GSW Wages of Wins updates

[Sleepy51]

The Moose turned me on to this discussion regarding a pre-season WP prediction for the Warriors, and specifically David Lee's likely WP contribution to our fortunes this season. The comments section includes a very interesting debate about the accuracy, value and flaws of the WP metric. Some of our RealGM regulars participate. (Moose is "ohreally" and he's wrecking shop.)

http://dberri.wordpress.com/2010/07/14/ ... n-50-wins/

The discussion continues on our own more neutral stats board here:
viewtopic.php?f=344&t=1037764&start=0

Just yesterday, the original author threw lots of W's under the bus in defense of his flagging predictions

http://dberri.wordpress.com/2011/02/08/ ... d-instead/

My take?
WP is inherently vulnerable in that it is based on a formula of boxscore stats and therefor CAN NOT account for defense in any meaningful way. Because the traditional boxscore stats do not account for defense in any meaningful way. We don't count the right things to measure defensive performance. We don't count how often you need help or switches. We don't count how often you change shots without blocking them. We don't count box-outs. We don't count blown rotations. We don't even count things as simple as shooting fouls conceded vs. non-shooting fouls or and+1's conceded.

WP is based upon regression analysis of boxscore statistics combined with an adjustment factor for team defensive performance. While regression analysis a powerful analytical tool, the quality of conclusions you can draw from regression analysis is HIGHLY dependent upon the quality of your data-set (and the quality of your UNDERSTANDING of the data being described.) If you do regression analysis on the wrong data points, your results will still correlate highly to wins, but not for the reasons you think. When data that you don't count starts to figure prominently into a particular player's game to game performance profile you get aberrations and major flaws in the analysis, like the prediction that David Lee would contribute 19 wins to this Warriors season.


The gaping hole in the WP analysis comes down to how the formula addresses defense. After accounting for a player's individual offensive production, an adjustment is made based on an incorporation of team defensive boxscore stats. It makes no distinction between team and individual contribution.

http://www.wagesofwins.com/CalculatingWinsProduced.html

It would probably produce a significant improvement in the WP formula to incorporate on/off differential in that defensive adjustment and would probably reduce the margin of error for the kind of player that has torpedoed the Warriors prediction, but Berri & cult are extremely hostile to 82games.com data in any way shape or form. That blindspot is probably the biggest Achilles heel in their understanding of the game. Some things you HAVE to account for by differential because of the nature of team concepts (like rebounding and defense.)

Folks who don't realize the grave inadequacy of the traditional boxscore in describing individual defensive contributions to the team result on the defensive end are missing the boat, and building a model based on bad/highly incomplete data set. The ability to disregard or overlook the inadequacy of that data set implies a certain lack of understanding of the game.

Discuss?

[floppymoose]

I knew you would be into that discussion.

[Sleepy51]

Unfortunately, no one else on our board apparently likes to read or do math.

[Visigoth]

Unfortunately, no one else on our board apparently likes to read or do math.

I just wrote a PDE mid-term solving this as one of my problems: http://img511.imageshack.us/img511/8984/mathh.jpg


Excuse me for not reading your post. =P

[floppymoose]

nice! despite being a math major, I don't think I ever had much PDE. I was more the abstract algebra type.

[Visigoth]

nice! despite being a math major, I don't think I ever had much PDE. I was more the abstract algebra type.

I'm envious! Abstract math seems so much more interesting.

I'm not even a math major, I'm an Engineering major, which makes learning it all the more balls because of how irrelevant it is to my work.

[Coxy]

Unfortunately, no one else on our board apparently likes to read or do math.

I just wrote a PDE mid-term solving this as one of my problems: http://img511.imageshack.us/img511/8984/mathh.jpg


Excuse me for not reading your post. =P

Is that Chinese?

[marthafokker]

Unfortunately, no one else on our board apparently likes to read or do math.

I just wrote a PDE mid-term solving this as one of my problems: http://img511.imageshack.us/img511/8984/mathh.jpg


Excuse me for not reading your post. =P

Is that Chinese?

You should see the steps it takes to solve this 5 variable PDE equipment. I rather write an essay. So glad I have been out of school for awhile. Not missing something like this.

Another (Applied) Math major here.

[floppymoose]

But the best part of this thread is that it begins with the words "The Moose turned me on".

[and1GS]

Unfortunately, no one else on our board apparently likes to read or do math.

