Italian Stallion
PostPosted: Mon Oct 04, 2010 5:38 pm Post subject: Mark Cuban's latest on Advanced Stats Reply with quote
http://blogmaverick.com/2010/10/03/buil ... averick%29
Back to top
View user's profile Send private message
acollard
Joined: 22 Sep 2010
Posts: 56
Location: MA
PostPosted: Tue Oct 05, 2010 10:32 am Post subject: Reply with quote
It seems to me that he is rejecting quantitative approaches because he has tried them and they have failed in his eyes. But to have a team that has been so solid for so long doesn't seem like a failure to me at all.
And on the other hand, they haven't been opting for a combo of youth, value and foreign players like the majority of "well managed" quantitative leaning teams have been doing (Portland, OKC, Hou, SAS). Instead they've signed or traded for a lot of not quite all stars (Butler, Marion, old Kidd) and not that great centers (Hayword, Chandler, Dampier, etc.) for fairly large and long contracts, while trading away first round picks. I think more than anything Mark Cuban's willingness to spend has hurt the team, rather than being patient and letting anything/anyone develop.
Back to top
View user's profile Send private message
Crow
PostPosted: Tue Oct 05, 2010 3:02 pm Post subject: Reply with quote
Youth, value and foreign players are good elements to look at and have been emphasized
by some "quant"-influenced teams, at least for stretches.
Dallas has hardly any youth right now, trying to win with vets in the later stage of Dirk... and
Carlisle. But thru the 2000s Dallas had regular injections of youth (including Howard, Daniels,
Najera, Harris, Diop, etc. and now Beaubois) around a young then maturing Dirk.
They've typically had at least 3 foreign born players and sometimes 4 - 5. They have 3 right
now . I think league average is around 2.5 - 3.
Value is a bigger question I won't try to review at the moment, though you could with any of
several productivity measures vs salary and perhaps using 50% - 100% better value than
league average as the breakpoint for "good value".
Some other "quant"-influenced teams were light on parts of the triology of youth, value and
foreign players.
I don't think Boston has used any foreign born players in their rotation. Rondo, Davis and
Perkins gave them some youth but that by itself seems more like average than elevated and
overall they have a high average age. They have brought in other youth then traded them
(including Bill Walker) or let them go. Recently they have brought in some more but remains to
be seen when and how much they'll use them.
The Cavs went from 4 foreign-born in 07-08 down to 2 last season. Probably will be the
typical 2 - 3 this season. They were a bit above average on team age last season.
Houston is younger now on average team age though the starting lineup is probably above
average in age. They went from 3 to 2 on foreign-born.
The Thunder are young for sure, for good and perhaps not for winning in the playoffs right
now. 3 foreign-born is average but the minutes are probably above average.
Portland is young. Heavy on foreign-born in the draft. 3 on the roster now but it might be
headed to 2 or 1 soon.
San Antonio went for some youth recently, kept some, let others go. Just 3 foreign-born but
probably near the top in minutes played.
Last edited by Crow on Wed Oct 06, 2010 8:08 pm; edited 3 times in total
Back to top
View user's profile Send private message
Crow
PostPosted: Tue Oct 05, 2010 3:44 pm Post subject: Reply with quote
Italian Stallion, the original article link could be replaced by this, if you are willing to adjust it
http://tinyurl.com/2br2od2
On Cuban's blog post, it was interesting to see the excitement 6 1/2 years ago about counting
the then largely uncounted. A lot of that detail has been captured since. Maybe half of it
publicly and probably more privately, though it is obviously hard to know from the outside how
much further the counting has gone on select teams.
The post is one of several recent signs of the statistical revolution undergoing some digestion
especially by long-time enthusiasts. His use of the phrase "the last rites" in the post title
struck me as overdone and not consistent with talk in the body about statistical elements still
playing a partial role.
Coaching / system and chemistry certainly affect past and future stats. You have to look deeper
than just the raw numbers but they are still the main input.
Team building is an optimization challenge that should use both the quantitative analysis of
basketball statistics and whatever you can find and reliably and expertly use regarding
psychological and social aspects. The latter
analysis is probably on the rise and perhaps deservedly but it too will have to be used
digested and used judiciously. I see a partnership more than the new analysis usurping
the analytic helm entirely.
