Everyone wants to correctly predict the outcome of November's election, but how do you do that when nationwide polling tells you the race is close thing and a state-by-state approach tells you Obama is winning big? Why can't these two methods seem to agree?
National polls seem to show the race as a virtual dead heat, with just one or two points separating the candidates and the leaders occasionally flip-flopping. Yet, a state-by-state analysis seems to show the president leading by three-to-five points. Not a huge jump, but significant enough to matter, particularly when you look at the most competitive states — the ones that will eventually swing the Electoral College to the victor.
So how to you figure out who is really winning? Well, Nate Silver and his FiveThirtyEight crew did some number crunching — which is basically all they do — to try to come up with a model that can meld the two approaches and see if there's a way reconcile them. This gets a little nerdy from here, so the math deficient may want to just skip the end.
Their new method starts with something called the "relative popular vote," which breaks down each state by matching their vote total to the national result. If the Democratic candidate wins the national popular vote by 10 points, but wins a state by only five, then that state's "relative vote" leaned 5 points toward the Republican.
According to Silver, those numbers are historically very consistent. So if you knew what the total national vote was four years ago and what the national polls say today, you can make a good guess to what each individual state's vote would be today. For example, if Washington is normally 10 points more Democratic than the rest of the country and Obama leads by two percent nationally, then you can infer that he leads by 12 points in Washington.
It works both ways too: A state poll that has a 13 point lead for Obama would translate to a 3-point lead nationally. So now if you take the site's "modified polling average" (which is a weighted collection of all the polls taken in a given state) and compare it to that state's "relative popular vote" you can guess what the national vote would be.
In New York, for example, the modified polling average put Mr. Obama 23.3 points ahead as of Tuesday’s forecast. In 2008, New York was 19.6 points more Democratic than the country as a whole. By subtracting 19.6 points from 23.3 points, we get what I call an implied national popular vote. In this case, the result is Obama plus 3.7 points.
Now do that for every state, average them together, and you get a "Implied National Popular Vote" without looking at a single national poll. FiveThrityEight adds in some of its own statistical magic to adjust for anomalies — the home state of the candidate; the reliability or unreliability of certain states; and the fact that no one even bothers to poll in South Carolina — and you end up with "FiveThirtyEight's Patented Presidential Voting Index Wieghted Implied National Popular Vote."
The verdict? President Obama leads by 3.6 points nationally, about two points better than the latest average of all the national polls. Which leaves us right back where we started — unable to agree on anything. Kind of like the election itself.
This article is from the archive of our partner The Wire.