There are many different ways to predict Presidential elections. Some prognosticators rely on broad economic and political indicators like GDP growth and Presidential Approval Ratings, while others like Nate Silver at FiveThirtyEight aggregate state and national polls to predict the final outcome. Models like Nate’s have had great success in the past, and I’m not here to delegitimize models based on polls.
But relying on polling leaves a lot to be desired. Not only do polls only sample a tiny fraction of the electorate, each individual polling company has their own biases and inaccuracies, and polls can be volatile over an election cycle. Additionally, many polls contact people either exclusively or partially through landline phones, a technology largely unused by younger people, an obvious mistake that led to a massive 23 point polling error in this year’s Michigan Democratic primary.
That’s why I’m proud to introduce a new way of predicting Presidential elections. During the primary season I created a statistical model based on Facebook “Likes” to predict the results of the Democratic primaries and caucuses. The accuracy of the model was astounding, exceeding my wildest expectations. My model correctly predicted the statewide winner in 46 of 52 major Democratic primaries and caucuses, including Washington D.C. and Puerto Rico. When my model and the polls disagreed about the winner of an individual state, my model was correct in 8 out of 11 disagreements, including the infamous Michigan primary.
Given this success, I decided to transition the model from predicting primaries and caucuses to the general election. Before we move on to state by state predictions, let’s explain how the model works (and how it doesn’t work):
- Having more likes than the other candidate in a given state does NOT mean said candidate is favored to win that state. Nationally, Trump has more Facebook Likes than Clinton in all 50 states. That does not mean that we should expect Trump to win in all 50 states. It simply shows that Trump has a larger presence on Facebook than Clinton.
- Statewide support for each candidate is estimated by the ratio of Facebook Likes in each state. Trump, for example, has roughly 3 times as many likes as Clinton nationwide. In California, however, Trump only has roughly 40% more likes than Clinton, much lower than the national average. In West Virginia, meanwhile, Trump outnumbers Clinton in Facebook Likes by a roughly 7 to 1 ratio, much higher than the national average. Unsurprisingly, the model has Clinton doing very well in California and very poorly in West Virginia.
- Nationwide support for each candidate (the popular vote) is estimated by using FiveThirtyEight’s “polls only” model. While I don’t always agree with FiveThirtyEight’s predictions, they are excellent at aggregating polls. In order to translate a Facebook “Likes” ratio to statewide results, a national baseline is needed. As of this writing, Clinton currently leads Trump nationally by 2.5 points in their polling average.
Before we look at the model’s current projections, a few caveats. First, the model does not take into account third party candidates. While I don’t expect third party candidates to significantly tip the election one way or another, they probably do make a small difference in states like Utah and New Mexico where Gary Johnson is polling very well. Additionally, another possible issue is that the Facebook data the model is using is old, dating back to April 18th. Fortunately, I have contacted FiveThirtyEight (the source of the Facebook data), and they have let me know that they will most likely be updating their data once more before the election. When they do so, I’ll make an updated Election Preview post with the new data.
With that being said, let’s look at the current state by state odds, assuming that Clinton is ahead of Trump by 2.5 points nationally. Listed in the table are Clinton and Trump’s projected share of the vote in each state, Clinton and Trump’s respective chances of winning each state, and Trump’s Z-Score, defined as how many standard deviations above or below average Trump is in a particular state compared to the national average. The standard deviation was calculated by using data from the Democratic Primary model.
