Super Tuesday Prediction Time!

Vote buttons

Phil Burch Phil Burch, Former Contributor

Since we last tried to analyze social data to predict the New Hampshire Primary, Donald Trump and Hillary Clinton have both begun to cement their status as front runners in their respective races. While both candidates have question marks in advance of the general election, Super Tuesday will go a long way to determine if they can hold off their rivals and win the nomination.

Benchmark Performance:

Let’s take a look at how well our predictions actually went in New Hampshire:

Republican Candidates Trump Cruz Rubio Bush Kasich
# of Delegates Expected by Polling % 7.061 2.852 3.312 2.599 2.99
# of Delegates Expected by Twitter Volume 8.632 8.931 2.848 1.544 1.045
# of Expected Delegates by Adjusted Facebook Volume 13.032 5.145 2.558 2.263 0
Actual Delegate Count 11 3 2 3 4

 

Democratic Candidates Sanders Clinton
# of Delegates Expected by Polling % 12.397 9.361
# of Delegates Expected by Twitter Volume 11.764 11.236
# of Delegates Expected by Adjusted Facebook Volume 12.607 10.393
Actual Delegate Count 15 9

Ultimately, the initial prediction was fairly accurate (with exception to Kasich’s mentions being redacted), but I do think there is room for improvement.  

 

Review of Methodology

After taking a look at my previous methodology for Facebook data, it was clear that I could optimize my prediction by tweaking the numbers further. I realized that instead of just minimizing all Facebook volume in my previous post, what I need to do is discount younger Facebook traffic while maximizing older Facebook traffic.

The reason for this becomes clear when you begin to look to historic voter turnout and apply that concept to our numbers. Here is a look at Facebook usage by age bracket:

  • 18-29 year olds: 87%
  • 30-49 year olds: 73%
  • 50-64 year olds: 63%
  • 65+ year olds: 56%

Now for added context, let’s examine historical voter turnout from the previous Presidential election in 2012:

  • 18-29 year olds: 45.0%
  • 35-44 year olds: 59.5%
  • 45-64 year olds: 67.9%
  • 65+ year olds: 72.0%

When adding these supplemental data points, it becomes clear that not all Facebook mentions should be considered fully equivalent. In order to further enhance our analysis, we should consider this additional data in order to provide more context to our prediction. These additional data points could get us much closer to a predictive measurement.

Optimization

Because my ability to create a weighted average of Facebook mentions has greatly diminished since high school, I sought out my Uncle Ray to get his opinion. I figured this was a worthy exercise considering he once taught Advanced Calculus for fun(!).

By taking the data above, he advised me to come up with a “skew” factor which allows us to discount younger Facebook mentions based on the historical voter turnout. Here are the skew numbers using his methodology:

Age Turnout Using Facebook Frequent Facebook Discounted Turnout Weight Factor Turnout Skew
18-29 0.45 0.87 0.644 0.2898 1.000 0.290
30-44 0.595 0.73 0.767 0.456365 1.322 0.603
45-64 0.679 0.63 0.889 0.603631 1.509 0.911
65 0.72 0.56 1 0.72 1.600 1.152

For each age group’s volume, we will multiply our results by the Skew factor in order to maximize Facebook content from older age brackets, and minimize for younger volume.

 

Prediction Time!

 

Republican Races:

 

Arkansas

(40 Delegates)

Trump Cruz Rubio Kasich
# of Delegates Expected by Polling % 9.2 10.8 9.2 1.6
# of Delegates Expected by Twitter Volume 19.2 13.6 6.4 0.8
# of Expected Delegates by Adjusted Facebook Volume 22.8 10.7 6.5 0.0

 

Alabama

(50 Delegates)

Trump Cruz Rubio Kasich
# of Delegates Expected by Polling % 19.0 7.4 10.2 3.4
# of Delegates Expected by Twitter Volume 24.0 16.3 7.5 1.0
# of Expected Delegates by Adjusted Facebook Volume 28.6 12.8 8.0 0.7

 

Tennessee

(58 Delegates)

Trump Cruz Rubio Kasich
# of Delegates Expected by Polling % 23.2 12.8 11.0 3.5
# of Delegates Expected by Twitter Volume 27.8 19.1 9.3 1.7
# of Expected Delegates by Adjusted Facebook Volume 32.2 16.0 9.0 0.8

 

Vermont

(16 Delegates)

Trump Cruz Rubio Kasich
# of Delegates Expected by Polling % 5.1 1.8 2.7 1.6
# of Delegates Expected by Twitter Volume 8.6 4.2 2.4 0.8
# of Expected Delegates by Adjusted Facebook Volume 9.9 3.9 2.2 0.0

 

