2. Without looking at all the states, we don’t know how to interpret state polls.

When incumbent presidents run for re-election, the electoral map tends to be fairly consistent with their first campaign. States that were competitive in 2008, for example, were also competitive in 2012 (with a couple of exceptions). Statistically, President Obama’s 2012 vote share in a state can be predicted within 4.5 percentage points, 95 percent of the time, just by knowing how much the national popular vote changed and how well Obama did in that state in 2008.2

But the map is far more likely to change in open elections, like 2016, when no incumbent president is running. In 2008, the state that voted most like the country overall was Virginia; Obama won the nation by 7.3 percentage points and Virginia by 6.3 points. Virginia also held that honor in 2012 — Obama won both the nation and the state by 3.9 percentage points. But in 2004, President George W. Bush won Virginia by over 8 percentage points while winning nationwide by a little less than 2.5 points. Nevada was actually the national bellwether in 2004, but Obama won it by double digits in 2008. Statistically, Obama’s vote share in 2008 could be predicted only within 6.4 percentage points, 95 percent of the time, based on 2004 state-by-state data.3 That error rate is nearly 2 percentage points higher than when Obama was running for re-election.

The point is that the 2012 electoral map may not apply in 2016. Let’s say Clinton leads Trump by 6 percentage points in North Carolina, as an average of the three polls conducted there in the past month suggests. So what? We don’t necessarily know what that says about the race overall.