Should the marginal effects be plotted as additive components (i.e., centered at zero mean), or absolute trends over time?