***UPDATE*** I take back everything I said!!

I am compelled to go where the data takes me. One of the reasons I spent all the money to buy the game data back to 1950 was so I could use my DPPI to calculate probabilities (very accurate) vs. using things like seeding (not very accurate). In all my previous NCAA projections I used probabilities based on curve fits to actual results by seed since 1985. In other words, I looked at the winning percentages of each seed against other seeds and tried to find a reasonable correlation. There never was a very good correlation to any seed other than the No. 1 vs. No. 16. All other correlations were unsatisfyingly poor. Then then las couple years I tried to adjust those correlations based on actual DPPI probabilities of the actual matchups in the NCAA. It verified my previous attempts to build correlations. At the root of the problem was, the seeds are not always well correlated to my DPPI. In other words, very often in my opinion teams were mis-seeded and sometimes badly. This season for instance, I have Kansas as the 13th ranked team and USC currently as the 7th ranked team. Yet KU received a 3 seed whereas USC received a 6 seed. Thus, based on the forced correlations, KU was given a 62% probability of winning whereas based on my DPPI, KU had only a 45% probability of winning. This is one game. When you multiply this across 17 seasons for Coach Self, if can add up to a significant difference IF (and this is a big IF) KU is consistently seeded higher than they should OR ( and this is a big OR) KU is matched against teams who are consistently seeded lower than they should. Remember, this is a zero sum game. For every team that gets the benefit of a tougher opponent than projected another team is getting an equally weaker opponent. Probabilities for every game always equal 1 for the two teams.

Well, and this is the shocking blockbuster hot off the presses (I mean I have just finished vetting the data and running all the seasons) guess what? KU has been generally seeded slightly higher than they should have been and overall, KU has faced opponents that were seeded slightly (on average) lower than they should have been. The net result is that when you calculate the probabilities of every game based on DPPI (highly accurate) and not on some force-correlated probability (not very accurate) the numbers TELL A DIFFERENT STORY!! And this different story is leaving me with egg on my face but I am a man of stochastic and mathematical integrity. I will stick with what the data says. Here is the new headline.

Coach William Eugene Self Jr. is NOT and underachiever in the NCAA tourney.

So for all of you who have taken issue with me on this claim, you were right and I was wrong. In my weak defense, I have averred the weakness of my method, that weakness being that NCAA seeding is not based on actual team strength. At least not totally and often not even close. Plus, the difference in the top No. 1 seed and the 4the best team who would also be a No. 1 seed can be pretty significant. For instance, in my previous method, a No. 1 seed versus any other No 1 seed was a 50-50 probability. But this year, If Michigan were to play Gonzaga, Michigan would be something like a 40% probability based on the DPPI.

So without further adieu, what does the data say. Well it is great news and great news if you are a KU fan and a Bill Self fan. First, also in my defense, I have been Coach Self's biggest fan in terms of regular season. Well you will be happy to know that based on my hot off the presses analysis (and this is groundbreaking and I challenge Ken-Pom and the other gurus to match my work here because as of yet, none have) Bill Self is easily the winningest coach in the regular season since 2004. If you look at the DPPI projection, Self should have won 408 games. Yet he has won 477 games. That is 69 games better than the probability. Or looked at as a percentage, Bill Self has won 117% of the games he was projected to win in the regular season. The 69 games and the 117% are both tops of any school during that time. Not only that, but in the NCAA tourney, Self has won 39 games while projected to win 33.9 games. That is 5.1 games better than projected or about 115% which is very close to the regular season. So Bill Self has not underachieved in the NCAA. In fact, he has overachieved. He has simply been the victim of his regular season success receiving seeds slightly higher than he deserved (based on DPPI rating) and has faced opponents that, on average, have been better than their seed.

I have not yet added in the coaches to my database but I will. Currently I just have run this by school. I will break it down to coaches as well. But you may be interested to know that Coach K has won 103% of the NCAA tourney games he should have. That is good but not as good as Self. Boeheim is 140%. The leader with 6 or more wins in Porter Moser at Loyola (IL) at 173%, Roy Williams is 125% at UNC. But he is only 112% in the regular season which is awfully good but below Self by 21 games. I will attach the link to the table. I'm not sure how it will appear inn the post but if you click on the table you can download it if you like. (as a png).

Corpus - I appreciate all the effort and resources expended to break down the data.

I find it hard to believe that in 2011, a #1 seeded KU team and #11 seeded VCU can be explained by over seeding or under seeding.

Five of Self's Tournament losses came against schools from what is now the FCS division of NCAA Division I (Bucknell, 2005; Bradley, 2006; Northern Iowa, 2010; Virginia Commonwealth, 2011; Wichita State, 2015).

Self has lost four times to double-digit seeds (#14 Bucknell, 2005; #13 Bradley, 2006; #11 VCU, 2011; #10 Stanford, 2014).