| State | Projected Trump Vote | Projected Clinton Vote | Trump Z Score | Trump Win State% | Clinton Win State% |
|---|---|---|---|---|---|
| Alabama | 67.9% | 32.1% | 4.6 | 100.0% | 0.0% |
| Alaska | 61.3% | 38.7% | 2.9 | 99.8% | 0.2% |
| Arizona | 51.2% | 48.8% | 0.3 | 62.3% | 37.7% |
| Arkansas | 55.0% | 45.0% | 1.3 | 90.2% | 9.8% |
| California | 28.1% | 71.9% | -5.7 | 0.0% | 100.0% |
| Colorado | 43.8% | 56.2% | -1.6 | 5.5% | 94.5% |
| Connecticut | 44.8% | 55.2% | -1.3 | 8.9% | 91.1% |
| Delaware | 50.2% | 49.8% | 0.1 | 52.4% | 47.6% |
| Florida | 53.1% | 46.9% | 0.8 | 78.8% | 21.2% |
| Georgia | 56.1% | 43.9% | 1.6 | 94.3% | 5.7% |
| Hawaii | 44.3% | 55.7% | -1.5 | 7.1% | 92.9% |
| Idaho | 63.2% | 36.8% | 3.4 | 100.0% | 0.0% |
| Illinois | 39.1% | 60.9% | -2.8 | 0.2% | 99.8% |
| Indiana | 57.7% | 42.3% | 2.0 | 97.7% | 2.3% |
| Iowa | 43.0% | 57.0% | -1.8 | 3.4% | 96.6% |
| Kansas | 54.1% | 45.9% | 1.1 | 85.7% | 14.3% |
| Kentucky | 61.9% | 38.1% | 3.1 | 99.9% | 0.1% |
| Louisiana | 66.6% | 33.4% | 4.3 | 100.0% | 0.0% |
| Maine | 53.4% | 46.6% | 0.9 | 81.1% | 18.9% |
| Maryland | 37.6% | 62.4% | -3.2 | 0.1% | 99.9% |
| Massachusetts | 33.5% | 66.5% | -4.3 | 0.0% | 100.0% |
| Michigan | 50.2% | 49.8% | 0.0 | 51.8% | 48.2% |
| Minnesota | 41.8% | 58.2% | -2.1 | 1.7% | 98.3% |
| Mississippi | 67.5% | 32.5% | 4.6 | 100.0% | 0.0% |
| Missouri | 57.7% | 42.3% | 2.0 | 97.7% | 2.3% |
| Montana | 60.5% | 39.5% | 2.7 | 99.7% | 0.3% |
| Nebraska | 51.8% | 48.2% | 0.5 | 68.4% | 31.6% |
| Nevada | 45.8% | 54.2% | -1.1 | 13.5% | 86.5% |
| New Hampshire | 48.9% | 51.1% | -0.3 | 38.5% | 61.5% |
| New Jersey | 45.4% | 54.6% | -1.2 | 11.5% | 88.5% |
| New Mexico | 45.3% | 54.7% | -1.2 | 11.1% | 88.9% |
| New York | 37.2% | 62.8% | -3.3 | 0.0% | 100.0% |
| North Carolina | 54.4% | 45.6% | 1.1 | 87.1% | 12.9% |
| North Dakota | 61.9% | 38.1% | 3.1 | 99.9% | 0.1% |
| Ohio | 55.2% | 44.8% | 1.4 | 91.4% | 8.6% |
| Oklahoma | 63.3% | 36.7% | 3.5 | 100.0% | 0.0% |
| Oregon | 42.3% | 57.7% | -2.0 | 2.2% | 97.8% |
| Pennsylvania | 54.9% | 45.1% | 1.3 | 89.8% | 10.2% |
| Rhode Island | 47.9% | 52.1% | -0.6 | 29.0% | 71.0% |
| South Carolina | 62.0% | 38.0% | 3.1 | 99.9% | 0.1% |
| South Dakota | 59.3% | 40.7% | 2.4 | 99.2% | 0.8% |
| Tennessee | 62.9% | 37.1% | 3.3 | 100.0% | 0.0% |
| Texas | 48.3% | 51.7% | -0.4 | 33.4% | 66.6% |
| Utah | 52.7% | 47.3% | 0.7 | 75.5% | 24.5% |
| Vermont | 44.4% | 55.6% | -1.4 | 7.4% | 92.6% |
| Virginia | 47.7% | 52.3% | -0.6 | 27.6% | 72.4% |
| Washington | 36.4% | 63.6% | -3.5 | 0.0% | 100.0% |
| West Virginia | 66.9% | 33.1% | 4.4 | 100.0% | 0.0% |
| Wisconsin | 49.1% | 50.9% | -0.2 | 40.6% | 59.4% |
| Wyoming | 66.9% | 33.1% | 4.4 | 100.0% | 0.0% |
| Tipping Point State | Projected Trump Vote | Projected Clinton Vote | Trump Z Score | Trump Win State% | Clinton Win State% |
|---|---|---|---|---|---|
| New Hampshire | 48.9% | 51.1% | -0.3 | 38.5% | 61.5% |
Click the map to create your own at 270toWin.com
To view the raw Facebook “Likes” data in visual form, click here, and select Trump and Clinton as the two candidates.
Trump’s Rust Belt Strategy Is Working
While polls have Clinton with a comfortable lead in Pennsylvania and show a tight race in Ohio, the Facebook “Likes” model sees Trump winning each state by roughly 13 points, a huge shift in each state from 2012. The main factor? Trump is uniquely popular in Ohio and Pennsylvania for a Republican nominee. Specifically, Trump is especially well-liked in eastern Ohio and Pennsylvania outside of Philadelphia. To prove this point, if you run the model again with Ted Cruz, a more traditional conservative, as the Republican nominee instead of Trump, Clinton would actually be favored to win both states. Places like Trumbull County, Ohio and Luzerne County, Pennsylvania – counties that voted overwhelmingly Democrat in 2012 – are now Republican strongholds thanks to Trump. There are dozens of counties like those all across Pennsylvania and Ohio.
