The Big Picture: Re-calculating the Save

Two weeks ago I proposed a new way of defining saves. The save had been instituted to reward the evolving role of the fireman, the reliever who came into a game in a tight situation to save the day. However, today’s closer rarely faces a game on the line situation, unless he creates it himself. The rule defined the role over the years, turning saves into relatively passive activities. It’s time to refine the rule, or introduce a more active statistic to move the closer back to the role intended, the pitcher who saves the day.

Which is where I came up with my proposal–rather than awarding saves to the person who finishes the game, the reward goes to the pitcher with the best performance in terms of increasing the team’s chance of winning while also meeting a threshold increase of 0.1. Unlike the 1970s, when baseball defined the save rule, we now know the probability of a team winning at every point in the game. Given the score, base situation, and inning when the reliever enters and leaves the game, we can tell how much he helped shift the probability toward winning.

That was the theory. This week’s column puts the theory into practice to find out how the proposed rule changes the number of saves awarded and which pitchers benefit and decline under this scheme. The play-by-play data used comes from Baseball Info Solutions and covers 2002-2006. William Burke and Keith Woolner supplied the Win Expectancy data, and I’m grateful for their help.

For every game in that time frame, queries determined the last event in a game in which a team was tied or trailing. Relievers entering the game after that point were eligible for a save, provided they did not get the win. For each reliever eligible, the inning, half-inning, outs, base situation, and score difference was recorded both for the event in which he entered the game and the event in which he left the game. Queries then calculated the win expectancies for both points, and the difference became the score for the pitcher for the game. If the score was…

• greater than 0.1 and the highest score for the game
• or greater than or equal to 0.3

… then the reliever was credited with a save. Note that the team’s chance of winning has to increase by at least 0.1–to spell it out, that means if a team already has better than a 90 percent chance of winning, the reliever can’t get a save.

The data shows that the new proposal eliminates about 40 percent of current saves:

Save Stats, 2002-2006

Current Rule	6108
New Proposal	3623
Difference	2485
Pct. Change	-40.7%

A much higher percentage of saves under the current rule are eliminated, almost 65 percent. Over one-third of those lost are shifted to other pitchers with the new proposal, however:

Saves gained and lost, 2002-2006

Current Rule, Saves Lost	3964
New Proposal, Saves Gained	1479

The data shows that a large number of saves are going to pitchers who enter the game in high-probability win situations. This rule shaped strategy and contract negotiations for a generation, and yet it is mostly applied in situations where the probability of winning is already better than nine in ten. The distribution of the various parameters shows us how the new proposal shifts save opportunities. First, the distribution of inning entered:

Distribution of Innings

Inning	Current	New
1-5	2	18
6	34	331
7	99	737
8	811	960
9+	5162	1577

As you can see, many of the ninth inning saves are eliminated. Basically, if you enter the start of the ninth and the lead is two runs or more, there’s no save opportunity. Notice, however, that pitchers in the sixth and seventh inning get many more chances to pick up a save.

The next table shows the distribution of outs when entering the game:

Distribution of Outs

Outs	Current	New
0	4922	2467
1	419	576
2	767	580

The distribution is not all that different. A higher percentage of saves comes with outs already recorded in the inning, but the great majority still occur with a pitcher entering with no outs. Pitching a full inning without allowing a run raises the win expectancy by over 0.1, and that gets higher the later in the game a pitcher enters.

Now for the base situation the pitcher faces entering the game:

Distribution of Base Situations

Men on Base	Current	New
0	4901	2187
1	357	408
2	120	162
3	44	43
12	327	392
13	150	133
23	59	97
123	150	201

p>

Again, the number of saves awarded in the relatively easy situation of entering the game with no one on is reduced. But while the number of saves rewarded in tougher base situation goes up, the current rule actually does a pretty good job of rewarding pitchers who enter games with runners on base.