Self's teams have suffered double-digit defeats in their NCAA Tournament losses in five of the past six tournaments, going back to 2015. The only exception was the 2016 Elite Eight loss to Villanova, which was a five-point loss. That means Self's teams have lost by double digits four years in a row and if you look at the opponents shooting numbers, 4 of those losses in 2017,2018, 2019 and this year rank among the top 10 opponent’s shooting percentage in a game all time against Kansas ( Jesse Newell tweeted this out).. So HCBS have won more games in the tourney than expected but, when his team lose it is often when they should not. I am not saying KU should have won Monday but they sure as heck should not be losing a game by 34 pts...

Further, do these numbers account for injuries or other variables. Everyone cites the 2012 run to the title game as an example of Self delivering in March, let’s not forget the injury to uNc’s stud point guard and how KU caught and took advantage of a hug break..We no all to well the flip side of that in 1997 or 2014 with Embiid...

Only VCU made the FF, Bradley I thought made it to the S16.

Hey Konza, got ya covered. I downloaded all the coaches back to 1950 so I could do the analysis by coach and not just by school. Below is the list of the 61 currently active head coaches who win at least 105% of their probable projected wins. Let me explain. When you project the outcome of every game (based on the DPPI) that will by definition balance out over the course of all games in a season. But some teams (coaches) win more than the probabilistic projection and some win less. There are several reasons for this and all the reasons for winning more are indications of better coaching. So the bottom line is, the best coaches will, by definition, win more than their probable projection. That is obviously born out in the data. The table below is for the period 2004 through present. There are three result metrics in this table.

1) Win% -- A correlation to coaching but also a correlation to recruiting and the school one coaches at.
2) +/- to Probable -- A direct relation to coaching but also the number of games coached
3) +/- % -- Or the percent of probable games won. This is a direct measurement of coaching effectiveness. The drawback is any coach can have a particularly good year so coaches with more games will be more impressive with the same %. In other words, Bill Self has won 109.5% of the probable games and has done that over 18 years and 634 games. Stan Johnson coach at Loyola Marymount, has won 109.4% of projected games but has only coached 22 games. That makes Self's record much more impressive since he has coached about 30 times as many games. I doubt Stan Johnson will have anything close to 109.5% when (and if) he gets to 634 games.

Let me clearly state, using this analysis, there is no one that can touch Self's record. Here is where Self stands in the three metrics.

Win% -- 81.4% -- Only Mark Few is better (84.4% but with a vastly lower SOS. You asked about Jay Wright. He has a win% of 74.1%.

+/- -- Self is has far and away the highes +/- of 44.8. Few is 2nd at 34.2 and Jay Wright is 3rd at 32.1. In 7th place, just behind Coach K in 6th, is Mark Turgeon with 29.6.

% -- 7 coaches have a higher probable win% than Self. But none has coached near the number of games Self has. As a matter of fact, all 7 combined have coached only 442 games to Self's 634. This list does have some names that bear watching, but all have coached for only a few years. Only 2 have had more than 2 years as head coach.

Bill Self has never had a season where his probable win% was less than 100%.

He is not on this list below because he is not active in the NCAA, but the only coach who can lay claim to a CV close to Self is Brad Stevens. His CV was quite impressive.

FYI, the shaded coaches are those with a 70% or greater actual win% and a 105 or greater probable %.

Also, there are a few coaches who I seriously raised my opinion of based on this analysis. I think Ed Cooley at Providence is way under the radar. I think Bruce Pearl (Auburn), Chis Mack (Louisville), Mark Turgeon (Maryland) and Dana Altman (Oregon) are also under the radar a bit. And watch out for David Richman at North Dakota St. and Terrence Johnson at Texas State. I would bet money their names will be more in the news for some big jobs in the next couple years.

Konza, sent you a spreadsheet with all coaching records 2004-present, 1950-present and NCAA only 1950-present.

Here is the really good news. Doing all coaches 1950 to present, the top 25 is literally a list of HOF coaches in terms of total +/- to probability. No. 1 on the list for total +/- is Beoheim. But bear in mind, he has coached since 1827 or something. He has coached in 1481 games. His +/- is 68.8/ I don't know if anyone will ever catch that. Dean Smith is 2nd with 59.1. Our own Bill Self is 3rd at 49.8. Here is the total list from 1950 forward. Pretty impressive. Jay Wright is 18th, between Tom Izzo and Eddie Sutton. Coach K, who has coached 187 more games than Self is in 5th at 46.4. This is really quite astounding for Self. Of the 2243 coaches who have coached since 1950, 128 have a better probabilistic % than Self. But of those 128, only 25 have won 100 games or more. And of those 25 only 4 are active coaches, Ed Cooley, David Richman, Will Wade and Bruce Pearl. And of those 25, only 3 have a better win% than Self, Brad Stevens (Butler), Al McGuire (Marquette) , and John Kreese (College of Charleston).



When you limit to NCAA games only, Self is not quite in the stratosphere. His % in NCAA is 105.1 compared to 107.5 total and 109.5 at Kansas. Jay Wright actually has a lower win% in NCAA than Self, 68.2% to 70.0%. But Jay's probabilistic % is much better 113.8% to 105.1%. Below is the table for coaches who have coached in at least 30 games in the NCAA since 1950. Poor Rick Barnes is unquestionably and not arguably the worst NCAA Tournament coach in history.