Tangotiger Blog

Sunday, December 14, 2014

Path to Game Score 2.0 - part 3 of 3

We're simply going to merge our equation from part 1 with our equation from part 2. The question is how much to balance each one. We know that a component-based metric is more indicative of the pitcher's talent than a run-based metric. In our component-based metric, we did our best to account for the fielders, while our run-based metric will be heavily influenced by the fielders. The split that we're looking at is going to end up somewhere around 70/30 or 60/40. Going with a 70/30 split will give us this (rounded):

• 40
• +3 K
• +2 other outs
• -2 walks
• -2 nonHR hits
• -3 runs
• -9 HR

One thing about the HR: because of all the rounding, and because of the run environment, its value should be somewhere between 4.2x and 4.7x of the walk (using Linear Weights or RE24), so, closer to -9.

We also know that a straight K minus BB metric works just as good as FIP. Which really indicates that the HR has a good amount of random variation to it. However, the HR did in fact happen, and we can't really credit it to anyone other than the pitcher. Well, a bit to the park. And I think it's reasonable to make some adjustment, however tiny, to acknowledge that. To that end, I'm going with a value of 4x to the walk, and so I'll use a value of -8 for HR.

And a little tidyup regarding strikeouts v other outs: giving +3 for the K and +2 for other outs is the same thing as giving +2 for all outs and an extra +1 for the K. So, that's how I'll present it.

And the same with nonHR hits being -2 and HR being -8. We can make all hits as -2, and HR get an extra -6.

Finally, while I give the starting point as "40", that should be set to that particular season so that the league average is "50". So, maybe it'll be 42 one season and 39 another season.

Therefore, here's my proposal to you guys:

• 40
• +2 outs
• +1 K
• -2 walks
• -2 hits
• -3 runs
• -6 HR

We can compare this to the Bill James classic (though adjusted) version. The outs value is the same. The K value is the same. The nonHR hits value is the same. James gave -4 for ER and -2 for UER (for a weighted average of around -3.8, so fairly inline with my -3, whereas mine has a bit more DIPS behind it). He gave -1 for walk and I have -2, and as I explained in the past, I think -2 works better. And while James has the classic version at a starting point of 50, his adjusted unofficial version is 35. I have it around 40. And the HR, which James didn't have at all.

***

Also, don't run correlations. At least, not the ones I think you are going to run. The best-fit would be something like IP/2-RA. And that's it. Why? Because runs is what correlates to wins in that same game. If you have runs, you don't need K and BB and HR and hits. You don't need any of it. As well, a SP that allows 1 run in 2 IP will likely have his team win as many games as giving up 2 runs in 4 IP or 3 runs in 6IP or 4 runs in 8IP. That is, the bullpen will cover 1 or 7 innings just about at the same performance level. But, we don't want a metric that equates the starting pitcher the same in all these cases.

Therefore, to better accurately judge the pitcher, we need to make it look more like WAR than like WAA (that is, wins above REPLACEMENT v wins above AVERAGE). Please, be careful with your correlations.

Using the terrificly presented data at Fangraphs:
http://www.fangraphs.com/leaders.aspx?pos=all&stats=sta&lg=all&qual=0&type=0&season=2014&month=0&season1=1993&ind=0&team=0,ss&rost=0&age=0&filter;=&players=0

We can easily calculate our base point:
Year base
2014 36
2013 38
2012 39
2011 38
2010 39
2009 41
2008 41
2007 41
2006 42
2005 40
2004 42
2003 41
2002 40
2001 41
2000 43
1999 43
1998 41
1997 40
1996 42
1995 41
1994 41
1993 40

The average is 40, so, I’d recommend going with that for something quick.

Naturally, in lower run environments, the base would be lower. The base point in 1968 was 30.

Top 10 seasons in Game Score, since 1993:

GameScore Year Pitcher
79 2000 Pedro Martinez
76 1999 Pedro Martinez
75 1994 Greg Maddux
74 1995 Greg Maddux
73 1997 Pedro Martinez
72 1997 Roger Clemens
71 1999 Randy Johnson
70 2014 Clayton Kershaw
70 1995 Randy Johnson

***

There were 233 seasons with at least a Game Score of 60 (out of 2840 seasons of 18+ starts). Pitchers with at least 4 such seasons:
10 Pedro Martinez
10 Randy Johnson

9 Curt Schilling

8 Greg Maddux (remember, since 1993 only)
8 John Smoltz
8 Roy Halladay

7 Roger Clemens (remember, since 1993 only)
7 Kevin Brown (most underappreciated pitcher of our generation)
7 Mike Mussina

6 Johan Santana (highly appreciated, but won’t be remembered as such)
6 Roy Oswalt (see Johan)

5 Clayton Kershaw
5 CC Sabathia

4 Felix Hernandez
4 Justin Verlander
4 Adam Wainwright
4 Cliff Lee
4 Jake Peavy

Missing:
3 Glavine
2 Pettitte

***

The one-hit-wonder belongs to Jeff D’Amico. Came into the majors very young, and held his own as a young pitcher. Then lost two full seasons. And came back to have the year of his life (one of the top 10% seasons of 1993-2014).

And then…. four years at replacement level play, and it was over.

The best one-hit wonder of the last 20 years?