View previous topic :: View next topic |
Author |
Message |
Ed Küpfer
Joined: 30 Dec 2004 Posts: 785 Location: Toronto
|
Posted: Fri Feb 18, 2005 1:10 pm Post subject: |
|
|
jkubatko wrote: | The RMSE using this method to predict team wins was 4.2 wins, which I thought was decent. |
Which it is. The RMSE with Pythagoreanish methods is between 3 and 4, so that sounds right.
jkubatko wrote: | You know, I didn't like the 3 WS = 1 win approach when James first preseneted it, but it grew on me. Here are two reasons why I like it: |
There's probably no reason to drive this one into the ground, so I'll just drop the whole thing. It's an argument about presentation, basically, not methodology, so it really isn't important in the grand scheme. _________________ ed |
|
Back to top |
|
|
jkubatko
Joined: 05 Jan 2005 Posts: 702 Location: Columbus, OH
|
Posted: Fri Feb 18, 2005 2:33 pm Post subject: |
|
|
Ed Küpfer wrote: | jkubatko wrote: | The RMSE using this method to predict team wins was 4.2 wins, which I thought was decent. |
Which it is. The RMSE with Pythagoreanish methods is between 3 and 4, so that sounds right. |
I also verified this empirically by trying values of x from .001 to .100 (incremented by .001). I think the one that produced the smallest RMSE was x = 0.078, but the differences between x = 0.08 and x = 0.078 were miniscule. I also was also happy that this backed up my intuition about a marginal team.
Ed Küpfer wrote: | jkubatko wrote: | You know, I didn't like the 3 WS = 1 win approach when James first preseneted it, but it grew on me. Here are two reasons why I like it: |
There's probably no reason to drive this one into the ground, so I'll just drop the whole thing. It's an argument about presentation, basically, not methodology, so it really isn't important in the grand scheme. |
Thanks for saving me from posting a similar thing. I would really like feedback on the methodology. Also, thanks for making me explain the 0.92 and 1.08 multipliers. I will add that to the explanation on my web site soon. _________________ Regards,
Justin Kubatko
Basketball-Reference.com |
|
Back to top |
|
|
Kevin Pelton Site Admin
Joined: 30 Dec 2004 Posts: 978 Location: Seattle
|
Posted: Fri Feb 18, 2005 4:37 pm Post subject: |
|
|
Mike G wrote: | If 2 players show 8 WS, I have to think "something between 7.5 and 8.5". |
Can you really find a tangible difference between a player with 7.5 and 8.5 win shares? At some level, differences are overwhelmed by the fact that these are merely estimates.
If you're going to create a rating called Win Shares, you are, for better or worth, trading on Bill James' work. That makes it necessary to follow James' conventions, including 3WS = 1 win and integers only.
I find it fairly interesting that the leaders in NBA Win Shares, at a glance, seem to have similar totals to leaders in MLB. |
|
Back to top |
|
|
HoopStudies
Joined: 30 Dec 2004 Posts: 705 Location: Near Philadelphia, PA
|
Posted: Fri Feb 18, 2005 6:56 pm Post subject: |
|
|
admin wrote: |
I find it fairly interesting that the leaders in NBA Win Shares, at a glance, seem to have similar totals to leaders in MLB. |
First of all, the method is definitely kinda cool. Another theoretical way to try to put a player's production in terms of something real (even if James did decide to go with win shares, rather than wins). Nice work, JK.
Second, I had noticed with individual win-loss records that they were comparable to what you get in baseball. Not that I'd ever seen such a thing in baseball, but James always talked about the top players in baseball being worth 10+ extra wins, which is about what I see in indiv w-l records. Justin's calcs reproducing baseball winshare numbers isn't a huge surprise.
The fun, though, is in asking why. Baseball has 2x the games. Baseball plays 9 guys on the field, and maybe 12 guys overall in a game with relief pitchers. We play 5 guys on the court and typically 9 guys overall. So that means distributing the credit (zeroth order) among 1.5 to 2x the guys. 2x the games divided by 2x the guys gives you about the same games per guy. 2x the games divided by 1.5x the guys gives you more games on the baseball side, but then you might say that the top guys in baseball don't have the opportunities to influence a game like the top guys in basketball. A big possession user in basketball gets about 1.5x the number of offensive possessions as an average guy and maybe 1.2x the number of defensive possessions (though most big offensive users are low on the defensive side). So 2x the games divided by 1.5x the guys and 1.2-1.5x the opportunities brings it again right back to about equivalent.
