Interesting article on Haralabob

[Jeff Fogle]

You probably saw the link at Henry's TrueHoop, or you're already a regular visitor to Negative Dunkalectics

http://www.negativedunkalectics.com/201 ... n-nba.html

Included this section on modeling that might be interesting for discussion here...

"I am pretty sure I pay more attention to detail to what is going on away from the ball (especially in terms of defense) than your casual fan. That along with realizing pretty early on that I was better off trying to predict what a coach would do in a given game versus predicting what he should or could do.”

The distinction, he points out, is crucial. “A lot of times I felt as though a team would have success in a given matchup because they could exploit X, Y, or Z only to watch a team do something completely different. I started having a lot more success when I started focusing on what a team would do and how that would work instead of what they should do.” This, as it turns out, is the basic point behind predicitive modeling.

I asked Voulgaris about something he said on the forum twoplustwo: “Database handicapping is extremely overrated, somewhere along the lines people mistook what I was doing for database work and I never corrected them. You have to have a lot of data, but the data just helps you build models either through strict regression or through supposition.” He explains that the difference between database handicapping and predictive modeling is partly what sets Voulgaris apart from the pack in the world of NBA gamblers. “I think a lot of people have databases and they do things like ‘team x does x in a back to back situation’ and they mine their databases for information like that, but that information is by and large useless.”"

[EvanZ]

Thanks, that was a really great interview/article.

[Jeff Fogle]

Any thoughts from anyone on how to mix in the things Haralabob is talking about with statistical work?

It seems on one extreme we have a position that coaching hardly means anything at all when analyzing NBA teams. Haralabob might be considered another extreme (though maybe not) when he talked about how much success he's had by focusing a lot on how coaches handle matchup challenges. Predictive models are improved when you properly anticipate how coaches will handle challenges (I'm paraphrasing...and please, anybody, suggest a correction if you think I'm not capturing it right)...and it's not that hard to anticipate what coaches will do once you've studied them awhile.

If we accept that the market prices are a strong predictive market...and this approach has shown success against that market...how can metrics be developed that measure what coaches do when presented with challenges. Can it even be done? Or, do you need to have every game downloaded as H does to eyeball the footage?

[Bobbofitos]

Any thoughts from anyone on how to mix in the things Haralabob is talking about with statistical work?

From past exchanges with him: (From twoplustwo)

Out of curiosity, coltrane, do you think (or, I should say, does your system purport) KLove is the wolves best player? do they have any other above average players on that team? just how bad is his supporting cast?

The supporting cast isn't nearly as bad as the coaching and the way they utilize their players. For whatever reason we end up having an inordinate amount of bets on the Twolves so I have watched nearly all of their games this year.

They do a bunch of really dumb things like taking shots early in the possession when leading late (game at home vs SAS). Crashing the offensive boards instead of stressing floor balance and on and on.

I think they do have a few above average players on that team most of their players are either above avg offensively, or above avg defensively but not both.

I do have Kevin Love as their best player, but its not like he far and away the best player on a team full of misfits.

Interesting post. This suggests that both fit and coaching system is far more important then any model has previously suggested - because when I see the Wolves roster, I just see a pile of trash. (And KLove, as in the diamond in the rough, obviously!) The wolves play a super fast pace and perhaps the triangle is a poor system for their personnel... I wonder what other decisions they've made or have enacted that has basically crippled their winning efforts.

I guess I'll try one more slice at this, because maybe you can answer this type of question: You referenced fit will impact a player, but only marginally. (something like Klove "could" be slightly better on some other teams, but only slightly) Why is it that on the T'wolves, who are probably making more in game and out of game mistakes then any other team, klove is only a small net positive, and yet on another team, like the Bucks for example, who we can guess are making the correct decisions, he'd still only be a slight positive? (My own answer would be something like, "Klove has certain deficiencies that cannot be hidden, period, so his ceiling is relatively low")

He'd be much more valuable on a team like the Bucks because they play a more regimented system and are poor shooting team. That would be the one team where I think he'd shine the most.

