When it comes to political warfare, the Republican party has the superior ground game—literally.
A glance at the map of any recent presidential contest confirms this. Due to their dominance in large, low-population states like Wyoming and the Dakotas, Republicans appear visually to command a dominant lead even when the actual results are very close.
To remedy that problem, Gott and fellow cosmologist Wes Colley invented a map that visualizes each state with blocks according to the number of electoral votes it receives, such that the area of red and blue exactly corresponds to the results of the electoral college. TIME developed an interactive version of this map below that you can populate with a variety of different forecasts for the 2016 election.
Maps like this look weird at first, but they’re quite useful in producing an accurate picture of the race. “It gives you a visual sense right away of who’s winning,” Gott says.
There are many variations on such diagrams, known as “cartograms.” The site FiveThirtyEight used hexagons to create a similar map, while the Wall Street Journal also takes a block approach. The physicist Mark Newman has created even stranger looking maps by treating the population like particles that diffuse until they’ve reached uniform density.
The advantage of this particularly cartogram, Gott argues, is that preserves the shape of each state as well as possible while also preserving which states border one another—hence the hole in the upper center for Lake Michigan, since Michigan’s Upper Peninsula connects with northern Wisconsin. The map even preserves the Four Corners where Arizona, New Mexico, Colorado and Utah intersect at a single point.
“You just have to fiddle with it until you get it right,” says Gott, who described the map as more art than science. (By contrast, the diffusion maps that Newman produces are all created algorithmically.)
In addition to the cartogram, Gott and Colley also have their own method for predicting the outcome of the election—the default predictions loaded in the above map—and it’s remarkably simple: For each state, they take every poll from the past month and look at the median—that is, the survey for which half of the other polls show stronger figures for Trump and the other half show better numbers for Clinton, regardless of who’s actually in the lead.
For example, as of this writing, there have been seven polls conducted in Ohio in the past month, based on those listed on RealClearPolitics. Of those, Trump has won five, Clinton has won one, and one was a dead heat. The median value—the poll in the middle if you were to sort them from most pro-Trump to least pro-Trump—is a poll that shows Trump ahead by 3 points.
This method is much better than taking the average of the polls, Gott says, because it discounts extraneous outliers. In Ohio, for example, the one poll that shows Clinton ahead reports a 7-point lead for the Democratic nominee, which is considerably far off from the rest of the polls. But that outlier would pull down the overall average to about a 1.6-point lead for Trump, while 3 points is arguably a much better estimate.
“We’re basically asking who’s ahead in more polls,” Gott says. Discounting outliers, he says, “is just what you want when you’re dealing with polls. One of the polls may be biased or too partisan.”
The crowded field of forecasters all have different ways of aggregating polls, sometimes using complicated mathematical models or large numbers of computer simulations. One of the attractions of Gott’s model is that it is very simple and transparent.
“As a predictive measure, it’s going to be a bit behind. But I’m a pretty big fan of the idea in general,” says David Rothschild, an economist at Microsoft Research who runs PredictWise, a site that makes a large number of predictions based on online gambling markets. (Full disclosure: I worked with Rothschild in 2012.)
In addition to PredictWise, you can also populate the map with predictions from University of Virginia political scientist Larry Sabato, the Cook Political Report and The Crosstab by G. Elliot Morris.
In 2004, the method that Colley and Gott developed correctly predicted every state except Hawaii, where only two polls were conducted. In 2008, they missed three states and one electoral vote in Nebraska, one of two states that splits its electoral votes by congressional district. (The other is Maine, which is why the above map specifies district numbers for those two states.) In 2012, they only missed Florida.
This year, Gott points out, Clinton has a “firewall” of 273 electoral votes—three more than she needs to win—in which she has won nearly every poll conducted. The most vulnerable of those is Colorado, according to his method. So don’t be surprised if you see her heading there in the coming weeks. Virtually every forecaster has Clinton with better-than-even odds of winning, but that gap is narrowing. And there are many more polls to come.