Just because something doesn't make sense to us laypeople doesn't mean that it doesn't make sense in a court of law, which is where these types of details are decided in fine. The FSF presumably has good reason to write this...
The first point to remember is that documenting copyright years serves to determine the date at which works fall out of copyright.
The Berne Convention leaves "implementation details" up to individual countries. In the US (which is the relevant country for FSF documents), copyright notices are defined by Title 17, Chapter 4; for "visually perceptive copies" (which includes software), they take the form
© year name
- © is specifically that symbol (not "(c)", which has no legal value), and can be replaced by "Copyright" or "Copr." (§ 401(b)(1)),
- year is the year of first publication of the work, or the year of first publication of the compilation or derivative work if relevant,
- name is the name of the owner of the copyright.
What counts is years of publication; for software this is generally considered years in which the software is released. So if you release a piece of software in 2014, and release it again in 2016 without making changes in 2015, the years of publication would be 2014 and 2016, and the copyright notices would be "© 2014" in the first release and "© 2016" in the second release.
In practice new software releases are considered similarly to derivatives, so the copyright notices accumulate; and when the name of the owner doesn't change, notices are merged, e.g. "© 2014, 2016".
The "© 2014-2016" doesn't correspond to anything documented in Title 17 (as far as I can see), which is why its use needs to be explained. To avoid this use being open to interpretation later (when it comes up in court...), it should be documented at the time of use, alongside the use. If it's not documented, it isn't clear whether the notice gives only the years of first and most recent publication (and you don't consider intervening publications as important), or if it specifies that the work was published only in these two years, or if it specifies that the work was published in every year in the range. Hence (I think) the FSF's recommendation.
Note that other communities document different approaches. For example, the Eclipse Foundation uses "Copyright (c) date owner", where "date" can be a single year or the first and last years of publication separated by a comma: "Copyright (c) 2014, 2016" (with no representation as to years of publication in the meantime).
The US Copyright Office's circular 3 has more information on all of this, albeit only single-year forms.
(In practice, these instructions are rarely followed, and copyright notices often have incorrect dates.)