Comparing NCAA Ranking Systems For 2021 NCAA Men’s Basketball Tourney Picks

One of my favorite things in the world is watching the NCAA Men’s Basketball Tournament with my Dad. More specifically, the first weekend blitz of 48 games is a personal holiday. I always take vacation days from work, travel home, and watch the tournament.

With. My. Dad!

COVID-19 stripped that away from me. It stripped a lot away from a lot of people. However, I’m beyond blessed to be able to return to our lifelong tradition this week. I’m beyond excited. But what does this mean beyond food, screaming at the television, and racing hearts?

Basketball picks!

My Dad and I make basketball picks annually during this tournament, and we pick games by the round instead of the whole bracket pre-tournament. The one with the most correct picks wins. Normally, I pick based on intuition; and I correctly pick a decent amount of games (probably around 70%). What would happen if I picked using something more scientific? This year, I’m going to find out by comparing NCAA ranking systems during the 2021 NCAA men’s basketball tournament. Exciting!

NCAA Basketball Ranking Systems

For this test, I’ll use four well-known ranking systems–Sagarin, KenPom, RPI, and BPI. While I’ve followed men’s college basketball for several years, I’ve primarily paid attention to RPI and the newer BPI. I’ve been aware of Sagarin and KenPom for a number of years, but I’ve certainly not referenced them nearly as much. That being said, here are the rules for this test:

• In each of these branded rankings, I’ll take the higher ranked choice between two teams for a game.
• For all rounds, I’ll make my picks before referencing any of the respective rankings (though I’m not sure if these rankings will change during the tournament). I’ll use intuition, records, and talking heads to generate my picks (which should be worse than ever since I’ve barely watched any basketball all year).
• I’ll pick the same as my family tradition in a round-by-round fashion, summing to 67 games total.
• Finally, COVID-19 could eliminate a team at any point which isn’t a factor in any of these ranking systems. For test neutrality, any COVID-19 eliminations will not be counted toward a win or loss in any of the win-loss counts.

And with that, follow me for the next three weeks as March Madness ensues. Let the games begin! (And Roll Tide Roll!)

The First Four

The First Four games of the NCAA Tournament are usually not ones I can watch since they are on Tuesday. Apparently, COVID-19 changes and delays tournament games so I am able to watch a little this year. My picks included Texas Southern, Drake, Norfolk State, and Michigan State. I had a bit of a delay in accessing and recording comparisons, but I think they are still accurate (KenPom perhaps being the exception). Here are the early verdicts:

Immediately, this is Day 1 but shows that I’m in the top echelon with a strong three out of four correct performance. Should we just check KenPom, BPI, and Melissa for future tournaments? Time will tell! Onward to the First Round games.

First Round

Day 2 of the NCAA Men’s Basketball Tournament ought to come with some excitement in the bracket (as always). The picks are as follows. Overall, 9 games were selected exactly the same across the group resulting in 7 difference-maker games. Of those 7 difference-maker games, the RPI picked 3 games fully different from the rest of the group, I picked 2 games different, and KenPom picked 1 game different. Based on these picks, the most-interesting game of the day may be Rutgers-Clemson.

Immediate post-tournament reactions include that KenPom is starting to show some separation from the field and that the RPI is a terrible performer. If memory serves, the tournament selection committee stopped using the RPI. This test, if results continue in this manner, may demonstrate the reason. Sagarin is also an underperformer but starting to rise to the occasion after a terrible First Four. Rutgers-Clemson was, in fact, a very tight game throughout. Finally, I’m personally demonstrating how barely watching any basketball all season negatively impacts results (based circumstantially on prior years).

Day 3 features 12 games that have exactly the same picks across the board resulting in 4 difference-maker picks. Of those 4 difference-maker games, 2 games feature the RPI picking different from all other sources. Based on these picks, the two most-interesting games may be St. Bonaventure-LSU and VCU-Oregon.

Post-First Round reactions include that I feel bad about VCU being eliminated due to COVID-19 and hope that is the only such instance of virus impact. Second, a distinction is starting to form with the RPI appearing to be the worst ranking system. KenPom also fell back toward the rest of the pack slightly on Day 2. Finally, the predictions about the most interesting games possibly being St. Bonaventure-LSU and VCU-Oregon did not happen. LSU fairly soundly won, and VCU-Oregon didn’t happen.

Round of 32

Day 4 of the tournament features games that can start springing more upsets. However, the rankings for Sagarin, KenPom, BPI, and RPI didn’t change much. For example, the top teams in the First Round are still the top teams in the Second Round. That means that my participation is now a little less scientific due to that new knowledge and prior bias. That being said, I predicted more upsets (5) than the four ranking systems. In fact, Sagarin predicts 1 upset, KenPom predicts 2, BPI predicts 3, and RPI predicts 2. My performance depends on picking the right upsets (which I didn’t do in the prior round).

