clock menu more-arrow no yes

Filed under:

Replace the win

New, 3 comments

It's one thing to denigrate the pitching win as a meaningful stat, quite another to propose a useful replacement. It's just our luck that a couple of people have.

Yet more ways to illustrate how great a pitcher Clayton Kershaw is
Yet more ways to illustrate how great a pitcher Clayton Kershaw is
Scott Rovak-USA TODAY Sports

My last post discussed how the win has ceased to be a meaningful statistic with which to evaluate pitchers simply because pitchers don't have the opportunity to accumulate them like they did in the past. The cumulative effects of fewer starts per year and fewer innings pitched per start combine to reduce the chance a pitcher can get a decision, let alone a win. Pitchers receive no-decisions in almost thirty percent of their starts, and when a stat isn't used to that extent, it's time to replace it with something else.

Luckily there are two baseball geniuses who have created alternatives. The Bill James Game Score has been around for some time and looks like this:

50 + (3 * innings pitched) + (2 * complete innings after the fourth inning) + strikeouts -       (2 * hits) - (4 * earned runs) - (2 * unearned runs) - walks

This is the Game Score for Clayton Kershaw's no-hitter on June 18th, 2014:

50 + (3 * 9 IP) + (2 * 5 complete innings after fourth) + 15 KO - 0 hits - 0 earned runs - 0 unearned runs - 0 walks = 102 Game Score

Choosing a different game to get all of the elements involved, this is from Kershaw's no-decision against the Giants on May 11th, 2014:

50 + (3 * 7 IP) * (2 * 3 IP after fourth) + 9 KO - (2 * 7 hits) - (4 * 3 earned runs) - 0 unER - 0 BB= 60 Game Score

The Game Score is scaled in such a manner that 50 means pitchers win around fifty percent of games and so on. I've always liked the Game Score because it breaks down the elements a pitcher can control and eliminates those he can't. In addition, he's evaluated in every start, not just those in which he's lucky enough to have received enough offensive support to record a win or threw poorly enough for a loss.

Tom Tango added not one, but five additional Game Score methods, the latest of which he derived last December and can be found at his blog in three parts (Part 1, Part 2 and Part 3). This table lists these methods:

Methods Equation
Runs Allowed 40 + (6.4 * IP) - (10 * R)
K - BB 40 + (0.4 * IP) + [3 * (SO - BB)]
FIP 40 + (2.5 * IP) + (2 * SO) - (3 * BB) - (13 * HR)
Linear Weights 40 + (8.4 * IP) - (3 * BB) - [5 * (H - HR)] - (8 * HR)
Tango 2.0 40 + (2 * outs) + KO - (2 * BB) - [2 * (H - HR)] - (3 * R) - (6 * HR)

A couple points -- these are my interpretations of his equations, so any mistakes are mine. In addition, it's very important to point out he defined a walk, which is actually (walks - intentional walks + HBP).

I don't pretend to understand the underlying math behind the constants in Tom's equations, but I trust him so it works for me as a method with which to evaluate pitchers using something other than wins. This table shows the top 20 pitchers with complete careers since 1920 and at least 200 starts (100 starts for current pitchers), ranked by the Bill James Game Score:

Pitcher From To GS GSc Rnk TT1 Rnk TT2 Rnk TT3 Rnk TT4 Rnk TT5 Rnk
Sandy Koufax 1955 1966 314 62.8 1 61.0 5 57.2 6 57.2 5 63.9 1 64.6 2
Clayton Kershaw 2008 2014 209 62.2 2 62.4 1 57.2 5 59.1 1 63.4 4 65.0 1
Pedro Martinez 1992 2009 410 61.2 3 58.5 17 60.5 1 58.0 2 61.8 8 63.3 5
Bob Gibson 1959 1975 482 61.0 4 61.9 2 54.1 30 57.7 4 63.5 2 64.4 3
Tom Seaver 1967 1986 647 60.3 5 61.4 4 53.2 45 55.8 18 63.4 3 63.4 4
Randy Johnson 1988 2009 603 59.9 6 55.5 75 60.0 2 56.0 14 59.4 37 61.2 14
Nolan Ryan 1966 1993 773 59.8 7 56.3 59 54.3 27 55.6 21 60.6 20 61.3 12
J.R. Richard 1971 1980 221 59.6 8 58.0 21 52.5 67 56.5 13 61.5 9 62.1 7
Andy Messersmith 1968 1979 295 59.3 9 59.3 11 50.4 148 52.9 76 62.6 6 61.5 11
Juan Marichal 1960 1975 457 59.2 10 59.8 7 53.1 46 55.7 20 63.2 5 62.6 6
Roger Clemens 1984 2007 707 59.0 11 57.8 24 56.3 12 56.8 9 60.2 23 61.7 8
Johan Santana 2000 2012 284 58.9 12 57.4 28 56.9 8 54.3 42 60.0 27 60.9 19
Stephen Strasburg 2010 2014 111 58.7 13 55.4 76 58.7 4 56.5 12 58.6 57 60.3 26
Jim Palmer 1965 1984 521 58.5 14 61.8 3 48.1 291 52.2 111 62.0 7 61.3 13
Curt Schilling 1988 2007 436 58.5 15 56.7 42 58.9 3 56.8 10 60.1 25 61.6 10
Sam McDowell 1961 1975 346 58.4 16 56.0 65 52.2 76 54.0 49 58.5 62 59.7 40
Felix Hernandez 2005 2014 303 58.4 17 57.5 27 56.8 9 56.9 8 59.1 45 61.1 18
Don Drysdale 1956 1969 465 58.4 18 58.8 14 53.4 41 55.1 26 60.8 14 61.2 15
Bob Veale 1962 1970 255 58.1 19 57.8 25 51.6 96 56.5 11 58.7 51 60.7 20
Adam Wainwright 2007 2014 217 58.0 20 58.8 16 55.0 20 57.1 6 59.2 41 61.2 16

