Vocab time:
Resampling = changing the sample rate (96 kHz to 44.1 kHz)
Dither = changing the bit depth (24-bit to 16-bit)
I'm having a hard time following what you're using to try to accomplish whatever it is you're doing.
Since we're trying to be precise, dither does not = changing the bit depth
Changing the bit depth is just that. Others call it word length reduction or whatever. It is simply throwing away the least significant bits, kind of like rounding off numbers. You see 13.345679 on the calculator and you say 13.3 (except for that little rounding rule we use). Same thing but in binary.
Dither is a mathematical technique used to capture detail that would not normally be captured at the reduced word length. When you truncate from 24 to 16 bits, you're basically chopping off the last 8 bits. On their own, these are only used to capture the very quietest sounds. During a peak, they capture the finest info about the event, like the last decimal place in the number example above. You can't chop the 8 bits on the other end of the 24 bit word because they cover all of the musical peaks so you chop the fine detail.
Dither involves adding some noise to raise those little bits up to the point where they will have an effect within the 16 most significant digits of the 24 bit recording. Then you chop off the last 8 and you have some of that detail you wouldn't have had without the dither. Noise shaping is about how you choose to apply the dither (out of scope!)
That's the theory on it. With respect to levels, I try for peaks around -6 to leave room for the unexpected. There's absolutely no point in going over.
Simply put, the digital world allows 6 db of dynamic information per bit. So, a 16 bit recording allows 96 db of info, while a 24 bit recording allows 144 db. Sooooo... when folks say you don't need to push the levels in 24 bit recording, this is simply because even if you're peaking out at -18 db you're still capturing 126 db of info which is still a higher resolution than a perfectly set 16 bit recording.
True, but 24 bit recording benefits aren't limited to the additional dynamic range in the sense that you can record louder sounds or anything like that. It's more like using a fine grain film or a higher image density in a digital photo. You get a more detailed account of each sampling event.
That said, I have yet to prove its worth to myself. I've been recording in 24 bit for over a year and just kind of do it on faith. I can do 24 bit playback but I'm not sure that live shows are really the way to showcase the sonic capabilities of 24 bit. Right now I can only playback 24 bit through an emagic A62 - I really don't know how good the d>a is. Got something else coming soon but I guess I need to take a show and run A/B on the 16 and 24 bit through the same d>a. hmmm...maybe I'll get on that.
<edited for spelling>