 |
 |
It’s been some years since you’ve seen one of these election analyses here at HA. Given that Biden and Trump will almost certainly be the nominees of their parties, and given that there are more than 300 state head-to-head polls released, it is time to start these up again.
These analyses use state head-to-head polls and mimic the electoral college process as well as following the rules of each state on how electors are awarded in the state. Most states (and D.C.) use the rule “winner takes all,” but Maine and Nebraska allocates two electors to the state victor and the rest go by the vote of each congressional district.
There are a few states that have no polling yet. These are D.C. (not a state, but it gets three electors anyway), Hawaii, Idaho, Illinois, Louisiana, Nebraska, Oregon, Rhode Island, South Dakota, Vermont, and Wyoming. Clearly, these are not swing states. I’ve simply averaged the result from the past 4 elections and always award electors from the state that way. The results are exactly what you would expect–Illinois goes to Biden, South Dakota goes to Trump, etc. Interestingly, Nebraska’s 2nd CD just barely went for Obama in 2008, but NE-2 will be assigned to Trump in these analyses until we get polling that allows the state to be rolled into the simulations.
New polls come out almost daily, so I’ll post new analyses when I have time to collect polls, conduct the analyses, and post results.
So, how do these analyses work? I am using a statistical method called Monte Carlo analysis to, essentially, conduct mock elections in each state. The outcome of each state election is a probabilistic outcome that reflects the number of people polled recently in the state and the fraction of polled individuals saying they would vote for each candidate. Currently, I consider the last two months of polling as “current”, but that window will shrink as the pace of polling picks up. I usually do 100,000 mock elections to simulate what would happen if the election was held now. The FAQ has the details, although I need to update the FAQ for this election cycle.
Note that I do not claim these analyses predict the election outcome. Rather they reflect what we would expect for an election held now. It serves as something like the score in a basketball game. The half-time score only tells you who is leading. A lopsided score may give you an inkling of who will win, but it is just the half-time score. Watching a basketball game is far more interesting when you are allowed to know the score. Likewise, knowing the “score” in an election contest is much more interesting. In developing these analyses, I have tried to be scrupulously unbiased. If all the polls were to be flipped, Trump’s results would look like Biden’s, and Biden’s Trump’s.
Okay, so what would the outcome be if the election was held today? After 100,000 simulated elections, Biden wins every simulated election. Biden received (on average) 359 electoral votes to Trump’s 179 electoral votes. In an election held now, Biden would have a greater than 99.9% probability of winning and Trump would have a less than 0.01% probability of winning.
The long term trends in this race can be seen from a series of elections simulated every seven days using polls from 06-May-2019 to 06-May-2020, and including polls from the preceding two months. When the purple line is above the dashed line, it means Biden is expected to win. The red and green lines provide “confidence intervals”. For example, at each time point, there was a 95% chance that any simulated election fell between the green lines. If the lower green line dips down to touch the dashed line, it means Trump would have a 2.5% chance of winning.
Here is a plot showing the distribution of electoral votes [FAQ] that resulted from the 100,000 simulated elections for the analysis done today (essentially, a vertical slice from the previous graph at today’s date):
[There is much more, including state-by-state analyses below the fold….]
