal traits when using stochastic herd simulation models. In this study multiple regression and model building with mediator variables representing other traits in the breeding goal were evaluated to avoid double counting. EV were derived from data simulated with the SimHerd dairy herd simulation model. Scenarios were simulated to represent dairy herds with low and high levels of metritis and cow mortality. The simulated data was analyzed statistically with the economic net return per cow-year as the dependent variable and either the incidences of metritis or the incidence of cow mortality as the independent variables. In the model with metritis we corrected for mediator variables representing the direct effects of metritis on milk yield, fertility and occurrence of other diseases. The EV was estimated as the marginal change in economic net return in response to a change in the trait of interest. To avoid the multiple regression models to correct the EV for structural herd effects (changes in distribution of parities and lactation stages) we used a single animal based indicator variable for each trait of interest, such as incidence rate of cow mortality 1–100 DIM in multiparous cows.The EV value of improving the trait ‘incidence rate of metritis 1–100 DIM in multiparous cows’ by 0.01 was estimated to be €0.93. The importance of avoiding double counting was demonstrated as the EV of metritis was overestimated by 82% when no mediator variables were included in the multiple regression analysis. And by ignoring structural herd effects for the EV of metritis we demonstrated an underestimation in the order of 9%. Further pitfall of underestimation was demonstrated for EV of cow mortality.The EV of improving the trait ‘incidence rate of cow mortality 1–100 DIM in multiparous cows’ by 0.01 was estimated to be €46.4. Correcting for the independent variation in mortality between simulation replicates within the individual scenarios was found to be important.The results of this study suggests a new method for designing simulation experiments and analyzing simulated herd effects for estimation of EV of traits in a breeding goal. This deals with a number of previous and new concerns of how to correct for double counting and at the same time still include the structural herd effects.