From Jonathan Bernstein:
Since it’s the Electoral College that matters, shouldn’t we should stick to a handful of potential tipping-point states?
State surveys have two big problems: A larger percentage of them are from lower-quality outfits, and there just aren’t enough of them. Remember, polls will be wrong for two reasons.
Some aren’t designed well; it’s like flipping a weighted coin that will average, say, 6 heads out of every 10. But even the best polls are subject to random error, just as if you flip a normal coin 10 times you can easily get 4 or 6 heads; you’ll sometimes get 3 or 7 heads; and you’ll even get more extreme results once in a while.
The power of polling averages is that it essentially corrects for random error, but only if there are enough surveys in the pot. So given the choice, I’ll stick with the far more plentiful national polls to get a sense of what’s happening. Especially since more of them tend to be high-quality efforts.