Tim Sheehan
github | Google Scholar | Baseball Blog | Science Blog | Resume
Photo by Evan MaynardThe focus of my PhD was uncovering how observers (humans) make sense of a constantly changing world given limited & unreliable information. In short, we use prior expectations to shape our perception in a Bayesian manner.
In particular, I explored how humans are predisposed to see the world as more stable than it actually is. On this website, I visualize a few different forms of biases exhibited by umpires. The most obvious, and possibly most consequential, is the bias in the called strike zones between left and right handed batters. Essentially, umpires are much less likely to call an inside pitch a strike resulting in a shifted zone (relative to home plate). Critically the shape and size of strike zones seems to be stable for each umpire across games and seasons, suggesting idiosyncrasies in what each umpire calls could be strategized around in determining batting orders or what side a switch hitter will bat from.
I have also explored how the strike zone changes with the count. Umpires are more likely to call a strike on a 3-0 count than a 0-2 count. Pitchers are also more likely to throw a true strike on a 3-0 count than a 0-2 count. This suggests that umpires are biased in a manner that will overall improve their accuracy. Lastly we observe a "switch" strategy where umpires are less likely to call a strike if they recently called a strike (and vice-versa). This is further evidence of a Bayes optimal strategy. Note that history models are fit while accounting for the strike zone itself and are not simply a reflection of where the pitch was thrown.
To understand how models are fit and how they can be interpreted, see related blog post.