Alan Schwarz of the New York Times wrote a great article about the use of simulations in baseball. in doing so, Schwarz looked at the impact of specific aspects of the game. One of those was the stolen base. Specifically, Schwarz showed that if a team such as the Tampa Bay Rays were to steal fewer bases, they would actually score more runs.
The stolen base. Advancing from first to second puts the runner in scoring position, but he — and the rest of your hitters — will have a hard time scoring if he gets thrown out. Mr. Kraemer looked at a recent team that ran wild (the 2008 Tampa Bay Rays) and one that barely stole at all (the 2005 Oakland A’s) and switched their mind-sets to see what happened. The A’s scored 20 runs fewer, which probably says more about their players’ inability to run in the first place. But when the speedy Rays stole sparingly, they increased their scoring by 47 runs per season — suggesting that perhaps the Rays were running too often in real life.
This caught the attention of Rob Neyer at ESPN.
On the other hand, the note about the Rays’ steals is truly surprising. The generally accepted break-even point for steals is something between 70 and 75 percent (depending on the scoring environment). Well, last season the Rays stole 142 bases and were caught 50 times for a 74 percent success rate, comfortably within that break-even range. I don’t know how to square 74 percent with those theoretical 47 runs … but if I were running the Rays, I sure would want to know.
First of all, we love baseball simulations. But we are the first to tell you that there are inherent limitations. Simulations, such as the highly respected “Diamond Mind” look at trends and averages and uses those to project what will happen over the course of a season. What they cannot do is project what will happen in very specific situations.
In the case of stolen bases, simulations look at a team like the Rays and see how often they steal and how often they are successful. What the simulations don’t see is when the Rays steal and what impact those steals have on other aspects of the game.
More specifically, the simulation will see that the Rays attempt 192 steals in a season or about once every 30 plate appearances. So when the simulator does its thing, it will have the Rays attempt a steal about once every 30 plate appearances without any further rhyme or reason.But even for a team like the Rays, there are times in the game when they are more likely to steal and times when they won’t because stealing a base would hurt the ballclub.
Furthermore, Diamond Mind only considers a stolen base as moving forward a base or being called out. But stealing a base is so much more than this simplistic view. A simulator does not consider the psychological impact on the pitcher and how he can be distracted. It does not consider whether the pitcher is throwing from the stretch or going to the plate with a quick pitch because somebody like Carl Crawford is on first base. It does not consider that a pitcher may throw more fastballs with BJ Upton on first base, and how that benefits the batter.
This is fine over the course of a full season because these type of idiosyncrasies will tend to even out. In other words, the negative impact of facing a fast team like the Rays is balanced out by all the positive impacts of facing a team that does not steal bases at all, like the A’s.
So what does this tell us? It suggests to us that simulations severely underestimate the impact of base stealers on a baseball game and on the success of a team over the course of 162 games. This could be one reason that the Rays beat even the most optimistic projections last season by 10 games, despite not having any players play above their expectations.
We love projections and we love statistics (we used to teach a statistics course) but with great power comes great responsibility. One has to be careful how they are used, and one has to understand the limitations.
Answering Baseball’s What-Ifs [New York Times]
Computer simulations sometimes offer surprises [ESPN]
