We may first separate datasets featuring 3D spatial information (e.g., 3D geometric models, 3D volume data) from those datasets featuring 3 or more variables but not “naturally” 3D spatial information. In the former cases, 3D visualization usually allows users to relate what is displayed to what they know more easily. For example, when a doctor sees the visualization of a heart (in a 3D-to-2D projection), he/she can reconstruct the 3D shape/structure (including the occluded parts) without much effort. For an unfamiliar object (e.g., an engine block), the doctor may have to use more interactions (e.g., rotating and zooming) to figure out the shape/structure.
For the latter cases, when the 3D spatialization appears to be unnatural for the datasets, users have to use more cognitive load to construct a representative mental impression of the data. So the 3D spatialization may stimulate undesired distortion during the reconstruction, e.g., due to humans’ natural intuition to interpret distances, sizes, and occlusions with perspective projections in mind. However, in some cases, stimulating such intuition may be desirable (e.g., for memorization or visual metaphors in disseminative visualization). Occlusions are themselves not necessarily always a problem. Some occlusions (e.g., lines in parallel coordinates) are less problematic than others, which fundamentally depend on humans’ ability to reconstruct the whole shape from a partial, translucent, or deformed view, and the cost of using interaction to reveal the occluded objects. Dealing with occlusions in visualization is closely related to visual multiplexing (Chen et al. CGF, 2014).