Pennsylvania and Ohio aren’t the only places that Trump is faring better than Romney did in 2012. The model also has Trump within striking distance in traditionally blue states like Maine, Michigan, Delaware, and Wisconsin. While Clinton does well in the urban parts of these states, such as Detroit, Portland, Wilmington, and Milwaukee, she fares worse than previous Democrats in the rural parts of these states. Indeed, even though Maine is still somewhat competitive as a whole, Clinton is almost guaranteed to lose Maine’s 2nd Congressional District, which encompasses rural northern Maine, even though Obama won the district by more than 8 points in 2012.
What do all of these blue and purple turned red states have in common? A higher percentage of non-hispanic whites and a lower percentage of hispanics compared to the national average, as well as a geographic home in the northeast or midwest.
Clinton Gains on Trump in the Southwest
While Trump has made gains in many traditionally blue states in the northeast and midwest, my model also has him losing ground in Southwestern states like Texas, Arizona, Nevada, and Colorado. While it’s impossible to pinpoint what exactly is troubling Trump in these states, it’s definitely worth noting that Latinos make up at least 20% of the statewide population in all four states.
In Texas, minorities actually make up a majority of the state population, with Hispanics accounting for 38% of the state population, African Americans at 12%, and Asian Americans at 4%. Many speculate that Texas has only remained a reliably Republican state because of a sizable Republican Hispanic population as well as lower turnout among Hispanics in Texas compared to whites.
Donald Trump, however, changes the equation with Hispanics. A recent Latino Decisions poll, known for its excellence in polling Hispanics in both English and Spanish, shows Trump getting only 18% of the Hispanic vote, which would be a record low for a Republican. Republicans have typically gotten between 27% (Mitt Romney in 2012) and 44% (George Bush in 2004) of the Hispanic vote in Presidential elections. It’s also fair to wonder if Trump’s presence at the top of the ticket will motivate Hispanics to turn out at higher rates than in previous elections, deepening the impact of Trump’s anemic numbers with them. According to the same Latino Decisions poll, 48% of Hispanics say they are more enthusiastic about voting in 2016 compared to 2012, compared to only 31% who say the opposite.
Indeed, if a more traditional Republican like Marco Rubio were the Republican nominee, my model projects that Rubio would win Texas somewhat comfortably. With Trump as the nominee, large urban centers like Bexar County (San Antonio) and Harris County (Houston) – counties that voted similarly to the nation as a whole in 2012 – should become major Democratic strongholds in 2016. Hispanics make up 59% and 41% of Bexar and Harris counties, respectively. In addition, Trump takes a major hit in less hispanic, traditionally Republican suburban counties. Places like Denton County and Williamson County – suburban counties surrounding Dallas and Austin, respectively – go from Republican strongholds in 2012 to tossup counties this time around. All in all, it’s enough to push Clinton to being a slight 3 point favorite to win Texas, despite Obama losing Texas by 16 points in 2012.
Trump Knocks Down Democrats’ Mythical “Blue Wall”
After Obama won a dominant electoral college victory in 2012 – despite only winning the popular vote by four points – many pundits speculated (and continue to speculate) that Obama and the Democrats had built a “Blue Wall” of reliably blue states that Republicans couldn’t win – giving the Democrats an edge in the electoral college in the event of a close election in the popular vote. While statisticians like Nate Silver at FiveThirtyEight have tried to dismantle this questionable theory, it has persisted in many political circles.
While the Facebook “Likes” model has Clinton leading Trump by 45 or 26 electoral votes – depending on whether or not you include tossup states – as well as winning the tipping point state of New Hampshire, Clinton’s position in the electoral college is perhaps even more precarious than it initially appears.
If you look only at states where the projected winner has a 70% chance or greater of winning, assuming that the other states are toss-ups, Trump actually leads Clinton by 10 electoral votes – 223 to 213. Additionally, it’s probably fair to question if Clinton really can flip deep red Texas blue. Remember, the Facebook “Likes” data is from way back in April, so perhaps some previously anti-Trump Republicans in Texas have warmed on him since the primaries.
Meanwhile, Trump’s popularity in Ohio, Pennsylvania, and Florida might actually give Trump the electoral college advantage, not Clinton. Either way, Trump’s excellent position in those swing states is sure to make this an extremely close election, even if Clinton’s lead in the popular vote grows over the coming weeks. Who knows, we might even have another 2000 election on our hands.
Lefty 
Jill 
Madeline Anbinder