5 minutes seems sort of arbitrary. How do they know they have a representative set that describes the whole population? What if in the next 5 minutes they get a rush of users who prefer the other headline? On the other hand, if they are getting second by second updates and they are looking at the standard deviation of the entire minute, then maybe it would be easier to tell which dataset was more statistically significant...no?
It all depends on the amt of traffic they get. I'm sure they've experimented enough to realize that 5 minutes is enough time to get a representative sample of their general audience.