Just fun, but I've always found comparative sport stuff interesting... I have to imagine that football players can't be responsible for more than about a win or two in a season. Maybe it's more because a quarterback is involved in about 50% of all offensive plays. So maybe 4 games for a QB, which is a big deal. _________________ Dean Oliver
Author, Basketball on Paper
The postings are my own & don't necess represent positions, strategies or opinions of employers. |
|
Back to top |
|
|
Yyzlin
Joined: 31 Dec 2004 Posts: 27 Location: North Carolina
|
Posted: Sat Feb 19, 2005 2:08 am Post subject: |
|
|
admin wrote: | Yyzlin wrote: | Depends; do you actually believe that McGrady conciously just decided to stop playing defense and driving to the basket, or perhaps the team context had a large part to do with that? |
But how do you decide (statistically) when a player is really worse and when he's just playing at half-speed because his team is so bad? How do you design a rating system that takes that into account? Is the point of a rating system to evaluate how players did that season, or how they're going to fare going forward? |
That's my point though. If McGrady, whose peak was much higher than Daugherty's suffered such a supposed drop-off because of the poor play of his teammates, would it not be assumed than Daugherty might experience similar troubles? |
|
Back to top |
|
|
Yyzlin
Joined: 31 Dec 2004 Posts: 27 Location: North Carolina
|
Posted: Sat Feb 19, 2005 2:09 am Post subject: |
|
|
jkubatko wrote: | Yyzlin wrote: | Depends; do you actually believe that McGrady conciously just decided to stop playing defense and driving to the basket, or perhaps the team context had a large part to do with that? |
I believe the former, as McGrady so much as admitted in Sports Illustrated about a month ago that he dogged it at times last year. |
So, right at the beginning of the season, McGrady just decided to not go full speed? Please. It was obviously only through a good portion of the season had passed when the team was obviously not going anywhere that McGrady would dog it. |
|
Back to top |
|
|
Ed Küpfer
Joined: 30 Dec 2004 Posts: 785 Location: Toronto
|
Posted: Sat Feb 19, 2005 4:37 am Post subject: |
|
|
Just for fun. I have no comment.
_________________ ed |
|
Back to top |
|
|
Ed Küpfer
Joined: 30 Dec 2004 Posts: 785 Location: Toronto
|
Posted: Sat Feb 19, 2005 4:45 am Post subject: |
|
|
Justin: Any chance you could post a complete season's worth of Win Shares for all players so's I can fiddle? Any season will do. _________________ ed |
|
Back to top |
|
|
Mike G
Joined: 14 Jan 2005 Posts: 3578 Location: Hendersonville, NC
|
Posted: Sat Feb 19, 2005 6:52 am Post subject: |
|
|
Ed -- the graphs are really cute. The 5-game clustered stats are more, uh, viewable even. What about doing season-cumulative averages, so's we could see how a player's whole year is adding up. For the current season, Vince Carter's move Up has been noted. You could see when Garnett "peaked". |
|
Back to top |
|
|
Ed Küpfer
Joined: 30 Dec 2004 Posts: 785 Location: Toronto
|
Posted: Sat Feb 19, 2005 12:11 pm Post subject: |
|
|
_________________ ed |
|
Back to top |
|
|
kjb
Joined: 03 Jan 2005 Posts: 864 Location: Washington, DC
|
Posted: Sat Feb 19, 2005 12:44 pm Post subject: |
|
|
Eyeballing the cumulative averages graph, it looks like McGrady may have started slacking off defensively in late December. |
|
Back to top |
|
|
Yyzlin
Joined: 31 Dec 2004 Posts: 27 Location: North Carolina
|
Posted: Sat Feb 19, 2005 1:25 pm Post subject: |
|
|
Yeah. Very nice graphs. It does appear that McGrady had a noticable drop in defense starting around the end of December. |
|
Back to top |
|
|
S.K.
Joined: 18 Feb 2005 Posts: 61 Location: Toronto
|
Posted: Sat Feb 19, 2005 4:25 pm Post subject: |
|
|
Interesting that his offense traveled upwards from that point at almost the same rate - so not a general malaise, just a concerted avoidance of doing anything he personally wouldn't benefit from (ie stat padding). _________________ No books - no articles - no website.
