Why is it a good idea to try to reduce the chances of errors happening in an experiment?