Elite8 Final4 Final Champ 1E Villanova 72.5 48.5 31.1 23.3 2MW Duke 81.1 54.0 28.8 20.4 4W Gonzaga 68.9 40.0 25.9 10.7 2E Purdue 61.1 26.3 13.7 8.7 3W Michigan 62.3 32.7 20.0 7.8 5S Kentucky 63.6 37.7 17.8 5.8 1MW Kansas 53.9 22.2 8.5 4.8 7S Nevada 56.9 27.6 11.4 3.2 5MW Clemson 46.1 17.5 6.1 3.2 5E West Virginia 27.5 12.4 5.1 2.8 3E Texas Tech 38.9 12.8 5.2 2.8 7W Texas A&M 37.7 15.1 7.3 1.9 11S Loyola Chicago 43.1 18.2 6.4 1.5 9W Florida St. 31.1 12.2 5.6 1.3 9S Kansas St. 36.4 16.4 5.6 1.3 11MW Syracuse 18.9 6.3 1.4 0.5
Rd2 Swt16 Elite8 Final4 Final Champ 1E Villanova 97.4 81.9 62.5 43.7 28.7 18.1 1S Virginia 96.7 78.1 59.9 41.4 28.9 17.6 2MW Duke 95.9 80.3 51.8 36.9 20.7 12.1 2S Cincinnati 87.3 62.8 42.1 21.8 13.1 6.6 2E Purdue 94.3 68.7 45.8 22.7 12.4 6.6 2W North Carolina 95.9 70.9 43.6 25.8 12.4 5.7 3MW Michigan St. 89.4 63.8 31.9 20.5 10.1 5.2 4W Gonzaga 82.6 53.1 33.4 18.6 8.9 4.0 1MW Kansas 88.8 59.4 35.5 15.2 6.5 2.9 3W Michigan 76.5 46.8 24.5 13.1 5.8 2.4 1W Xavier 97.9 63.5 31.3 15.2 6.2 2.3 3S Tennessee 88.1 57.5 27.6 11.5 5.5 2.2 3E Texas Tech 85.4 53.3 24.7 9.5 4.0 1.8 5W Ohio St. 74.4 35.5 19.2 9.2 3.6 1.4 4MW Auburn 84.1 47.6 24.4 8.8 3.1 1.2 5E West Virginia 71.4 42.6 14.4 6.8 2.9 1.2 6W Houston 65.3 33.0 15.0 7.2 2.7 1.0 5MW Clemson 65.7 35.6 17.8 6.5 2.4 0.9 5S Kentucky 62.4 36.7 12.5 5.7 2.5 0.9 4E Wichita St. 84.1 43.6 13.7 5.9 2.3 0.9 4S Arizona 71.8 37.5 12.2 5.3 2.2 0.8 6E Florida 63.8 30.5 11.8 3.7 1.3 0.4 7S Nevada 56.5 20.1 9.7 3.2 1.2 0.4 6MW TCU 61.6 22.8 7.6 3.5 1.1 0.4 10E Butler 55.8 18.2 8.6 2.6 0.9 0.3 8MW Seton Hall 55.5 22.5 10.4 3.1 0.9 0.3 7W Texas A&M 61.9 19.9 8.4 3.2 1.0 0.3 8S Creighton 56.0 12.4 5.8 2.2 0.8 0.2 6S Miami FL 51.5 20.8 7.1 2.1 0.7 0.2 8E Virginia Tech 57.1 11.1 4.7 1.7 0.5 0.2 8W Missouri 49.1 17.9 6.5 2.2 0.6 0.2 11S Loyola Chicago 48.5 19.0 6.5 1.9 0.6 0.2 9W Florida St. 50.9 18.4 6.5 2.2 0.7 0.2 10S Texas 43.5 13.6 5.8 1.7 0.6 0.1 12S Davidson 37.6 17.1 4.3 1.5 0.5 0.1 7E Arkansas 44.2 12.1 5.0 1.3 0.4 0.1 9S Kansas St. 44.0 9.1 3.8 1.3 0.4 0.1 9MW North Carolina St. 44.5 15.8 6.2 1.5 0.3 0.09 7MW Rhode Island 49.8 9.9 3.0 1.1 0.3 0.07 11W San Diego St. 34.7 12.6 3.8 1.3 0.3 0.06 12MW New Mexico St. 34.3 13.1 4.5 1.0 0.2 0.05 10MW Oklahoma 50.2 9.0 2.6 0.9 0.2 0.05 9E Alabama 42.9 6.5 2.3 0.7 0.2 0.04 11E UCLA 22.9 8.6 2.5 0.6 0.1 0.04 11MW Arizona St. 20.5 5.7 1.4 0.5 0.1 0.04 12E Murray St. 28.6 11.0 2.1 0.6 0.1 0.03 11MW Syracuse 17.9 5.1 1.2 0.4 0.09 0.02 10W Providence 38.1 8.6 2.6 0.8 0.2 0.02 14W Montana 23.5 7.6 1.9 0.5 0.1 0.02 13S Buffalo 28.2 8.7 1.4 0.3 0.07 0.01 12W South Dakota St. 25.6 6.3 1.8 0.4 0.08 0.007 13W UNC Greensboro 17.4 5.1 1.4 0.3 0.06 0.007 11E St. Bonaventure 13.4 4.1 1.0 0.2 0.03 0.004 15S Georgia St. 12.7 3.5 0.9 0.1 0.02 0.001 14MW Bucknell 