Around the league- some quick comments

[v-zero]

If there's interest I could write code to compute playoff probabilities. No need to run simulations, as it's easy enough to compute exact probabilities under the model assumptions.

-Chris

I'd certainly be interested to know how you'd go about doing it.

[DSMok1]

If there's interest I could write code to compute playoff probabilities. No need to run simulations, as it's easy enough to compute exact probabilities under the model assumptions.

-Chris

I have an Excel spreadsheet floating around that does this. Somewhere...

[talkingpractice]

If there's interest I could write code to compute playoff probabilities. No need to run simulations, as it's easy enough to compute exact probabilities under the model assumptions.

-Chris

You have MIA 6th in team strength, and 2nd in the East. So when you say "under the model assumptions", are you implying that under your model/assumptions, the Heat would be at best the 2nd most likely team in the East to win the Title, and likely only the 3rd or 4th most likely team in the NBA to win it?

How do you deal with the DRose situation in this type of model?

[Crow]

Hollinger lower on title chances for Knicks and Heat than BR. Way lower on Heat. (But his team win predictions are currently last in the contest here.)

His (ESPNs) odds for the Griz are half of BRs.

BR has 6 teams with 10+% chance at title, Hollinger has 3.

[talkingpractice]

I guess to be fair, though the ranking/probabilities are all wrong, the Betfair lines do have the correct 8 teams in their #1 through #8. ie the 6 "good" teams, plus LAL and CHI (who may be just Nash or D-Rose away from being good teams).

[Eternal]

How do you deal with the DRose situation in this type of model?

Models like these assume constant team strength subject to random fluctuations. If this is no longer true (Rose recovers from his ACL injury), you have to revise the model. It'd be easy enough to add an indicator for Rose playing/not playing, for example. If you add indicators for every player and break the data down by possession, you have a more sophisticated version of +/- (but you're now evaluating individual player offense/defense, not team).

-Chris

[talkingpractice]

How do you deal with the DRose situation in this type of model?

Models like these assume constant team strength subject to random fluctuations. If this is no longer true (Rose recovers from his ACL injury), you have to revise the model. It'd be easy enough to add an indicator for Rose playing/not playing, for example. If you add indicators for every player and break the data down by possession, you have a more sophisticated version of +/- (but you're now evaluating individual player offense/defense, not team).

-Chris

Right, and some of what you're doing sounds somewhat similar to some of what we're doing. But you have to make so many assumptions (dozens of them, such as how well/much D-Rose is going to play when he returns), that doesn't this end up in the 'needs tons of qualitative input/judgment' pile, instead of the 'code/model/exact' pile?

I'm not exactly hiding that I think CHI trading at 1-3% on some of these sites is insane...

[Eternal]

You can include subjective input in statistical models, too. It's all just information.

-Chris

[DSMok1]

Models like these assume constant team strength subject to random fluctuations. If this is no longer true (Rose recovers from his ACL injury), you have to revise the model. It'd be easy enough to add an indicator for Rose playing/not playing, for example. If you add indicators for every player and break the data down by possession, you have a more sophisticated version of +/- (but you're now evaluating individual player offense/defense, not team).

-Chris

That's just called adjusted plus/minus. And it's not as easy as it sounds.

But using some sort of player rating system is critical when doing true team ratings, to account for absences. I plan to develop such a system soon--I have the ASPM plus/minus system; that plus knowing exact playing time in every game should get me pretty far.

EDIT: I don't know how much basketball stats background you have, but adjusted plus/minus is a pretty important section of the research. I wrote an overview here: http://godismyjudgeok.com/DStats/2011/n ... ilization/

[v-zero]

I guess to be fair, though the ranking/probabilities are all wrong, the Betfair lines do have the correct 8 teams in their #1 through #8. ie the 6 "good" teams, plus LAL and CHI (who may be just Nash or D-Rose away from being good teams).

That's true, but anybody with decent prior knowledge (i.e. knowledge outside of simply this season) would pick the same top eight, with the exception of perhaps Boston swapping in for LAL.

[talkingpractice]

I was going to do this very soon anyway, so I'm going to run Playoff Power Ratings over the next couple of days (maybe even today), and will post them in this thread when done. It'll be based on expected Playoff rotations in terms of injuries (ie DRose in, Dirk in, etc), and also in terms of shortened rotations and increased MPG for the better players. I'll ofc use IPV for each player, with some rotation/lineup tweaks.

I don't think there's tons of value lost by my publishing these here, as these are small markets, and team strength is just one of several factors for calculating playoff winprobs at this point of the season. I'm pretty sure that this is the most logical initial step for something like this, but from here we'd have to take these power ratings, adjust for likely home/road locations in the Playoffs, adjust for how many good teams you may have to beat to get there (West > East, obv)... and then we'd have the winprobs. I'm not going to do the other parts now, but I'll enjoy doing the fun (initial) part, so will do that straight away so that we can compare them against what anyone else posts here.

[v-zero]

I will probably do something similar over the holidays once the next version of SEBF (based off of a 26-variable box score derived from PBP data) is complete. In the past I found that classifying players into brackets added further information as to the likelihood of that team winning in the playoffs, specifically finding that teams with good role players and a few superstars would generally fare better than teams completely made up of great-but-not-superstar players, even if the sum of their player strengths was equal.

[talkingpractice]

I was going to do this very soon anyway, so I'm going to run Playoff Power Ratings over the next couple of days (maybe even today), and will post them in this thread when done. It'll be based on expected Playoff rotations in terms of injuries (ie DRose in, Dirk in, etc), and also in terms of shortened rotations and increased MPG for the better players. I'll ofc use IPV for each player, with some rotation/lineup tweaks.

This is finished and uploaded to our site in the "Playoff Power Ratings" sheet. Here is the top tier, with some comments.

Rank Team Rating
1 Oklahoma City Thunder 10.66
2 Miami Heat 10.61
3 Los Angeles Clippers 9.66
4 Los Angeles Lakers 9.51
5 San Antonio Spurs 8.96
6 New York Knicks 8.54
7 Denver Nuggets 6.91
8 Memphis Grizzlies 6.88
9 Chicago Bulls 5.67

So, team performances thus far (measured by srs or similar) have pointed to a top tier with 6 teams (OKC, MIA, LAC, SAS, NYK, MEM). The wagering markets currently have a top tier of 8 teams (CHI, LAL), due to obvious reasons. We're finding here that DEN may belong in that elite tier. If they play an 8-man rotation, then they don't play anyone with an IPV < 0. Well.... Corey Brewer is -0.01... but whatever.

LAC sits at 3rd in our ratings, though they're a clear 1st right now in srs. This is due to their great bench not being nearly as valuable in the playoffs when rotations shrink down.

[Crow]

Rubio with 2nd lowest eFG% in the league so far among players over 50 minutes. 23rd worst offensive rating amongst same group.

Top 8 teams on talkingpractice's list all have 5-7 wins against top 10 teams. Knicks with the best win % against that level of opponent. Chicago dramatically weaker against top 10 teams than those ahead of them with a record of 2-7.

Rockets paying $15 million this season for players waived and of course have the 1st round pick that has yet to pay or deliver expected value.

[Crow]

Review of recent Celtics trends:

http://celticshub.com/2013/01/04/the-di ... s-offense/