That's a good question, and it highlights the fact that special relativity describes far more than just motion at a constant speed in a straight line, and therefore the Lorentz transformations. Confusion about time dilation is almost invariably due to taking an overly simplistic view of special relativity.
The only way two observers can directly compare their clocks is if they are at the sample place in space. Two observers in linear motion can directly compare their clocks only once because by definition they can never meet again. That's why both can conclude the other's clock is running slow without fear of contradiction.
However accelerated observers can compare their clocks more than once - the obvious example is the infamous twin paradox (which isn't a paradox, but that's a rant for another day). When two observers can directly compare their clocks twice, or more, there can be no ambiguity about which clock has been running slow.
The ISS moves in a circle around the Earth. In principle I can put an observer hovering in space in the path of the ISS, so the ISS passes the observer once every orbit. That observer and the astronauts on the ISS can compare their clocks whenever they meet, and when they do all parties will agree that the clocks on the ISS have run slower than the clock held by the hovering observer. That's why we say the clocks on the ISS are running slow.
If however you have one frame of reference passing by another frame of reference, and it is assumed that in each of these frames of reference that all clocks are synchronized, you will use just one clock while examining at least two clocks in the other frame of reference as that frame passes by. This is what allows this strange phenomena to occur in which it always seems to be the "other" guy whose clocks are ticking slower than yours, a view which is shared from both frames of reference. See http://goo.gl/fz4R0I
– Sean Dec 13 '15 at 13:22