stanleybostitch wrote:

SupremeCommander wrote:The only way to come up with a statisitc and correlate it to something (the article says "using the metric "Wins Produced" that predicts how statistics correlate to winning.") that I know of is to run a linear regression.

essentially, without knowing this dude's model, it would look something like:

(y) = (b1)(x1) + (b2)(x2) + (b3)(x3) + ... + a + e

or

(wins) = (b1)(points) + (b2)(rebounding) + (b3)(blocks) ... + (unexplained) + (error)

where the "b" is the correlation, or slope of a line, between the "x" and the "y"/wins

there are a whole slew of ways to **** one of these models up. There is a ton of theory on this too, and it's called econometrics. Basically, it attempts to forecast something based upon the relationship between two variables. There are a lot of statistics to weigh in basketball, and that leaves a lot of room for error, because the less variables the better the model.

It is impossible to explain what the output of the model means without knowing how the model is constructed and what the inputs are. But what he's essentially trying to draw a line through a bunch of data points

Thank you Nate Silver!

haha... well thanks. I wish I understood it like he does...

TMS wrote:

Bonn1997 wrote:Gallo is significantly below average for a starter in the NBA though. Does he rebound well? No. Does he create shots for his teammates well? If so, I see no statistical evidence of it. Does he block shots well? No. Does he get a good number of steals? No. The only two things he does better than the average starter are shoot from the perimeter and draw fouls. Although he has good skills in those two areas, he only utilizes those skills well enough to put up 15 PPG.

first of all i disagree w/u if u think Gallo's a below average starting NBA player... but let's stick to the topic here... if you think Gallo's so below average, then why even hedge on trading him, AR & filler for Carmelo Anthony? does that make sense to you? what's your point here, that u wanna lend some kind of validity to this chart that ranks guys like Carlos Delfino ahead of Amare Stoudamire? or Thabo Sefolosha way above Tony Parker? do u really think those guys contribute more to their teams' wins? the list is below for your reference... go ahead & peruse it & see what kind of conclusions u can reach:

http://arturogalletti.files.wordpress.com/2010/12/untitled37.png

ben wallace > kobe... BWAHAHAHAHAHA.

Bonn1997 wrote:One could teach a whole class on test validation but one way to validate a measure (which it looks like the guy did with this measure) is to correlate it with the outcome(s) you're trying to predict--in this case, correlating the total of each individual's WP score on a team with the team's overall win-loss record. (Teams whose 12 players combine for the high wins produced totals should have more wins.) In addition, you'd want to show that your measure correlates better with that outcome than do other measures like Hollinger's PER. Normally, when you get an anomalous piece of data, you don't dismiss it but you don't obsess over it either. If you get a surprising blood test result, you repeat the test. Here, if a player has a surprising WP score, I'd look at his score in each of the past several seasons. I'd also keep in mind that a test doesn't have to be perfect to be useful. If a blood test is correct 60% of the time, it's more useful than flipping a coin, and in the long-run you'd be better off going with the blood test than coin flipping to make decisions. That means a test can be wrong sometimes and still be useful.

Bonn, I understand what you are saying but I have to disagree with the assessment. Some things are so fluid and have so many variables that a test measuring an individual aspect is not useful as is the case with this. This seems to skew in favor of multidimensional players and seems to value as a per 48 minute stat. I can best explain it as a hypothetical example:

Player A:
40 mins per game. 22 pts-8 rebs-3 ast-1 stl-1blk. The player scores his points on say 7-16 from the floor(on average) and 4-7 from the line (on average). Say the player averages 35% from 3. Say this player is the main cog on a .500 team

Player B:
20 mins per game. Say 12pts-4 rebs-2 ast-1 stl-.5 blk. The player scores his points on 5-10 from the floor, 2-3 from the line and shoots 37% from 3. Say this player is a role player on a .600 team.

This stat seems to value efficiency and multidimensional ability. I bet if you plugged in these numbers player B would have a higher score. There are guys who put up numbers who just don't seem to be winners (Jamison, Abdur-Rahim) and there are guys who don't put up numbers who seem to win all the time (Robert Horry comes to mind). Just because a guy happens to be on a winning team doesn't mean he should be measured in forming a baseline for a test on winning (Hi Nate).

I think the guy who formed this test had good intentions but at the same time I think he limited his information and did not include ALL variables.

Oh and before you look. Being worse than Lebron in something does not mean you are a worthless POS basketball player. You can probably be slightly worse than him and still be a max player and possible HOFer.

1. Effective field goal %age. Basically just FG% adjusted for 3 pointers. Not too crazy of a stat. We already use FG% in evaluating players. Someone who scores 20 pts a game but shoots 40% may not be as efficient or valuable as someone who scores 20 pts a game shooting 50%.