They also have one of the best defensive C's in the league in Bogut so thats another plus as he'd mask a bunch of Love's inadequacies on defense.

Put Love on the Spurs in place of Dejuan Blair and the differences would be negligible imo.

I also think the Lakers would be worse with Love vs Odom.

In response to counterpart data:
It appears my C+P didn't work and I'd rather not deal w/ the search engine there, but basically Haralabos does not believe in counterpart data (I do).

Originally Posted by napolewan View Post
Is it your goal to evaluate defenders, or is it your goal to determine optimal situational strategy based on who is most likely to be in what place at what time in what situation?

Vaguely, My goal is to accurately predict as best as possible what will happen in a given basketball game. How much better on offense is a team with this player on the court, how much better are they on defense is a team with this player on the court etc.

For my purposes its much more important to try and predict how a given team WILL play vs how a team should play. On that end I don't spend too much time worrying about optimal strategy.

I raved about JE's coaching APM earlier this year. Haralobos dismissed it:

Uh this coach as 6th man thing wouldn't be able to separate the differences between being a great lockerroom coach and blundering in game decisions -

Whoever said there wasn't enough player / coach turnover for this to be worthwhile was pretty much spot on.

Apparently I don't know optimal foul strat:

i think the hornets are fouling too early here. almost a minute left.

No its correct to foul early. Its a 7 point game with 51 secs left.

In order to win the game Denver is going to have to miss free throws, (you can only stop 2x max if they are going to run out the clock before having to foul).

You want the information (whether they miss or not) early vs late, teams in general wait to long to foul.

This is very analogous to teams in the NFL down 15 going for the extra point first, then going for two the next time (also wrong).

The teams that understand late game fouling strategies best are ORL, HOU and MIL, i liked what I saw of NOR did tonight late (a lot of good that did them).

From last year:

Went to the LeBron game tonight, it was the second time I had seen him play up close and he hasn't disappointed yet.

Interesting observations;

We were sitting courtside center and LeBron puts on quite the warmup show. He spends the first 5 or so minutes under hand throwing the ball from center court. And the last 5 mins doing foolish dunks.


Cle should really run a high screen with Lebron as the ballhandler every time down, its simply unstoppable. They ran it less than 5 times in the entire game and not once the entire first half. They ran it 3 plays in a row and got and And1, A dunk and a Lebron bobble that could have been another dunk.

Cleveland should never play Varajao and Shaq together (you can probably argue that they should never play Shaq at all) but I actually think he does one thing well and thats draw some early fouls and get you into the bonus earlier. People rave about Varajao's +/- and he is a great player, but in addition to benefitting from playing against backups part of the time, he also has the benefit of being on the floor when his team is more likely to be in the bonus.

I read a quote from M. Brown that claimed he didn't care if Varajao and Shaq played together, that he disagreed with the assertion that the offense bogs down with them on the court together - I don't think he is watching the same games I am.

I think Cle is the best team in the league, I wish they had a better coach though. LOL moment of the game was Mike Brown taking Lebron for the rest of the 1q because he picked up two fouls. I bet my gf that he wouldn't have more than 3 by the end of the game.

He ended with 3.

He also took him out with 1 min left in the 1st half and LeBron actually shot him a dirty look sat down on the scorers table before finally retreating to the bench.

Cle is around 4 to 1 to win the title, which is an awfully short price but I'd bet my whole net worth on them winning if I could coach them.

Here are two pics

[Jeff Fogle]

Thanks bobbo. Interesting selection. Enjoy reading your website too...

[Crow]

“Cleveland should never play Varajao and Shaq together”

The Cavs were +7.8 per 48 minutes on raw +/- when they were on the floor together last regular season.

"I read a quote from M. Brown that claimed he didn't care if Varajao and Shaq played together, that he disagreed with the assertion that the offense bogs down with them on the court together - I don't think he is watching the same games I am."