The comment by bryanb426 — October 5, 2010 @ 1:19 pm on Cuban's blog gets back to the
presumably previously untapped territory of counting that you could explore with massive
application of cameras and precise analysis of that data. Haven't anything further about the
league experimenting with more cameras since it was leaked / announced a year ago.
What data did they get and are they going to continue / expand?
You could conceivably, if you can get to this point technically, measure down to things like defensive
slide speed in thousands of a second across entire games or even seasons and goodness of offensive player coverage
during those slides. You could perhaps measure
jump height on rebounds and correlate the timing of max jump height with when the ball
actually reached a location where it became recoverable- in general given the athletic max in
the league or given the specific players in position or the one particular individual under
study- for a season. And so on.
Back to top
View user's profile Send private message
acollard
Joined: 22 Sep 2010
Posts: 56
Location: MA
PostPosted: Tue Oct 05, 2010 8:34 pm Post subject: Reply with quote
Crow, I definitely overgeneralized there, without going into specifics. Dallas certainly does go big into the foreign market. My main point was that they seemed to spend a good deal of money and draft picks to acquire older role players and centers. It seemed to backfire and really sets them apart from those other teams.
Sure not every "quant" team (or whatever you would like to call it) goes with youth, but the sucessful ones seem to target the other market inefficiency, max players. Boston was able to (and had to because of Pierce's age), LA also did with Gasol.
I feel like Dallas and Cleveland both had the same problem in filling out their rosters around their superstar with 8-10 mil/year players instead of finding a solid second star for their team. Part of that was because they didn't have any or enough young trade chips, part of it was impatience, and part of that was bad luck or timing I think.
Back to top
View user's profile Send private message
Crow
Joined: 20 Jan 2009
Posts: 806
PostPosted: Tue Oct 05, 2010 10:07 pm Post subject: Reply with quote
AC, I understand you were making a very brief comment and that constrains the amount of
ground and nuance you could cover. I meant to build on your lead, even if I mainly checked Dallas'
degree of fulfillment of the points of focus and
searched for light spots and exceptions for teams noted for pursuing those 3 strategies. My
comments are still just dealing with the topics in a fairly broad-brush way as well. More could be done
and your latest comments hint at some of the other details and angles that could be followed
further.
mtamada
PostPosted: Wed Oct 06, 2010 1:11 am Post subject: Reply with quote
Crow wrote:
The comment by bryanb426 — October 5, 2010 @ 1:19 pm on Cuban's blog gets back to the
presumably previously untapped territory of counting that you could explore with massive
application of cameras and precise analysis of that data.
Yes, I think quant analysis of basketball is only beginning (and not in its last rites as Cuban claims -- although I doubt that he really believes that). As I've mentioned before, I think there've been three revolutionary steps in basketball analysis: coherent quant analysis of boxscores; analysis of play-by-play data; and the upcoming analysis of the multi-camera video data. KevinP may've been wrong when he composed the APBRMetrics epigraph "The statistical revolution will not be televised". With those 6 hi-def cameras that STATS LLC is putting up, maybe the third revolution will be.
BTW, late in August Sportvision had its third Pitch f/x summit. Though baseball analysis remains a simpler problem than basketball, I'd have to think that the people quantifying and analyzing that video data are creating an analytical, or at least data architecture, foundation that basketball analysts might find useful.
Back to Cuban: he's right of course in pointing out many of the limitations of quant analysis, the various important things that are difficult or impossible to quantify. I think it's no coincidence that he's no longer a client of Winston & Sagarin's WinVal system (but I do think useful research on adjusted plus-minus continues to be done, not just the regularized APM but the multi-level modelling that RyanP had been doing). But from his post it sounds like one of the areas that he's drifting into is psychological profiling of players? I.e. to get at the aspects of chemistry and to identify the "nuckleheads"? I'm a bit dubious; I've heard rumors of at least one team that tried that approach. The NFL has been doing psychological profiles of players for years ... my impression is that they have not proved to be useful, but that's based on superficial reading of news articles, I suppose if they're still doing it, it's because they're getting useful info.
