Stats and Reality Two interesting happenings over on ESPN.com today. First, John Hollinger has augmented his baby, P.E.R., with two new formulas. One is similar to VORP in baseball, the other uses the VORP number to project wins created. And just when he gets done explaining his new methodology, Henry Abbott takes his system to task (sort of) on True Hoop. When it comes to this type of argument, I fall firmly on both sides. I have to admit, every new advanced stat that attempts to quantify a basketball player's value completely fascinates me. Truthfully, I dive in and try to find the exception that disproves the methodology. They usually stand out like a sore thumb (Check out where Lou Williams is on the PER list, for example). When I'm feeling particularly geeky, I toy around with formulas of my own, but I never really get too far down that road. Ultimately, I agree with Abbott's takeaway: "Real basketball knowledge is king. Statistical systems like PER are tools to help cut through the reams of available information." I'd also add that advanced statistical analysis in basketball makes what we do here a lot more fun. Now, let's get to the fun stuff, Hollinger's new stats. You can see the leaders in VA here. (VA stands for value added, and it's supposed to represent the number of extra points the player has given his team over a replacement-level player). And the EWA leaders are here (EWA stands for estimated wins added. This stat is simply VA divided by 30, assuming 30 extra points from a player would equal 1 win). You could call these stats Hollinger's reaction to Marreese Speights. He's been taking heat for touting Speights among rookies. Marreese has been leading all rookies in PER for most of the season, even though he only plays 15 minutes per game. These new formulas take minutes played into account. So while, on a per-minute basis, Speights kicks Derrick Rose's ass in PER, according to EWA, Rose has added 5.7 wins to the Bulls while Speights has only added 4.1. Here are the new numbers for the Sixers:  Latest Posts• Focus • The Difference • Took You Long Enough • Time To Assert Yourself • Stats and Reality • Three Weeks, One Day • Disappearing Act • The Sixers Redefine Consistency • Accountability • Gotta Have It • Against All Odds, Again • A Dangerous PremiseSearchBlogs in The NetworkSixers BlogsPassion and Pride Sixers 4 Guidos Liberty Ballers Recliner GM PhillyArena Heard In The Cheap Seats | ||||||||||||||||
I'll paste the link in here as well
http://apbr.org/forum/viewtopic.php?p=6864
This is an excerpted article from nba.com about Harvey Pollack (written march 10th on the forum) with information on how to purchase the book Abbot references.
Yeah, I'm definitely going to be sending Harvey a check (possibly with a resume attached :)
That was actually a dream I had a couple years ago (if i was much much younger) - interning for Pollack and learning everything I could and maybe being 'groomed' as his replacement - i mean - he's going to retire someday I would think.
Here's my issue/thing
Basketball isn't baseball - you can't sum it up as easily in the numbers like you can baseball - the reason baseball is so 'ripe' for statistical transformation is that almost everything has been tracked since the beginning of time and any one event is really only between 2 people at most - it's a team sport but made up of many individualistic moments that are easy to quantify (that's how i see it - i might be explaining it wrong)
Other sports are not like that so can't be approached the same way.
That doesn't mean basketball doesn't have room for improvement on the way they look at things - by no means am I saying that - I'm just saying I don't think it's as simple as looking at the historical data in a vacuum.
And even in baseball there isn't one agreed upon 'to rule them all' - i'm not sure why the quest seems to be that in basketball - or at least the believed quest - that isn't the quest at all - and it wasn't the point of money ball either (anyone who thinks money ball was about OBP wasn't paying attention) - it's about looking at things in a different way and maybe finding an under appreciated resource and using it to your advantage.
In basketball there's just too much movement, fluidity, what happens over there affects what happens over here..in baseball - if the right fielder is out of position - it has no effect on a ground ball to the 3rd baseman - individualistic events...in basketbal - if all your guys are spaced one way - you'll get an open shot whereas if one guy is too close, allowing for the easy double team, it can effect others.
So - with things like PER - while i think they are 'helpful' - i don't thing they are comprehensive (and if Hollinger WASNT working for ESPN I don't think PER would have as much a following and 'weight') - you can't take one formula and evaluate everything - hell you can't do it in baseball and it's much easier to isolate events - so why do they try to do it in basketball.
This ignores the 'old schoolers' who dismiss 'sabremetrics' as useless and for state geeks and see it as an all or nothing (zero sum?) game - they think that their 'old ways' are being dismissed and ridiculed whereas I always see it as integrating with the new ones to make leagues better.
If you come up with better ways to evaluate talent - you'll have less draft misses - you'll have less bad contracts - and possibly you'll get a better game on the court
"Basketball isn't baseball - you can't sum it up as easily in the numbers like you can baseball - the reason baseball is so 'ripe' for statistical transformation is that almost everything has been tracked since the beginning of time and any one event is really only between 2 people at mos"
Also, baseball's events are much more isolated than basketball. If a guy strikes out, it's likely his fault, not that there is a lack of floor spacing because the guy next to him can't shoot.