10.6 2.6 0.3 0.07 0.009 0.001 14S Wright St. 11.9 2.8 0.3 0.04 0.006 0.001 13E Marshall 15.9 2.8 0.2 0.02 0.004 0.001 13MW College of Charleston 15.9 3.7 0.7 0.08 0.01 <.001 14E Stephen F. Austin 14.6 3.5 0.5 0.05 0.008 <.001 16MW Penn 11.2 2.3 0.4 0.03 0.006 <.001 15MW Iona 4.1 0.9 0.09 0.02 0.003 <.001 15E Cal St. Fullerton 5.7 1.0 0.1 0.009 <.001 <.001 15W Lipscomb 4.1 0.7 0.07 0.006 <.001 <.001 16S UMBC 3.3 0.4 0.04 0.006 <.001 <.001 16E Radford 2.4 0.5 0.06 0.005 <.001 <.001 16W Texas Southern 1.5 0.09 0.004 <.001 <.001 <.001 16W North Carolina Central 0.6 0.04 0.002 <.001 <.001 <.001 16E LIU Brooklyn 0.2 0.01 <.001 <.001 <.001 <.001
If you are looking my annual conference tournament probabilities, they are being posted to my twitter account.
Hey everyone. Now that the beginning-of-the-season rush has settled down, it’s a good time to go over some of the new features on the site.
The short version is:
– New and improved win probability graphs (with FanMatch updates)
– Team records
– Age and position information on the player pages (when available)
– Better position information
Now, the long version:
Win probability graphs
The biggest project of the offseason was upgrading the win probability charts. I hadn’t really touched these since their initial unveiling in the offseason of 2010. For one thing, they look more professional now, almost like you could put them on a real web site.
I also refreshed the algorithm used to calculate the win probabilities which was last updated in 2012. The method is the similar as back then, but I am making use of the results of the tens of thousands of games played since then. One change in methodology is that the initial state is determined by the algorithm itself and not by the percentages given by the ratings. Often, these differences are small.
In addition, each win probability chart contains additional measures to describe the game. These are comeback, excitement, tension, and dominance. The charts now show the ranking in each category among all games played in its respective season. I will briefly explain each of these.
Comeback measures the minimum win probability for the winning team. Please remember that the win probability accounts for the relative strength of the competing teams. Therefore, a team can overcome a low win probability without necessarily overcoming a large deficit on the scoreboard.
Excitement measures the length of the win probability graph assuming both teams are equal. (This is not the same as the plot you see on the graph which accounts for the relative strengths of the teams.) The most exciting game of the season as of this writing was Friday night’s Florida/Gonzaga game. That game was pretty exciting and I think would win a vote of sportswriters so far this season.
