Printer-friendly

June 19, 2008

Lies, Damned Lies

Trying to Get LOESS-t

by Nate Silver

In playing around with the McCain-Obama polling data over at my political blogging project, I came across a procedure known as a LOESS regression. LOESS is a way to create smooth-looking curves out of sequential data. Some of LOESS’s virtues are aesthetic; the curves it produces seem to look and feel just right. But it can also be considerably more robust than something like a moving average in handling time series data, performing more strongly in the presence of outliers. Since we look at all sorts of time-series data in baseball, LOESS ought potentially to have all kinds applications to the game.

Today I am going to start with a simple application: how the eight primary fielding positions have progressed since the 1950s. The metric I am using is the OPS accumulated by all players at a given position each season relative to that of the league as a whole. I don’t include pitchers or designated hitters in the league-wide averages, as offense at the former position is mainly a fun little curiosity, whereas the latter would skew the results, as the DH position was created solely for offense. (Ironically, designated hitters as a group have quite frequently been outhit by both corner outfielders and first basemen.)

Let’s begin behind the plate:

Rest assured, the other graphs are a little bit more interesting than this one, but it says something that the catcher position has had roughly the same strength for the past 50 years. At every other position, teams can consider shifting their personnel around to strike the offense-defense balance that is fashionable in their day. But a catcher is just a catcher (unless he’s Craig Biggio), and the sort of John Russell / Josh Willingham / Robert Fick experiments have never really caught on.