But even aside from the high-tech video world or even the play-by-play world, back in the boxscore world I think there's still useful work to be done. As Cuban noted in his article, and as all of us have been bemoaning for years, there's a ton of basketball actions that could be simply counted and added to boxscores, so we're no longer limited to assists, TOs, etc.
Back to top
View user's profile Send private message
schtevie
PostPosted: Sun Oct 10, 2010 4:49 pm Post subject: Reply with quote
The Cuban blog post is interesting but should also be considered in conjunction with what he has said at the last two MIT Sloan Conferences in the Basketball Analytics panels (http://www.sloansportsconference.com/panels-2/).
What I find most interesting about his remarks is not so much the direct point that there are limitations to existing techniques when it comes to team building. (Isn't that already pretty much the conventional wisdom among those familiar with the issues?) And I am sure that despite his having traded the employment of an arch-APM adherent for the more eclectic Roland Beech, Mark Cuban still has available and still considers all, what is now conventional, analytical output.
What I find interesting is the inference that I draw from his remarks, regarding the potential upside of investing in analysis. Specifically, that there isn't much. In his 2004 blog post to which he refers, and in his remarks in the last two years of the Sloan Conference, his expressed view is that (presumably much) useful information could be gained from more detailed statistics gleaned from game video (and he lists at least some of these in the 2004 post). And indeed he was encouraging the league to collect it, so as to share the cost, and to this common database he would apply his "special sauce".
Now perhaps he, the owner of Synergy, has already done this work and found the revelations wanting. (But if so, why discuss the matter publicly all along?) Alternatively he hasn't arranged for this data to be generated, which doesn't seem to make much economic sense at all.
There are 1230 games in the season. How much would it cost to teleconference an agency in Mumbai or Shanghai and arrange for each game to be broken down by, say, three, specially-trained scorekeepers? You have them cross check the data and end up with exactly what you want. $100k, max for the labor? Say, twice that, $200k. Say, five times that, $500k.
Compare this to the additional revenue that might be expected if the competitive advantage resulting from such information leads to just one additional win. (Never mind the psychic joy of getting one's team over the hump, to a championship.) What is that $1 million, $1.5?
If the most forward-thinking owner, a relatively free-spender, who has invested greatly and continues to invest in improving his team, doesn't think that generating data that he has claimed over many years could prove useful, what does that say?
Back to top
View user's profile Send private message
Crow
PostPosted: Sun Oct 10, 2010 5:15 pm Post subject: Reply with quote
It might, in part, say something about how fully, consistently and well GMs, coaches and players
have used whatever that analysis showed and suggested. But there are major factors that
would take substantial research to sort out and probably an inside view. Can't say what gains or
losses were left by analytics on the table unused.
It probably also means that strategic and tactical changes get met with opponent
responses and might pull down the expected gains of the first moves. Not sure if teams plan
as well for the response to the response as for the first move or do as well with it after they are
there.
Not that you can say anything definitive from it because of large error estimates but in 2008-9
regular season the Mavs' top 5 most used lineups were all estimated positive on 1 year BV
Adjusted +/-. In 09-10 just 3 of 5. In 2008-9 playoffs 4 of the top 5 most used lineups
were estimated positive. In 09-10 playoffs... just 1 of 5.
But expand to top 15 and in 2008-9 regular season the Mavs' top 5 most used lineups were
estimated positive on 10 of 15. In 09-10 just 11 of 15. In 2008-9 playoffs 8 of the top 13 most
used lineups were estimated positive. In 09-10 playoffs... just 6 of 13 (with ratings).
The amount of difference varies using the different cutoffs but 09-10 was not better for top
units on this metric.
Overall they were 5 spots better on defense efficiency in 09-10 than in 08-09 but 5 spots
worse on offensive efficiency. I might have expected the reverse in a move from more
emphasis on Adjusted +/- to more emphasis on discrete stats but I don't know what discrete
defensive stats they might be tracking or using much. Regular season they won 5 more but
exited one round earlier. Not sure I'd draw sweeping conclusions just from these 2 years.
