How Accurate Is WAR At Predicting Team Wins?
Now, I won’t attempt to explain the ins and outs of the calculation of the WAR statistic, as I’ve attempted to study the calculation and, like many others, have come away with a general understanding of the statistic rather than a technical one.
Every year, 30 MLB teams play 162 games, resulting in a total of 2,430 games played (81 games x 30 MLB teams), and therefore 2,430 total wins in a given year.
FanGraphs’ WAR formula resulted in a total of 999.6 cumulative WAR for all players in MLB in 2012.
To find the number of wins for a “replacement level” team, subtract 999.6 from 2,430 to come to 1,430.4 wins, then divide by 30 MLB teams, to find that the replacement level team would win an estimated 47.68 games in 2012.
Therefore, if you start every MLB team with 47.68 wins, then add the cumulative WAR of their team’s players for the year, the result should be close to the number of wins accumulated by that team for that given year (in this case, 2012).
Basically, what I’m trying to find here is as much a “prediction” as it is a measure of accuracy. If WAR truly values each player’s contributions somewhat accurately, then the teams with the most accumulated WAR for that year should also have the most wins, right?
Here are the results:
From these results, we can find that the standard deviation of the difference between expected wins and wins accumulated to be 6.05 wins, and the correlation between cumulative WAR and team wins is .86.
For non-nerds, that means it’s pretty darn accurate.
This means that in 2012, 21 of the 30 teams fell within the standard deviation with their win totals, and only one team (Baltimore) fell outside two full standard deviations.
How much different would the standings have been? Let’s take a look:
So according to the WAR numbers generated by teams in 2012, the WAR metric would’ve expected the Yankees, Tigers. Rangers, and Angels to represent the AL in the playoffs and the Nationals, Braves, Cardinals, Brewers and Giants to represent the NL. In short, WAR overvalued the Angels and Brewers, and undervalued the Athletics and Reds, but correctly predicted 8 of 10 playoff teams. Not bad.
So how is WAR doing so far this year? Well, here are the numbers as of 9/6/13:
This year the WAR metric has been even more accurate thus far. In 2013, the standard deviation between expected wins derived from each team’s cumulative WAR and their actual wins is just 4.81, and the correlation is .89.
Here are the comparisons between the expected and actual standings thus far: