Fw: [閒聊] The 20-80 Scale, SABR Style

看板Prospect作者tanaka0826 (田中鬪莉王)時間11年前 (2013/02/21 16:52)推噓0(0推 0噓 0→)

留言0則, 0人參與討論串1/1

※ [本文轉錄自 MLB 看板 #1H9IQY1A ] 作者: tanaka0826 (田中鬪莉王) 看板: MLB 標題: [閒聊] The 20-80 Scale, SABR Style 時間: Thu Feb 21 03:40:01 2013 The 20-80 Scale, SABR Style 以數據觀點探討球探報告之給分方式 http://www.fangraphs.com/blogs/index.php/the-20-80-scale-sabr-style/ When scouts evaluate the players on the field, they use a 20-80 scale as shorthand to describe a player’s tools and/or his overall ability. Receiving a 50 on the scale means that one is major-league average, and for every 10 points up or down the scale, the scout believes the player is one more standard deviation above or below major-league average. An 80 is incredibly rare because one would have to be 3 standard deviations above the mean (or in the top 0.1-0.2 percent of players), and it’s a representation of the truly, truly elite. But the question becomes what those grades represent. When someone says that a player is an [insert grade], what should we actually expect them to do statistically at the major-league level? Armed with some advanced statistics and z-scores, I went to find out. 球探評量一名球員時，會以20-80之間的分數來速記球員的tools和(/或)整體的能力。50 分代表大聯盟平均，每往上十分與往下十分都代表著相差一個標準差。拿到80分是相當難能可貴的，因為80分代表著該球員的能力比平均值高出三個標準差(大約在前0.1%~0.2%之間)，是個菁英中的菁英。問題在於這些分數代表什麼？如果某個球員拿下某個分數，代表球探預期他會在大聯盟拿出甚麼樣的數據呢？以下我們將從進階數據以及 z-scores (得分-平均)/標準差來下手。 (Note: I used statistics from 2010-2012 because the timeframe is large enough to get a sample and small enough to stay within the recent run environment, and I used Russell Carleton’s measurements for each statistic to get a sample at least indicative of skill.) (註：以下採用的數據為2010-2012，因為時間長度得超過「小樣本」；同時也得取得夠短，才比較能看出一位球員在近期的表現。以下參考自 http://www.baseballprospectus.com/article.php?articleid=17659 ，以取出足夠看出某項技能之等級的樣本。) Hit Tool Hit Tool z BA Player BABiP Player 80 3 .336 Miguel Cabrera .383 - 70 2 .313 Josh Hamilton .357 Dexter Fowler 60 1 .290 Martin Prado .332 David Wright 50 .267 Rafael Furcal .306 Asdrubal Cabrera 40 -1 .244 Vernon Wells .280 Shane Victorino 30 -2 .221 Brendan Ryan .254 Mark Teixeira 20 -3 .199 - .228 - H - HR ( BABiP = Batting average on balls in play， ──────── ) AB - K - HR + SF You can certainly argue with the statistics I choose for each of these tools, but I preferred to use statistics with which you are already familiar to simply give you an idea of the spread. In regard to the hit tool, I could have used Contact%, but while the hit tool is defined as the ability to make contact, we usually imply some semblance of production with it -it doesn’t matter that the player makes a lot of contact if he doesn’t do anything with it. Looking at the chart specifically, Cabrera was the only 80 hitter, and he was the only hitter within 15 points of the .336 mark. All the way at the bottom, you see a lack of examples for a 20 grade, but it shouldn’t be too surprising that a 20 hitter wouldn’t get enough PA (1000 for BA, 1500 for BABiP) to be on the list. It certainly doesn’t mean 20 hitters don’t exist. 