I have never done that with Audacity, but there is a serious flaw in the technique! The sound of the difference (subtracted files) is NOT the difference in the sound! Weird, eh?
MP3 makes phase & timing changes that may not be audible, but they show-up in difference testing…
It’s true, that if you subtract two files and you get absolute silence, the two files are mathematically identical, and of course they will sound identical. And, we already know that MP3 is lossy, and we know that low bitrates are more-lossy than high bitrates…
But for example, if you add 10 milliseconds of silence to the beginning of a file it will sound identical to the original. But, if you subtact the two files you will get a comb-filtered file showing a drastic difference that bears no resemblence to a tiny delay… The sound of the difference is not the same as the difference in the sound.
Another example - If you invert the phase the file will sound identical (or very-nearly identical). Now if you “subtract” that file from the original, you are of course actually adding the two files together (subtracting a negative), resulting in a difference-file that’s twice as loud as the original, but otherwise identical (assuming no clipping).
After repeating this process with 192 and 320 kbps MP3s, I’ll be able to see exactly how much noise each bitrate adds, which should better help my decisions in the future about ripping media to MP3.
There is software available for performing blind [u]ABX tests[/u] to determine if you hear a difference in your program material.
It turns-out that program materal has a lot to do with it… Some songs easier to compress than others, and the variable bitrate LAME “V” settings try to take this into account, using a higher bitrate when necessary and a lower bitrate when possible. So rather than choosing a constant bitrate like 192, it’s usually best to choose a “V” setting that averages-out around 192.
Like I said, I’ve never done this with Audacity, but take a look at [u]Mixing[/u]. Mixing is done by addition (both in analog mixers and digitally), so inverting and mixing with the original will give you the result you are looking for.