It is true that scoring per 48 minutes did fall to 95.4 per 48 minutes compared to the team average of 102.1, but points allowed fell to 87.6 from 95.6 for all team minutes. The overall team point differential was +6.5. So Shaq +Varajeo on the court had a better differential than the team average.



“(you can probably argue that they should never play Shaq at all)”

If Varejao and Big Z could have played a lot more minutes, without reduced effectiveness or foul problems, maybe. (Depending on if you compared Z to Shaq on raw team +/-, Adjusted +/-, other things or a combination.)


But the PFs besides Jamison were big negative of raw +/- and Hickson was a big negative on Adjusted +/- (the others didn't rated enough minutes to be rated on 1 year APM), whereas Shaq was a modest positive of raw +/- and a smaller negative on Adjusted +/-.

It would have been better to have Varejao replace minutes by the other non-Jamison PFs at PF first, before replacing Shaq minutes, if he could so with similar efficiency and without foul trouble or excessive foul trouble.


Here is a table of player combinations with and without players or combinations.

S= Shaq
V= Varejao
J= James
Z= Z. Ilgauskas

Two initials together means on the court together.

Anything subtracted removes the minutes that player or combination played from the first broader value.


Min +/- Pts for Pts against (per 48 min)
sv 468 7.8 95.4 87.6
v 2164 11.3 102.6 91.3
v-s 1,696 12.3 104.6 92.3

s 1240 3.7 98.2 94.5
s-v 772 1.2 99.9 98.7

svj 304 8.8 95.7 86.8
j-svj 2,661 10.7 106.6 95.9

vj 1534 16.1 107.1 91.0
sj 1034 3.6 98.8 95.2

sv-j 164 5.9 94.8 89.0
sj-v 730 1.5 100.1 98.6
jv-s 1,230 18.0 110.0 92.0

j 2965 10.5 105.5 94.9

v-j 630 -0.4 91.5 91.9
s-j 206 4.0 95.1 91.1

zj 889 16.1 106.6 90.4
z 1339 8.1 100.3 92.2
z-j 450 -7.9 87.8 95.7


Shaq without Varejao (+1.2) was worse on team +/- than with Varejao (+7.8) because V is a very beneficial player (+11.3 overall).

Varejao benefited slightly (1 pt per 48) when without Shaq.

James was slightly better when not with S+V (by 2 pts).

Shaq with James and no Varejao was close to an even proposition (+1.5).

S + V - J was surprisingly good (+5.9). V+J was very very good (+16.1) , only exceeded by V+J-S (+18).

Shaq was better overall without James (+4) than Varejao without James (-0.4), probably since he can score.

Z+J (16.1) was as good as V+J.

Z without J was a disaster (-7.9). That would have been the first change I would have looked at hard and probably made at Center. Replacing PFs besides Jamison with Varajeo would also have been a priority. Replacing Shaq would have been a third option, with a focus on eliminating his weakest on team +/- minutes, where possible within available, more efficient than Shaq, big Z or V minutes.

The playoffs are another dataset. Similar analysis could be done for that or any split of the regular season with more time and a reason to do so. Looking at it just briefly Shaq was worse than his regular season overall and with Varejao too but only slightly negative on each. So the evaluation of playing S+V varies between the two datasets. Eliminating Shaq and S+V would have have more beneficial in the playoffs. Looks like Z without James was also slightly negative and Z with James was only slightly positive. Would have to look at minutes very carefully to see how much of these players and combos could have been trimmed. Varejo fined on team +/- in the playoffs, only slightly off his regular season overall rates, though when without James he was negative and much more negative than the regular season. Hickson and the other PFs not named Jamison were by far associated with the worst team +/- results. that remained problem #1 (at least among big men).

Of course raw +/- should not be the sole criteria, but I didn't take the time to look at boxscore stats for confirmation or contradiction. APM or RAPM results for these players and combinations would add information. How to weigh raw +/- with Adjusted +/-, that is another issue. Each is affected by sample size and may not very accurately estimated and may not necessarily remain stable, especially as opponent lineups face changed, but you can either try to use this information presented or go further or not use it as one prefers.