當然各項挑選出的數據並不是正確解答，但我偏好使用大家熟悉的數據以便大家了解。關於hit tool，我其實也能選用Contact%，但是hit tool的定義是make contact的能力，而Contact%可能會被假產物影響─例如常常碰到球但是都打不好。回到上表，Cabrera是唯一一位hit tool拿到80分的球員，打擊率.336甚至領先第二名.015 。至於最底下的20分那欄為什麼沒有人呢？因為20分的球員通常活不久，無法拿到足夠的打席數；而並不代表世界上沒有20分的人。(註：BA最低PA數需要1000；BAPiP則是1500) Power Tool Power Tool z ISO Player HR/PA Player 80 3 .294 Jose Bautista 6.6% Jose Bautista 70 2 .242 Joey Votto 5.2% Edwin Encarnacion 60 1 .191 Buster Posey 3.8% Kevin Youkilis 50 .140 Nick Markakis 2.4% Shane Victorino 40 -1 .089 Ryan Hanigan 1.0% Jose Altuve 30 -2 .038 Ramiro Pena -0.4% - 20 -3 -.013 - -1.8% - ( ISO = Isolated Power，SLG - BA ) ISO was the stat of choice here as it is the most commonly used power metric, and I used HR/PA to give a look with a stat that didn’t involve speed (SLG, and therefore ISO, give singles, doubles, and triples different weights when speed could be the deciding factor between getting one or the other). Again, these are here to give you an idea of what a certain grade would merit in the majors. Back to the list, Bautista is the only player in either list to get an 80, but Giancarlo Stanton is so close in both (.282 and 6.2%) that you could go ahead and throw an 80 on it. As for the negative numbers, it’s mainly just a glitch in the numbers (those negatives obviously aren’t possible). Emmanuel Burriss had the least amount of power with a .007 ISO and a 0.0% HR/PA. Again, 20 power guys don’t stick on MLB rosters very long (180 PA – I took out pitchers – was the restriction here). ISO是量測力量方面最廣為使用的數據，因此這裡採用ISO。我也在此使用HR/PA，藉此排除速度的影響。(ISO會用到SLG，而SLG的是由一壘安打、二壘安打、三壘安打以及全壘打組成，並各自給予加權─因此，能否多跑一個壘也會成為一項因素。)總而言之，這些東西主要是要讓大家了解某成績的球員之後在大聯盟的表現如何。上表中的Jose Bautista是唯一一位拿下80分的球員，不過Giancarlo Stanton也相當的接近(.282、6.2%)，所以給他80分也是很合理的。至於表中的ISO出現負值，主要是來自於統計上的小瑕疵(與平均值與標準差的大小有關；最爛應該是0)。 Emmanuel Burriss是ISO最差的球員，只有.007；同時，HR/PA是糟糕的0.0%。與Hit tool 相同，20分的球員通常無法站穩大聯盟，一下子就再見了。(註：扣掉投手後，至少 180 PA) Speed Tool Speed Tool z SB/3 Player BsR/3 Player 80 3 38 Ichiro Suzuki 9 Michael Bourn 70 2 29 Elvis Andrus 6 Rajai Davis 60 1 19 Chris Young 3 Andrew McCutchen 50 9 Jon Jay Yunel Escobar 40 -1 0 - -3 Jason Kubel 30 -2 -10 - -5 Adrian Gonzalez 20 -3 -20 - -8 Ryan Howard ( BsR = Base Runs ) Speed isn’t much easier to isolate outside of the traditional 60-yard dash and home-to-first times, but as I said earlier, we expect a certain production from the tool by the time a player reaches the majors. BsR/3 (I divided the cumulative BsR by 3 to give you an idea of what it would take per season) gives you a better spectrum of players in this instance as it can go into the negative range (0 SB players are slow, but they aren’t necessarily equally slow), and it incorporates other instances involving speed, such as going first-to-third. 