initial Big 12 projection

Just to provide some variety, Asteroid uses the Sagarin algo, this is from the DPPI. Sort of similar to last season only KU was in 3rd with an 18.4% probability of winning. Let's hope this season goes similarly.

Why such a large discrepancy do yo think?

Tech's got a good record, but they haven't beaten a single top 25 team, Aside from losing to Duke the only other team of any quality on their schedule thus far is Nebraska. I think their SOS is somewhere around 340 out of 353 teams in D-1.

Meanwhile, we're 3-0 vs the top 25, 4-1 vs. the top 50. Wins include MSU (#10), Tenn (#5), Nova (#17), with the loss to #18 ASU.

Racking up 50 more points than Incarnate Word, or blasting Abilene Christian by 34 might look good on the stat line but it would seem to me, at least , that the term "quality wins" and Texas Tech shouldn't be used in the same sentence when describing Tech's non-con schedule.

Not-O, you may end up being correct. But the whole idea of a computer algorithm is to find the magic sauce that allows you to compare teams that play different schedules. First of all, as I've said several times, the human factor trumps all predictors. On any given night, the human factor will make a team 8-10 points better or worse based on so many things. The idea of the algorithm is to find trends that are meaningful, repeatable and thus predictable within the confines of the data available. I use only scoring margin. This computers that use box scores have far more data and can derive meaningful trends that can in my data set is noise. They can diminish the signal to noise ratio to where the human factor may only be 5-6 points one way or the other.

So to answer your question, Texas Tech has actually played the 257th ranked schedule. But that is a skewed way to look at it. The average normalized loss factor of their opponent is 71.50. KU's SOS is ranked 3rd but the normalized loss factor is 78.46. That sort of equates to the point differential of the opponents. So KU has played a schedule that is on average about 7 points better. (it is more complicated than that but that is a close assumption). Yet Texas Tech is winning by 23.1 points on average whereas KU is winning by 11.8 points. So KU has a scoring margin that is 11.3 points less than Texas Tech against competition that is only 7 points worse. Now without going into the gory details suffice it to say that there are adjustments for blowouts and for teams trending one way or another and another couple adjustments. When all is said and done, Texas Tech rates out at 87.38 and Kansas rates out at 85.34.

Now to complicate matters even more, when you calculate the probability of a win, I use the consistency factor. This, along with the mental toughness factor can have a huge impact in the conference race. The consistency factor goes something like this. If a team is very consistent, never playing much above or below their projection, the probabilities are going to have a high peak at the mean. Another way of saying this is, for any given scoring margin projection, a consistent team will have either a higher chance of winning or losing depending on the projected margin. Say KU has a consistency factor of 8.44 (which they do). Say they are projected to win a game by 8 points as in Oklahoma. Their probability of winning is 86.7%. But a team with a consistency factor of 4 (very consistent) might have a probability of winning of 96%. West Virginia has the best consistency factor in the Big 12.

The other factor that I do question but have not found a better algorithm yet, is the mental toughness. Simply stated, that means a team plays better against tougher opponents. Texas Tech thus far has the 3rd highest mental toughness behind Kansas St. and TCU. Kansas is 4th. Oklahoma is last. This will skew the conference games since all Big 12 teams are in the top 1/3rd and most are in the top 50.

Last season my DPPI was on average 1.5 games off for the 10 teams. For the last two years, KU has been about 1 game better than projected. That means KU has tended to improve more than the average throughout the conference season. That is amazing given theinjuries we have had. But right now, the 12 game non-conference season lends itself to favor Texas Tech at least in the DPPI. Asteroid uses Sagarin and KU is favored. I like that projection more. But I do not do the DPPI to make KU look good. I project all 12,000+ games and I do the DPPI to bring the lowest overall standard deviation of projection to all games.

Thanks for the explanation, Corpus and I don't doubt your methods. It just seems to me that on the surface, Tech hasn't been tested nearly as stringently as KU has.

