Humans are Good Pattern Recognizers

Sam Miller notes that drafting unusual body types is one thing that did not stick from Moneyball.

So, with all those caveats in mind, we looked at the 2002 draft, the one where the A’s were all alone on the island of misfit toys, the season that preceded the wave of analytics that would eventually convert more or less every front office into a stathead-friendly one. We compared the listed heights and weights of players picked in the first five rounds of that draft with those picked in the first five rounds of last year’s draft.

The average player picked in the first five rounds then: 74.2 inches tall, 198.3 pounds. The same five rounds last year: 74.2 inches tall, 198.2 pounds. If we break it down by hitters and pitchers, by college kids and high schoolers, there remains very little daylight. The 2016 pitchers were slightly taller — about 0.2 inches — and the high schoolers were leaner, both of which we might describe as positive physical attributes.

There were five pitchers listed under 6-foot who were drafted in the first five rounds of the 2002 draft. There were five in 2016. We count five hitters listed over 220 pounds in 2002; there were four in 2016.

This reminds me of reading the about stolen bases in The Hidden Game of Baseball. At the time, in the mid-80s, Pete Palmer used linear weights to show that the break even point in terms of runs for stolen bases was a 67% success rate. If a team stole at a rate less than that, they were costing themselves runs. What I later learned is that managers implicitly understood this. In season when the break even point was lower, they attempted more steals. When the break even point was higher, they attempted fewer steals. Baseball managers were not doing calculations, but positive and negative feedback on winning and losing when attempting a steal helped drive the decision. The extra-base power behind the runner helped drive the decision, and managers and linear weights came to the same conclusion.

Now we see that scouts knew something about players based on looking at their bodies, and it was a good indicator. Scouts watch 1000s of young men play ball. They make recommendations that work out, and recommendations that don’t. The scout learns from each of those, even if it is not quantified in a spread sheet. The thing to remember with these pattern recognition models, be they natural or artificial intelligence, is they simply give you a probability of being correct. The trick is knowing when a team should ignore the body model and take Dustin Pedroia or Jose Altuve. Or even Aaron Judge.

Hat tip, BBTF.

from baseballmusings.com http://ift.tt/2sL0ehV