Average small forward 48.2 %
Lebron 50.9% career.
Melo 47.6% career.

This surprised me because Melo is a nasty shooter.

Possible explanations:

Melo, like other # 1 options may have to take shots at the end of the shot clock. I was watching last night's game against the Heat and I noticed he'll often take a contested shot in that circumstance but he'll go get his own rebound and put it back in. Melo will also be the guy at the end of quarters who will get the ball in an ISO situation and have to create. This might bring his FG% down because he's not just getting open looks of screens or from his man playing off him to double a post presence. Nene is actually not that bad in the post and can command a double team against smaller teams but he's no Tim Duncan or Amare on offense.

See..wasn't that much more of a fun conversation than

Melo sucks because his effective FG% isn't 50% like Lebron! He is worse than Jared Jeffries!

No effective FG% sucks! This is saying Melo is worthless. Rawr!!!!

2. FT % FTs attempted

Melo is better at getting to the line and hitting than most SFs. Obviously some of it has to do with the fact that Melo (and Lebron) shoot more than the average SFs and they should be. I like the fact that Melo is strong and can finish through contact.

3. Rebounds per 48.

For his career Melo is an above average rebounder for a SF. For this season he is a beast on the boards.

4. Steals

Slightly below average. Would it be nice if your franchise player was an above average thief? Sure. But again hard to be above average in EVERYTHING unless you're Lebron or Jordan. And even Lebron has that FT% below average.

5. Turnovers.

Somewhat worrying but again, when you have the ball in your hands a ton and you're driving a lot it happens. This is why ideally I prefer seeing even Amare catching the ball on his way to the hoop of a PnR or from off the ball movement because he is nasty and defenders can't try to poke the ball away.

6. Blocked shots

Let's hope what he's doing this season becomes the norm. There's no reason why it can't be, right?

I mean overall if you look at the formula it's not that crazy.

Win score =

- How many points you score (per 48 minutes I think)
- How many rebounds you get
- How many steals you get

These I have no issue with.

- Blocks/2 . Blocks, I get. Why divide it by 2? Anyone know
- Assists. Ditto. Only reasonable explanation is that you need whoever you throw the ball to, to finish.

I get why you subtract turnovers. Fouls I kind of have an issue with because you can argue it's either due to being a lazy defender and reaching instead of moving your feet or a sign that you're actually trying on defense.

Don't understand why you subtract FGA and FTs.

So obviously not a perfect formula but the standards ARE the same for every player and it does for the most part use stats we already pay attention to.

Bonn1997 wrote:

joec32033 wrote:

Bonn1997 wrote:One could teach a whole class on test validation but one way to validate a measure (which it looks like the guy did with this measure) is to correlate it with the outcome(s) you're trying to predict--in this case, correlating the total of each individual's WP score on a team with the team's overall win-loss record. (Teams whose 12 players combine for the high wins produced totals should have more wins.) In addition, you'd want to show that your measure correlates better with that outcome than do other measures like Hollinger's PER. Normally, when you get an anomalous piece of data, you don't dismiss it but you don't obsess over it either. If you get a surprising blood test result, you repeat the test. Here, if a player has a surprising WP score, I'd look at his score in each of the past several seasons. I'd also keep in mind that a test doesn't have to be perfect to be useful. If a blood test is correct 60% of the time, it's more useful than flipping a coin, and in the long-run you'd be better off going with the blood test than coin flipping to make decisions. That means a test can be wrong sometimes and still be useful.

Bonn, I understand what you are saying but I have to disagree with the assessment. Some things are so fluid and have so many variables that a test measuring an individual aspect is not useful as is the case with this. This seems to skew in favor of multidimensional players and seems to value as a per 48 minute stat. I can best explain it as a hypothetical example:

Player A:
40 mins per game. 22 pts-8 rebs-3 ast-1 stl-1blk. The player scores his points on say 7-16 from the floor(on average) and 4-7 from the line (on average). Say the player averages 35% from 3. Say this player is the main cog on a .500 team

Player B:
20 mins per game. Say 12pts-4 rebs-2 ast-1 stl-.5 blk. The player scores his points on 5-10 from the floor, 2-3 from the line and shoots 37% from 3. Say this player is a role player on a .600 team.

This stat seems to value efficiency and multidimensional ability. I bet if you plugged in these numbers player B would have a higher score. There are guys who put up numbers who just don't seem to be winners (Jamison, Abdur-Rahim) and there are guys who don't put up numbers who seem to win all the time (Robert Horry comes to mind). Just because a guy happens to be on a winning team doesn't mean he should be measured in forming a baseline for a test on winning (Hi Nate).

