This story is about our predictive statistics, which we continue to analyze in order to improve our models. We have analyzed them over the last five years of high school varsity football in Illinois, and we hope to develop a new and improved model for the 2013 fall season. If you want to see how our mathematical models predict a certain playoff game might turn out, please visit our football pages by following the link in the blue strip above and then choose “Select Team.”
Illinois varsity football teams played their quarter-final games last weekend, all but one of them on Saturday. Our statistics don’t perform as well during the playoffs as they do during the regular season, first because the number of games is much lower, especially as teams are eliminated, and second because the playoffs tend to bring out the best in teams that had mediocre regular seasons.
In Rounds 1 and 2, for example, both Downers Grove North and Notre Dame High School in Niles were defying all the predictions. Both teams won both their early playoff games, against strong predictions from our algorithms that they would lose. In the quarter-final games, however, both teams lost, as our mathematical models predicted they would.
Out of 32 predictions …
Our computer predicted winners in all 32 games, and we got very mixed results.
| Winner | Loser | Models’ Predictions for Winner |
| Neuqua Valley, 23 | Waubonsie Valley, 20 | -0.96, -0.34 |
| Lemont, 21 | Oak Forest, 14 | 2.11, 1.44 |
| Loyola Academy, 19 | Palatine, 7 | -2.39, 1.99 |
| Glenbard West, 3 | Wheaton North, 0 | 3.47, 4.45 |
| Newman Central Catholic, 28 | Wilmington, 26 | -3.61, -3.05 |
| Evergreen Park, 37 | Brooks, 14 | -4.02, -7.38 |
| Triopia Coop, 22 | LeRoy, 20 | 5.35, 4.64 |
| Stark County, 50 | Forreston, 20 | 6.27, 6.31 |
| Morris, 28 | Washington, 27 | 6.55, 6.02 |
| Lincoln-Way East, 40 | Edwardsville, 10 | 6.71, 6.48 |
| Maroa-Forsyth, 48 | Carrollton, 14 | 6.99, 7.02 |
| Joliet Catholic Academy, 21 | Lincoln-Way West, 20 | -7.48, -4.27 |
| Lake Forest, 31 | Notre Dame (Niles), 19 | 9.03, 9.30 |
| Greenville, 49 | Carterville, 38 | 9.21, 8.92 |
| Glenbard North, 29 | Maine South, 23 | -9.63, -8.47 |
| Harrisburg, 41 | Marquette (Alton), 7 | -11.11, -7.86 |
| Sacred Heart-Griffin, 33 | Glenwood, 21 | 11.76, 9.18 |
| Alleman, 28 | Coal City, 13 | 12.29, 12.50 |
| Lake Zurich, 21 | Boylan Catholic, 6 | -14.44, -9.83 |
| Crete-Monee, 42 | Ottawa Township, 6 | 14.58, 14.93 |
| Cary-Grove, 7 | Crystal Lake Central, 0 | 15.85, 15.47 |
| Clifton Central, 48 | Annawan-Wethersfield Coop, 27 | 16.96, 15.15 |
| Unity (Tolono), 10 | Williamsville, 7 | -19.69, -15.83 |
| Central-Southeastern Coop, 38 | DuQuoin, 25 | 20.43, 18.52 |
| Stockton, 48 | Lena-Winslow, 14 | 20.75, 16.21 |
| Rochester, 46 | Notre Dame (Peoria), 21 | 21.56, 20.19 |
| Althoff Catholic, 19 | Casey-Westfield, 17 | -22.74, -19.78 |
| Aurora Christian, 49 | Winnebago, 14 | 23.43, 20.26 |
| Benet Academy, 26 | Downers Grove North, 24 | 23.62, 21.62 |
| Montini, 42 | Marian (Woodstock), 27 | -27.40, -17.95 |
| Mercer County, 33 | Amboy-LaMoille Coop, 21 | 28.54, 25.83 |
| Mt. Carmel, 45 | Lyons Township, 10 | 33.55, 28.89 |
Colors: Green = winner in close game, red = miss, pink = miss on close call.
A negative prediction indicates that the model predicted the actual loser would win the game.
Of the 64 predictions we made (two different models × 32 games), we got 21 wrong, for a mark of 67 percent correct. That gives us a D, but 12 of the 21 incorrect predictions came in games that ended up being very close. We assume that teams playing in the quarter-finals are good teams, and our predictions tend to fall apart when two very strong teams play each other, as we note in our explanation. That’s because these games sometimes come down to one or two plays, and anything can happen on one or two plays, regardless of how strong the teams are on average.
Let’s recompute: we have nine blatantly incorrect predictions out of 64 total predictions, and then we have 12 incorrect predictions in close games. Of those 12, however, we didn’t predict all of them would be close, so let’s count those games we thought would be comfortable wins for the actual loser a little worse for our average.
- Total predictions: 64
- Blatant incorrect: 9 (14.1%)
- Incorrect in games that were close: 12 (18.8%)
- Incorrect in close games we thought would be close: 6 (9.4%)
- Incorrect in close games we thought would be comfortable: 6 (9.4%)
What we’re going to do to compute our average for the quarter-finals is add all the correct predictions, half the incorrect predictions in close games we thought would be close, and one-eighth of the incorrect predictions in close games we thought would be comfortable. This gives us 43 + (0.5)(6) + (0.125)(6) = 46.75 out of 64 = 73 percent, which is a low C.