最能獨立出速度的數據就屬傳統的60碼衝刺與跑到一壘的時間囉。不過如同前面所說的，我們要的是預測球員站上大聯盟之後該tool的能力等級。BsR/3(除以三的意義是把三季總合變成單季平均)更能讓你看出球員的速度對比賽造成的影響；同時結合了其它與速度有關的因素─例如從一壘衝到三壘。(備註：零盜壘的球員不快，但是不一定代表零盜壘的人通通都一樣慢。) Defense Tool Defense z UZR/150 Player Fld/3 Player 80 3 22.8 - 17.9 Brett Gardner 70 2 15.7 Adrian Beltre 12.4 Michael Bourn 60 1 8.7 Giancarlo Stanton 6.9 Carlos Gomez 50 1.6 Casey McGehee 1.5 Drew Stubbs 40 -1 -5.4 Dan Uggla -4.0 Rickie Weeks 30 -2 -12.5 Asdrubal Cabrera -9.5 Michael Morse 20 -3 -19.6 - -14.9 - UZR = Ultimate Zone Rating in runs above average (Arm+DPR+RngR+ErrR) (150：每150場) Fielding = Fielding Runs Above Average based on UZR (3：三年) Defense might be the hardest tool to look at in this situation, and while it might have been better to look at this position-by-position, the sample (needing 2500 innings) was already pretty small. According to UZR/150, there are no 80 defenders in the game, and although Fld/3 names Gardner, he was the only one on the list. Perhaps 80 defenders are usually bad enough at offense that they don’t get the playing time necessary for this query, or we may simply need a bigger sample. 防守應該是最難用這種方式分等級的tool。雖然說每個位置分開來看比較好，不過我們取的樣本數已經夠小了(2500局)。根據UZR/150，沒有任何80分的防守者。而從Fld/3來看的話也只有Brett Gardner一個而已。或許80分的防守者通常攻擊力都不怎麼樣，無法拿下足夠的打席。也或許我們的樣本(年數)需要擴大。 Arm Tool Arm Tool z rARM/3 Player ARM/3 Player 80 3 8.8 Alex Gordon 8.5 Jeff Francoeur 70 2 5.9 Jose Bautista 5.7 Alex Gordon 60 1 3.0 Jayson Werth 3.0 Jayson Werth 50 0.2 Austin Jackson 0.2 Matt Kemp 40 -1 -2.7 Ichiro Suzuki -2.5 Matt Holliday 30 -2 -5.6 - -5.3 Ryan Braun 20 -3 -8.4 - -8.1 - Arm - Outfield Arm runs above average (UZR) rARM - Outfield Arm Runs Saved runs above average (OF) I added this in just to show you I wasn’t ignoring it. I used these arm ratings, but they only exist for outfielders and include accuracy as well as arm strength. Radar gun measurements and/or FIELD f/x velocity measurements would probably be more helpful for objectively measuring this tool. 我列入這個表格只是要告訴大家我沒有忽略這部分。我會用這些Arm tool的數據，不過這只適用於外野手，而且同時牽涉到力量與準度。用測速槍和(或)球場的追蹤系統來進行調整的話可以讓這個數據更客觀的呈現外野手的Arm tool。 Fastball Velocity FB Velo z SP Velo Player RP Velo Player 80 3 97 - 100 - 70 2 95 Stephen Strasburg 97 Daniel Bard 60 1 93 Mat Latos 95 Drew Storen 50 91 Adam Wainwright 92 Ramon Ramirez 40 -1 88 Mike Fiers 89 Michael Wuertz 30 -2 86 Mark Buerhle 86 Pat Neshek 20 -3 84 Jamie Moyer 84 Livan Hernandez This one is the easiest to isolate. While none of the pitchers have 80 velocities on average, several of them are obviously able to touch or even sit in that range for a period of time. As for how I split up the data, I originally did it for both SP vs. RP and RHP vs. LHP. Using SP and RP demonstrated the differences between the velocities necessary for starting and relieving, and while I hoped the LHP and RHP would show the differences between the two, I got some weird results. The mean for the two were 91.2 (RHP) and 90.5 (LHP), which was expected, but when I applied standard deviations, the SD for LHP was larger (probably due to the much smaller sample). A 60-80 necessitated a higher velocity from a LHP than a RHP, which didn’t seem to make sense. Using the mean velocity difference of about 1 mph, you can dock the grades shown above by 1 mph and use that for lefties if you so choose. 