Which Candidates Have an Electoral College Advantage?

Posted: March 2, 2016 in Uncategorized

FiveThirtyEight in conjunction with Facebook recently released a map of Facebook likes for presidential candidates by county.   It’s interesting data, and gives a relatively high resolution map of where candidate loyalties lie.  Although it’s clearly not a random sample of likely voters it can be used to fill in some of the gaps of traditional polling.

Here I attempt to use this data to get an idea of what the electoral college could look like under different candidate matchups.  The assumption I use is that the election will be a close race, with essentially 50% popular vote for each candidate.  Facebook likes are going to be a biased sample of the voting population, but we can adjust for some of that bias by looking only at the candidates performance in a state relative to their national average.  I assume each candidate remains popular in states they are already popular in, then assign each state to the candidate that has the strongest relative support in that state.

There is some danger of reading too much into this data. I’m under no pretenses that this will be a full prediction of how the states will vote on November 8th. The number of likes at this stage in the primary will not match fully with the number of votes in a general election eight months from now.   I expect the effect of partisan voters lining up behind their party’s nominee will make the final electoral map appear like an average of these maps and historical swing state maps.   So solidly Republican states like Alaska will likely ultimately remain red, even if Trump is unpopular there.  Additionally there are likely large biases still remaining in the data – for instance, if Hispanic voters are less likely to use Facebook, are these maps accurate for those states with large Hispanic populations?  Still I do believe there is a value in this data, in showing states and regions where a candidate will be surprisingly competitive. This can be seen as a rough prediction of tipping point states, which can be seen to vary dependent on the candidate matchup. I expect the political campaigns are looking at their own versions of similar maps as they chose where best to spend their resources.  Enjoy!

Sanders vs Trump

Click the map to create your own at 270toWin.com

Advantage Trump. This means Trump would win the electoral college if the popular vote were tied (which it may not be).  In a more lopsided election, the candidate with the higher popular vote would win.

This map is the most interesting because it shows a matchup between two very unconventional candidates. Trump is a political outsider and not consistantly a Republican, while Bernie spent his Senate career as an Independent. Both candidates appeal to voters outside of their traditional party bases. Trump’s main advantage geographically is his strong support in Democratic mid-Atlantic states. Trump is very weak in the West overall, allowing him to lose the traditional Republican strongholds of Alaska and Utah in this model. Unfortunately for Sanders many of these Western states he performs well in are not worth that many points. Bernie’s current weakness with black voters means he is not able to penetrate into the South to bring states like North Carolina and Virginia into play. Bernie performs well in states near Vermont, but loses some key high value states.  It may not be a coincidence that Trump does well in the casino and resort states of Nevada, New Jersey and Florida.

Sanders vs Cruz

Click the map to create your own at 270toWin.com

Advantage Cruz.

Cruz generally performs very well in traditional Republican states and the south, leaving the battle to be fought primarily in standard swing states. In all of the matchups I look at Republicans have the advantage in Florida, a key high-value swing state. Bernie’s lack of competitiveness in North Carolina and Virginia means he starts at a disadvantage in swing states.  Sanders holds the Northeast but is unable to pry away states in the midwest and loses some crucial swing states.

You’ll notice Montana is colored a lighter shade of blue. Since I’m relying not on the raw Facebook data, but on rounded data from the fivethirtyeight website, for some states it’s difficult to tell which candidate has the advantage. For that reason I don’t attempt to predict the margin of victory for each candidate. Montana is a state that appears to lean towards Bernie, but is within my margin of error so is colored a lighter shade. Candidates which have lower overall numbers of Facebook likes have a larger margin of error. Since Bernie and Trump are the most popular on Facebook, their map is the most accurate. I also generated a Clinton- Bush map, but the accuracy was too poor to predict states accurately due to Bush’s relative unpopularity on Facebook.

Sanders vs Rubio

Click the map to create your own at 270toWin.com

Advantage Sanders.

Republicans seem to be losing their grip on certain Western states, and Sanders is well positioned to take advantage of this.  Sanders comes out ahead due to strength along Rocky Mountain states and the NorthEast.  Marco Rubio’s support falls mainly in traditional Republican states in the South, with potential to secure states with large Hispanic populations like Arizona and Florida.  Rubio captures the south but doesn’t expand the map to put any traditional Democratic states on the table.

