Updates / Notes

No updates other than an unprecedented week of failure by the model. This was particularly shocking as, by this point in the season, I expected teams to have statistics that would serve as strong indicators of their future success (given the number of conference games played at this point in the year)¹. This sort of result is something I would have expected much earlier in the year, but to see it in week 10 is shocking.

Week 10 Statistics

Home/Away Model:

1. Base Version Record: 30-21
2. Base Version Accuracy: 58.82%
   –
3. Slimmed Version Record: 30-21
4. Slimmed Version Accuracy: 58.82%
   –
5. Agree-Disagree Between Versions: 49-2
6. Disagreement Percentage: 3.92%

Neutral Site Model:

1. Base Version Record: 1-0
2. Base Version Accuracy: 100.00%
   –
3. Slimmed Version Record: 1-0
4. Slimmed Version Accuracy: 100.00%
   –
5. Agree-Disagree Between Versions: 1-0
6. Disagreement Percentage: 100.00%

Combined:

1. Base Version Record: 31-21
2. Base Version Accuracy: 59.62%
   –
3. Slimmed Version Record: 31-21
4. Slimmed Version Accuracy: 59.62%
   –
5. Agree-Disagree Between Versions: 50-2
6. Disagreement Percentage: 3.85%

Notable Predictions

#20 Texas over #9 Vanderbilt (78.76% base, 78.38% slimmed)

Most Unlikely Outcomes

West Virginia over Houston (6.36% base, 4.79% slimmed)
Kentucky over Auburn (8.72% base, 9.78% slimmed)
Oregon State over Washington State (17.65% base, 18.54% slimmed)

Biggest Nailbiters

#21 Michigan over Purdue; 21-16 (Michigan Win: 96.51% base, 96.11% slimmed)

Running Totals

Home/Away Model:

1. Base Version Record: 219-94
2. Base Version Accuracy: 69.97%
   –
3. Slimmed Version Record: 216-98
4. Slimmed Version Accuracy: 68.79%
   –
5. Agree-Disagree Between Versions: 301-12
6. Disagreement Percentage: 3.83%

Neutral Site Model:

1. Base Version Record: 3-1
2. Base Version Accuracy: 75.00%
   –
3. Slimmed Version Record: 4-0
4. Slimmed Version Accuracy: 100.00%
   –
5. Agree-Disagree Between Versions:3-1
6. Disagreement Percentage: 75.00%

Combined:

1. Base Version Record: 222-95
2. Base Version Accuracy: 70.03%
   –
3. Slimmed Version Record: 220-98
4. Slimmed Version Accuracy: 69.18%
   –
5. Agree-Disagree Between Versions: 304-13
6. Disagreement Percentage: 4.10%

Full Breakdown of Results

Given that the Personal Plan in WordPress doesn’t allow for plugins to be installed, I can’t embed data tables into these pages without running into a ton of formatting issues.

To see the full breakdown of results, please follow the Google Drive link to the Excel document below:

Full Results

1. The model was trained on season-end statistics. For more information/context, see the Constraints section of this article. ↩︎