On Cantu and the “little things”, Pt. 1

I’ve been harping on the whole Jorge Cantu versus Dan Uggla debate on the side of Uggla for the last few weeks, even though the point is moot because both players will be on the Marlins for 2010 and both will be getting significant playing time. Reader jrhana mentioned something of interest to me though, with regards to Cantu’s 2009 WPA/LI. The figure had Cantu with a WPA/LI of 2.2 wins above average, while traditional context-neutral linear weights from wOBA had Cantu at around eight runs, or 0.8 wins above average. That difference of around 1.4 wins put him second in 2009’s King of Little Things award, as handed out by friend of the Maniac Matt Klaassen.

What gives? Usually WPA/LI and linear weights are fairly close because the distribution of all events is random enough that the difference would not be so significant. Still, how could Cantu be earning so much credit when looking into one system as opposed to the other? We’ll define and do some light investigation after the jump.

Definitions

First off, we should probably define everything. Linear weights systems, as I have mentioned before, give credit to hitters for events based on the average value of each event. In other words, every event is weighed the same regardless of the situation in which it happens. WPA/LI is, in actuality, game-state linear weights, but without the leverage component.

Now, that may sound a little like WPA, a highly context-sensitive stat that has wild swings and is not worth using for talent analysis. But WPA/LI is actually quite different. For example, it is clear that a double is worth more with a runner on second than it is with no one on, but part of the reason why it is worth more is that the event has a naturally higher leverage index. WPA/LI divides the leverage index away, so that the “importance” of the situation (the leverage) is taken away. What is left, however, is the relationship between the importance of the events.

Imagine, for instance, a classic bases loaded, two out, bottom of the ninth, tied game situation. In such a case, the value of a walk is just as valuable as a home run. While linear weights would still value the homer as approximately four times the value of the walk, WPA/LI in this case would measure it as the same value. This is an extreme example, but it is meant to show that WPA/LI can differentiate the relative value of events to each other.

So what’s the difference again?

As pointed out in the linked article and in various attempts at explanation by Tom Tango, WPA/LI can be really useful in examining a player’s “situational hitting.” In creating a context-neutral stat, we assume players hit their various events randomly enough to use a blanket average value, but in doing so we undoubtedly throw a little bit of talent out with all the random luck. WPA/LI over a long period of time could be useful in determining those differences.

Having said that, this is not about defining WPA/LI, but rather about talking about Cantu’s anomalously large difference in WPA/LI and regular linear weights. How well did Cantu do the so-called “little things” in 2009? Tomorrow, I’ll take a closer look at that answer.