I just wrote a PDE mid-term solving this as one of my problems: http://img511.imageshack.us/img511/8984/mathh.jpg


Excuse me for not reading your post. =P

Is that Chinese?

seriously...F that. So glad my major only requires 1 math class.

[floppymoose]

So Sleepy, you know the over the back shot Monta swished from the tunnel? Well, I just hit one of those, stats style, starting right here:

viewtopic.php?p=26740379#p26740379

Big ups to mysticbb who took my idea and turned it into a WP world beater.

[Sleepy51]

WP got served.


I had a passion for math and physics but didn't have the discipline to go far with it. I got my BA stat work done and moved on before I went crazy and ended up living in steam tunnels under the dorms.

I have had to resurrect some of that stuff for my business, but just the broad strokes. I couldn't hold someone accountable for the veracity of what they are claiming their math says. I just don't have the vocabulary, but I can follow the logic to critique and debate the value of what they are saying it says.

And I can tell that what Berri & cult are saying that WP says is flawed. Its flawed because the dataset is inherently flawed as we've all discussed. If you are trying to make any holistic statement about baskeball performance and team contribution with only the boxscore to characterize defense, then you are fail.

[Twinkie defense]

WoW accounts for defense as if it were a team result... which it is. Plus/minus-based models are much worse because they purport to measure every players direct impact on the defensive end, which is ridiculous.

Do you think Monta and Steph would be better defenders if the Warriors center was Dwight Howard?

What I want to know is how come only three of sixteen players to play for the Warriors this season are playing better than last season - and one of them only barely.

[floppymoose]

It's interesting that people so frequently prefer a "clean" answer over a much more accurate but "dirty" answer.

WP makes no attempt at all measure the defense of players. If you remove the team defensive adjustment from WP calculation, it changes the correlation with wins *a lot*, but it changes it's rankings of players *almost none*. You can be as bad a defender, or as good a defender, as you want, and you will still end up ranked about the same by WP.

On/Off analysis is noisy and suffers from several problems, but it gets players defensive abilities ranked much better than WP does, because WP doesn't do it *at all*.

[Carl_Monday]

Floppy - you win. That was freakin awesome.
I spent my "extended" lunch today at work reading through all of that.

As for people preferring a simple "clean" answer as opposed to a more complex, or "dirty" answer, as you put it, I chalk it up to general intellectual laziness. Christ - look at the recent economic collapse: the vast majority of the US attributes it to solely either Wall Street/mortgage industry shadiness/greed or lack of government regulation (with the "correct" answer dependent upon pre-conceived views). It couldn't possibly have been due to something that takes more than seven second to explain.

While I totally appreciate what the advanced metrics folks do, and value the results to an extent, the fact that there's a refusal to acknowledge the fundamental flaws associated with box score metrics annoys the heck out of me - and Dave Berri is by far the worst offender. It's argued as if it's the only stat that matters and that it's infallable. Just state that it can be used as one of many statistical tools in evaluation to be used in conjunction with watching the games, and be done with it.

Anyways, that was a fantastically entertaining read, even if some of it was a bit beyond my comprehension (engineer with a strong math background, I just haven't done any advanced statistical analysis since my college days, and that was many beers ago). By the way, per your suggestion in the "Backtracking of 50 Win Prediction" post, I googled "FM 48 floppymoose" (eventually finding the right link to teh Statistical Analysis forum), but the next two links were for the 8" Floppy Moose and the 5" Floppy Moose. For some reasons I just found that really funny.

[Twinkie defense]

The "clean" answer is the one that purports to do what it can not do. +/- is very, very "clean" in this regard - it says "defense? No problem, here is your individual number for defense."

The "dirty" answer is the WoW answer - it says what you are asking for is impossible, so we're not going to give you a neat answer. What they DO say is we've regressed the model and can accurately account for 90-something percent of what factors, in what percentages and combinations, account for wins - and the individual defensive performance, outside of measurables such as defensive rebounds, blocks, steals (which directly impact possessions, which are hugely important, and which ARE accounted for by WoW), account for a very small portion of wins and losses.

Again, take my example - if the Warriors had Dwight Howard (to use an extreme example), are they a better defensive team? Of course. Are Steph and Monta also better defensive players as a result? OF COURSE! Having that defensive backstop is going to allow them to be more aggressive defenders, and even when they're blown by, Dwight is going to erase many of their mistakes. But if defense is to any large degree a personal venture, how is something outside of yourself supposed to improve your defense?

