The recent resurgence of talk about WAR due to Sam Miller's article has got me thinking about the meaning of WAR itself: how many wins did a player contribute to the team over a replacement player. A replacement player is basically who the team would add to the roster (at the same position(s)) if the player under consideration was removed from the team. But this got me thinking... what if we removed all of the players? What if an entire team was filled with only replacement players? If WAR is doing a good job, shouldn't the number of wins from an all-replacement team (RT) be fairly fixed?
So I decided to run some numbers (for the 2012 season), to see how a team of only replacement players would do. For each team, I added the offensive WAR, i.e. WAR from the hitters, and added it to the pitching WAR, WAR from the pitchers. Then I took this total number of Wins Above Replacement and subtracted it from the actual wins each team had, calling this result Replacement Team Wins (RT Wins). For this exercise, I used Fangraphs WAR, since they had listings of team WAR more available.
|Team||W||L||Off WAR||Pit WAR||Total WAR||RT Wins|
The results are a bit surprising. Now, the average RT Wins came out to 43.32, which is right on Fangraphs definition of a replacement level team. However, there are some broad deviations from this. Four teams are below 35 wins, with the Brewers the lowest at 32.7. Three teams are above 50, with the Orioles the highest at 61.1. The distribution seems pretty random with respect to actual wins, offensive WAR, pitching WAR, or total WAR. That is, I had trouble finding correlations.
The interesting thing is, some teams appear to be getting more wins from their WAR than others. For example, the Orioles managed 93 wins from 31.9 total WAR, yet the Brewers only got 83 wins from a whopping 50.3 WAR.
This makes me think that WAR doesn't represent Wins accurately. Perhaps runs created/saved isn't being calculated well, or the conversion from runs to wins isn't valid. I did some more calculations to look at how many wins a RT would have based on the number of wins determined from the PythagenPat algorithm, and there is the same, if not more, variation. I won't post the data, but have it available, if desired.
So why the variation? To be honest, I'm not really sure, and it embarrasses me to make a FanPost without any clear conclusions, so I'll at least throw out some thoughts.
- The definition of a replacement player still needs some work.
- WAR is merely a tool to compare players with each other. Cumulative WAR should not be used to compare teams.
- WAR actually converts a player's value in Runs to Wins using an approximation/variation of the Pythagorean Win formula, which itself often does not match the amount of games a team actually wins.
- There is some data I'm missing or not considering that influences the number of games a team wins other than offensive and pitching WAR (and luck).
- The variation is due to one or more of the issues described in Fangraph's explanation of WAR.
I'm curious to hear what you make of this. Does this discredit WAR in any way, or am I trying to read too much into it?
One more fun fact: The 2003 Detroit Tigers was pretty darn close to an all-replacement player team. Posting 43 wins, they had an offensive WAR of -0.7 and a pitching WAR of 3.8 for a Total WAR of 3.1. Nate Cornejo (1.9 WAR), Jeremy Bonderman (1.4), and Mike Maroth (0.6) had a combined 57 losses and yet had the top three WAR among pitchers on the team. Dmitri Young was the only other player on the team with a WAR over 1.0 at 2.1.