Many films are telecined or scanned to HD resolution before the DVD is made. Let's say, for argument's sake, that the resolution is 1080p -- 1920 x 1080. Or, let's say that we're looking at a frame from Episode II or III, which originates from 1080p. Whatever: we're starting with 1920 pixels wide, by 1080 pixels high. Those pixels are square, by the way.

Now we go make a DVD.

The horizontal width goes from 1920 pixels to 720 pixels -- a compression of 8:3, or 2.66:1. The vertical height goes from 1080 pixels to 480 pixels -- a compression of 2.25:1. Ergo, exactly analogously to film, the image is squashed horizontally more than it is squashed vertically, and everything in the frame appears too tall and skinny.

Assuming the 16:9 flag is set in the MPEG-2 stream (duh...), this video plays back correctly by applying a horizontal stretch to the pixels (PAR is 1.18 for NTSC) -- again, exactly analogous to the use of an optical lens to stretch the picture horizontally. Not quite as much as the 2:1 stretch I've heard is used for an anamorphic lens, but analogous nonetheless. QED.

Actually, I just wanted to use QED. I don't know if I really can in this case...

Also, I should point out that at no time do any DVD players simply toss out one out of every four lines for display on a 4:3 set. That would simply look way too crappy. There's a resizing algorithm that's used. The net effect is cramming 4 lines of info into 3, but it's more like using some sort of linear resizing in Photoshop than simply failing to display every fourth line.