Automate the Strike Zone, Unleash the Offense

Hello World! As a software developer, automation is my way of life. It kills me to see the tedious yet important job of calling balls and strikes performed at less than 90% accuracy. Worse, catcher framing is now a thing, which is essentially baseball’s equivalent of selling the flop.

Today, I want to talk about how automating the strike zone would affect the MLB run-scoring environment. Don’t we all want to save the environment?

Let’s pretend that before the 2014 season, home plate umpires were fitted with earpieces giving them a simplified Pitch f(x) feed of balls and strikes. They heard a high beep for a strike, a low beep for a ball. They then called balls/strikes exactly as they were told, resulting in a perfect zone.

Experiment 1: Walks/Strikeouts overturned

The most damaging ball/strike errors happen when ball 4 or strike 3 was thrown but not called. Sometimes the umpire is redeemed by luck, and a walk/strikeout happens eventually anyway, but not nearly every time. Think of how many times you’ve seen a 3–0 count where a ball was called a strike, only to have the hitter swing and ground out harmlessly on the 3–1 pitch.

For these experiments, let’s look at short description of the situation, the number of instances of that situation in 2014, and net runs that would have been added if a perfect zone had been called.

Data courtesy of Baseball Savant; click on a situation to see the query I used.

Situation Instances Net Runs (Rough)
Strike 3 thrown, batter safe 146 -88
Ball 4 thrown, eventual out 691 415
Difference 545 327 (.07 team runs per game)

Are you surprised? The umpires made 545 more extra outs than extra ‘safes’. Using a rough walk minus out run differential of 0.6 runs, we see that a perfect zone would have added 0.07 runs per game. Interesting, but not huge.

But think again—this effect isn’t limited to plate appearances that should have ended with a bad call. We all know that the count affects the expected run value all on its own. So let’s expand this to all ‘bad calls’ in 2014.

Experiment 2: All balls/strikes called correctly

Balls and strikes don’t obviously translate to runs. So I’ll use someone else’s much more careful research and use a ball minus strike run value of approximately 0.14 runs. Here’s what happens when we apply a perfect zone to all balls and strikes. Brace yourself!

Situation Instances Net Runs (Rough)
Strike thrown, ball called 8724 -1212
Ball thrown, strike called 40557 5633
Difference 31833 4422 (.91 runs per game per team)

Whoa. Are you kidding me? If we’d run last season with a perfect strike zone, the run environment would go from 4.07 runs/game to nearly 5! That’s the highest level since 2000. I know what you’re thinking: this is crazy, and probably wrong.

Sanity checking

I also found this result to be larger than expected, to say the least. So let’s back up, check the mirrors, and look at the frequency of called strikes vs. balls.

Called Ball 233421
Called Strike 123922
Difference 109499

There are a ton more called balls than called strikes. This makes sense because batters are more likely to swing at strikes. But the ratio of balls to strikes is only about 2:1, that doesn’t account for the 5:1 ratio among ‘mistaken’ balls/strikes! How do we account for this?

A possible explanation

Here we dive into speculation, but stay with me for a minute. Maybe there’s a logical explanation.

What sequence of events must occur in order for a Pitch f(x) strike to become a ball?

  1. Pitcher throws in strike zone: ~45% (Zone %)
  2. Hitter takes said pitch in the strike zone: ~35% (100% – Z-Swing %)
  3. Umpire makes bad ‘ball’ call: ~10%

By this ridiculously rough method, we would expect bad ‘ball’ calls about 1.5% of the time (0.10 * 0.35 * 0.45). Compare that with the observed value of 1.2%

Conversely, the sequence for a Pitch f(x) ball becoming a called strike is as follows:

  1. Pitcher throws out of zone: ~55% (100% – Zone %)
  2. Hitter takes said pitch outside the strike zone: 70% (100% – O-Swing %)
  3. Umpire makes bad ‘strike’ call ~15%

We therefore expect bad ‘strike’ calls about 5.7% of the time (0.15 * 0.7 * 0.55). Again, compare that to the observed value of, wait for it, 5.7%. Boom!

