Every March, as the NCAA Tournament rolls around, sports fans get ready to watch 67 games where the winner advances and the loser goes home. There are buzzer beaters, crazy upsets, and unexpected Final Four runs. Many people make brackets to try to attempt to predict what will happen, although no one has succeeded in getting all 63 picks correct. However, this year I will attempt to make some predictions for March Madness, using my college basketball predictive metric: College Basketball SABER (Schedule Adjusted Basketball Efficiency Ratings).
How the SABER Metric Works
In order to reduce the length of this post, the full explanation of the SABER ratings can be found here.
In summary, the SABER metric works by finding each team’s offensive and defensive rating, then adjusting those values for difficulty of opponent, recency, location of games, and difficulty of conference in order to create values that accurately capture the team’s offensive and defensive ability. Then, the offensive and defensive values are combined to create a total team rating that can be used to predict future games.
Analyzing Past Predictions
In order to analyze the past predictions by the SABER metric I created, I had to find the probability of a game using the difference in xNet of two teams. Therefore, I ran a logistic regression using the 2019-20 season as the sample in order to generate win probabilities for any game. Then, I examined how the predictions made by SABER on the Selection Sunday before each tournament fared in predicting first round upsets and the ultimate Final Four teams.
First Round Upset Predictions
Extreme Upsets (Seeds 13-16 win against seeds 1-4)
Using the last 5 NCAA Tournaments, I analyzed how the SABER metric predicted extreme upsets, which I consider to be when a top 4 seed loses in the first round. First, we can start with games that SABER predicted to be not close at all. The table below shows the games in which 1-4 seeds had the greatest chance to win. In order to include a variety of teams, I used half of the results from games with 1 or 2 seeds and the other half from games with 3 or 4 seeds.
As expected with such high seeded teams, the ones that were predicted to win won most of the time, with the only exception in the list being Iowa State’s loss to UAB. Additionally, the model did not see UMBC’s win over Virginia or Middle Tennessee’s win over Michigan State at all, even though they are not shown.
The next table shows the games including seeds 1 through 4 in the games they were least expected to win. The model was unable to predict any upsets by 15 or 16 seeds, so it is not very reliable when it comes to picking those teams to win in the first round. However, it does have success in picking 3 or 4 seeds that may be vulnerable, including West Virginia in 2016, Baylor in 2015, and Wichita State in 2018. Not many predicted those teams to lose in the first round, but the model had them on upset alert.
More Common Upsets (Seeds 9-12 win against seeds 5-8)
As we saw above, the model has had previous success picking 3 or 4 seeds that could be upset in the first round, but not 1 or 2 seeds. Now, we can see how it predicts first round games including seeds 5 through 8, which are games in which upsets are much more common.
First, we can look at the games that the SABER metric greatly favored the higher seed. As seen by the table below, when the win probability based on SABER is especially high, the higher seed is very likely to win. The only miss of the games included below was when Minnesota upset Louisville in 2019. However, it successfully predicted that teams like Iowa in 2016, Indiana in 2016, and Northern Iowa in 2015 did not have a large reason to worry about losing in the first round.
Next, we can look at the games that were expected to be upsets based on SABER. For games with 7 or 8 seeds, the model was very good at predicting the winners. Of the five games in which the lower seed had the highest probability to win, the lower seed won each time, proving the model’s accuracy for these types of games. Additionally, the model successfully predicted 5 and 6 seeds that were upset, including Minnesota in 2017, Miami in 2018, and Seton Hall in 2016. The only miss of this group was when Michigan lost to Notre Dame despite being favored as the lower seed.
Based on the tables above, the SABER metric does well in predicting first round upsets for seed 3 through 8, but not for 1 and 2 seeds. Therefore, the SABER ratings can provide reliable predictions for the first round matchups of this year’s tournament.
Final Four Predictions
The SABER metric can also help predict the Final Four teams. Below is a list of the final four teams from the last 5 tournaments along with their seeds and ranks in the SABER metric.
All of the final four teams except for the three most unexpected (2016 Syracuse, 2017 South Carolina, and 2018 Loyola Chicago) were in the top 15 of the SABER rankings. Additionally, 36 of the 40 teams were in the top 10 of either offense or defense. Every year with the exception of 2018 had at least 2 of the top 5 teams in the Final Four. Even in 2018, when only 1 of the top 5 teams made the final four, Michigan was ranked 7th overall. Also, it is important to note that the number 1 team based on SABER made the final four every year except for 2016, when Michigan State was the number 1 team based on SABER but they lost in the first round.
We can also see how the very best teams according to SABER performed in the NCAA Tournament. The top 5 teams based on SABER are shown for the 2015 through 2019 seasons.
The teams that are the best according to SABER usually perform well in the NCAA Tournament. Of the top 5 teams every year, usually 3 or 4 teams made it to the Elite 8 or farther. On average, about one of the top 5 teams each year is upset in the first or second round. These teams include Villanova in 2015, Michigan State in 2016, Villanova in 2017, and Virginia in 2018.
