2016 General: The Model is Here

Hillary Clinton is favored to win tomorrow’s election since the beginning of the 2016 cycle. Since after the primaries, Donald Trump has only lead five non-Rasmussen or Ipsos, non-tracking national four-way polls. This possibly points to an overwhelming Democratic win tomorrow, but our model, largely based on state-level polling averages, suggests only a modest Clinton win.


The model has a simple methodology: combining all polls published on Real Clear Politics conducted within 2 weeks (for the same pollster, only their three most recent surveys are included; RCP is not used because of its extremely narrow inclusion guidelines) and adjust their two-party numbers by their historical partisan leanings (provided at FiveThirtyEight). If such number is not provided, the poll’s self-reported partisan registration number is compared to statewide partisan registration numbers * 0.75 to “unskew” the poll (factored at 0.75 considering LV/RV gap). If partisan registration numbers were not provided in a poll, such poll is discounted from the model. On top of this first adjustment, average two-party polling error in a specific state is adjusted as well, according to numbers published by Real Clear Politics for the 2008 and 2012 elections.



Two party average vote is calculated using equation 1 (P denotes party, D for Democratic, R for Republican)



Systematic polling average error is adjusted by equation 2: (TPA denotes true polling average)



State AvgPE
Arizona 3.23
Colorado -3.64
Florida -1.68
Georgia -0.72
Iowa 1.7
Maine -2.85
Michigan -3.68
Nevada -4.9
New Hampshire -1.1
North Carolina -0.9
Ohio -1.03
Pennsylvania -2.1
Virginia -2.7
Wisconsin -2.58

Table 1: Average polling errors in several states


Table 1 provides us how off polls were in these battleground states. A positive number indicates poll in favor of Democratic candidate; a negative number indicates poll in favor of the Republican candidate.


From table 1, we developed table 2 (which is not shown here due to its massive size). Table 2, containing our full methodology  can be downloaded in the link below. Due to time constraint (it’s 1AM Pacific on November 8), I cannot to explain my full methodology here.


In fact, my prediction model is fundamentally flawed because of its reliance on RealClearPolitics data. If my model included more polls, it would be more predictive. My model would be most predictive in Florida, Pennsylvania and Virginia because of these states’ abundance of polls, and less predictive in Wisconsin, Ohio and Maine, where polls are not as abundant.

Here are my numerical predictions for each battleground states:

AZ: 18% D, 82% R (HRC 46.55 DJT 53.44)

CO: 86% D, 14% R (HRC 54.05 DJT 45.94)

FL: 64% D, 36% R (HRC 51.29 DJT 48.71)

GA: 32% D, 68% R (HRC 48.09 DJT 51.91)

IA: 31% D, 69% R (HRC 48.15 DJT 51.85)

ME: 88% D, 12% R (HRC 54.19 DJT 45.81)

MI: 91% D, 9% R (HRC 54.77 DJT 45.23)

NV: 69% D, 31% R (HRC 51.93 DJT 48.07)

NH: 58% D, 42% R (HRC 50.77 DJT 49.23)

NC: 52% D, 48% R (HRC 50.07 DJT 49.93) — Pure Tossup

OH: 29% D, 71% R (HRC 47.94 DJT 52.06)

PA: 76% D, 24% R (HRC 53.29 DJT 46.71)

VA: 90% D, 10% R (HRC 54.20 DJT 45.80)

WI: 86% D, 14% R (HRC 54.51 DJT 45.49)

Raw Data

My electoral college ratings:


%e5%be%ae%e4%bf%a1%e6%88%aa%e5%9b%be_20161108005714 %e5%be%ae%e4%bf%a1%e6%88%aa%e5%9b%be_20161108005837


Enjoy today and live with whoever we get tomorrow for four years!

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes:

<a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>