Minnesota

(38 Delegates)

Trump Cruz Rubio Kasich
# of Delegates Expected by Polling % 6.8 8.0 8.7 .8
# of Delegates Expected by Twitter Volume 19.4 11.8 6.1 1.1
# of Expected Delegates by Adjusted Facebook Volume 22.0 10.6 5.4 0.0

 

Colorado

(37 Delegates)

Trump Cruz Rubio Kasich
# of Delegates Expected by Polling % No Data No Data No Data No Data
# of Delegates Expected by Twitter Volume 18.5 12.2 5.2 1.1
# of Expected Delegates by Adjusted Facebook Volume 21.1 10.1 5.3 0.4

 

Texas

(155 Delegates)

Trump Cruz Rubio Kasich
# of Delegates Expected by Polling % 42.9 56.7 27.3 10.7
# of Delegates Expected by Twitter Volume 71.3 60.5 20.2 3.1
# of Expected Delegates by Adjusted Facebook Volume 80.9 48.7 23.3 2.1

 

Georgia

(76 Delegates)

Trump Cruz Rubio Kasich
# of Delegates Expected by Polling % 28.2 16.6 16.6 5.6
# of Delegates Expected by Twitter Volume 38.0 25.1 11.4 2.3
# of Expected Delegates by Adjusted Facebook Volume 43.0 19.5 12.4 1.1

 

Massachusetts (42 Delegates) Trump Cruz Rubio Kasich
# of Delegates Expected by Polling % 19.0 4.6 7.8 6.4
# of Delegates Expected by Twitter Volume 21.4 12.6 6.3 1.3
# of Expected Delegates by Adjusted Facebook Volume 25.4 10.0 6.0 0.6

 

Alaska

(42 Delegates)

Trump Cruz Rubio Kasich
# of Delegates Expected by Polling % 7.8 6.7 2.0 .6
# of Delegates Expected by Twitter Volume 23.1 12.6 5.0 0.8
# of Expected Delegates by Adjusted Facebook Volume 25.3 11.9 4.9 0.0

 

Democrats:

Massachusetts

(116 Delegates)

Sanders Clinton
# of Delegates Expected by Polling % 51.9 55.7
# of Delegates Expected by Twitter Volume 88.2 27.8
# of Expected Delegates by Adjusted Facebook Volume 64.0 52.0

 

Texas

(252 Delegates)

Sanders Clinton
# of Delegates Expected by Polling % 80.6 157.0
# of Delegates Expected by Twitter Volume 178.9 73.1
# of Expected Delegates by Adjusted Facebook Volume 123.8 128.2

 

Georgia

(116 Delegates)

Sanders Clinton
# of Delegates Expected by Polling % 32.8 73.4
# of Delegates Expected by Twitter Volume 77.7 38.3
# of Expected Delegates by Adjusted Facebook Volume 56.7 59.3

 

Vermont

(26 Delegates)

Sanders Clinton
# of Delegates Expected by Polling % 22.0 2.5
# of Delegates Expected by Twitter Volume 22.9 3.1
# of Expected Delegates by Adjusted Facebook Volume 17.1 8.9

 

Tennessee

(76 Delegates)

Sanders Clinton
# of Delegates Expected by Polling % 25.1 44.8
# of Delegates Expected by Twitter Volume 55.5 20.5
# of Expected Delegates by Adjusted Facebook Volume 36.7 39.3

 

Arkansas

(37 Delegates)

Sanders Clinton
# of Delegates Expected by Polling % 10.5 21.1
# of Delegates Expected by Twitter Volume 24.0 13.0
# of Expected Delegates by Adjusted Facebook Volume 17.4 19.6

 

Oklahoma

(42 Delegates)

Sanders Clinton
# of Delegates Expected by Polling % 16.8 18.6
# of Delegates Expected by Twitter Volume 30.2 11.8
# of Expected Delegates by Adjusted Facebook Volume 21.3 20.7

 

Minnesota

(93 Delegates)

Sanders Clinton
# of Delegates Expected by Polling % 23.3 54.9
# of Delegates Expected by Twitter Volume 68.8 24.2
# of Expected Delegates by Adjusted Facebook Volume 50.8 42.1

 

Colorado

(79 Delegates)

Sanders Clinton
# of Delegates Expected by Polling % 21.3 43.5
# of Delegates Expected by Twitter Volume 58.5 20.5
# of Expected Delegates by Adjusted Facebook Volume 44.4 34.6

 

Alabama

(60 Delegates)

Sanders Clinton
# of Delegates Expected by Polling % 16.2 39.0
# of Delegates Expected by Twitter Volume 39.0 21.0
# of Expected Delegates by Adjusted Facebook Volume 23.5 36.5