Source = 5.1 AC3 @ 448kbps

I decode the AC3 to WAV using HeadAC3he v0.24-a13, using these settings:
- Dest. Format: WAV - 16 bit init
- 2-Pass Mode
- Normalize to 95%
- Downmix Type: Surround 2

Then, I edit the WAV using Audacity, using sampling rate of 48000hz, and 16-bit.

Finally, I encode the WAV to MP2 using tooLAME with the following as my encoder settings:
toolame -s 48 -m s -p 1 -b 384 -d n -e input.wav output.mp2

My question(s) is/are, are my settings amongst all the programs set properly?

For instance, when I decode the AC3 to WAV, do I want 16 bit init? Or would 24 bit init be better? Or even 32 bit float?

HeadAC3he uses Surround Mode 2 by default as the Downmix Type, does that have any factor as to which psymodel in tooLAME to use? I've been using psymodel 1 during my encodes.

tooLAME also lets you choose which stereo mode to use. When going from AC3>WAV>MP2, is it preferred to use -m s or -m j (stereo/joint stereo)?

Finally, would it be best to use 16 bit init, 24 bit, 32 bit, when editing the WAV in Audacity?

by default audacity does its stuff in 32bit (i think).. since you are editing the files, it would probably be better to export from headac3he in 24bit
I would export in 16bit from audacity though.

personally, I can't tell any difference between stereo/joint stereo and the psymodels in toolame, especially at 384kbps ... so I can't really help ya there.

ps: headac3he

Audacity does, by default, use 32 bit float. I changed it to 16 bit only because I thought the end result MP2 would have to be 16 bit anyhow so why not let Audacity do it.

As for HeadAC3he, it can handle up to 32 bit float.

Do the end-resulting MP2 files have to be 16 bit in order to be dvd-compliant?

No, you can safely use the highest bit depth that your mp2 encoder can handle. (e.g. 24 bit integer or 32 bit float)

(note: the encoded MPEG audio file does not have a "bit depth" property in that sense)