Bracketology: Projecting the Future Leads to Better Results

Bracketology: Projecting the Future Leads to Better Results


Bracketology: Projecting the Future Leads to Better Results

I’ll be coming out with my first 2018 bracket projection for the NCAA tournament shortly, but I thought it would be good to look back and learn. My philosophy with my brackets is to project what the bracket will look like on Selection Sunday, and not just based on results that have already happened. I can see the argument either way–only rewarding what has happened versus trying to peer into the future–but I think the value of these projections is seeing which teams are on track for the tournament, and where they will be seeded, if they play like they have so far, against their remaining schedule.

The selection committee largely cares about RPI, and how teams have fared against other teams ranked highly in the RPI. “Top 50 wins” get outsized importance. Well, if you take a team like Clemson, then most of their big games are yet to come. They’ve played three teams in the Top 50 (Louisville, Ohio State, Florida) and won all three. Maybe you want to rank only what they’ve accomplished so far. 3-0 is pretty good, along with a 15-1 record. But using advanced stats like Ken Pomeroy or power ratings like Sagarin give us a better picture than 3-0.

I can tell you that using something like RPI Forecast, Clemson is expected to win about 3.4 out of the 9 games against likely Top 50 teams remaining, using Jeff Sagarin’s Predictor ratings, and 4.5 out of 9 games using Ken Pomeroy (there is a slight difference on where they are ranked in the two). That puts their expectation entering the ACC Tournament at around a 6-6 or 7-5 record against Top 50 teams, and a 22-8 or 23-7. RPI Forecast puts the average final RPI at 20.7. Using those projections allows us to compare Clemson to similar teams.

Do that for every potential at-large team and then we’ve got our projections, using both the banked wins and losses already plus what is still to come.

Looking back on last year’s initial January projections at mid-month, it did pretty well, especially with the upper half of the at-large pool. Only one of the 24 teams ultimately seeded as a 6 or better was projected more than 2 seeds off, two months before Selection Sunday. That was Iowa State, who was 12-6 on the date of those projections, but won 9 of their final 10, including the tough Big 12 tournament.

Here’s a comparison of how my method did, comparing it to the bracketology predictions from ESPN, Sports Illustrated, and Fox Sports on or near the same date, using the 46 teams that were seeded 11 or better in the actual NCAA Tournament (so, the at-large pool).

We all missed teams that far out, seven each, though not always the same teams. It’s not a perfect exercise, but it’s a good way to see which teams are likely to be on the bubble. Six of my final 10 at-larges made it, while of the teams that made it but I did not have in the field in January, five of them were among my Top 8 just out. (And 3 of the first 4 out made it). Vanderbilt, a team that was 8-10 on the date of the initial release, was the only big surprise.

Meanwhile, I had 26 of the 46 projected within one seed line of where they would eventually be placed. Other brackets were more likely to have teams that did not power rate as highly, but had an early number of Top 50 wins, seeded higher than they ended up (Baylor as a #1, Florida State as a #2) while also having traditional powers projected down based on some early January losses (Arizona and Duke as 5 seeds).

Here’s an interesting thing about the projections as well. If you know where the RPI has the conferences ranked, you can largely project on a broader basis how many teams or the quality of combined seeding that will go to those conferences. The non-conference is 95+% done, so any movements now come with one conference member moving up at expense of another.

For example, if you used a simple scoring system where a 12-seed got 1 point, and a 1-seed got 12 points, and 1 point difference for each seed in between, I would have projected the following in 2017, versus the actual results:

I underestimated how strongly the committee would reward the ACC, since I had them already projected to get the lion’s share. But other than that it was pretty spot on. The identities may have changed (Michigan surged while Indiana collapsed; Iowa State and Oklahoma State went up while Texas Tech and TCU dropped; Creighton and Xavier’s seeding fell while teams like Seton Hall and Marquette improved) but it was pretty stable.

Based on that, and knowing how the RPI is this year, we can expect the ACC, Big East, Big 12, and SEC to get about 60-65% of their members in, while the Pac-12 and Big Ten are rated way lower, along with the American, and those conferences are likely to get about 35-40% of their teams in. The identities again may change, and it might be hard to predict whether it is Georgia or LSU or Alabama or Mississippi State, but some of those teams will make it.

The first projections in the year 2018 will hit tomorrow and it looks like a tough and wide-open year, but taking into account conference strength and the quality of games yet to come will provide a better result. I feel good in predicting that.

Latest Leads