The media’s mock NCAA tournament bracket was released this weekend, and the Stat Geek Selection Committee released one of our own soon afterward.
Here’s some analysis on what the main differences are, and why those differences exist.
Who Got Into Only One Bracket?
Four teams were selected by one of the two groups, but not by the other. The first two are closely related:
Arizona Wildcats (Media: #12 / Stat Geeks: OUT)
Miami Hurricanes (Media: OUT / Stat Geeks: #12)
We actually originally had both of these teams in the field, but due to some surprise mock conference tournament windows, the bubble shrank at the last minute. As a result, we had to drop a team, and our choice came down to either Arizona or Miami. This turned out to be the longest debate we had in the entire process, and in the end it was decided by a single vote. In other words, pay no attention to this difference. It almost didn’t exist.
Illinois Fighting Illini (Media: #10 / Stat Geeks: OUT)
Illinois was in one of the media’s play-in games, which means they were one of the last four in. According to Bracketville, game was originally slated to be on the 11 seed line, but got moved up for bracketing purposes. The Illini were the third team out for us (behind Arizona and Cincinnati), so this is another close one. Ultimately, I don’t think you can really fault us for excluding a team that has now lost 8 of their past 9 games — and only scored 42 points in that lone win. Especially when the team that made it in instead was…
Middle Tennessee State Blue Raiders (Media: OUT / Stat Geeks: #11)
The Blue Raiders are 13-1 in the Sun Belt, and played Vanderbilt close before ultimately falling by 7 in a late January nonconference game. While their biggest win is probably at home versus Belmont, simply racking up wins the way they’ve done is more difficult than people realize.
Biggest Seeding Disagreements
There were 12 teams for whom our seed was at least two lines higher or lower than the one given by the media. Here’s the full list. The media liked the teams on the left better than we did, while we rated the teams on the right higher than the media did.
(It’s important to remember that our goal was not to try to predict which teams will be selected by the actual committee. Rather, our goal was to pick the best teams. There’s a big difference between those two. If we wanted to predict what the real committee would do, we would have payed close attention to the RPI. Instead, I don’t think it was mentioned once in all of our emails.)
|Team||Media||Stat Geeks||Diff||Team||Media||Stat Geeks||Diff|
|Murray St||7||11||+4||Saint Louis||9||5||-4|
|San Diego State||6||9||+3||California||11||8||-3|
The team with the biggest seed difference between the media bracket and the stat geek bracket was Mississippi State. This comes as absolutely no shock at all, as the Bulldogs have been listed for weeks on our polls comparison page as one of the most overrated teams. This is a team whose last three games are losses to Georgia, LSU, and Auburn. Sorry, but Mississippi State is simply not good enough to be a #7 seed.
Murray State is an interesting case. The Racers have had to make furious comebacks to keep from losing three Ohio Valley games, which isn’t a great sign. Their lack of blowouts against a very weak conference schedule has kept their predictive power rating relatively low — they were #37 when we conducted the voting for our bracket, but keep in mind that was before they beat St. Mary’s. However, there’s something to be said for simply winning all the games you’re supposed to.
It’s interesting that the media includes two Pac-12 teams compared to our one, yet only rated California as a #11 seed while we had them up at #8. My guess is that there are two reasons for this. First, the media placed far too much emphasis on California’s 39 point loss to Missouri. Yes, it was a bad loss, but #2 seed UNC lost by 33 to a worse team. Second, the Pac-12’s general stink is rubbing off on Cal in most people’s minds. We all say we don’t consider conference affiliation when making these decisions, but it’s impossible not to let it cross your mind.
Better To Be Lucky Than Good?
There is one overarching theme to the differences between the two brackets — the media highly rewarded teams for scraping out close wins, while the stat geeks tended to place more emphasis consistently dominating opponents.
The two groups of teams above have roughly the same overall record. The teams seeded higher by the media are 119-44 (73%), while the teams seeded higher by the stat geeks are 123-40 (76%). But take a look at their records in games with a double digit margin of victory: the media darlings are 69-16 (81%) while the stat geek favorites are 95-12 (89%). Now take a look at their records in games decided by single digits: the media’s teams are 50-28 (64%) while the stat geeks’ preferences are only 28-28 (50%).
Basically, the stat geeks are assuming those close-game records will start to even out over the long run, but the big-win advantage will remain. The media swings the other way, crediting those close games to guts, clutch play, knowing how to win, etc.
Another way to look at this is from the point of view of our Luck Ratings, which tell you how many “extra” wins a team has, compared to how many we’d expect based on their schedule and power rating. If a team’s win-loss record is better than their average margin of victory would suggest, then they’ll have a high Luck Rating.
The average luck rating of the six teams seeded higher by the media is +0.9. The average luck rating of the six teams preferred by the stat geeks was -1.0. That means that if we played the season all over again, we’d expect the media’s preferred teams to win one less game, and ours to win one more.
This is a classic case of choosing between rewarding teams for their wins and losses, versus using all the information available to predict which team is better. While the NCAA guidelines actually instruct the selection committee to vote for the best teams, better seeds often go instead to the most accomplished. The difference may be subtle, but the effects on seeding can be large.