On choosing a license for open scientific data

I (and my collaborators) am currently in the process of releasing an open-access and open-source data set. In doing so, I've had my first brush with the semi-logical process of selecting a license to stamp on that data. To make the process of selecting a license easier for others who may be in the same situation, I'll walk through my thought process when deciding which license to pick, and which elements to include in the license.
This is only a single example; any generalizations are drawn at the reader's own discretion. (Also, be wary of taking legal advice from a physicist!)

My thought process went as follows:

Should I share my data?

This question relates mostly to the nature of the data set: Will others benefit from having my data? Can I see questions being posed to the data outside of the ones I'm intending to ask? Is my data original in either type (fairly rare!) or scope (very common!) than other, currently available data sets?

For us, the answers are yes, yes and yes. Very simple, we can move on to...

Which license should I choose?

Here's where we hit the first section of the alphabet soup: MIT, MIT OCW, CC, GPL, BSD 3-clause, BSD 2-clause, Apache, etc. etc. Opensource.org has long diatribes on this, but little of the text relates directly to how the licenses are applicable for scientific data sets.
A cursory examination revealed that two licenses consider explicitly how databases work: The MIT license, and the different flavors of Creative Commons (CC) licenses.

For our purpose, the MIT license is too permissive, since some of the data used to generate our data set restricts some uses (see below). We thus went with a CC license. This leads us to the alphabet soup of:

Which CC license should I choose?

To choose the type of CC license you want, there are 4 choices to make, which are, in increasing order of difficulty to answer:

BY: Do I want to require any users of the data set to cite me?
ND: Do I want to prohibit users from altering/updating the dataset?
NC: Are the users forbidden from using my data set for commercial purposes?
SA: If someone alters/updates and re-publishes the data, do they have to put the same license on the new work?

In our case, the first three were easy to answer:
BY: Yes. This is standard scientific practice, and this choice makes the user legally obligated to do so.
ND: No, users can alter the data. We want our users to be able to update and extend the data set using their expertise and own data, to create a better data set for the years to come.
NC: Yes, users may not use the data set for commercial purposes. This is a restriction due to the upstream data used in the data set's construction.

We've gotten to a CC BY-NC license, so far, so good. But choosing whether to apply SA or not deserves a section on its own.

Should I use the Share-Alike license?

When reading the description of SA above, my knee-jerk reaction to this question was to say yes, of course I should restrict the re-licensing of my data. After all, a big part of why I'm building open data sets is to encourage more open data use in science! If someone wants to use open data, they should give back to the community in the form of more open data!

Reality turns out to be somewhat more muddy than this.

Suppose someone has created a data set Y with, say, measurements on chairs, and released it under a CC BY-SA license. You are a researcher on chairs and chair-like objects, and it just so happens that you have a great idea for a research article cross-correlating the measurements on the chairs with their maximal carrying weight. Unfortunately, Y did not have this data and you don't have the budget for sufficient chairs and weights, but you do have friends in the industry that are able to collect the data on carrying weight for you. Your industry friends give you a data set Z on the carrying weight, on the condition that you can only share this data with fellow sitting-implement researchers who explicitly ask for it.

You write your article, it gets published to great fanfare, and several of your peers write to ask for the data you used so they can reproduce the results and extend your work.

You are now stuck in a legal bind. If you say yes to sharing, the SA part of Y's license says you must stamp any redistributed work with the same license, but the agreement you made to get Z means you cannot put a CC license on the combination Y+Z. You are thus not permitted to distribute the combination Y+Z. Various grey areas exist here, such as distributing Z and pointing your fellow researchers in the direction of Y, but to stay on the straight-and-narrow, you are really in a bind.

Thus, the creators of Y, in an attempt to encourage more sharing of data, have put you in a situation where your research is less reproducible. Adding the SA tag restricts which works can be mixed in with yours, leading to a lower quality of research from your work. It may also hamper adoption of your data sets, as some of your peers may not want to deal with the hassle of complying with the SA tag.

How significant you consider this burden to be is an individual assessment, but the complications should be taken into account when weighing the option of the SA tag.

For what it's worth, I decided to not apply the SA tag to our data set release, unless my collaborators have serious objections. I rate the ability to combine our data set with proprietary (or limitedly shareable) sources to be a higher upside than the potential loss of openness which not having the SA tag may incur, especially in a field that still has a way to go in the adoption of open data. Your mileage may vary.

(Please point out my misinterpretations and any nuances I will inevitably have missed in the comments.)