The True Value of a Save

This post will be a little different than the previous post, in that it is much more difficult to quantify the value of a save than the value of a home run. For home runs, we can fairly easily calculate the differences between each one, as there are only 24 different ways for a home run to occur (the base-out states). There are many, many more ways for a save to unfold. There are one and two inning saves; one, two, or three run leads; and a number of different base-out combinations throughout a save attempt.

To determine the true value of a save, we are going to look at all 1,204 saves in 2010, and determine the WPA of each save. A description on WPA can be found here, but put simply, it is the probability of a team winning after an event subtracted by the probability of a team winning before the event. It will show how much a player contributed to his team winning the game. Every save from last year can be found here, sorted by WPA. The true value of a save will then be the average WPA for all saves.

The most valuable save last year was recorded by Andy Sonnanstine of Tampa Bay, with a WPA of 0.662. The least valuable save was recorded by Matt Harrison of Texas, with a WPA of only 0.001 (he actually pitched 3 innings in a blow out game, which is one of the obscure ways a reliever can get a save). The average of all saves last year was a WPA of 0.114. What this means is that the average reliever recording a save will increase his team's expected win probability by about 11% (from 89.6% to 100%).

Unfortunately, there are many debates going on (such as here) as to whether or not WPA is an accurate measurement of a relief pitcher's value. The probability of a team winning when leading going into the 9th inning has not changed whatsoever from 1952 to 2010 (which is pretty amazing!). Naturally, this calls into question the value of the modern day closer. So instead of using WPA, many sabermetricians use WPA/LI, otherwise known as Context Neutral Wins, which is described here. LI is the leverage index of a certain play, as a tie game in the 9th inning will have much more pressure than a play in the 1st inning of a game. Simply using WPA will not account for the context of the situation, so the value of a reliever could be drastically overvalued merely because they pitch in higher-leverage situations.

WPA/LI takes care of this problem by neutralizing the leverage of the situation. As a result, a player's contribution will almost always be less, especially for relievers. If we look at the WPA/LI for all of the saves from 2010, the average is WPA/LI is now only 0.061 (almost half of the WPA value).

So the problem now becomes, which statistic do we use? WPA or WPA/LI? This really depends on your own beliefs. If you believe that closers are really good pitchers who can do things other relief pitchers cannot do, especially in high-pressure situations, then you would want to use WPA. However, personally I believe that closers are only marginally better pitchers than their bullpen counterparts, and as such, are getting some undue credit. So I believe that using WPA/LI is better, especially considering that many closers are failed starting pitchers. However, I will use both statistics in comparing home runs and saves. My next post will finally answer the question of whether 30 home runs or 30 saves are more valuable.