Predictive modeling does seem like an intelligent approach for this application. How players and teams perform against similar players, teams, defensive strategies (the shots they try to prevent the most and best vs the ones they steer you to or can't prevent). Points per possession for players and lineups and the team as a whole by shot location and by opponent defensive strategy at the level of zone vs man to man and more specific defensive details, and perhaps specific counterparts to be faced too (and maybe game time and situation and other factors), would provide at least a lot of the basics, if not the main basis, for a predictive model that might have betting edge, if prepared well and tuned well and constantly over time.

Coaching plans and in-game decision-making to certain results could be evaluated by data analysis and observation and these coaching actions are probably pretty important. Probably would have a tougher time getting and maintaining betting edge unless you understood what coaches were going to do or likely to do as well as what the players are going to do or likely to do.

[Jeff Fogle]

Very interesting breakdown crow.

Finally saw a clip of the Sloan Conference session on sports gambling. Not everyone's cup of tea, but they did talk a bunch about analytics...and I think some of you would find it an hour well spent if you weren't there in person.

http://www.sloansportsconference.com/pa ... nnovation/

Panel was hosted by Chad Millman of ESPN, who first became prominent I believe with a book he wrote many years ago involving a basketball bettor known as Alan Boston who did his analytics by hand (each team got a page in his notebook). On the panel were Michael Konik and Jeff Ma who have written books in the field (Konik used to be a runner for Billy Walters who was recently profiled on 60 Minutes, Ma was part of the MIT Blackjack team as I'm sure many of you are aware). Andrew Garrood is listed as the Executive Director of Las Vegas Sports Consultants, a place that supplies lines in Vegas among other responsibilities (and is the smartest person I've ever seen speaking on behalf of LVSC by a factor of about a gazillion-billion, and that's not just because British accents make people sound smarter). Mark Warkentein, listed as Director of Pro Personnel for the Knicks is also on the panel. I thought he'd be a corporate type, but he seemed like more of a gambler than anyone else! That's because he was with UNLV back in the Tarkanian days and knew a lot about sportsbooks and casino's.

That link has links to other Sloan presentations too. Maybe there's one listed that you'd be interested in even if you want to skip this one because you're not into pointspread stuff...

[Bobbofitos]

Very interesting breakdown crow.

Finally saw a clip of the Sloan Conference session on sports gambling. Not everyone's cup of tea, but they did talk a bunch about analytics...and I think some of you would find it an hour well spent if you weren't there in person.

http://www.sloansportsconference.com/pa ... nnovation/

Panel was hosted by Chad Millman of ESPN, who first became prominent I believe with a book he wrote many years ago involving a basketball bettor known as Alan Boston who did his analytics by hand (each team got a page in his notebook). On the panel were Michael Konik and Jeff Ma who have written books in the field (Konik used to be a runner for Billy Walters who was recently profiled on 60 Minutes, Ma was part of the MIT Blackjack team as I'm sure many of you are aware). Andrew Garrood is listed as the Executive Director of Las Vegas Sports Consultants, a place that supplies lines in Vegas among other responsibilities (and is the smartest person I've ever seen speaking on behalf of LVSC by a factor of about a gazillion-billion, and that's not just because British accents make people sound smarter). Mark Warkentein, listed as Director of Pro Personnel for the Knicks is also on the panel. I thought he'd be a corporate type, but he seemed like more of a gambler than anyone else! That's because he was with UNLV back in the Tarkanian days and knew a lot about sportsbooks and casino's.

That link has links to other Sloan presentations too. Maybe there's one listed that you'd be interested in even if you want to skip this one because you're not into pointspread stuff...

I'll check out that link now - for whatever it's worth, the book Millman wrote (I believe it's called "The Odds") is a really good read. I recommend it.

[Crow]

Jeff, thanks for the read, compliment and reminder about the Sloan panels.