[外電] Stats geek pokes holes in Mark Cuban's proof that Jason

[popstarkirby (POP先生)]

http://tinyurl.com/cs9a6j


ESPN.com's John Hollinger is probably the premier basketball stats geek. He's
also a guy who thought the Mavs got ripped off in the Jason Kidd deal from
the get-go.

So it's no surprise that Hollinger finds flaws in the Mavs' player and lineup
evaluation system, which ranks Kidd as the league's second most effective
player this season, per a recent Mark Cuban blog post.

Here's a sample of what Hollinger had to say about the system (ESPN Insider):

    The reason is something statheads call multicollinearity; in plain
English, that means that since the same players tend to play together most of
the time, it's tough to tease out to what extent each is impacting the unit's
results unless you have a massive number of observations.

    Kidd, for instance, hardly ever plays without Dirk Nowitzki (look at his
10 most common lineups), and rarely takes the court without Jason Terry at
his side. A statistical method to differentiate Kidd, Nowitzki and Terry
would rely on the relatively small samples of minutes when they're apart --
all while trying to solve the same equation for the other nine players on the
court. As a result, it takes multiple years of data to get anything
approaching sound results.

    While the Mavs undoubtedly feel like they've gotten a pretty fine season
from Kidd, the fact remains that they gave up two first-rounders to pay three
times as much money to a player who isn't as good. Half a season of adjusted
plus-minus data doesn't make a compelling case for altering that assessment.

Can Mark Cuban argue with multicollinearity?



一位自稱Mark Cuban的人的回應:

Posted by mark cuban @ 10:59 AM Wed, Feb 11, 2009

Hollinger makes my point. If you actually read my post, maybe you and john
havent ? It says that this program measures how well coaches put their
players in a position to succeed and how well they perform when placed there.

That is multicollinearity. The issue is that not all players perform when
placed in that position. Whether its coaching or the player, JKidd does a
better job of performing in the lineups he is placed in than all but one
other player to that point in time

And as i said in the blog "It also helps us understand what combinations of
players work well together, and which dont. "

Which is why Jkidd and dirk play alot together, but bass and damp dont.

He then goes on to say it takes years of data... well guess what, we have
8years of data. Regular season, post season.


If you actually read my post Ben, then you would see that John is making my
point. Even if he doesnt know it


--
※ 發信站: 批踢踢實業坊(ptt.cc)
◆ From: 220.136.180.120
→ popstarkirby:文章下面戰很大，Cuban會回應嘛
※ 編輯: popstarkirby    來自: 220.136.180.120      (02/12 05:56)
推 JimmyHsu0420:樓上那麼早起啊? 是還沒睡 還是太早起啊?
→ popstarkirby:算沒睡吧，一點就醒了
→ Sadosi      :←逛ptt到現在還沒睡
→ popstarkirby:剛打完球回來@@
→ Sadosi      := =好熱血 我是沒有人可以找 之前都會打一整個凌晨