UConn's KenPom ranking 2-02

[DRBackup]

It doesn't have to be nefarious to be a glitch in the model.

I work with a lot of models in my job, and most of them have implicit guardrails that tend to influence the model's interpretation of new data, i.e. the model's starting point never gets completely removed from the model. I also find that models frequently equally weight data, including outliers. I imagine this is unavoidable with a model based on per possession efficiency, which is fine for measuring offense and defense, but can skew outcomes when looking at the team as a whole.

By focusing on efficiency, KenPom has to be leaving the margin uncapped, even if the model is adjusting for quality of opponents. There is no other way to incorporate efficiency with a capped margin because how would the model know which possessions to throw out?

I'm not sure I'd call what KenPom does a "model". The only thing "subjective" that he does is adjust for opponent and tempo. You could call that a "model", I suppose, but I imagine he has a literal ton of empirical data to back that up. I'd be curious how much these adjustments even affect the overall ratings, honestly.

Our defense has not been impressive this year. Combine that with a lot of teams that played very well on offense, keyed up to play the 2 time defending champs.

 

[Get a Job]

Some dude once said, "All models are wrong, but some are useful."

 

[auror]

That said I’m not sure I’d call KP predictive as it’s basing its metric on historical data. It’s simply a measurement that can imply trend, and quality of play against a variable schedule, if you want to call that predictive. NET is simply measuring quality of resume more than quality of play, which is then used to evaluate body of work for seeding purposes. The difference between the two is valuing how a team wins and loses.

There are two types of college basketball team metrics and they are usually referred to as "predictive", aka future facing. Or "resume", backwards facing. They both look at what a team has done, but 1 uses that to model how the team will perform in the future, and the other just looks at how the team has done relative to others. So obviously the models will be variously good at doing the predicting, but that's the umbrella term they fall into.

Popular metrics:

Predictive:
KenPom, T-Rank, Miya Relative Rankings, BPI, Haslametrics

Resume:
SOR, KPI, Torvik and NET WAB, the new Miya Resume Quality, Haslametrics "Deserves"

The NET is a blend of both, but primarily predictive. Based on comparing it to other models, I look at it as predictive with a bonus for teams that have a lot of raw wins.

 

[Orange]

UConn is only 35 as of right now, with Gonzaga still playing. There are 14 teams with 6 or more losses ahead of UConn in KenPom, including 3 that UConn beat head to head. What is driving UConn's bad ranking? Is it the 3 games against teams ranked in the bottom 10 of D1?

Unlike BPI, Pomeroy completely removes preseason projection stuff from his rankings around the time that conference play begins. That's part of why BPI always rates us considerably better than KenPom does. We're better than how we've actually played so far this year.

 

[nelsonmuntz]

The only reason we’re interacting in this thread is you replied to one of my posts which was addressed to a completely different individual. Even the post you replied to here was not addressed to you, so maybe you should stay out of the conversations I'm having with other posters. You should stick to going into the “other NCAA games” threads and complaining about how people are discussing other NCAA games, that’s more your strength. Things involving math just aren’t for you in this life.

You are the neediest poster on this site.

 

[BigErnMcCracken]

You are the neediest poster on this site.

If you don’t know, then shut up. There are posters that really understand this stuff, and there are posters like you, who just just like to take cheap shots at other posters.

 

[superjohn]

This thread is worth it to watch Waylon dunk all over himself, it never gets old.

We get the Waylon trifecta here.

Dunks all over himself while being a clueless jerk, plays the victim, and then accuses others of exhibiting behavior he exhibits here daily.