Further, and to account for this fundamental error, plus/minus proponents will say, well, we'll measure the TEAM'S defensive results with player X, as opposed to without player X, and the differential - positive or negative - will tell you how good a defender is *relative to his teammates*. Well, first off, is defense relative? Again, someone is a better or worse defender depending on who their teammates are? There is a disconnect there guys. And also, we don't play basketball in a lab where all other variables are controlled for - if Steph comes out and is replaced by Reggie, Reggie is not going to be in the same game situations, with the same four other players surrounding him, and the same five players opposite him. Again - that type of model pretends it is measuring what it says it is measuring, and then gives you a number that perfectly quantifies that players defensive contributions.

The ultimate lie is this: compare players net plus/minus contributions over time, as they are on different teams, and with and against different players. Are their contributions stable over time? I think you will find they are not. So then are players radically swinging back and forth from being good and bad defenders? I would say that's ridiculous - do you think if you put Jamal Crawford under Larry Brown he's suddenly going to become a good defender?

Maybe if he has Dwight Howard.

[floppymoose]

If the correlation with win% is why you like WP, then you should really move to my FM model. It correlates with wins better than WP. It is every bit as stable year to year as WP is.

And the claim by WoW crowd that defense account for a very small part of wins and losses is completely false. WP does not correlate well with wins and losses at all without the "defensive adjustment" applied to each team. It's a completely mandatory part of the model.

[Twinkie defense]

Yeah I didn't say that defense doesn't account for wins and losses, I said that INDIVIDUAL defense outside of things like rebounds, blocks, and steals don't matter very much in the scheme of things, for your average player.

Defense is a team activity, and the discrete individual defensive contributions for players can be accounted for fairly well by looking at measurables such as blocks, steals, taking charges, etc. I wish we had even more individual measureables to consider - for example, making someone make a tough pass or a difficult shot - but we don't, as far as I know. My problem is trying to deduce those aspects for an individual based on what happens as a whole with the team. That's like telling what the weather is like outside without having heat-sensing eyes - and I learned the hard way in NY that just because the sky is blue doesn't mean it's warm out.

But I would love to see the Floppy Moose model, assuming it is a real thing... and not "the team with the higher scores wins"

I would also love to hear any WoW detractors/plus/minus proponents address the specific points I made above...

[cellomac1212]

Is this one of those smart people only threads?

What the hell are you guys talking about?

Lol/JK...

Can you post the WP formula? I would like to know how they calculate it.

[old rem]

Unfortunately, no one else on our board apparently likes to read or do math.

Not so, I read many of John Steinbeck's classics during Trigonometry class.

Stats of any sort are but a way to quantify history..the facts of what DID happen,in a format allowing one to presume to have an idea of what WILL happen.

Lee was pretty much the lone soldier on offense inside,and rebounding for last year's Knicks. Now,he's in a tie for #3 option scorer for GSW. D Wright may not hit the contested midrange J at a high %,but his role is MUCH changed. His impact shows well in rebounds,asst,steals,blocks. He can be a + in a lot of ways. Reggie is pretty much THE bench...but last year was often playing starter minutes,was in the flow,and now he might play 30 one game...average 12 min the next two. Reggie is the only non starter averaging 10+ points. last year 5 non starters averaged 10+.
S Curry is a bit below avg on D relative to most G's. Lee is quite a bit below avg relative to most PF's and across the board,GSW's players get handicapped by below par defensive tactics that they are just stuck with.

Yes...a player who "scores a lot" needs to be efficient......BUT...there are guys who can be efficient on their 5-6 shots a game,and can't GET the shots to produce more. You COULD get another 6-7 ppg from Biedrins but that would be pushing it,as he'd need to be taking some that for him are low%. Ellis? I don't doubt you COULD tell the W's Monta has to score 50 tonight,and he's able to create the shots to GET 50. It would not,likely,be an efficient 50,but he'd MAKE a lot of shots most guys can't make.

What I trust most is the old Mark 1 Old Rem Eyeball,which has served well for decades. i DO like stats but there's no type of stat in basketball that tells the whole tale. You take from each,something. GSW getting 50wins? You could be droppin acid and not expect that. GSW gave offense,defense and depth to get Lee and expiring trash. Only having way fewer injuries has kept this group ahead of last year. Just the same,it's a pretty good starting 5 with a weak bench and a coach who's still learning the ropes. I figured .500 would be the MAX and where GSW is,is close to what seemed probable.