Sunday, September 11, 2016

Clinton vs Trump.
11 September 2016,
58 days from the election

OMG, it really is less than two months until the election. It's feels like it's been an eternity and it might never end.

First things first. It was a good week for Trump and he now has his best numbers since July. Those numbers aren't good by any rational standard, but he currently has the momentum.

And now the numbers, state by state, on in the case of Maine, district by district. (I'm still waiting for such precise polling from Nebraska, which also splits its electoral votes by district.)

Solid Trump (more than 95% Confidence of Victory [CoV]): ND WY AL ID TN LA MT SD WV MS OK TX KY UT AK IN KS MO SC AR  
Total: 159

Leaning Trump (between 50% and 95% CoV for Trump): NE GA ME-2 AZ IA   
Total: 39

Toss-up: FL
Total: 29
Leaning Clinton (between 50% and 95% for Clinton): OH NC NV CO WI RI MI NJ PA NH
Total: 116

Solid Clinton (more than 95% CoV for Clinton): ME VA MN CT OR DE NM WA IL DC ME-1  MA MD NY VT HI CA
Total: 195

The current count 



Grown-ups will be made nervous by Florida being a toss-up and Ohio being the weakest leaning Clinton state. All I can say to calm those worries is that right now, Trump has to run the table of all the states where he currently leads, and then win in Florida, Ohio, North Carolina, Nevada and Colorado.

Mathematically, that's hard to do, which is why my system now counts him as being far behind. I personally would hate to see him win Florida and Ohio, because it would give his candidacy
legitimacy it does not deserve, but winning those two states alone won't make him president.

A change this week. I not only list the 15 battleground states, but also the probability Clinton will win, the probability Trump will win and the number of electors, which in statistical terms are p, q and n. I also give Trump's count of electoral votes that are not in the battleground mix. This gives any keen nerd a chance to check my work.


Trump's non-battleground count of electors: 160

The battleground states ranked by pqn
(Clinton%, Trump%, electors)

1. FL: 50%, 50%, 29
2. OH: 54%, 46%, 18
3. NC: 55%, 45%, 15
4. AZ: 34%, 66%, 11
5. GA: 11%, 89%, 16
6. CO: 79%, 21%, 9
7. IA: 39%, 61%, 6
8. NV: 66%, 34%, 6
9. WI: 87%, 13%, 10
10. PA: 94%, 6%, 20
11. MI: 94%, 6%, 16
12. NJ: 94%, 6%, 14
13. NE: 10%, 90%, 5
14. RI: 91%, 9%, 4
15. VA: 98%, 2%, 13

Current probability of victory if the election were held today:

Clinton 96.5%
Trump 3.5%

I set up my system to have 15 battleground states, but if I were using my own judgment instead of an algorithm, my current view of the true battlegrounds would be Florida, Ohio, North Carolina, Arizona, Colorado, Iowa and Nevada. If Georgia becomes closer for Clinton or Wisconsin closer for Trump, they would be on my true list as well. Using the shorter list, Trump's chances improve to 94.6% to 6.4%, still much longer odds than what Nate Silver is broadcasting to his much larger audience.

Fun stuff for data nerds only.
1. As I said earlier, I'm waiting for more info on the district by district numbers in Nebraska. Often, the Omaha district goes Democratic and the rest of the state goes Republican strong enough to overwhelm the "big city".

2. There are now three companies polling at least 40 states each, Ipsos, Google Consumer Surveys and SurveyMonkey.  Ipsos posts every week, but I decided my system wouldn't let any one company have more than one poll in the mix at any time, so only their freshest poll gets counted. Ipsos and Google have the unfortunate tendency to take very small polls in low population contests, which messes up my Confidence of Victory (CoV) method. For example, a 5 point lead in a sample of 500 voters give a CoV in the 90% range. If the lead is 5 points but the sample size is 200, that shrinks to 80%. Due to these discrepancies, I have made the arbitrary decision that the lowest acceptable data set size is 256, which is two to the eighth. I'm an OP (Original Programmer), and that's how I roll. 

3. The three big companies don't show a major bias in either direction, sometimes showing Trump closer than you might expect, sometimes doing the same for Clinton. On the other hand, there are pollsters at the state level, notably the newcomer Emerson and the once reliable Quinnipiac, who now skews Republican by assuming white people always vote and non-whites not so much. I take them at their word, reporting the numbers they publish, figuring the market will correct for them over time. Nate Silver's 538 has decided instead to "adjust" numbers for pollsters they don't trust completely. I am morally opposed to doing this. I can read cross-tabs as well as anyone who understands addition, subtraction, multiplication and division, but I don't adjust the numbers a company produces as long as the sample sizes are large enough.

In any case, I'll be back next weekend with more info on the Senate races and the electoral college. Thanks for stopping by.


  1. I come every week to read your number Prof. very good work, and thanks for explaining your work. Good job Prof.

    1. Thanks, Luis. I can check the number to see how many have visited, but I have no way of knowing the number of actual people vs. the number of bots. Comments let me know folks are reading.

      Let your friends know as well.