Just opinions.
Ill-informed opinions. |
|
Back to top |
|
|
Mike G
Joined: 14 Jan 2005 Posts: 3578 Location: Hendersonville, NC
|
Posted: Mon Feb 21, 2005 11:54 am Post subject: |
|
|
Having hand-entered every 2004 "win-share" and "player win" number at b-r.com, I went looking for players who have the biggest difference (after multiplying PW by 3). PW favors these players, relative to WS:
D-J PW3 WS player
9.1 29.4 20 Shawn Marion
8.7 30.0 21 Donyell Marshall
7.7 24.0 16 Marcus Camby
7.6 33.9 26 Ben Wallace
6.5 22.8 16 Samuel Dalembert
6.4 23.7 17 Kenyon Martin
6.3 36.6 30 Andrei Kirilenko
6.1 32.4 26 Jermaine O'Neal
5.6 27.9 22 Lamar Odom
5.4 38.7 33 Tim Duncan
All these guys are big rebounders, or they have big defensive numbers (steals, blocks), or both.
Here are players that get more credit in the WS method than in PW:
D-J pw3 ws
-6.9 35.4 42 Predrag Stojakovic
-6.5 28.8 35 Kobe Bryant
-6.4 21.9 28 Reggie Miller
-4.9 23.4 28 Gary Payton
-4.8 25.5 30 Mike Bibby
-4.5 31.8 36 Sam Cassell
-4.5 22.8 27 Steve Nash
-4.3 21.0 25 Jeff Foster
-3.9 14.4 18 Stephon Marbury
-3.7 0.6 4 Milt Palacio
The negative sign just means this is the bottom of the list, inverted. All these guys are high-TS%, high Ast/TO, or named Milt.
I can't seem to come up with an equivalent system that is as close to either of these as they are to one another. And I'm not sure that I want to. But what I've got right now is something I could call "expected win shares with an average team". And this just means there's no a priori bias toward guys on good/bad teams. A great player on a rotten team should produce substantial wins on an average team.
Here are players that look better in my "T2" system than in Justin's WS column:
M-J mT2 WS
14.1 29.1 15 Vince Carter
12.9 20.9 8 Allen Iverson
11.4 26.4 15 Lebron James
11.1 29.1 18 Paul Pierce
10.4 20.4 10 Jamal Crawford
9.9 28.9 19 Tracy Mcgrady
9.2 13.2 4 Drew Gooden
9.0 29.0 20 Zach Randolph
8.7 22.7 14 Jason Richardson
8.2 28.2 20 Baron Davis
As expected, all players from mediocre-to-bad clubs.
Comparing my little list to the PW list:
M-D mgT2 PW
11.9 26.4 14.7 Lebron James
11.4 20.9 9.6 Allen Iverson
11.3 29.1 18.0 Vince Carter
9.1 20.4 11.4 Jamal Crawford
8.9 18.1 9.3 Joe Johnson
8.9 13.2 4.5 Jalen Rose
8.7 28.9 20.4 Tracy Mcgrady
8.3 19.3 11.1 Antoine Walker
8.0 14.8 6.9 Quentin Richardson
7.9 12.9 5.1 Ronald Murray
Again, the PW figure is 3x that shown at b-r.com. Lebron's number is almost identical in the other 2 lists. |
|
Back to top |
|
|
Ed Küpfer
Joined: 30 Dec 2004 Posts: 785 Location: Toronto
|
Posted: Thu Feb 24, 2005 2:46 pm Post subject: |
|
|
For years I've been using a Wins Above Replacement measure that gives out pseudo-Win Shares results. I calculate is like this:
Code: | WAR = Games * (Win% - x) |
where Games is Oliverian wins plus losses, Win% is the ratio of wins to Games, and x is a replacement level. I typically set x at 10% or 20%, calling the results WAR10 and WAR20 respectively.
I compared WAR*3 to player WS over the last 5 season, and the results were very close. In the WAR20 sample, 1381 of the 2401 players (58%) came within +/- 3 WS. WAR10 was even better: 1846 players (77%) came within 3 WS. Finally I tried WAR5, using 5% as my replacement level WIN%, and got the best results of all: 2040 players (85%) were within 3 WS. I consider this supprt for the the WS methodology. _________________ ed |
|
Back to top |
|
|
|