Cuban can make conclusions and state what he wants, as it is his team and generalship. I
wonder if it is a mix of fatigue and perhaps wanting to discourage new entrants to the
analytic group or greater competition for depth. With the emphasis on the importance of
coaching and chemistry I'd think coaches and players are on notice they will be held
responsible. As they should but so should the the front office.
Back to top
View user's profile Send private message
Crow
PostPosted: Sun Oct 10, 2010 11:07 pm Post subject: Reply with quote
Game coaching is important. As to how important, I hear various views and while I side
with very important I don't know what exact magnitude to put on it.
In addition to what is done looking average stats for the season at lineup and team level and
across seasons, I'd think it could be productive to look at play by play level data with a focus on
coaching. What sequence of plays were called or
went down. What sequence of substitutions were used compared to their normal. I'd think you
could characterize the apparent motive of a good share of them- going big, small, more offense,
more defense, more post, more perimeter, more beef, more speed, etc. and aggregate the data at
some balance of detail vs roll-up. If you were
good with programming I'd think you could probably get the computer to handle a large
batch quickly once the right criteria were specified.
Were any "sucker" player substitutions or play actions used to bait the opposing coach and then
beat them with the next punches or is it usually just right at it? You could look at the sequence of
dueling substitutions and look at the results of small segments and larger ones. Do some
coaches use the same sequences over and over
again? Do they modify the sequences effectively if the productivity declines? What was the best
opponent response on average and what was the best response to the response? The computer
could find the smaller and larger sequences and tabulate the results. So much you
could do with the time and skill and the right clients.
Back to top
View user's profile Send private message
Rhuidean
Man. I read that post by Cuban, and was a bit surprised.
Quote:
You can take all the PER, WS48, WP48, Adv plus/minus and the rest and when you add them together the day before the season starts you still know nothing more than the minute before you added them all together. They are meaningless when it comes to putting together a team.
Why is this the case ? Two reasons: coaching and chemistry.
Each coach has a different style and each player contributes differently depending on the players around them and the style of play and coaching. How a player on another team will fit into the coaching environment and system of your team can not be answered by stats.
So is he in essence saying, let's solve a harder problem rather than the easier one of figuring out who is good and who isn't? Because there is far more data available about individual players than about coaching, chemistry, etc. Or stated differently, whatever methodology he is using to figure out things like coaching and chemistry, if his approach is reasonable he should be able to apply those same techniques and figure out who is good and who is not.
Quote:
Until you can quantify coaching and chemistry, you can not use the numbers to build a team. Period end of story. You can use them as partial input along with scouting and other elements, but there ain’t no Moneyball solution or the NBA and I don’t see Bill James walking through that door with a solution. Stats will continue to play a role in lineups, matchups and trends, but teambuilding, not so much.
I don't get this either. If I understand lineups and how they compare against each other, do I not understand teams? Teams are just a bunch of different lineups playing against each other.
Honestly, I am starting to wonder if Cuban felt he was burned by +/-, and thus now thinks less of quantitative approaches as a result. For example, was the horrendous Kidd deal primarily motivated by just looking at +/-? Or the Diop bad contract? Or Marion's bad deal?
The Mavericks made a series of bad deals over the past few years that have realistically made them not have a chance of being a championship contending team. But if these bad deals were informed by whatever variant of +/- Cuban uses, that doesn't mean necessarily reflect poorly on all variants of +/-, for example.
I do agree with the latter part of his blog. Measure everything, and then build algorithms to suck out useful information from what you measure afterwards. If adjusted +/- isn't cutting it for you, figure out more stuff to measure so you can have more data and hopefully a more powerful statistical model.
Back to top
View user's profile Send private message Send e-mail
jim
Joined: 01 Aug 2009
Posts: 13
PostPosted: Mon Oct 11, 2010 5:09 pm Post subject: Reply with quote
Quote:
For example, was the horrendous Kidd deal primarily motivated by just looking at +/-?
Why do you think the deal was "horrendous"?
Back to top
View user's profile Send private message
Rhuidean
PostPosted: Mon Oct 11, 2010 6:44 pm Post subject: . Reply with quote
From a team construction perspective, a second option type who can get to the FT line and create his own offense is exactly what Dirk has always needed. Jason Kidd is not this guy.