Nevertheless, both teams are playing tonight, which takes us out of the "theoretical" and into the "practical".

KU will be favored over OU, but OU is a top 25 team and nothing can be take for granted.

Tech also has their hands full. While WVU isn't as good as they've been in the past, I highly doubt that Huggins' team will just lie down and let anybody come in to beat them in Morgantown, and to add insult to Tech's weak schedule, this is their first true road game.

It wouldn't shock me at all if KU wins tonight and Tech loses on the road, which I'm sure would change things dramatically. It also wouldn't shock me (but would likely infuriate me) if Tech wins tonight and KU loses it's home opener in conference.

Suffice it to say, I'll be paying attention to what's happening in West Virginia as I begin my pre-game rituals. The outcome there should be known right around the time we tip off.

I have read numerous comments over the years along the lines of a weak schedule should be punished arbitrarily for some reason or other. There seems to be this assumption that a weak schedule means the team is weaker. That is the whole idea of an algorithm, to try to find how you can connect teams that play not only different schedules but oftentimes different strength of schedules. Here's a hypothetical. Say Duke, the No. 1 team this year,for some strange reason played the 353rd weakest schedule. Maybe they signed some contracts that just forced them into that. It is certainly arguable that such a schedule doesn't prepare them for tougher teams down the road, but should they be penalized simply because they were fulfilling contracts to play mediocre or poor teams? That is the magic that all of us quants try to figure out. How can I filter through the noises and the limited data variance to get to a true measure of a teams ability to perform going forward. The fact that Texas Tech plays a weaker schedule in and of itself means nothing as to their team strength. It does start to mean something in projecting against future competition but that is a very problematic correlation at best. The bottom line is that Texas Tech has beat just about everyone by more anyone else is beating them (on average). It is hard to punish them simply because the opponent is mediocre or poor.

And that's where I think the data can get screwy. If you're up 20 with 5 minutes to play, you could theoretically win by 25 or 30. But maybe you put in your bench and win by 15. Or maybe the coach puts in some arbitrary point of emphasis as "practice" ("ball must go inside every time" or "3 passes before anybody shoots", etc.). It's an artificial manipulation of the data.

Or what about KU vs EMU. Lots of bench guys got lots of minutes. Maybe we would have won by more if we'd played the starters rather than doling out minutes.

Point being, in cupcake blowouts, the data may or may not be as accurate a measure than it is in games decided by single digits.

I don't see SOS as a penalty as much as it is a factor for judging how accurate the data might be. Tech's 340 rank (current sagarin SOS) is putrid and, at least as I see it, put some of the data in question.

Finally, we need to look at what the data is and is not. I know I fall into the trap of using it to gauge who is better than who (which it is, kind of). In reality, it's probably better to look at it one game at a time, rather than projecting 16 games into the future which introduces all kinds of improbability.

NotO, you have to think that if you have a theory on a data validity or trend issue, than we AI guys have not only had that same theory, but we have exhaustively worked through seasons of data on how to correlate that data. The issue you cited of late game relaxation in blowouts is a simple AI issue. Considering there are over 12,000 games per year and a large number of them are large delta games (meaning large projected margins), working this into the algorithm is not a very challenging task. This is actually a pretty straightforward correlation. There are far more difficult correlations that stretch credulity. My DPPI has a subroutine exactly for this large delta, late-game situation. Conversely, there is a small delta, issue that arises when a tight game results in a protracted FT shooting contest down the stretch in the last minute or two. These are all fairly easy for the AI algorithm to decipher with a reasonable correlation. I ask that you give credit that if you have an awareness of these reasonable correlations that the weighty guys like Sagarin from MIT and other smart guys can figure it out as well. Maybe an AI hobbyist from a mid-level academic institution like the University of Kansas is not up to the task like myself, but you have to think that guys who went to real academic institutions and have real backers like USA Today or ESPN can figure this stuff out.