Last year marked the first attempt to try predicting the winners of the games all the way through the championship game just by using a team’s season statistics. The formula used takes a look at 11 different statistical categories and weights each category in relation to every other category by how well that statistical category correlates with tournament success in previous year’s championships. Among the categories we find that field goal percentage, free throws made per game, and steals per game seem to be key indicators of a team’s success, however they are not the only categories looked at.
So how well did we do last year? Unfortunately not as well as we thought we would. After testing the formula with data from the 2008-2014 tournaments we had a rate of accuracy of just under 69% including picking 3 of 7 championship winners. Our control model of picking the higher seed in every instance scored a 72% accuracy rate, just a tad bit better.
The 2015 tournament model didn’t come close to living up to the expectations of the success of how well the model did in testing at least in terms of overall correct numbers. For the whole tournament the model correctly picked just 48% of the games, over 20% worse than in testing. The model did correctly have Morningside playing in the championship game, but the team picked to win, Davenport, ended up losing in the second round. In defense of the model, Davenport was a #1 seed that had gone 31-0 entering the tournament, so that one upset which most people probably wouldn’t have expected either ended up hurting the numbers quite a bit.
There were some other areas the model could have done better in as well, most notably in the 3/6 seed matchups. This was something that had also popped up in testing though so it wasn’t necessarily as surprising of an outcome.
Part of the concern of an off year can be put into some perspective however. The control of just selecting the higher seeded team also suffered a worse year than usual, fueled in part by early exits by the prior mentioned Davenport as well as first round losses by #2 seeds Marian and Indiana Wesleyan. We can also argue that last year’s outcomes may have been an outlier and that this year we could return to a success rate similar to the test rate.
So now let’s take a look at this year’s bracket prediction and break it down a bit. The model likes Morningside to repeat as this year’s national champions. Not a bad pick considering that the Mustangs bring back their starting lineup from last year’s title game and have lost just a pair of games this season. They won the GPAC tournament, something that we also have said is a decent predictor of success in winning the NAIA Championship.
On the other side of the bracket, the model likes Southern Oregon to make it to the championship game. Also not a terrible pick since the Raiders enter the tournament with a 29-1 record and rank in the top 10 in several statistical categories. However, the Raiders fell to Cinderella team Tabor last year in the second round in an early morning game for anyone, let alone a team two time zones behind Sioux City, Iowa. While game times shouldn’t be an excuse this year, Southern Oregon finds themselves in a spot eerily similar to Davenport last year. Should the Raiders win their first round matchup, they would find themselves playing the winner of the 4/5 seed matchup. Who could that be? None other than Briar Cliff, the team that knocked out Davenport last season. Of course the Chargers would need to get past a solid St. Ambrose team first, but the intrigue of such a second round matchup is interesting to think about.
Some first round games that the model thinks could be the closest: 2Jamestown vs. 7Oregon Tech (Wed. 9:15 p.m.), 3Dakota Wesleyan vs. 6Haskell (Wed. 8:30 a.m.), 4Concordia vs. 5Goshen (Wed. 12 p.m.), and 4Briar Cliff vs. 5St. Ambrose (Thurs. 12 p.m.).
In terms of the newly named regions, the model shows the most parity between teams in the Liston bracket. The Liston bracket also seems to be the strongest overall even though a couple individual teams from other brackets project to go further. The weakest bracket appears to be the Cramer bracket according to the model. The best team according the model in this bracket is College of the Ozarks, although they would be just the 5th best team in any of the other brackets.
Check out the full bracket prediction of the statistical model below: