I think so, if you can inject synthetic errors with the same distribution as the errors you care about.