As we explained last year, state polls are more useful during a primary. That's because states vote one after the other during a primary, with the outcome of early contests influencing the outcome of later contests. Also, candidates only campaign in the early states (at first), and early state voters pay attention to the race earlier. None of these factors apply in a general election, since the entire country has the same voting day--this year, on November 6th.
For the general election, there are many reasons why state polls are actually much less useful than national polls, even if the state being polled is a swing state. First, state polls tend to be less accurate than national polls. Most pollsters are geared toward polling the country; they do not have as much experience polling one state, and they may not have very good infrastructure in place for it. State polls are less frequent--and it's better to rely on multiple polls rather than a single poll. The infrequency of state polling also leads to the use of out-of-date polls: A newspaper column may categorize a state as being a "swing state" or a "safe Democratic state" based on a poll from a few months ago, simply because there aren't any recent polls there. In addition, desperation for state polling can result in less-reputable pollsters getting their state poll widely published, while national polls are generally reported only if the pollster is well-established.
Even when state polls are frequent and accurate, there are other problems to consider. No one state decides the outcome of a general election. The electoral college total for each candidate is determined by the outcome of all 50 states (and the District of Columbia). If one's focus is narrowed to the swing states, there are still always multiple combinations that can result in the same winner. Thus, cobbling together an electoral college prediction will depend not only on good polling for one state, but for many states. Moreover, when states are very competitive, the uncertainty increases. For example, if polls show a dead heat in several states, one would need to create electoral college projections that somehow manage to call each race accurately.
All of this means that judging the state of the presidential race is far more difficult using state polls than simply using national polls. But what about the fact that the winner is determined by the electoral college rather than the popular vote? For all practical purposes, it doesn't matter. History demonstrates that if a candidate wins the popular vote by at least one point, he is virtually guaranteed to win the electoral college. Even in the 2000 race, where Al Gore won the popular vote by half a point, he was a few hundred votes from victory in Florida. The difference in the popular vote in that race was largely due to low voter turnout in the solid red states of the South where George W. Bush was guaranteed to win; Republican get-out-the-vote efforts in those areas increased in subsequent elections to prevent another mismatch between the popular vote and electoral college winner.
In other words, it's extremely unlikely that the winner of the popular vote will fail to win the presidency, even in a very close race. Since state polls are less accurate, and since electoral college projections rely on calling multiple races, it makes far more sense to follow national polls.