Here are the NCAA Men’s Basketball Tournament Round of 32 picks:

Picks were only 48% accurate on Day 4 which was significantly lower than the rest of the tournament. Interestingly, half of the games were seed upsets. KenPom continues to lead by a slim margin.

For Day 5, Oregon-Iowa, LSU-Michigan, and USC-Kansas are the games of note that could create separation. The rest of the picks will either elevate or deflate the group’s cumulative percentages and counts.

At the end of Day 5, three models–KenPom, BPI, and RPI–led with six correct picks. This did help bring the RPI model closer to the rest of the models while enabling KenPom to create greater separation from the rest. With only 15 games remaining, I will likely need to center further on art as opposed to science to have a chance of catching up with KenPom. Of course, that also risks me falling further behind…

The Sweet 16

Interestingly, in The Sweet 16, all professional men’s college basketball ranking models rank in the exact same way. I deviated with two upset picks. However, this may be an indication that as the tournament continues with some upsets, the rankings do not shift consequentially enough to influence and predict additional upsets. In other words, rankings do not shift enough to meaningfully change direction from the higher level picks. We’ll see if this continues to hold true.

The two games I predicted to be upsets are Florida State-Michigan and Oregon-USC. We’ll see where things stand after two days of games.

As an immediate reaction, I picked two upsets. There were two upsets; and I chose poorly which set me back dramatically in my quest to catch up with KenPom. KenPom now has a lead which is nearly insurmountable based on the number of games remaining. Comparing NCAA basketball ranking systems at this stage almost certainly seems to hold the final result.

The Great Eight

For the NCAA Men’s Basketball Tournament Great Eight, all games were picked exactly the same based on the four ranking systems. The four higher-seeded teams seem to have mostly large numeric separations between themselves and the lower-seeded teams. If we are to trust KenPom (seemingly the most trustworthy model at this point in this study), the most likely upset will be USC over Gonzaga (5-rank separation in KenPom) followed by UCLA over Michigan (12-rank separation) and Arkansas over Baylor (16-rank separation). The least likely upset will be Oregon State-Baylor (41-rank separation). Regardless of the results, all systems will move the same due to the same picks.

As we end the regional finals, we maintained roughly the same scores with a slight boost across all models. UCLA demonstrated how incredibly shattering an upstart can be to correctly picking the tournament, and KenPom fought through that dynamic to correctly pick over 70% of the games. With three games remaining, the results are likely solidified.

The Final Four

And as we enter The Final Four weekend, I was correct about the results likely being solidified. With the exception of the RPI picking Houston, all picks are Gonzaga and Baylor. There is a slim possibility that RPI may catch up with me or Sagarin. There is a slim possibility that Sagarin or I may separate from currently being tied. KenPom and BPI have both cinched first and second place, respectively.

Gonzaga won in an instant classic, and Baylor won in a game in which Houston never competed. The final rankings are set with KenPom winning handily. But more to come on that after the National Championship.

The National Championship

As we reach the final game in our quest comparing NCAA ranking systems, we have much to discuss. But first, we’ll see if we can improve the cumulative percent correct.

With Baylor winning, every model dropped by a percentage. With that, I’ll cover lessons learned.

Lessons Learned

I’ve had roughly a week to think about what transpired during this pseudo-scientific test comparing NCAA ranking systems. Here are four lessons learned:

• KenPom slightly outperforms Sagarin and BPI. Eliminating The First Four removes the advantageous 4-0 start that KenPom leveraged for the rest of the tournament (43 correct picks compared to 42 each for Sagarin and BPI without this starter group of games).
• RPI is the worst model. The RPI, which is no longer used by the Selection Committee, appears to perform worse than all models, including mine.
• KenPom, Sagarin, and BPI select few, tightly-seeded upsets. One difference-making decision is when a person or a model picks upsets. Sagarin, KenPom, and BPI only picked 4-5 tournament seed upsets across 62 games of the main tournament (does not include First Four). RPI picked 8 seed upsets, and I picked 10 seed upsets. (Apparently, I’m a risk-taker!) The rankings for this selection of picks is very similar to the final results–KenPom (3-2), Sagarin (2-2), BPI (2-2), Melissa (3-7), and RPI (1-7). For KenPom, Sagarin, and BPI, the seed differential in these picks was 1-3 seeds.
• Upset picks can really make or break performance. If I had not selected ten upsets, my performance would have been the same as KenPom. No matter how good it felt to correctly pick Oregon over Iowa and Loyola-Chicago over Illinois, the upset strategy failed me in the end. After all, there is a reason for the seeds these teams earned.

So that completes our analysis for the 2020-2021 NCAA Men’s Basketball Tournament. I hope this has been informative and helps you with your future picks. Maybe we’ll do something similar next year with some twists…stay tuned!