However, the excitement index doesn’t care about the quality of the teams involved. By the end of the season, the most exciting game is as likely to be from the SWAC as it is the SEC.
Excitement is scaled for overtime, so it is measuring something like excitement per minute. Although the most exciting things happen when the game is on the line, so overtime games will usually be quite exciting by this measure. But some overtime games are not that exciting until very late in regulation.
The value listed for excitement is not terribly meaningful. I like to think that a value of one or more means you got your money’s worth and a negative value means you should demand a refund. The entity which sold you the ticket may or may not agree.
Excitement will also have a bias towards high-possession/high-scoring games. For fans of low-possession/low-scoring games, there’s the tension measure.
Tension measures the area between the win probability plot (assuming equal teams, so not the one shown) and the 50% line. The most tense game to date was an overtime affair between Sacred Heart and Maine on November 19. Neither team led by more than six over the entire 45 minutes.
Dominance is nearly the inverse of tension. Behind the scenes, I have a model that predicts the final margin at any point in the game assuming the teams are equal. Dominance is measured using the average predicted margin during the game for the winning team. It is not terribly useful, but I needed another score for some symmetry. Build a big lead early and continue to run up the score during the game, win by like 60, and you’ll post a high dominance score. The most dominant game so far was last Thursday’s late-late-late night game between San Diego State and Sacramento State. The Aztecs built a 49-14 halftime lead en route to an 89-52 victory. Go team.
These measures are not meant to settle arguments nor designed to be publication-worthy algorithms. They are simply useful guides to get a feel for how a game was played. If you feel that you witnessed a game more exciting than Florida vs. Gonzaga this season, you probably did. Congrats!
The excitement and comeback measures are also on the FanMatch™ page now with top 100 season ranks noted where appropriate. These are mainly for reference. If you couldn’t stay up to see the Florida/Gonzaga game (no shame in that because it ended at about 2:30 AM ET) then you could have reviewed the FanMatch™ page the next morning and noticed that you missed a crazy game. You can check out the win probability and box score to get an idea of what happened. Then consult the work of local beat writers for quotes and such. You will never miss anything important in college hoops again!1
Team records page
There is a team records page available from the team history page. It lists various game records for each team over the history of the site. Wondering where your team’s recent game ranked all-time in turnover percentage since the 2002 season? Well, if it’s in the top ten best or worst, that page will tell you the answer.
Player page enhancements
Player birthdate and age is now listed on the player page when that information is known. Shown is the player’s age as of January 1st of the appropriate season along with their age relative to their class average. In addition a player’s position for each season is shown provided the player played at least 10% of his team’s available minutes and the season is in the play-by-play era which begins with the 2010 season.
Position identification improvements
The position identification for the depth chart has been improved by leveraging play-by-play data to better identify the point guard that is on the floor. I’m sure you can find teams where the algorithm mis-identifies the point guard, but it is slightly more difficult than before. This has been applied to previous seasons and after a few more games, the 2018 season depth charts will take advantage of play-by-play stats as well.
[ + ]
|1.||^||Assuming timely dissemination of play-by-play data by the home team.|
Some of you who have used the kenpom app lately may have noticed that you are unable to log in. (You may be lucky for a while if you have an Android.) Without getting into the technical causes of this, it’s the symptom of a larger issue.
The talented people that originally designed the apps have moved on to better things and as the features on the site have grown, the functionality of app has lagged behind. Given the constant evolution of the site, it makes more sense to make the site itself more mobile-friendly. Going forward this is a better solution, allowing mobile users to more easily access all the features available on the site, especially as new ones are added. Therefore, the apps will no longer be supported.
This will likely not be a fast process but you should see improvements gradually rolled out during the season. Thanks for your understanding and patience.
In our last installment, we discussed the degree to which past home-scoring advantage predicts future home-scoring advantage. We can look at other stats and derive home advantages from them. For an example, let’s use home steals advantage. Like home-scoring advantage, this is the difference between a team’s home steal margin and road margin.
