2009 Projections with Hit Tracker
Oh, no, not another projection system! Why would someone want to join the logjam of current systems? In no particular order, we have ZiPS, CHONE, Oliver, Marcel, Bill James, PECOTA and no doubt some others I haven’t stumbled across (sorry). All of these systems are designed to tell us how MLB players will perform next season, but none of them can convincingly claim to be more accurate than all the rest. When I look at any particular player’s projections in the various systems, I see a lot of similarity, which makes me suspect there must be some degree of groupthink going on. I believe there is some potential to improve performance forecasting by doing something different.
In the following paragraphs, I will outline a system for forecasting using Hit Tracker, an aerodynamic model for flying baseballs that is well-known for providing accurate home run measurements. I can guarantee that the Hit Tracker system will be different. Better? I won’t be able to say for sure until the 2009 season is over.
Background: How We Forecast Now
Why is it so difficult to forecast a player’s performance accurately? One huge reason is that every one of the current systems for performance projection starts from a set of data — the player’s prior year’s “box score stats” — that is positively riddled with statistical noise (chief among these uncontrolled noise factors are the dramatic differences in ballpark configurations and playing conditions across the 2,430 games played in 30 different parks over the course of six months).
Let’s consider another familiar form of forecasting: weather. In the 19th century, after the invention of the telegraph, weathermen began to form their predictions by first learning the weather “upwind,” and then adjusting those measurements to come up with a forecast. “How hot will it be tomorrow? Well, it was 85 degrees today in the state where our weather seems to be coming from, so we’ll start with 85 and then adjust it up or down according to our experience. It’s usually a little hotter there than it gets here, so let’s say 82 degrees?” They didn’t call them “city factors” back then, but they could have.
After computers became available in the mid-20th century, weathermen became meteorologists, and the process of forecasting weather has continued to become more involved and mathematical as the years have gone by. Contemporary meteorologists now monitor a much larger array of parameters, and they feed these lower-order parameters into elaborate computer-based models to arrive at predictions for the higher-order outcomes like temperature, or winds, or precipitation. Thanks to more accurate measurements, and more detailed models, weather forecasts are dramatically more accurate today than those of even only 10 years ago.
In my opinion, baseball forecasting systems resemble the “19th century weatherman” system described above: to forecast something, measure something (well, in baseball we should say “count” something) that has happened already, then adjust this number to predict what hasn’t happened yet. So, to predict a player’s home runs, for example, the starting point is always his prior year’s total for home runs (or perhaps a weighted total from several seasons). From this starting point, various adjustments are applied to arrive at a final projection. Never mind where those home runs were hit, or how far they flew, or how much help or hindrance the weather may have provided them. Just count and adjust.
Starting from last year’s total assigns an equal value to what may in reality be very different events. For example, Jeremy Hermida hit two radically dissimilar fly balls last year, each of which cleared the home run fence: first, a windblown 321 foot homer in San Francisco on Aug. 20th, and second, a 443 foot rocket in Miami on July 19th. In a game context, they count the same, but when we are trying to measure the likelihood of future home runs, we should acknowledge that the outcome of one of those fly balls (the short one) was entirely dependent on its ballpark and weather context, while for the other fly ball, the ballpark and weather were irrelevant to the outcome. The short fly ball could only have become a home run in a park with a very shallow RF fence like AT&T Park, and only with the help of a tail wind. The long one would have been a homer in every park major league baseball has ever been played in, in any wind short of a hurricane blowing towards home plate.
38 kkk breast
Any system that cannot recognize the difference between two events such as these Hermida home runs cannot hope to consistently generate highly accurate predictions. I don’t mean this as a criticism of anyone who has created a projection system, don’t get me wrong. But I do believe that those systems have reached the limit of their capabilities, with average errors of around 60 …
Related posts: Key Democrats blast Obama stimulus plan, In good company, Michael jordan, Wipeout tv show, Chinese zodiac