Sure, they'll face good pitchers at times, and bad pitchers other times, which influence individual event results, but more or less, the sample size is large enough that competition is relatively comparable by the end of the season.
It's much harder to isolate culpability in basketball, so even if you collected more in depth stats, it's still not proving cause.
"so why do they try to do it in basketball."
I don't think any stat is meant to replace analytical thinking, or be taken into a vacuum. Just like you you wouldn't take a list of the league's highest scorers and say "these are the best players in the league", you shouldn't take a list of PER and say they are without any analysis of your own either.
Defense is also much, much easier to quantify in baseball. For example, they have the data to tell you if a ball that gets in the right-center gap would've been caught by an average center fielder. There's nothing in basketball that tells you if a shooter would've been open against an average defense, and I'm not sure if there ever could be.
Baseball has essentially static parts, and you can isolate almost any action to one hitter, one fielder and one pitcher. The only thing you can equate that to in basketball is a foul shot.
Yup - but baseball has also had some technology revolutions - for instance - i don't get it - but i understand this pitchfx data is amazing - and it is provided in game on certain 'java' game logs.
I think an upgrading in the technology of the cameras and the court couldd help immensely - i wonder what marc cuban would be willing to do :)
PitchFX data is really cool, but they have a ton of kinks to work out. Check out fangraphs.com to see some awesome applications of the data.
I've seen a lot of it - i don't get it all but i understand some of it can be pretty compelling - i really need to go back and take some statistics courses i only took one as an undergraduate
Getting the data is one thing, interpreting it is another. Will all the millions (according to Simmons) that these teams spend on gathering esoteric stats, GMs still do the swing and a miss thing come draft or free agent time.
There's still an art to the science.
p.s. man, wouldn't that be cool tho to get paid to pour over the stats for a team? drewl...
That would be ideal.
People swing and miss all the time in baseball too - billy beane (and sandy alderson who gets less credit) miss as well - thus proving while helpful - it's not perfect - in fact most people believe Beane is onto a new 'untapped' statistic now that everyone knows that getting on base matters regardless of how.
To Derek - I agree with you that no stat should be taken in a vacuum or seen as the end all and be all i'm just syaing it's how it seems 'mainstream' media deals with it - for instance espn baseball games will refer to 'moneyball' players which pretty much means to me that they don't really get what it means - and the 'old schoolers' will dismiss all the new stats by dismissing it as geeks who believe that the stats are supposed to be the be all and end all.
Which was the primary motivation for my site idea that is slowly percolating - to get 'both sides' to understand that it's not an either or but a partnership - both are better together than separately - though i have no idea HOW to do that.
I did email simmons my 'league pass' idea because I think if ESPN wanted to they could 'run' the revolution just by their base as opposed to the way it's pretty much ignored on their baseball folks except for the guys who they limit to the chats and ESPN News appearances or the 'fantasy show'
Sure, and that's my point: stats be tricky bidniz.
I think these team-collected stats can at least identify areas of strength (say, reliable jumpshot) and/or weakness (say, commits too many fouls). But stats that purport to affix some sort of objective value to the player are on shakier ground. So many other factors go into that, like system, coach, the other four players on the court, etc. They may be great for fantasy/simulation games, but real world power is limited.
There's always a story behind the numbers.
Some things you just can't adapt stats to completely reflect. This is an even bigger problem looking at college ball or comparing different leagues. You can adjust for things like pace of play, but how do you adfjust for level of competition or the effect of your teammates.
Even basic stats like points have a lot to do with your teammates (playing on a lousy team or with a great PG helps) and assist have a lot to do with whether you have teammates who are good finishers (in the post or jump shooters.)
No one will find the Holy Grail of stats, but they are useful and help overcome natural bias.
But the best, and most reliable stats are the kinds that interest teams more and fans less. We like stats that compare players, where stats that describe players are probably more useful. Stats like those show a player shoots better off a screen or off the dribble (like the one's Derek gets from Synergy.) Those types of stats help coaches and players maximize their talents, even if they don't tel;l you who is 'better.'
There are two basic stats that I'd like to see improved upon, or at least broken down to a greater degree.
1. Assists
2. Turnovers
I think these stats can be misleading, possibly more than any others on the typical box scores. We've talked about how an assist for a dunk is usually a harder assist than an assist for a jumper, but you can go further than that. Dean Oliver dives into that in his book, so I won't elaborate more.
Turnovers are another. You can apply the same type of math to them.