In conference games from 2002-2009, Alabama averaged 1.32 more steals than their opponents at home and averaged 1.98 fewer steals than their opponents on the road. That gives a home-steals advantage of 1.66, which led all teams during that time.
Does home-steals advantage predict future home-scoring advantage? Well, no, not at all. And most box-score stats don’t. But there is one stat that actually outperforms points. If we just use home-foul advantage by itself to predict future “home-court advantage”, it does significantly better than home-scoring advantage. Here are the predictions compared to the observed values based on using previous home-scoring advantage (HCA) and previous home-foul advantage (HFA).
It’s not a huge difference, but our cloud of data points is getting stretched out along the line of perfect predictions a little bit. Of course, using both home-foul advantage and home-scoring advantage is even better to predict future home-court advantage. The plot on the right below illustrates the predictions of future home-scoring advantage made from such a model.
So it’s pretty clear that fouls drive home-court advantage more than any other thing recorded in the box score.1 Another piece of evidence that foul bias is an important source of home-court advantage is the identical trend in home-court advantage and home-foul advantage2 over the past decade-and-a-half.
As far as what’s causing the decrease in home-foul advantage, one can only speculate. As officials and their supervisors get more and quicker access to video from games, it stands to reason that they would become more fair about making calls. But it’s also possible that less enthusiastic fan support may be decreasing the home crowd’s influence on officials.3 Or players themselves may be more prepared in road games than they used to be for some reason. Whatever the source, the result appears to be a fairer whistle towards the road team.
The next step is to make predictions for the home-court advantage of each team. So far, I’ve been using eight years of training data and eight years of target data. It obviously helps to have more seasons to stabilize some of the data, but just using half of our 16-year sample to predict the other half limits us to 325 data points.4 We could use the previous two years to predict the next season of home-court advantage. This has the drawback of increasing the error of predictions, but it has the benefit of increasing our sample by a factor of 15.
After considerable experimentation, I’m using the past six seasons of data for each team to predict its next two seasons of home-court advantage. This gives the value for each some stability while allowing flexibility for home-court numbers to change over time.
The increase in samples for the model revealed two other stats of importance: non-steal turnovers and blocks. Home advantages in both of those categories influence the model as well. And finally, a team’s elevation is included in the model.
I must warn you that any predictions of team-specific home-court advantage will be noisy. I mean, just look at the plot of predicted vs. actual HCA based on home-foul and home-scoring advantage. A lot of those points stray pretty far from the line of perfect predictions. And those are in-sample predictions. Any predictions for the next few years will undoubtedly be more noisy than that.
The result is that determining which team has the best home-court advantage is impossible with any degree of certainty. Even distinguishing between the tenth-best and 60th-best home-court advantage is on shaky ground. It’s probably best to think of of home-court rankings in groups of three. Teams rated in the top-third probably have an above-average home-court advantage and teams in bottom-third probably have a below average home-court advantage.
Nonetheless, I have posted a team’s home-court advantage computed from this method at the bottom of each team’s schedule. In light of the previous paragraph, the rankings and values are mostly for conversational fodder at your next cocktail party.
Along those lines, the team that currently owns the top ranking in home-court advantage is Air Force, at 4.5 points. An unlikely choice to be sure, and it’s worth repeating that we can never know which team has the best home-court advantage. It is probably not Air Force, a team that plays in a dingy 5,800 seat arena that requires clearing a security checkpoint to get to.
However, it’s illustrative of the fact that home-court advantage and team quality do not have to be related. The Falcons own the second-longest active losing streak in road conference games at 22. And over that time they’ve gone 13-10 in conference games at home. It’s a safe bet they benefit from an above-average home-court advantage.
[ + ]
|1.||^||This is largely consistent with findings offered by Tobias Moskowitz and Jon Wertheim in the book Scorecasting: The Hidden Influences Behind How Sports Are Played and Games Are Won.|
|2.||^||It’s important to note that a team’s home-foul advantage is not simply how many fewer fouls that team commits at home compared to its opponents. It’s the team’s foul difference at home relative to its foul difference on the road.|
|3.||^||There’s another question about what drives home-foul bias and given that teams from better conferences tend to benefit from home-foul advantage the most, one could infer that the crowd has a significant influence on this.|
|4.||^||Additionally, a team’s home-court advantage is probably not constant over an eight-year period.|
Who has the largest home-court advantage in college basketball? How many points is it worth? Well, I have some bad news for you: It is really impossible to answer these questions with much precision. We can look at the box score of a game and see how many points a team won by or how many points each player scored.
