The first part of this is how we define "unpredictable". While it might be difficult to predict a single game correctly, given a large enough sample size and an accurate enough predictor we should be able to know how many we should get right (see: expected value). As a simple example, think of flipping a coin and trying to predict its behavior. If we always guess "heads" then we'll be wrong roughly half the time. But if I tell you I'm going to flip a coin 1,000 times and I want you to tell me how many heads will come up, 500 would be a good guess. If the coin is fair the answer should be somewhere in the neighborhood of 500.
Similarly, if over time our college football predictor is correct 75% of the time and I tell you there are 400 ACC conference games in our set, then we should be able to get roughly 300 correct. There might be stretches of bad luck where the predictor is wrong, and there will be some big upsets. But over time we should get nearly 75% correct.
With that in mind, let's see how our predictor holds up for each of the "Big Six" conferences. For each conference we'll look at the
- total number of games,
- number we expected to get right,
- number we actually got right; and
- ratio of actual to expected.
| Conference | Games Played | Predictions | ||||
|---|---|---|---|---|---|---|
| # Expected correct | % Expected correct | # Actually correct | % Actually correct | Actual / Expected | ||
| ACC | 383 | 274.8 | 71.8 | 255 | 66.6 | 0.928 |
| Big 10 | 352 | 255.8 | 72.7 | 262 | 74.4 | 1.024 |
| Big XII | 392 | 287.1 | 73.3 | 294 | 75.0 | 1.024 |
| Big East | 195 | 135.1 | 69.3 | 136 | 69.7 | 1.007 |
| Pac-10 | 345 | 253.7 | 73.5 | 253 | 73.3 | 0.997 |
| SEC | 392 | 290.2 | 74.0 | 288 | 73.5 | 0.992 |
Over the course of eight years the ACC is (a) the second-most-difficult to predict, and (b) even then we don't get close to the expected value. For most conferences the predictions are within 2-3% of what we expect to get correct; even the Big 10 and Big XII are within 10, but for those we're under-estimating the number we should get correct.
Let's put this in perspective: the Big XII has 49 conference games per year (48 regular season plus the championship game) and the predictions miss about 2.4% of the time. That means every 6 years it gets a total of 7 games wrong. The ACC, on the other hand, has the same number of regular season games but hits the 7-wrong-games mark in less than two seasons. This 7-game deficit is compounded by the fact that the computer is already less confident about ACC games to start with. The raw data tells that tale well enough; the computer gets 3 out of every 4 Big XII games correct, while it struggles to get 2 out of every 3 ACC games correct.
Why does this happen? We're not entirely sure. The ACC tends to chew through coaches somewhat regularly, so a lack of consistency may lead to fewer predictable results. Similarly the ACC is offensively challenged, so a favored team falling behind may have difficulty catching up. Small, improbable events such as pick-6s, interceptions, fumbles, and other odd plays could have a greater-than-usual effect on the result.
Until we figure it out, we'll just have to rely on the ACC Wheel of Destiny.