I don't recommend reading this on your phone. GSc is the Bill James Game Score, TT1 is the first Tango equation and so on. The ranks shown are in a pool of around 600 pitchers that can be viewed in this Google Docs spreadsheet, and clicking any column heading will sort the columns however you wish.

It's no surprise players of recent vintage are well-represented since strikeouts and walks play a key role in all the equations, and there are more strikeouts than in the past. There are some anomalies, but typically, if a pitcher strikes out a lot of hitters and keeps the walks low, he'll do well using any of these methods.

This shows the dominance of Sandy Koufax, and shows how close Clayton Kershaw is to being his Dodger heir apparent. While this is a good start, it will almost certainly not take off as a new way to discuss baseball -- walk up to your friends and casually mention "You know, Sandy Koufax had an average 62.8 in his starts using the Bill James Game Score, best of all time -- I told you he was the goods!," and you'll likely just get a blank stare. Absent getting new friends (something I highly endorse), there needs to be a better way to discuss this.

I'm not the biggest fan of counting stats since they can be misconstrued to confuse a long career with an outstanding one, but by setting thresholds pitcher performance can be teased out, such as counting up all the games with Game Scores of 55 or higher. This is a screen grab from the Google Doc:

Game Score

Using Clayton Kershaw as an example, in his 209 starts he's had a Bill James Game Score of 55 or higher 150 times, or over seventy percent of the time. The sort tabs can be used to differentiate between the counting stat (Nolan Ryan has the most such games, 502 of his 773 starts, almost sixty-five percent) and rate stat to see how current pitchers fare, and you can also choose the Tango equation of your liking. In addition, I created a Tableau data viz that might help you visualize the data better and allows for all kinds of filtering, but be forewarned that I consider it kind of a mess. How's that for an endorsement?

Stabs at using some measure other than the win have already occurred, the most obvious being the Quality Start. I have no issue with that other than I find it too generous. Using any of these Game Score methods drills down into the constituent elements that players can control and helps separate good starts from great. In addition, using any of these methods easily separates pitchers with long careers from those with long successful ones. Taking a look at the spreadsheet shows this quite dramatically.

Educating the baseball public on better ways to measure and evaluate players is tricky. However, over the years, the Quality Start has gained a foothold, and if broadcasters were to choose one of these six methods and hammer it home, constantly explaining it as well as showing who the best are currently and of all time, it could gradually take hold. Looking at the six methods shows little difference, and fans already are familiar with the constituent elements.

I'll leave it to people smarter than me to determine what to name these units. Kershaw has 150, but 150 of what? Effective starts, good outings, plus games, I don't know, my knowledge of catchy stat names is only matched by my mathematical ineptitude. But it can be done -- Wins Above Replacement is working its way into the broadcasting lexicon, WHIP is gaining traction, and a concerted effort to relegate the win into non-use and replace it with one of these methods could succeed. In addition, Baseball-Reference and FanGraphs can easily incorporate them as both counting stats and rate stats on their main pages, broadening their availability and making them more accessible to readers and researchers. Go ahead and pop off a tweet to Sean Forman and Dave Cameron suggesting this -- they're both dying to hear from you!

One last thing -- my last post elicited a comment from Al Yellon, who runs Bleed Cubbie Blue, the SB Nation Cubs site:

I predict at least one pitcher active today will post 300 wins.

This chart shows the lockstep correlation between wins and starts since 1980, right around when the no-decision rate hit thirty percent:

Games Started

Pitchers get wins in around fifty percent of their starts, and that's being extremely generous. Ten players tied for most starts in 2014 with 34, meaning they'd have to start that many games for almost eighteen seasons and win half of those starts -- that's quite a bit, which is why I think the 300-win career is a thing of the past. I rarely say never, but I think that's an unrealistic standard to hold pitchers to, which is why we need new measures.

All data from Baseball-Reference. Any mistakes in obtaining and processing the data are the author's.

Scott Lindholm lives in Davenport, IA. Follow him on Twitter @ScottLindholm.