I would guess hearing plays a part, as a powerful human sense that works in 360 degrees.
Just as a hypothesis, I'd imagine that small sounds help you sense people behind you, and that this, plus chance, plus the very human capacity to see meaning in random events, gives the impression that there's a way to sense that someone not just behind you but actively looking at you.
In our every-day experience (that is, not in an artificial experimental setting), there are a lot of ways to bias our observations / data collection. For instance,
- a half-conscious glance behind you is quickly forgotten (data discarded) if you realize it was just somebody strolling past, but remembered (data retained) if you see someone staring fixedly at you. This will spuriously increase your positive results (that is, the impression that "every time I look behind me, my sense that someone is looking is confirmed")
- on the other hand, if someone is staring at you from behind, but you don't notice, you also do not retain that negative data, because, by definition, you didn't notice. So there is a strong effect that is spuriously decreasing your negative results.
So, as Brandon Invergo mentioned, we'd like to see a testing protocol along the lines of a blindfold and earphones playing white noise on the "senser", and the "watcher" behind the senser staring or looking away at random intervals, and a large number of trials.
This is not to diminish the fact that some people have much better situational/interpersonal awareness than others (say, a ninja vs. me), but I really imagine that this is more an effect of cognition (putting together small cues from hearing/peripheral vision/smell etc.) than of any difference in sensation.