Probabilities, stochastics and Kentucky

It pretty much sucks that Kentucky dominated us in Allen Fieldhouse. I hate losing at home. I hate losing to Kentucky. I hate getting blown out. I hate losing period. But that is life in college basketball. As you know I have run my DPPI back to 1950. So I now have the luxury of a much deeper stochastic data set than just about anyone. Wen you look at this game in terms of probability you find that a game like this is going to happen about once per season to every team. That is not a law it is simply the real probability. . I know that games like the Kentucky game elicit a number of questions about this team. And those are legitimate questions. But if it is any consolation, my DPPI as well as the other 50 or so computer power rankings measure a team based on their entire corpus of work. One game is part of that corpus but only a part. It is human nature to be hyper aware of the worst news and discount the best news. Remember the 2nd half against West Virginia. KU just completely and absolutely dominated in every way in that 20 minutes of basketball. Self even said at the time "that wasn't real." Well it was "real" it was just not "really indicative" of how the team would and should do night-in and night-out. That was the "best of" the Jayhawks. The Kentucky game was the "best of" the Kentucky Wildcats and probably close to the "worst of" the Jayhawks. It was a perfect storm. I can definitively tell you based on my own extensive evaluation not to mention my 50+ other computer quant compadres that that game was indicative only of what happens when one team is very close to their peak and the other close to their nadir. That will statistically not happen again. That game was almost 3 standard deviations out of the projection. Out of the last 515,836 college basketball games, that game ranks 19,737 in standard deviation. Okay, that was useless information. What does that really mean? That means this kind of game happens about once every 26 games. If you stick to 2022 it ranks 341st in 8836 games or 1 in every 26 games. So this is something that happens in college basketball. Unfortunately it happened to us in a game against Kentucky. So does it portend of something ominous that was heretofore unseen? I would say emphatically no. It is truly a probability thing. It would be nice to avoid those pesky probabilities but that is not how it works. So fear not. The boys came back to beat Iowa St. by 8 and led by 14 late. I don't think the Kentucky loss revealed any hidden weaknesses. I think it simply showed weaknesses we knew were there. But this team is the same team that dismantled West Virginia. So that probability cuts bot ways. It would not be freaky weird if KU defeated Kentucky by 10 in the NCAA tourney. In fact, that would be reasonable. So don't get too down about the Kentucky game. Below are the trend charts for Kentucky and Kansas. Kentucky has an interesting trend chart. They have had three blowout games where they played above their heads. They are North Carolina, Tennessee and Kansas. Notice a trend. These guys seem to get up for Big games. Conversely, KU's loss was their worst game of the season. It was a perfect storm against Kansas.

Konza, couple things First the easiest.


I am with you on Remy. Remy brings something to the table that is unique and valuable. I do believe our absolute ceiling includes a healthy fully integrated Remy. If we do not get that we still have a very high ceiling but Remy does pose the possibility of the highest ceiling. Getting him healthy is one thing but getting him fully integrated is another. I saw signs of that happening when he got hurt. Remy is unique and I think integrating him tested Self's formula. But Self is not a HOF coach for nothing and he would (will) figure it out). Here's rooting for a fully on Remy soon.

As for the "matchup" issue, trust me, it is fully integrated into the analysis. Well I would aver that it is virtually fully integrated. One of he beauties of a stochastic evaluation is that it can point out areas that need to be evaluated and hopefully compensated for. So all of the mature algorithms have done this over time. I, for instance, do some sort of modification to my DPPI every year based on measurable correlations I have learned from evaluating past seasons. Certainly "matchup" is an obvious issue that is close to the top of the list begging for compensatory correlation. Injury frequency is an example of another issue. Obviously basics like home-away-neutral or performance against top rated teams versus bottom rated teams or performance when the difference between the ratings of the two teams is small or large or correlations to good defensive teams or good offensive teams or even teams that have a larger delta between their offense and defense. As the data set grows (and my data set is huge going back to 1950 and comprised of over 250,000 games) you can really do some real serious data crunching looking for meaningful correlations (meaningful meaning predictable within a tolerable deviation). When read things like standard deviation and consistency it is meaningful within the context of the veracity or accuracy of all of these various correlations. So for instance, is it possible that KU and UK offer a unique matchup that inordinately favors UK? Of course. But the question becomes how unique. If it is so unique that the advantage that UK has over KU does not exist over any other school then it is not predictable because it is correlation of 1 point which is non-stochastic meaning it has no veracity. But does the advantage UK has over KU exist in other matchups? Well there is the magic of the stochastic approach and the search for meaningful correlations. The one things stochastics does not do and in fact cannot do is determine and actual cause. All it can do is measure the likelihood of a cause effect relationship between two variables. It is blind to the cause. But that is where the you and I come in. We hypothesize on potential cause effect relationships and we test the reasonableness with the stochastics. Then we seek to find a "meaningful" (that word keeps coming up because it is of paramount importance) correlation. So I assure you. that the "matchup" issue has been attempted to be correlated in a number of ways. And thus we get a test of reasonableness with the standard deviation. So when we see a game like Kentucky that was close to three standard deviations we can safely say that it violated the incorporated correlations beyond their predictable level meaning there are either other correlations that we need to find or just accept the fact that there will always be a certain amount of randomness in the data. Is KU a nightmare matchup for Baylor? Probably not. Games like that just happen with a certain frequency of randomness. We can predict the frequency but we cannot predict which game will be that data point. We simply know that that sort of random occurrence happens once every 26 games or so. Ideally we can isolate every variable and develop the ability to collect the proper data and predict with ever increasing accuracy. But no one, not KenPom, not Sagarin, not me have been able to reduce the randomness to zero. And the simple reason is, there is randomness in the universe and there is randomness in college basketball. True it is that everything is ideally predictable if you had infinite data but we do not have infinite data. And even with infinite data (things such as the mood of every player pregame, the humidty, the temperature in the arena and how that effects the stamina of every athlete etc are examples of the infinite data, these and a million more data points) randomness still will be with us through in-game injuries, balls bouncing one way versus another, bad calls by the refs, and a million other things that will never be predictable. So all this droning explanation to say, I agree with you that matchup is a valid and real factor and variable but it is one that has been correlated out the gazoo in many ways with the available data and is incorporated into the prognostication.