clock menu more-arrow no yes mobile

Filed under:

Occam's Razor, Predicting ERA, and the Astros

Explaining Glenn DuPaul's pFIP and how it relates to the 2013 Houston Astros in the simplest terms available.


I love Occam's Razor, especially when applied to Baseball statistics, which can get overbearing in their mathiness. I tend to get absorbed in the why's of baseball, and sabermetrics contain some great concepts to answer those questions. The problem is, it's easy to get lost in the numbers, which is a great way to make readers puke on their keyboards. The digger I deep into a complex question, the more complex the numbers become and the harder it becomes to make my point to an audience.

And so, my approach to this question of complexity versus explainability is best summed up by Thoreau:

"Our life is frittered away by detail. Simplify, simplify, simplify! I say, let your affairs be as two or three, and not a hundred or a thousand; instead of a million count half a dozen, and keep your accounts on your thumb-nail."

Thoreau's famous quote is the epitome of the theory behind Occam's Razor, which is essentially that if one is presented with several different explanations to a problem, the simplest answer is usually the best answer. My first exposure to the razor was in middle school, as I delightfully read through the Ender's Game trilogy. Orson Scott Card has a tendency to divert into philosophy amidst his novels and to apply those concepts to his plot lines.

"Occam was a medieval old fart. The simplest explanation that fits the facts is always, God did it. Or maybe - that old woman down the road is a witch. She did it."

--Orson Scott Card, Xenocide

Obviously, Card is being a bit facetious with this statement, as we know there is a wide gulf between the simplest implausible explanation and the simplest reasonable explanation. But the point remains the same - Keep It Simple, Stupid.

The other week, I read an interesting article by Glenn DuPaul on The Hardball Times about predictive Fielding Independent Pitching, or pFIP. Tracing the link in the first paragraph back to the origin of DuPaul's thought process that led to the development of pFIP, I discovered happily that his motive rested in Occam's Razor, Thoreau, and the KISS principle. He was attempting to find the simplest method to predict ERA, without resorting to head-splintering modifiers, ball-in-play rates, multidimensional calculus, quantum mechanics, or scatomancy. This was a subject right up my alley because, while fascinated by the why's, I have not the energy nor the time to devote myself to understanding the statistical modeling theory behind the how's that explain those why's.

The thing that most commoners like myself forget about FIP is that it is not really a predictive stat at all. One cannot simply say, "Bud Norris had a 4.23 FIP in 2012, therefore his ERA will be lower in 2013 than the 4.65 that he posted in 2013." Or rather, you could say that, but FIP and next year's ERA have been found to have varying levels of correlation, depending on circumstance. In short, FIP tells you what the pitcher would have done in the past, if only those dratted things that were not under his control were taken away - things like team defense, park effects, bad luck, and solar wind. This is interesting and useful because it allows us to valuate a player's skill in a vacuum, rather than blaming him for what is out of his control.

Variations of FIP or ERA predictors have been released throughout the years, such as SIERA, xFIP, and kwERA, but DuPaul wished to create a Runs Allowed predictor that would be accessible by the masses and easily explainable. A man after my own heart, and I believe he has succeeded.

The idea behind pFIP is that it is based in the Three True Outcomes for pitchers. These are the three things that a pitcher can absolutely be responsible for: Home Runs allowed, Base on Balls allowed, and Strikeouts. The equation DuPaul comes up with after a lot of math that he explains far better than I ever could is:

pFIP = (20*HR + 10*BB - 10*K)/TBF + 4.60

Where TBF is the Total number of Batters Faced, and 4.60 is a constant that just moves pFIP to a scale roughly equivalent to ERA, which we're already familiar with. Right?

See? That's not so hard. Home Runs. Walks. Strikeouts. And nice whole numbers. DuPaul did a lot of work comparing pFIP to other predictors and projection systems by plugging historical data into it and comparing the results against actual ERA's. Without delving into it too far, pFIP proved to be one of the best ERA predictors.

DuPaul was nice enough to include a spreadsheet containing 2013 pFIP's for 139 starting pitchers. Here are the Astros, along with comparisons to projection systems available through Fangraphs:

Pitcher pFIP Marcel Steamer Oliver ZiPS
Bud Norris 3.96 4.25 4.29 4.23 4.54
Erik Bedard 4.01 4.37 4.36 4.06 5.13
Lucas Harrell 4.12 3.87 4.65 4.16 4.75
Jordan Lyles 4.35 4.66 4.49 4.38 4.95
Philip Humber 4.77 4.76 4.55 4.45 5.27

For the sake of the Astros, let's hope that pFIP is indeed the most predictive stat. It's worth pointing out that many of the Astros' starting pitchers are successful in limiting walks and home runs while posting strikeout rates that are above league-average. The big question is, are those skills truly predictive of future success, or is there more to the equation? Is Occam's Razor a lie?

I choose to think not. It makes sense that the skills that are most under direct control of the pitcher should have the largest impact on his success in the future. In that sense, we can expect marked improvement from all of the Astros' projected starters listed above (health permitting, luck permitting, and defense permitting).

As with every new statistic, pFIP has its share of areas of possible improvement, such as accounting for more than the most recent season and testing against minor league statistics (which would need to be converted to major league equivalents first). But I am a fierce advocate of the 80-20 rule, and if pFIP in its simplicity can provide a reasonable accuracy and retain it's relative ease of explainability, then I applaud DuPaul for what he's contributed to the sabermetric crowd and urge him to leave well enough alone and move on to simplifying other concepts.