Maybe one could do a little more fancy maths and make an estimate on what each player contributed to the point total. But we can’t tell how many points the home court contributed. Maybe someday, but for the most part people don’t care as much about the value of home-court as they do about the value of a star player. So that’s where the research focus lies.
But it’s a fun analytic exercise to try determine the value of home-court. I’ve been thinking about it for a while, starting a few years ago I wrote this article about home-court advantage at ESPN. The analysis was rather primitive and I apologize to everyone for paying their hard-earned Insider money for that. Basically, it identified the teams that had the largest difference between their home and road scoring-margin in conference play. Those teams probably had the largest home-court advantage, right?
For instance, from 2002-2009, Texas Tech outscored its opponents by an average of 6.0 points in conference games in Lubbock and were outscored by 9.3 points per conference game on the road. The difference of 15.3 suggests a home-court advantage of 7.6 points per game relative to a neutral court. That figure was the highest among 320 teams that played at least 50 conference games over that time. (more…)
There is one final obligation to tend to before the books can be closed on the 2017 season and that is the revealing of the 2017 kenpom.com player of the year. This season’s honoree is Villanova’s Josh Hart who closed out a distinguished four-year run at Villanova in style.
This season, Hart posted a 122 offensive rating with a 27% usage rate, which included making 40% of his three’s and 58% of his two’s. He had just five games this season where his offensive rating was below 100. Along the way to leading Villanova to its fourth-consecutive solo Big East regular season title, he also posted career-highs in steal rate and assist rate. (more…)
There were 1,930 games between Division-I teams in February and March. Here are the wildest things that happened during the months:
[Tournament games only] March 2: #305 Campbell 81, #98 UNC Asheville 79 (OT) (11.7%) The biggest upset of single-elimination basketball was the seven-seed over the two-seed in Big South quarterfinal action. But the story of this game was Campbell’s Chris Clemons. He’s 5-9, occasionally dunks, and scored 51 points in this game. He led the nation by taking 42% of his team’s shots when on the floor, and he used an incredible 53% of the Camels’ possessions in this game. Normally, that’s the hallmark of a senior going rogue in some sort of Kobe-like farewell game.
But Clemons is a sophomore and was quite efficient in this performance, committing just three turnovers while going 18-of-32 from the field which included eight 3-pointers. Campbell’s eventually appearance in the Big South title game was the second least-likely team accomplishment in conference tournament action, given a 3.6% chance of happening before action began.
2. February 27: #323 Savannah State 74, #141 N.C. Central 73 (5.9%) I’m not sure what’s in store for Savannah State next season, but in the Tigers’ penultimate game of the season, they recorded their most impressive performance, a road win against the regular-season champs of the MEAC. Weirdly, it was the most conventional game the team played all season. They took an unremarkable number of 3’s (26) and played just 70 possessions of basketball, equalling their slowest-paced game of the season. (more…)
Hey coaches, my offer for scheduling help is back for a second season. I can provide you a spreadsheet with a first cut of expected ratings for every team next season which will include a list of all 351 D-I teams with overall ranking, and national rankings in offense, defense, and tempo. It will be updated every two weeks with the latest roster changes around the country. The projections are mostly based on the quality of returning players, transfers, and previous team performance, but there is more discussion about the ingredients here.
In addition, for those of you that are into gaming the RPI, teams are also ranked by projected conference winning percentage so you can identify opponents which may rack up plenty of wins but are relatively easy to defeat. This data can assist you in finding the type and quality of opponent you are looking for to fill out your schedule for the upcoming season and will not be shared with anyone outside the coaching community until October. Please send a message to ratings at kenpom dot com and we can discuss terms.