It's election morning, 12 more hours of GOTV to go. Work till you drop, then celebrate. I keep a running average of the top seven polling aggregators and election pundits and update it once a day or so. A flood of new polls came out over the last 48 hours. A lot of crappy right wing polls reverting back to reality based numbers in order to get the final polls of the election right. So a lot of polls that had been running with a 6 point Trump bias suddenly got on target. So Hillary "appeared" to shoot up in the polls jumping from a 3.0 point national lead to a 4.3 point lead. State polls were similar. The senate races are down for the last few days and one could argue that the same factor might be at work, that a lot of MSM polls have been putting a thumb on the scale for the Democrats in some races for the last couple of weeks to create a more exciting narrative, and now are reverting to un-skewed numbers. Whatever the reasons, the Senate is now closer to a toss up. With chances running from 50% up to 66% in most estimates. My guess in the Senate is that we end up with 50 seats, and then Susan Collins will switch parties a week or two after the election (how's that for a prediction?).
My data follows the 4-way race of likely voters. Hillary's national lead peaked up to 7.0 very briefly three weeks ago and then started shrinking at a slow but constant rate as almost all establishment Republicans threw integrity to the wind, held their collective noses and decided to vote for Trump. The FBI pissing all over the Hatch Act sped up the slide for a few days, but that appears to have played out by the end of last week. This morning Clinton's national lead is at 4.3%, up from 3.0% on Saturday, and her position in a few toss-up states got a bit better, so her odds of winning have jumped up to 84%. As always, Trump's chances of winning are almost all based on a hope that there is 3 to 5 point systemic error in all, or most, of the polling in his favor. Errors that large have happened in the past but they are unlikely. It is equally likely that such an error would be in Hillary's favor as opposed to in Trump's favor.
Other data:
Total 3rd Party support is at 7.3%, and undecideds are down to 5.1%, both very close to the 7% and 5% that I predicted weeks ago. Undecideds will obviously reach 0% by the time the polls close, but I don't expect them to break dramatically for one candidate or another. Most of them will probably not vote and the rest will split between Hillary, Trump and the third party candidates and write-ins, in unsurprising proportions (hence, little or no effect). Trump has picked up most of the voters fleeing Johnson in the last several weeks as establishment Republican resolve to not vote for a violent, racist, misogynistic, authoritarian con man, has proven to be as rock solid as all of their other moral and ethically grounded standards of behavior and belief.
Here are my projection for the final popular vote results:
Clinton = 48.77%
Trump = 44.35%
Johnson = 4.47%
Stein = 2.01%
McMullin = 0.29%
All others = 0.11%
The model shows that FL, NV, NC, OH, NH, and IA are all toss-ups right now, within 2 points. The model says Hillary is more likely that not to loose OH and IA, but I think her ground game in those states will be worth 2 points so I think she'll carry them both, but both may end up being close enough for lengthy recounts. The large Hispanic vote will push her over the totals shown by the model in FL, AZ and NV, by a point or two, but that will still be a tiny bit short in AZ. So with my "gut" feelings, unsubstantiated by hard data, added into the mix, my numbers look like this (I fudged the 2012 historic factor in five states to factor in my unscientific guess work):
That's it. Those are my predications for this election. 347 EV's for Hillary and 191 for Trump and four states potentially close enough for recounts. In map form:
About the Spreadsheet Simulator:
It's a Spreadsheet that uses iterative summations, and the NORMSDIST() and RAND() functions to do U.S. Presidential election simulations based on state polling data. If anyone wants a copy of it, feel free to let me know (ken@indianoakstables.com) and I'll email it to anybody that asks. I've had a lot of fun and success with it for the last 8 years. I average state and national polling data and expert opinion from a number of sources in every state and turn it all into 53 percentage chances of a Democratic victory in the upcoming General Election in each state, D.C., and two odd congressional districts in Maine and Nebraska. Then randomly simulate several thousand elections and come up with a percentage chance of winning 270 or more Electoral College Votes in November.
I also include in the simulation a "systemic error" in all polling that will raise or lower all of the polling in every state by a random amount. This takes into account the possibility of things like: Trump's supporters not fitting typical polling categories and being dramatically under-sampled by typical polling techniques; and/or Hillary's ground game being so overwhelmingly better than Trump's that her actual numbers on election day will out perform her polling numbers. The most likely outcome of this factor is a zero or near-zero bump in the poll numbers one way of another, with a bump of 1, 2 or 3 points being reasonably possible, and bumps of 7, 8 or even 9 points being possible, but highly unlikely. I establish the value spread of the systematic error empirically, using data from many elections in the past 20 years.
I don't get too fancy and try to add in a ton of factors for how far we are from the election, demographic changes from the last election, second guessing the pollsters, etc. As such, this calculation does not represent Hillary's chance of winning in November, it represents her chances of winning if the election were held tomorrow (assuming it has the most current data), or, more-accurately, this would be her chance of winning if the poll numbers on November 7th were the same as these). In that sense, it is similar to the 538 "now-cast" using polling data only. Thus, it may or may not be very accurate right now, but it will inevitably become more and more accurate the closer we get to election day. There are a number of factors that converge to increase it's accuracy as we approach election day. As examples: each day there is less time for people to change their minds, the number of undecided voters will continue to decrease, there is less time for new developments and news items to influence the opinions of voters, the number and accuracy of the publicly available polling data will continue to increase, and early voting will begin to lock in votes even before we reach election day. So, by the time we reach election eve, you can be pretty sure this sort of predictor will be about as accurate as possible.
Notes on the spreadsheet:
- Points shown are percentage points difference in expected vote share, or "point spreads".
- A 10 point spread might indicated an expected outcome of 50% for Clinton, 40% for Trump and 10% for "Others", or 46% Clinton, 36% Trump, etc.
- Points are positive for a Dem. lead and negative for a Rep. lead.
- Wherever possible, I use data from polls showing "likely voters" and a 4 way race including Clinton, Trump, Johnson and Stein.
- "EP" is data from www.electionprojection.com/...
- "538" is data from projects.fivethirtyeight.com/...
- "2012 +X.X" are actual results from the 2012 Presidential election adjusted to eliminate Candidate or Party bias in the Nationwide vote. Each State's vote spread has been adjusted by equal percentages to reflect an even 50/50 national vote in 2012, then the current 2016 nation-wide average poll spread is added to each state's normalized spread.
- "270" is data from www.270towin.com/...
- "NYT" is data from www.nytimes.com/...
- "HP" is data from elections.huffingtonpost.com/...
- "RCP" is data from www.realclearpolitics.com/...
- "EV" is data from www.electoral-vote.com
- "AVG" is an average of the estimated point spreads of the 8 data sources.
- The nationwide popular poll spread used to adjust the 2012 results is an average of the national poll spread from the same seven data sources above.
- I normalize the 2012 election results for HI and UT equally to try to factor out the "home-team" effect of them being candidate's home states.
- Blank spaces in the data matrix indicates there is no available data (usually in states with little or no polling).
- Point spreads are converted into percentage chances of electoral victory in a given state using a Cumulative Distribution Function.
- Percentage Chances of electoral outcome are all expressed as chances for a Dem victory.
- When I post a copy of this spreadsheet the data is as up-to date as possible, but some of the data sources update their numbers several times a day, so it is possible that some of the numbers in the data input area may be out-of date in the 30 minutes it takes for me to post it.