Now I'm confused -- you did truncation, or you did dithering? After thinking about my earlier argument, I realize it really didn't matter whether or not you truncated or rounded -- rounding is just truncating + 0.5. I disagree that it's inaudible -- why would they bother with dithering if it's inaudible?
Alright, so here's my attempt at explaining in concise form why one should use dithering instead of truncation:
The difference between the original analog signal and the unfiltered digital one is a good representation of the quantization error. If you picture this signal, you'll immediately intuitively see that it has frequency components that are not functions of any frequency in the music, but are functions of the sampling rate. If there is any regularity to this at all (and we're dealing with computers and other silicon-based lifeforms here, so there is), you'll also realize that these components could easily divide down into the audible frequency range. So one has to force a randomness to it, and the science of this forced randomization is called dithering.
You're a scientist -- you should perform Torpedo's experiment of recording the same exact analog source through a variety of different bit depths and sampling rates. I actually have the facilities (Alesis Masterlink 9600), I might have to set it up at the next meet. I'll also see if I can create dithered down and truncated versions of the same material and see if people can hear the difference between the three (originally recorded at 16/44.1, dithered down to 16/44.1, and truncated to 16/44.1...or something like that -- I might do 48 just because of grawk's point about sampling rate conversion, which is a legitimate one).