Balancing central and marginal rejection when combining independent significance tests