Clinton vs Trump

Click the map to create your own at 270toWin.com

Advantage Clinton.

This is a very close map.  Trump is actually more popular then Clinton in Bernie territory of New Hampshire and Maine. He also tends to do very strongly in mid-Atlantic states. Clinton served as senator in New York, and it appears this gives her the advantage necessary to keep Trump from stealing New Jersey. Clinton pulls out ahead in this scenario only by managing to win Texas.  For years there has been talk of Texas potentially becoming a swing state. This scenario shows Clinton actually contesting this large prize due to the difference in support between her and Trump in towns along the border.

Clinton vs Cruz

Click the map to create your own at 270toWin.com

Advantage Clinton.

This is the most conventional map. Each candidate represents the core of their party, with Clinton as the establishment Democratic candidate and Cruz as the self-appointed king of the Tea Party (the new Republican establishment). It is not surprising that the map follows traditional swing state boundaries, which I see as a vindication of this method. The only slight surprise here is Ted Cruz winning New Mexico. However there’s some possibility that New Mexico, with Hispanics making up 40% of the voting population, may not be modeled as well as other states by the Facebook data. While Ted Cruz solidly holds on to all areas of Republican territory, including the Western states that Trump and Rubio are weaker in, he fails to show much appeal outside these regions.  Cruz appears to have zero appeal in the Northeast and is generally outperformed by Clinton in most swing states.  Clinton wins by sticking to traditional Obama Democrat and swing states, staying competitive in Virginia.

Clinton vs Rubio

Click the map to create your own at 270toWin.com

Advantage Rubio.

This map is also fairly traditional, as an election between the two establishment party picks. Rubio does well in all traditional Republican states, and is able to deliver the South, while remaining competitive in traditional swing states. He is seen as winning New Mexico and keeping Clinton from making inroads into Virginia.  This is consistant with the perception that Rubio nullifies some of Clinton’s advantage among minority voters. Likewise, Clinton is able to hold the Northeast but not expand heavily into new territory in the West or win key swing states.  A slight bump is seen in Arkansas, home of the Clintons, but she is not anticipated to win the state.

General Conclusions

It would appear that Republicans have the electoral college advantage in North Carolina, Florida and Ohio this year, making a collection of smaller states crucial to determining the outcome of a close election.   For Clinton the path to victory is very similar to Obama’s 2012 path.   However if either Trump or Sanders are in the general election, the map changes considerably.  Trump’s areas of strength seem to be locations were his businesses have a presence.  The states where he received the most google searches between 2004 and 2014 match closely with the states he is shown as leading.

  1. jimmythehack says:

    So how’d the model do? Not great really. The Trump vs Clinton model got Nevada, Delaware, Iowa, Wisconsin, Maine and of course Texas wrong. Now Texas was always an unlikely flip & within the margin of error in this technique, but it wasn’t close to flipping in the end – traditional partisan politics prevailed. You could look at this map and say that it predicted Clinton’s difficulties in Michigan, Pennsylvania and North Carolina. But it failed to predict Clinton’s failings in Wisconsin. It’s not really clear to me that this method was much more predictive than a traditional swing state map. Still it is pretty neat that this very crude method using data collected over 8 months in advance of an election can produce a map that looks even vaguely like the results. I still believe that a more sophisticated analysis with access to the actual raw data could prove pretty insightful.

  2. jimmythehack says:

    So I actually went back to check the margins in the Wisconsin calculations. It turns out FiveThirtyEight updated their data in April after I did the maps. Trying the same method with the new data gives a better result. Here we can actually see some prediction of the Wisconsin surprise, and more of the states are predicted correctly. Virginia shows as a complete tossup. https://www.270towin.com/maps/yvvRk

    Why is this map more accurate? It uses data closer to the election, which we would expect to be greater in volume and more predictive. The data I’m working with is rounded to the nearest percent, and it’s possible some of the rounding errors were decreased this time. I’ve also had more of a chance to double-check the calculations this time. To calculate which candidate has the lead I took their popularity vs the US average and multiplied Clinton’s by 2.66 (to match her difference in likes overall). Any states which may be affected by the 1% rounding are marked in lighter colors. A few states moved in and out of this rounding uncertainty.

    The Facebook data also shows a weak correlation to regions won by Clinton and Sanders in the primary, but seems to be only partially predictive. https://www.nytimes.com/elections/2016/national-results-map

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s