For example, if player A is making a pass on the perimeter to player B, who is standing at the top of the key, that pass isn't increasing the team's chances of scoring, if it's completed. The odds of scoring on the possession were 50% before the pass and after the completed pass, so a turnover in that situation should be pretty damning. On the other hand, if a player is on the perimeter and tries to hit his center, who is being fronted, with a lob for an easy dunk, that pass has a higher reward attached to it. If completed, the pass would increase the team's chances of scoring on the possession from 50% to 80-90%. I guess what I'm saying is that there should be some kind of risk-reward assessment taken into account on both assists and turnovers.
If a player turns the ball over while trying to make a play which would greatly benefit the team, is that really as bad as turning the ball over when the pass isn't going to gain the team anything at all?
Stuff like this haunts my brain sometimes, sorry to rant guys.
One more quick thing on how this type of math could be applied.
Using the example above:
Player A has the ball in a position where an offensive score is likely on 50% of possessions, his center is being fronted, a completed pass over the top would result in an 80% chance of scoring on the possession.
If player A can make and complete that pass 62.5% of the time, it's a worthwhile risk. If he can't, it's not a worth it. (Meaning, if this play was run 100 times, and 62.5 times the pass was completed and the center scored on 80% of those possessions, it would equal 50 scores, or roughly the same as the average possession. Completing that pass at a higher rate means better offense. Completing it at a lower rate means worse offense than average).
Now, let's take another example directly from the Sixers. When Lou Williams drives baseline and leaves his feet, the odds of scoring on the possession probably increase from 50% to say, 60-65%. Now, when he kicks the ball out in mid-air, for a lower percentage shot (say a 40% jumper) he's decreasing the probability of a score. When turns the ball over in that situation, well, it compounds the error.
*nods*
You probably can't start changing how many assists a player gets, but you can track these kinds of things.
I pretty sure someone tracks what the TO % is for different type of passes, and what percentage of passes are of the riskier/high reward type.
To some extent the "hands rating" and "passing rating" as listed on 82games tries to take this into account. They are probably better weighted values than A, TO and A/TO.
For example, it gives a higher rating to a player like Iguodala, who's passes are often more of the playmaking, higher risk type.
82games had me track 'would have been an assist' for a couple games - quality passes that should/could have led to a basket but didn't - but whether or not it should hve been a basket is where you get esoteric.
The primary example i've always used is 'blocks' - there's nothing that says you can't 'split' blocks into change of possession/not change of possesison blocks - and that's something honestly that could pretty much be done from todays box scores and game logs I think. It might not make a difference - or it migh - but until you look you never know.
Also - and i think we discussed this earlier - in regards to turnovers - if a guy fumbles a great or easy to catch pass (yeah you stone hands haitian) doesn't the guy who threw the pass get called out for the turnover? I mean why isn't the guy who couldn't catch it called out for the turnover if he's wide open and it hits him in the chest
It just seems like the 'advanced metrics' that are useful are coming slowest in basketball and that basketball is encountering more resistance even in baseball (or just doesn't have as many advocates)
I read ESPN Footbal chats from i think the 'football outsiders' and they have a bunch of stuff that seems to be new analysis of defensive players that's pretty well accepted
Yeah, you see, that's what I was talking about before. In baseball, they have enough stats to tell you if an average centerfielder would've gotten to a ball in the gap. How in the hell can you ever quantify if an average center would've caught a low bounce pass in the post?
I think it's scorer's discretion on fumbled passes.
I expect there's enough film of enough centers to come up with something if someone would fund the time to do ti.
Cuban - you conned yahoo out of 2 billion - you gotta have a sizable chunk left - give back to the internet that made you what you are today...
Too many variables to take into account. Anyone who watches a ton of games could make an educated guess, but I think a fan of a certain team is going to have a bias when grading.
well that's why you assign fans of one team to different teams :)
And also why you have a 'double check' system with at least 2 people (maybe more) watching the recording the same film.
There's a lot of variables to take into account if a ball is 'catchable' in centerfield as well - they defined average by figuring out average break on the ball average speed etc...you have to work out who has average hadns, below average hands and above average hands the same way...it's not impossible i don't think - just a lot of time involved
And there's hella more archival footage i think of both baseball and football
You probably can't start changing how many assists a player gets, but you can track these kinds of things.
I pretty sure someone tracks what the TO % is for different type of passes, and what percentage of passes are of the riskier/high reward type.
To some extent the "hands rating" and "passing rating" as listed on 82games tries to take this into account. They are probably better weighted values than A, TO and A/TO.
For example, it gives a higher rating to a player like Iguodala, who's passes are often more of the playmaking, higher risk type.
Sorry, not a good enough post to be worthy of posting it twice...
I have an extra ticket for tonite if someone wants to go. I don't need money. email me at [email protected] before 5:30 if interested.
I need money
I'm just saying