By J. F. Kolacinski, Elmira College
You might recall that I published a few of these diaries over the summer, but Secretary Clinton's dominance in the polls made this exercise pretty uninteresting and that remained the status quo with a brief exception since the conventions. But the polls began tightening significantly in the last couple of weeks so this kind of analysis might have something more interesting to tell us. The last time I ran a simulation we had Clinton at 46.4% and Trump at 39.6% nationally. The national polling average is 46.6% for Secretary Clinton and 44.8% for Mr. Trump.
Data: When we went through this exercise last summer we were using a combination of the Real Clear Politics Polling Average and some historical data to estimate the polls for each state. I've continued to use the RCP Average where it’s available but I was surprised at the number of states that have little or no data on RCP. If I repeat this process in 2020, I'm likely to switch to some other site to find my polling data.
When the RCP average didn't appear to be available, I used the "Projected vote share over time" from fivethirtyeight.com to estimate the polls for a state. Specifically, I've used the 538 estimate for Alabama, Alaska, Connecticut, DC, Hawaii, Idaho, Kansas, Kentucky, Mississippi, Montana, Nebraska, New York, North Dakota, Oklahoma, Rhode Island, South Carolina, South Dakota, Tennessee, Vermont, West Virginia and Wyoming.
Poll data was adjusted to eliminate the possibility that a third-party candidate could win the race using the following formula:
% for Candidate A/(% for Candidate A + % for Candidate B).
The Simulator:
As before, we estimate our probabilities by simulating a number of state-by-state national elections using the “U.S. Presidential Election Calculator” which can be found here: < www.maplesoft.com/...>. This application runs on Maple, a professional-grade mathematics program and you can read more about how it works there.
Some items are worth pointing out.
There are known relationships between how the polls change in different states although the simulator assumes the states are operating independently. That will introduce some inaccuracy into the results but, I suspect the effect is small. Those interdependencies should be accounted for at least partially in the poll numbers themselves.
Additionally, the simulator does not treat the congressional districts in Nebraska and Maine separately. This wasn't an issue in 2012, but there's a real chance that the Electoral Votes in Maine could split this time around.
The Map:
The colors on the map are determined by The polling estimate rather than by the probability that a candidate will win a state. If a candidate is polling at 55%, the probability that he or she will win he state is actually quite a bit better than that.
In a state, based on the polling estimate, the conditional probability that an individual voter chooses Clinton assuming he or she chooses either Clinton or Trump is given by:
P(Clinton|Clinton or Trump) = % for Clinton/(% for Clinton + % for Trump)
The colors on the map are determined as follows:
Solidly Democratic (Dark Blue) 100% > P(Clinton|Clinton or Trump) > 55%
Weakly Democratic (Light Blue) 55% > P(Clinton|Clinton or Trump) > 52%
Toss-Up (Gray) 52% > P(Clinton|Clinton or Trump) > 48%
Weakly Republican (Light Red) 52% < P(Trump|Clinton or Trump) < 55%
Solidly Republican (Dark Red) 55% < P(Trump |Clinton or Trump) < 100%
Results:
One thousand simulated elections were run.
Clinton won 87.3% of the time, while Trump won 6.3% of the time and 6.4% of the simulations were ties. It’s reasonable to expect that some of the ties would be Trump victories if the 2nd Congressional District in Maine were treated separately.
The mean electoral result was:
Clinton: 296.7 EV, Trump: 241.3 EV.