I think Nigel's comments show the difficulty in distinguishing between the collection types.
As you say, "These Dublin Core definitions overlap (a dataset is a collection of data/facts in a defined structure). Not sure if that is a problem or not."
I think it is, see discussion at viewtopic.php?f=151&t=1511
where I have brought together discussion on this issue and the proposals for multiple collection types or multiple service types.
* What is a purpose of making the distinction? — from the ANDS perspective, to use in facet displays in Research Data Australia. Multiple types will allow resources to be listed under more than one facet where appropriate, which will improve discovery. From the contributor perspective, types assist in description, but, I think, only if this is a non-painful decision, hence the suggestion to allow multiple types to be assigned.
* Are we aiming for non-overlapping definitions?—as discussed in my other post, yes.
If the categories overlap this decreases data quality.
You would not be happy with a survey that asked you to put your age into a category and then offered 20-30, 30-40, 40-50 and so on, because if you are, for example, 30, you don't know which category to choose. Well-constructed surveys would offer 20-29, 20-39 and so on.
Categories need to be mutually exclusive and comprehensive - a place for everything and everything in its (one and only possible) place.
If we can't do this, and I think for collection types it looks increasingly as if we will not be able to, then we should offer multiple collection types as an option instead, to cater for the overlap.
Australian National Data Service
W. K. Hancock Building (#43)
The Australian National University
Canberra, ACT, 0200, AUSTRALIAhttp://www.ands.org.au
P:+612 6125 1176
M: 0466 579 618