Heck, what are Dallas Mavericks fans secretly hoping for with regards to Roddy B? They hope that he (rapidly) becomes that dynamic scoring combo guard that can be Dirk's sidekick. Well, the Mavs already had that in Devin Harris; we saw how he flourished offensively during the 2008-2009 season. There is absolutely no reason not to believe that he could not have had a similar positive impact in Dallas over the past 2 years, if used properly.
Look, Kidd is/was a great player. But as far as top team needs to get to that next level, he is not the sort of player you need next to Dirk. You need someone who can punish opposing team defenses, get to the FT line. Not just shoot jumpshots.
Back to top
View user's profile Send private message Send e-mail
kjb
Maybe I've been spending too much time dealing with conspiracy theories, but I didn't take Cuban's post as being an authentic representation of what he thinks. Which is a really polite way of saying that he might be lying.
It's so obviously wrong that I wonder if he's actually trying to dissuade other teams from investing in stat analysis for team construction.
Mogilny
acollard wrote:
Instead they've signed or traded for a lot of not quite all stars (Butler, Marion, old Kidd) and not that great centers (Hayword, Chandler, Dampier, etc.) for fairly large and long contracts, while trading away first round picks. I think more than anything Mark Cuban's willingness to spend has hurt the team, rather than being patient and letting anything/anyone develop.
If there exists a "lemon" effect on the player market perhaps trading picks for players is a flawed strategy on a systematic level because of information asymmetry?
Crow
Joined: 20 Jan 2009
Posts: 821
PostPosted: Tue Oct 12, 2010 3:39 pm Post subject: Reply with quote
There could be something of a lemon effect. But GMs have seen the vets play for many years and if they are actually good at player evaluation I'd think they should be able to pretty well even without the insider information. I don't think that much is really hidden from view if you really work it with tape and live scouting and intelligence gathering operations.
Draft picks probably have a fresh-squeezed orange flavor or smell. There are financial benefits at the top of the draft especially but it sometimes seems GMs value this opportunity for choice more than the average results suggest.
acollard
Joined: 22 Sep 2010
Posts: 56
Location: MA
PostPosted: Wed Oct 13, 2010 10:40 am Post subject: Reply with quote
I wasn't trying to imply a lemon effect. I've long heard the contention that draft picks and max players are the best way to build a team because both contracts are league limited to below market value. So trading the opportunity for a deal (draft picks) for non-max players (essentially guaranteed to be less cost effective on average),doesn't seem advisable. Even if the players they traded for at the time were a little better, the Mavs gave up assets, and now they have few assets anyone would really want. Going down their roster for this year. Butler (1y/10.5m per year left), Marion (4y/$8m/y), Terry (2y,11m/y), these aren't good deals really.
Update:
Looking at their transaction flowchart , it is not as if they gave up a lot of picks, and clearly they have been hurt by being so good for so long. At the same time, there is an opportunity cost to signing all these overpaid players, and it should be noted that the one time they traded for a high draft pick they acquired Devin Harris.
Back to top
View user's profile Send private message
Crow
Joined: 20 Jan 2009
Posts: 821
PostPosted: Wed Oct 13, 2010 1:09 pm Post subject: Reply with quote
Draft picks and max players are both important and quite valuable but I am not sure if you can build the entire core of a deep playoff contender on just those 2 legs.
When Dirk first went to the playoffs the average experience of the other 4 starters was 5.5 seasons. They made the second round. They got a little younger with Josh Howard and Marquis Daniels and Devin Harris in the middle of the decade but never went more than 2 starters being real young and generally averaged 5+ years of experience for the other starters and they had their decade of 50+ wins with that approach rotating in "new" guys with experience.
What team has made the conference finals with 3 starters under 5 seasons of experience? In the last 5 years only the 08-09 Magic and the 06-07 Magic and Jazz fully did that but they had a lot of guys with 4 years experience. None won it all. 3 of the 20 conference finalists in the period. A few others with a bit more experience did as well but they didn't win it all either. An average of 5 years for the starters seems close to the line of enough experience to go deep and maybe win it all. But perhaps not quite enough.