More reasons to automate

  1. Automatic things happen faster. As a professional automator, I guarantee this will speed up play, by more than you think. I bet the umpire thinks for about 1 second on every pitch. That’s just the obvious part.
  2. Set the umpires free. Focusing on something as difficult as calling balls/strikes squeezes out the umpire’s attention on other important matters, such as enforcing pace of play.
  3. Crazy cool things will happen. For example, we will finally see what happens to an insane control pitcher’s K-BB%. V-Mart might never strike out!

I welcome your comments, criticisms, or even praise 🙂

We hoped you liked reading Automate the Strike Zone, Unleash the Offense by foxinsox!

Please support FanGraphs by becoming a member. We publish thousands of articles a year, host multiple podcasts, and have an ever growing database of baseball stats.

FanGraphs does not have a paywall. With your membership, we can continue to offer the content you've come to rely on and add to our unique baseball coverage.

Support FanGraphs




Steve Lind software developer and baseball lover living in San Francisco

newest oldest most voted
Rawson
Guest
Rawson

If your analysis and assumptions are correct, this is incredible information. Thanks for doing it. However, it also starts the conspiracy-theory gears to grinding. Why on earth would MLB tolerate such incompetence? There is no way the league office could be unaware that the umpires are missing so many calls in ways that impact the same so markedly. Why would the league tolerate performance that artificially elevated the value of pitching at the expense of hitting? I’m baffled, especially seeing it laid out this clearly.

Green Mountain Boy
Guest
Green Mountain Boy

I was never a fan of potentially automating ball-strike calls, but this piece is eye-opening and is starting to get me to reconsider the benefits. What would be the basic parameters of your system? And specifically, how do you propose to deal accurately with different stances (upright vs crouched) and different heights of batters (5’6″ to 6’7″)? How would you deal with pitches that nip the front corner of the plate in the zone but end up 6″ outside or in the dirt when caught?

AdamD
Guest
AdamD

The Pitch f(x) system is able to handle this with the help of its operators. The operators set the height of the strike zone for each individual according to their stance, so they can stretch it for tall guys or guys that standupright, and shrink it for short guys or crouching Tigers. As for the second part of your question, the Pitch f(x) system doesn’t consider the strikezone a flat plane (e.g., where the ball touches the front of plate). The strike zone is a polyhedra (a pentagonal prism?), so if the ball touches that polyhedron at any point, it… Read more »

Peter Jensen
Guest
Peter Jensen

AdamD – Wrong on both your statements. Pitch Fx setting of lower and upper bounds of the strike zone has been shown to be an extremely inconsistent process that results in as much as a 6 inch difference from game to game. Because of this many of the studies on the Pitch Fx strike one have substituted upper and lower bounds calculated from the batter’s height and handedness as suggested by Mike Fast in a BP article several years ago. Pitch Fx doesn’t physically measure the ball’s location anywhere near the strike zone. Its measurements begin 35 to 45 feet… Read more »

Ok
Guest
Ok

let us also build cyborgs that don’t break and watch them play instead of humans all together

Steven
Guest
Steven

Hitters would ostensibly be able to tighten up the location of pitches they’re looking for and might see an increase in quality of contact or power.

rory
Guest
rory

I wonder if there’d be a way to cross-check the errors of the calls against the salaries of the players. My guess is that you’d also find that bad calls disproportionately favor stars over rookies.

Trolly McTrollerson
Guest
Trolly McTrollerson

Micro chips embedded in the lettering and knees of the uniforms broadcast signals that serve as the top and bottom of the zone while lasers project upward from home plate to create a real time grid that senses the microchipped ball as it Passes thru the zone.

Also, magnets or something…

Captain Tenneal
Member

I used Markov chains to predict league strikeout and walk rates if every call were made perfectly. I then ran the predicted rates through the FIP formula and came to a similar conclusion, that offense would be expected to rise nearly a run per game. So it’s not crazy to think that perhaps MLB should get on this if they want people to keep watching baseball.