Based on the tables above, we can make some conclusions about teams’ SABER ratings and their expected finish in the NCAA Tournament. Teams outside of the top 15 of the SABER rankings are going to be long shots to make the Final Four. Additionally, it is likely that 2 of the top 5 teams based on SABER are going to make the Final Four, and 3 or 4 of them will advance to at least the Elite 8.
Most importantly, we can form a narrow field of the actual champion contenders. In each of the past 5 tournaments, both championship teams were in the top 10. Additionally, the all the tournament champions were in the top 6 of the SABER rankings on the Selection Sunday just before the tournament: Duke was 6th in 2015, Villanova was 4th in 2016, North Carolina was 5th in 2017, Villanova was 1st in 2018, and Virginia was 1st in 2019. If these trends hold up these year as well, we can narrow the field of actual champion contenders to just 6 teams, the field of actual championship game contenders to 10 teams, and the field of final four contenders to just 15 teams.
Predictions for the 2021 Tournament
Now that the bracket for the 2021 NCAA Tournament has been released, I can analyze the most likely upsets and Final Four teams using SABER. First, the top 10 teams in SABER this season are shown below.
Each of the 1 seeds in this year’s tournament are in the top 5 of the SABER rankings, and all the 2 seeds except for Ohio State are in the top 10. However, ranks 8 through 10 are somewhat surprising, as Virginia is a 4 seed, USC is a 6 seed, and Colorado is a 5 seed. All three of these teams could make a longer run than expected as they have performed better than their seed would suggest.
First, we can look at the matchups in the West (top left) region of the bracket. According to SABER, this region is one of the strongest, as 6 of the top 20 teams are included. However, there are still a few possible first round upsets. Both the 14 and 15 seeds in this region are stronger than usual, meaning that Iowa or Kansas have a slight chance of being upset. The only other real upset candidate would be UC-Santa Barbara beating Creighton, although Creighton’s 73.4% chance to win is close to the average probability of a 5 seed winning. Based on the fact that the last 5 final fours have included mostly top 15 teams, the Final Four candidates from this region include Gonzaga, Iowa, Virginia, and USC. However, the only major championship contender (top 6 SABER) is Gonzaga.
The East (bottom left) region of the bracket is led by Michigan, who got the last 1 seed. The East region is also pretty strong, including Michigan and Alabama in addition to sleeper teams like Florida State, Colorado, and UConn. Some upset candidates include Abilene Christian over Texas and UCLA over BYU. Only three 3 seeds in the last 5 tournaments have had a lower chance to win in the first round than Texas: Texas Tech in 2018 (beat SFA), West Virginia in 2016 (lost to SFA), and Baylor in 2015 (lost to Georgia State). Additionally, UConn, the 13th best team according to SABER, is playing much better than their seed, so they could make a deep run. Final Four candidates include Michigan, Alabama, Colorado, UConn, and Florida State. The Championship contenders from this region are Michigan and Alabama.
The South region (top right) is led by Baylor, the second 1 seed. This region has a lot of depth, but isn’t very top-heavy. There are 6 top 16 teams, but the only top 10 team is Baylor. Additionally, this region could include many upsets. Arkansas has the lowest probability for a 3 seed in the first round since West Virginia in 2016, meaning they should watch out for Colgate. Additionally, Purdue has a low probability for a 4 seed. Although it is not predicted by SABER, Villanova could be on upset watch as they have been struggling without their star point guard Collin Gillespie (I was unable to include a method to account for injuries when I made the SABER rankings). The sleeper candidate from this region is Texas Tech, which ranks 11th overall in SABER but is only a 6 seed. The Final Four candidates include Baylor, Texas Tech, Villanova, and Arkansas, but the only championship contender is Baylor.
The last region is the Midwest (bottom right) region, led by Illinois. This is by far the weakest region according to SABER with only 3 top 20 teams. The 8-9 matchup is fantastic, as both Loyola and Georgia Tech are in the top 21 of SABER as low seeds. An upset candidate is Syracuse, having just a 50% chance to beat San Diego State. Illinois and Houston are the only Final Four candidates and only championship contenders from this region.
Using 1000 simulations based on each school’s SABER rating, I found the probability of each team winning the NCAA tournament and making the Final Four. The results for the teams with the best chances of winning the championship are shown below.
The only teams with at least a 5% chance of winning the championship or a 25% chance of making the Final Four this year are Gonzaga, Baylor, Illinois, Michigan, Alabama, and Houston. After those teams, there is a significant drop off as the next best team, Iowa, has just a 14% chance to make the Final Four and only a 3.5% chance to win the championship. Each of the teams in the top 6 of SABER are also placed in the top 8 of KenPom, the top 10 of ESPN’s College BPI, and the top 8 of 538’s Elo Ratings. Therefore, it is likely that the champion will come from one of these 6 teams. Using my SABER ratings, the 6 main contenders for the 2021 NCAA championship are Gonzaga, Baylor, Illinois, Michigan, Alabama, and Houston.