The distribution of the lead a pitcher inherits changes things quite a bit:

Distribution of Score Difference

Lead for Pitcher	Current	New
1	1918	2358
2	2023	903
3	1798	295
4	259	57
5	68	9
6+	42	1

Under the current rule, saves are pretty much evenly distributed over one-, two-, and three-run leads. Again, the rule for awarding a save by starting the ninth with a three-run lead or less shapes the distribution, but under the new proposal, the distribution is heavily weighted to games with a smaller score difference. Another way to look at this is by subtracting inherited runners from the lead, a measure of potential scoring:

Distribution, Lead Minus Base Runners

Lead-Men On	Current	New
-2	11	45
-1	116	260
0	313	615
1	1979	2248
2	2135	399
3	1479	52
4+	75	4

This distribution makes it clear that saves under the new proposal are shifted to much more perilous situations. Under the current rule, the distribution peaks at +2, but with the new proposal, 87 percent of the saves come in situations where this figure using leads minus baserunners is one or less, or the games that are really in jeopardy. It is also clear on examination of the average win expectancy when a pitcher enters a game for a save–the average win expectancy for a pitcher earning a save under the current rule is 0.89, but under the new proposal, 0.77.

The new proposal clearly makes earning a save more difficult. Just how much is evident in the leader board for the new proposal over the five years from 2002-2006:

Most New Proposal Saves, 2002-2006

Pitcher	Saves, New Proposal	Saves, Current Rule	Difference	Pct. Change
Billy Wagner	69	178	-109	-61.24
Mariano Rivera	68	198	-130	-65.66
Eric Gagne	66	161	-95	-59.01
Trevor Hoffman	61	168	-107	-63.69
Jason Isringhausen	54	173	-119	-68.79
Eddie Guardado	51	153	-102	-66.67
Francisco Cordero	51	133	-82	-61.65
Brad Lidge	51	104	-53	-50.96
John Smoltz	48	144	-96	-66.67
Danys Baez	47	111	-64	-57.66
Francisco Rodriguez	47	106	-59	-55.66
Armando Benitez	45	137	-92	-67.15
Joe Nathan	44	123	-79	-64.23
Todd Jones	42	80	-38	-47.50
Braden Looper	40	98	-58	-59.18
Bob Wickman	39	111	-72	-64.86
Robert Ryan	39	78	-39	-50.00
Jose Mesa	37	140	-103	-73.57
Jorge Julio	35	99	-64	-64.65
Troy Percival	35	114	-79	-69.30
Chad Cordero	35	91	-56	-61.54
Flash Gordon	35	52	-17	-32.69
David Weathers	34	34	0	0.00
Keith Foulke	33	101	-68	-67.33
Joe Borowski	32	80	-48	-60.00
Brian Fuentes	32	65	-33	-50.77
Ugueth Urbina	31	103	-72	-69.90
Tim Worrell	31	64	-33	-51.56
Mike Timlin	29	25	4	16.00
Damaso Marte	29	31	-2	-6.45

It seems the quality closers also do very well under the new proposal. What interests me is if this rule change was ever adopted, would managers start using these relievers differently? Would we see them more often in the eighth inning, or even the seventh? Some closers complain if they come in before the start of the ninth. If coming into a key situation in the seventh will earn them a save, are they more likely to want to pitch early? Would they eschew pitching with a two-run lead in the ninth? I’d suggest that closers pitch longer in games less often; better to save them for that tough situation in the eighth, and give someone else the three-run lead in the ninth.

There are, of course, pitchers who do face those early, tough situations. Here are the pitchers who increase their saves by at least ten under the new proposal; you’ll recognize a few quality set-up men in that group.

Most New Proposal Saves Gained (10+), 2002-2006

Pitcher	Saves, New Proposal	Saves, Current Rule	Difference	Pct. Change
Juan Romero	20	2	18	900.00
Guillermo Mota	24	7	17	242.86
Rheal Cormier	17	1	16	1600.00
Scott Eyre	15	2	13	650.00
Ricardo Rincon	14	1	13	1300.00
Steve Reed	13	1	12	1200.00
Brendan Donnelly	16	4	12	300.00
Alan Embree	16	4	12	300.00
Chris Hammond	14	2	12	600.00
Chad Qualls	12	1	11	1100.00
Travis Harper	13	2	11	550.00
Scot Shields	25	14	11	78.57
Jason Grimsley	12	1	11	1100.00
Joel Zumaya	11	1	10	1000.00
Ryan Wagner	10	0	10	Inf.
Scott Linebrink	13	3	10	333.33
Paul Quantrill	13	3	10	333.33

Among the many excellent comments on the original piece, one person suggested that since the chances of a change to the save rule were thin, perhaps this new proposal should stand by itself. He suggested, “The Rescue” be introduced as a new statistic. I like that idea very much; the difference between a save and a rescue is the difference between catching the person who tripped and rushing into a burning building to pull out the children. If a middle reliever ends up with more rescues than a closer, maybe fans and management will start to question that role more closely.