IMO.. dithering is a compromise so that people can easily listen.
If 16bit is your goal, I personally think a 16bit recorder would give you the best finished product.
If 24bit is your goal, same said for a 24bit recorder.
I haven't touched a 32bit recorder, but my assumption is the same.
Sorry but I must disagree.. on both points
First of all dithering is not a compromise. It is the correct
standard procedure which should be applied whenever truncating from a higher bit depth to a lower one. The alternative is to simply truncate without dithering which produces quantization artifacts. Granted that may be inaudible in many cases, but it is not the correct way to do it. Perhaps you meant the
reduction of bit depth in general (which is done correctly by applying dither first) from 24 to 16 bits or whatever, rather than
dithering. But I have to disagree there as well. As long as the total dynamic range of the program material fits within the delivery container, no additional bits are needed. They will contain only random values below the bottom of the actual range (noise) and/or zeros (silence) above the top of the actual range. Those extra bits contain nothing of value, they are wasted.
Extended dynamic range is beneficial when recording and processing. Once those processes are complete and the total dynamic range of the audio file can be easily determined and will not change, there is no harm in throwing away the excess bit which contain no useful information.. assuming it is done properly.