What does PSE do to get from 3200 to 500 pixels? >Are pixels combined,
eliminated, shrunk?
Elements will interpolate your information based on the interpolation method you select when downsizing your image. Usually you will use Bicubic interpolation, which downsizes and "sharpens" at the same time.
Interpolation affects how the choice is made between merging pixels. Pixels are a standard unit within the image (pixel = picture element), so they are not shrunk. In a sense they can either be combined (bicubic and bilinear interpolation do averaging between pixels to determine new values) and/or eliminated (Nearest Neighbor chooses from the best representation of pixels currently in the image without averaging); in going from a larger image to a smaller one, in some sense image information has to be removed.
Sizing down should not have as great a visible effect on the final image as sizing up would. While sizing down is common (repurposing for less demanding output) sizing up should be a relative rarity (when you size up, information is added to the image -- representing the application flat-out guessing at what should be there...). In other words, while both interpolate and 'guess', you see less of the image as it gets smaller--and what is dismissed by the application will weigh less on the result. Sizing up is like using a magnifying glass to zoom in on the flaws in the process.
Hope that helps!