Unofficial Big 12 Projection (we have a real challenger)

I will not do my official pre-conference projections until the last non-conference games are completed. But I thought it interesting that Texas Tech has hung in there. I thought they would fade a little by now but they have not. They are a real threat. No other apparent threat at this point. For a frame of reference on how accurate this could be, last season my pre-season projections averaged 1.6 games off. If you exclude K-State (4.5) and Oklahoma (4.2) the average was only 0.9 games off. I projected KU to win 11.9 and they won 13. This can clearly change between now and Jan 2. I also included last seasons win projection by round. Remember that amazing run from game 13 through 16. We won the Big 12 with that 4 game run. After 5 conference games last we were projected in 4th place. What a great conference season last season was.

No disrespect intended, Corpus, but I have to be a little skeptical about an analysis that has the 10-0, #1 ranked Jayhawks as underdogs to Texas Tech to win the conference championship. Are you certain that the bottom two bars in your histogram are correctly labeled?

Yes I am sure.

Our SOS (via sagarin) is #7. Tech's is #328. There are only 353 teams in D1. So we're in the upper 2% of D1 in terms of SOS and are undefeated. They've got a great record (and I don't know who can beat Duke right now) but they're SOS is in the BOTTOM 8% of all D1. In other words, 92% of D1 has played a tougher slate. In fact only 1 other team in the top 50 (NC State, SOS #345) has played a weaker schedule. Frankly, I'd be embarrassed if I were a Tech fan.

They're 1-1 vs. the top 25 (and top 50). We're 4-0 vs. the top 25 and 5-0 vs. the top 50. If we win tonight, we can get our 6th vs. the top 50. But boy howdy did they put a whuppin' on Incarnate Word, Miss Valley State, Ark Pine Bluff, Northwestern State (not "Northwestern" and Abilene Christian.

So while I don't doubt Corpus' stats, I think we've proven (so far) to be the better team. I certainly don't see Tech beating us twice in conference. That means it comes down to who you play. Our non-con schedule has been tougher to date and is tougher going forward.

In addition to the Rio Grande Vaqueros (#228 in Sagarin) they play Arkansas in the Big 12/SEC Challenge. We've got @ASU, Eastern Michigan, and @UK to round out our non-con schedule.

While the non-con schedule doesn't mean anything for the Big 12 race, we've certainly put a whole lot more "at risk" than they have, which makes me wonder how challenged they've actually been to date.

Without wanting to be critical of Corpus' analysis, I'm wondering if Tech's non-con SOS is SO out of whack that it's throwing off the projection formulas.

In any case, I welcome the competition and hope that Tech and KU are seeing the same numbers and that Tech is confident that the game they brought to Incarnate Word (#342) and Northwestern State (#344) is all they'll need to take KU down in Lawrence this year. I also hope our guys are using the same data as motivation to show Tech what real D1 teams play like.

By way of explanation and just to provide a frame of reference (in other words, nerd alert), here is some information on the SOS and other calculations.

My DPPI has KU with the 2nd toughest SOS (actual SOS rating is 79.2) and Texas Tech has the 256th SOS (actual rating 70.9). That means that KU's average opponent is 8.3 points better than Texas Tech's average opponent. Now KU is winning against this schedule by a average of 12.2 points whereas Texas Tech is winning by an average of 22.9 So Texas Tech has a scoring margin of 10.7 points more than KU against opponents that are only 8.3 points worse. Thus, Texas Tech is actually 2.4 points better than Kansas.

Okay, that is the basic skeleton of the algorithm but it is far more complicated. These relationships are not linear and there are a number of single and multi-variable adjustments within the relationships. For instance, there is an acceptable range (1 standard deviation) where there is no adjustment for a single game scoring margin. Any scoring margin outside of 1 standard deviation has a diminishing impact adjustment. For example, if KU is projected to beat ASU by 5 and by a stroke of wonderful fortune we win by 20, we would not get the benefit of the full 15 points above projection because it is outside the standard deviation (which is +/- 8 points more or less). So beating ASU by anything more than 13 points would trigger the diminishing impact adjustment. Gonzaga beat Denver by 61 last night. The projected margin was 46.1. Add in the STD of 8 and that is 54. So they will not get the full credit for the 14.9 point margin delta.

That is one of a number of adjustments. There is a mental toughness adjustment, a diminishing time adjustment (recent games count more than older games) as examples. The weak schedule issue is a real problem early in the season when the teams are not fully connected. As of the last week, all teams are now statistically connected. What this means generically is that the relationship adjustments that are made are based on statistically less valid data early on before connection, but once connected, adjusting for SOS becomes statistically valid.

Okay, I said this was a nerd alert post. I am very proud of my DPPI. I have followed guys like Sagarin, Massey, KenPom and others for years. I do not know all the gory details of their algorithms and certainly their cred is established. But the DPPI has been part of Massey's composite for three years now and has held it's own with these other guys. I am limited in what I can do mostly only because I don't subscribe to the highly detailed data that some of them have access to. I could do far more algorithmic adjustments, for instance, if I had the box scores for every game. I do not get the box scores for every game so I can't do the micro adjustments that come from data that is only available at the box score level. But having talked to several of them I have compared for the last two years my STD versus theirs on all 12,000 games played and my STD is in the pack of the top algorithms (meaning my game score projection accuracy is right in line with the better programs.)

In the final analysis (pun intended!), the utility of a model is determined by its ability to make robust, accurate predictions. I'm going to print this histogram up and put it on my bulletin board. I hope you don't mind me pulling your chain a bit in March when the Jayhawks win their 15th consecutive title going away.

I do my best to help people to understand what statistical projection is and what it isn't. I statistical projection by definition has a mean projection surrounded by a confidence range. Most people don't really understand statistical analysis so to make it more palatable I present only the mean projection. The real power and value of a statistical analysis you need to understand the nature of the bounded confidence range (defined in various ways, the simplest of which is a standard deviation that corresponds to the mean). Again when presenting to the casual observer, it is sensible to limit the analysis to the mean only (a discrete singular answer -- often referred to as most likely outcome). The last two years, the pre-season projection has had a standard deviation of 1.3 games more or less. So essentially if I project Kansas to win 13.2 games, anything from from 11.9 to 14.5 is really what the analysis says.

Got it. So, if you construct confidence intervals around the means for KU ant Tech, is the difference in those means statistically significant?

If you played the Big 12 season 100 times, there is a 90+% confidence that KU would win the conference outright 21 times and tie for the title another 14 or so times. The problem that while my computer can play the season on paper 1 million or more times, we only play it on the court once. So how do we characterize the analysis of who will win that one time. That is where the nuance of statistical analysis comes in. And it really only matters if you are putting money on a set of possible transactions you would do the analysis to optimize return based on intersecting confidence ranges. For our purposes (which is entertainment only) I provide the mean projection. Then I will characterize it like I did by saying that the last year or the last two years the average projection has been off by 1.5 games or something. So one way to look at it is to just look at the entire range which is to say KU will win 12.1 to 15.1 games, Texas Tech will win 13.3 to 16.3 games and TCU will win 10.1 to 13.1 games. That is a high confidence range (it varies but for the sake of brevity use 90%) . That would say that there is a pretty high confidence that only KU or Texas Tech will likely win the Big 12. Of course, this will possibly change a bit by 1/2/2019 right before we tip-off the 1st game. I put this one out there just to show how it can change early on. Feel free to debate or pull my chain. No harm no foul. All is meant for enjoyment so all is good.

> I will not do my official pre-conference projections until the last non-conference games are completed.

But then you'd have to wait until after January 26, when we play Kentucky!

Obviously the pre-conference non-conference games.