Atomic-scale elemental maps of materials acquired by core-loss inelastic electron scattering often exhibit an undesirable sensitivity to the unavoidable elastic scattering, making the maps counterintuitive to interpret. Here, we present a systematic study that scrutinizes the energy-loss and sample-thickness dependence of atomic-scale elemental maps acquired using 100-keV incident electrons in a scanning transmission electron microscope. For single-crystal silicon, the balance between elastic and inelastic scattering means that maps generated from the near-threshold $\mathrm{Si}\ensuremath{-}L$ signal (energy loss of 99 eV) show no discernible contrast for a thickness of $0.5\ensuremath{\lambda}$ ($\ensuremath{\lambda}$ is the electron mean-free path, here approximately 110 nm). At greater thicknesses we observe a counterintuitive ``negative'' contrast. Only at much higher energy losses is an intuitive ``positive'' contrast gradually restored. Our quantitative analysis shows that the energy loss at which a positive contrast is restored depends linearly on the sample thickness. This behavior is in very good agreement with our double-channeling inelastic scattering calculations. We test a recently proposed experimental method to correct the core-loss inelastic scattering and restore an intuitive ``positive'' chemical contrast. The method is demonstrated to be reliable over a large range of energy losses and sample thicknesses. The corrected contrast for near-threshold maps is demonstrated to be (desirably) inversely proportional to sample thickness. Implications for the interpretation of atomic-scale elemental maps are discussed.