球速是最容易獨立出來的數據。雖然沒人的「平均球速」能站上80分，不過有些投手至少能在一段時間之內碰到或保持在80分的速度。至於我怎麼分開這些數據呢？分成先發 vs. 牛棚以及左投vs.右投。前者的分法能看出先發與牛棚所需要的球速不同；蛋我希望能從後者的分法看出差異─我得到了奇怪的數據。右投手91.2mph、左投手90.5mph(這很符合預期)，但是在我帶入標準差後發現左投手的標準差較大(也有可能是因為樣本數較小才會這樣)─因此，左投手在60分-80分的球速是比右投手還要高的，感覺不大合理。或許你可以利用平均球速的差距(約1mph)進行調整，先將數據降低。 Control Control z SP BB% Player RP BB% Player 80 -3 1.7% - 2.0% - 70 -2 3.7% Roy Halladay 4.4% Sergio Romo 60 -1 5.7% Rick Porcello 6.8% Jason Motte 50 7.8% Derek Holland 9.2% Guillermo Mota 40 1 9.8% C.J. Wilson 11.6% Manny Parra 30 2 11.8% Danny Duffy 13.9% Tim Collins 20 3 13.9% Jonathan Sanchez 16.3% Carlos Marmol I considered using Strike% here, but there are strategic reasons to throw balls and BB% is more commonly used when talking about pitchers. Again, I split up relievers and starters, and as you might expect, starters walk fewer hitters than relievers, which is probably at least one reason why they’re starting as opposed to relieving. You’ll note that there are no 80 control guys, and no one was even within 1% of reaching that grade. Perhaps 80 control (in its literal definition) is too high of a standard here, but there ’s also the possibility that there is such a thing as “throwing too many strikes”, where pitchers somewhat choose to walk a guy even when they theoretically could avoid doing so. 其實我也有考慮過要用Strike%，但是Strike%會受投球策略的影響(i.e.配壞球)，而BB% 也是較常用來討論投手Command的數據。一樣的，這裡也成先發與牛棚兩邊。跟大家所想的一樣，先發投手的壞球率低於牛棚。從上表中可以看出並沒有Command 80分的投手─ 事實上，最接近的投手也比80分的BB%高出1%以上。大概是因為80分的標準過於嚴苛(以標準差來計算的話)；也有可能是因為所謂的「丟太多好球了」，選擇閃開而讓對手獲得保送機會才造成這個結果。 *** The point of all of this was simply to give us all an idea of what it would actually take to reach a certain scouting grade. How rare is a literal 80? How hard is it to sustain such elite performance? What does it mean to be “ plus” (60) in something at the major-league level? And how bad does one actually have to be to receive a 20? As prospect lists continue to roll out, you’ll hear these grades used frequently, and I just thought it was interesting and necessary to look at what it actually means to receive these scouting grades in our current environment. 這篇文章得主要目的在於告訴大家你在球探報告上看到的分數大概會是什麼樣子。80分有多難得？一直保持超高水準表現有多難？在大聯盟等級之下的"plus"(60分)是甚麼樣子？ 20分有多爛？新秀排名不斷的出現，也常常會看到誰的哪個tool幾分。所以我覺得告訴大家在這個時空背景之下拿幾分代表什麼意思應該蠻有趣的，同時也是大家該知道的。 -- "HARD WORK BEATS TALENT WHEN TALENT FAILS TO WORK HARD." -- ※ 發信站: 批踢踢實業坊(ptt.cc) ◆ From: 140.114.207.153 ※ 轉錄者: tanaka0826 (140.114.66.104), 時間: 02/21/2013 16:52:08 ※ 編輯: tanaka0826 來自: 140.114.66.104 (02/21 16:52)