Baseball analysts, gurus, pundits, and fans love to throw around pitching metrics (by metric I mean any statistic related to pitching) and their various importances. By all accounts, the five most popular pitching metrics to date are: Wins, ERA, FIP, WHIP, and WAR. Some prefer the simplicity of Wins, ERA, and WHIP; while others prefer the complexity of FIP (fielding independent pitching) and WAR (wins above replacement).
But how do we know which of these metrics matter, or if all of them do, or if none of them do?
I don’t have any way of determining which is the most important and will lead to the best accuracy in pitching prediction, and if I did I’d probably be hired by the Mets, but I can relate all these metrics to each other and find their correlations.
Before I throw up a chart that makes people cry, a very quick lesson in this one statistic: the coefficient of determination (or R-squared). Each number in the chart represents the R-squared value between two metrics (wins vs ERA for example) I’m comparing. I converted these values into percentages so they’re easier to interpret. Basically, the higher the percentage, the higher the two metrics are correlated, or affect each other. If the R-squared value is 100%, then the two metrics are the same (if I compared ERA to ERA). If it’s 0%, then I’m probably comparing ERA to the number of apples in Wal-Mart. The data I used is from qualifying starters in the NL in 2010 (44 of them). So here’s the chart, and then I’ll do a bit of analysis.
You’ll first notice that the highest correlation is between FIP and WAR, 88%, which makes sense because FIP is used to calculate WAR, so they better be highly related! We’re off to a good start.
The second highest correlation is more telling – ERA and WHIP have a 68% correlation. Neither is used to calculate the other, so we actually have something worth talking about here. What this means is that a pitcher’s WHIP is actually a fairly good determinant of his ERA! A 68% correlation probably isn’t good enough to actually make a pin-point prediction, but it’s enough to see a very solid trend.
Another really interesting thing I see is that Wins have low correlations with…everything! Wins aren’t highly correlated with ERA, FIP, WHIP, or WAR! None of them. What does this mean? It effectively means that the stat of “Wins” is useless. Any Cy-Young voters voting based on a starter hitting some arbitrary win mark, have just been proven to be quite silly. Either Wins are meaningless, or every other stat I’ve listed is. I’ll side with the other stats.
One last thing I’ve noticed is that WAR doesn’t have an incredibly high correlation with anything aside from FIP, again, which have similar calculations. What this means to me is that an analyst should probably either use ERA/WHIP or FIP/WAR to determine the worth a pitcher, but probably not a combination of the pairs, or things may become confusing. Notice how WAR and WHIP’s R-squared value is only 39%. Fangraphs may love WHIP and they may love WAR, but this proves that the two really don’t have much in common!