I think the guy who formed this test had good intentions but at the same time I think he limited his information and did not include ALL variables.

It's not clear there is much of a difference between those two players. We humans are just so used to looking at stats without considering mpg. I'd agree that multidimensionality helps players in any statistical analysis, but I think it *should* help them. I think the major reason Jamison and Abdur-Rahim is that statisticians haven't (yet) developed good ways to measure man-to-man and help defense (and you're right that this measure doesn't take into account every factor--I don't think any test in any field does). If Carmelo was playing outstanding man to man and help defense or hustling after every loose ball, I'd say that he's bringing a lot to the table that this test is missing. That's obviously not the case though. None of this is saying that he's a bad player, and I think some are interpreting it that way. His win production was not bad; it just wasn't at superstar/max contract level.

But there are huge differences in roles in those 2 players. If player A has to score and the other team game plans for that and doubles him vs player B who gets his points against the 2nd team. Player A has more value.

We agree that no test can measure everything, but there are tests that measure out so bad or whose method is severely flawed that they shouldn't be used as a measuring stick.

In this particular test/method, roles are not accounted for. Scores against double teams are not accounted for which could lend itself a little to Carmelo's inefficiency.

I can see where you say the players are similar but only in regards to a per minute statistical analysis. If teams are focusing on player A, and player B is getting his stats while a team is focusing on someone different the stats should be weighted as such.

I purposely made the stats look the way they do with no players in mind, but player A looks like a guy like Carmelo while player B puts up stats similar to a Carlos Delfino or Ilyasova.

This stay just has way to many variables out of the referenced players control for ot to have any meaning to me.

SupremeCommander wrote:I'm glad I have the stat I need to prove that, yes, Kevin Durant does in fact suck donkey balls

Kevin Durant had an incredible year last year, but he got off to a poor start for him (relative to last year) this year. His shooting efficiency was down sharply early in the season. Over the last few weeks he made some adjustments to his shot (according to HIM) and he is playing a lot better now. His total wins will probably be down for the year because of the bad start (on any model).

I don't agree with all the weights he gives to various factors to come up with his ratings. However, first you have to understand that these numbers are POSITION ADJUSTED. What that means is that each player's production is compared to what is typical for his position. That's why some of the ratings don't seen to make sense.

Here's an example:

Let's say Amare get 9 rebounds and Fields gets 7 rebounds. Most people would say Amare had the better game.

On this model, Fields will rate a LOT higher because 7 rebounds is GREAT for a SG and 9 rebounds is barely average for PF/C. Get it?

The most significant thing is that points per game is not weighted very high. What is most important is your scoring efficiency. If you score efficiently you get more credit for your scoring. If you score a lot of points efficiently you get a ton of credit. If you score a lot but not very efficiently, your scoring is not anywhere near as valuable as the average fan, media person or even former player thinks and you get less credit for your scoring.

Personally, I don't agree with his formula exactly, but I am 100% certain that efficiency is critical to a proper evaluation. Melo for example if not a very efficient scorer. So there's no way his scoring is as valuable as Lebron's and other high efficiency scorers. Its a mathematical certainty that Melo is overrated as a scorer based on what he is doing for Denver.

ItalianStallion wrote:I can tell most of you are totally unfamiliar with Berri's work and model.

I don't agree with all the weights he gives to various factors to come up with his ratings. However, first you have to understand that these numbers are POSITION ADJUSTED. What that means is that each player's production is compared to what is typical for his position. That's why some of the ratings don't seen to make sense.

Here's an example:

Let's say Amare get 9 rebounds and Fields gets 7 rebounds. Most people would say Amare had the better game.

On this model, Fields will rate a LOT higher because 7 rebounds is GREAT for a SG and 9 rebounds is barely average for PF/C. Get it?

The most significant thing is that points per game is not weighted very high. What is most important is your scoring efficiency. If you score efficiently you get more credit for your scoring. If you score a lot of points efficiently you get a ton of credit. If you score a lot but not very efficiently, your scoring is not anywhere near as valuable as the average fan, media person or even former player thinks and you get less credit for your scoring.

Personally, I don't agree with his formula exactly, but I am 100% certain that efficiency is critical to a proper evaluation. Melo for example if not a very efficient scorer. So there's no way his scoring is as valuable as Lebron's and other high efficiency scorers. Its a mathematical certainty that Melo is overrated as a scorer based on what he is doing for Denver.

you're right, I am unfamiliar with his work. But I don't think he understands "correlation does not imply causation" as statistics do not imply winning. See: Lee, David. See: Marbury, Stephon. He's suggesting you need to get certain stats from certain positions in order to win. I think that's a ridiculous premise to build any model upon and I think the numbers are worth as much as the nasty coffee deuce I'm about to drop