If you gather together a flight of rookie contracts and a max player or two and age them past 5 years experience that model can work to a very high level. But I not sure a draft pick kept into his second contract is that different in price or nature from a vet you can get by free agency or trade. If you have a big budget you have the option to do it the other ways repeatedly. If you have a smaller budget maybe the flight of rookie contracts and a max player or two are the main parts of the strategy though I'd imagine other players acquired at middling prices and by other means will figure into the core of that success to some degree.
mtamada
Joined: 28 Jan 2005
Posts: 377
PostPosted: Mon Oct 18, 2010 6:53 pm Post subject: Reply with quote
mtamada wrote:
Yes, I think quant analysis of basketball is only beginning (and not in its last rites as Cuban claims -- although I doubt that he really believes that).
kjb wrote:
Maybe I've been spending too much time dealing with conspiracy theories, but I didn't take Cuban's post as being an authentic representation of what he thinks.
Tying this in to the NCSSORS thread, not only is Roland Beech an example of Cuban's actions not matching his words, but until his talk at NCSSORS this weekend, I had not realized that RolandB actually travels with the team, probably the only quant analyst to do so.
Which does not seem consistent with Cuban abandoning quant analysis. Switching to different techniques and analysts, sure .... drop WinVal, pick up RolandB, get ready for the coming video revolution, etc. ... but surely not to drop the stuff.
I don't have time to do a conference write up ... RolandB's talk was probably the highlight of the conference for me. Ben Alamar says that the next NESSIS (i.e. the Harvard conference) will be in 2011, i.e. they're not waiting two years as in the past. So that creates an obvious opportunity for NCSSORS and NESSIS to take turns each year, or really to take them both and treat them as the annual conference of the American Statistical Association's Sports Section -- i.e. an annual conference which alternates between Harvard and Menlo College. BenA's in favor of the alternating schedule, but he's waiting to hear what NESSIS says.
Roland's was the only presentation about basketball, but five of the posters were about basketball. EdK did not show up in a moustache as he threatened.
Back to top
View user's profile Send private message
schtevie
Joined: 18 Apr 2005
Posts: 413
PostPosted: Tue Oct 19, 2010 11:05 am Post subject: Reply with quote
I think the point needs to be emphasized that Cuban's remarks in his recent blog post specifically pertain to the "team building" aspects of analysis, and not its potential for improving performance within a given team context.
His frustration appears to be specifically with estimates of player values not transferring across teams, which is understandable, but again, a bit odd, given relatively recent on the record remarks. In particular if you go to the videotape from the 2009 MIT Sloan Conference, you see him emphatically contesting John Hollinger's suggestion that the Jason Kidd for Devin Harris (etc.) trade didn't work out for the Mavs. And then this last year, Cuban seems to have been quite content with the way that transaction played out.
It would be interesting to know what revelation led to this recent epiphany, that is if he is not blowing smoke.
P.S. Do we know for sure that Ed Küpfer is not Keyser Söze? Two 'e's, an 'r', a 'K', and an umlaut in common: coincidence?
Back to top
View user's profile Send private message
mtamada
Joined: 28 Jan 2005
Posts: 377
PostPosted: Tue Oct 19, 2010 2:09 pm Post subject: Reply with quote
mtamada wrote:
Roland's was the only presentation about basketball, but five of the posters were about basketball.
Whoops I forgot about Steven Schmanske's presentation about effort by NBA players. Some good aspects to that presentation, he treated "effort" as a latent (i.e. unobservable) variable. His model did not explain much of the variation, I think not only because of the inherent difficulty in estimating latent variable models, but also because I'd have to think that psychology and a host of factors random to each individual player affect effort at least as strongly as the rational, planned factors that he examined. I'm not sure why he kept talking about sustainability so much ... it's certainly important with his fundamental observation that no player (or at least very few) can go all out with 100% effort 100% of the time: it's that fundamental, and very believable, assumption, that creates a situation in which effort is variable (rather than a constant) and something that might be measured and explained. So good job with bringing up sustainability there. But he kept going on and on with 3 different sustainability models and seemed to want to tie all this in to sustainability in the ecological sense, how much oil and petroleum-based fuel should we save for future generations, and all that other stuff, which I found to be a strange non sequitur.