Feng: The curse of small sample size in bullpen stats

Ed Feng
Special to The Detroit News

Bruce Rondon has shown us extremes in his pitching this season.

On July 10 against Minnesota, the Tigers reliever gave up three hits in the ninth inning in part of an epic collapse from a 6-1 ninth-inning lead to a defeat. Rondon threw hard that game, but he left pitches out over the plate. The Twins hitters made him pay.

One of my friends called Rondon a fireballer, meaning a pitcher who throws hard but doesn’t have the ability to locate pitches. His early struggles made Tigers fans doubt his ability to recover from Tommy John surgery.

In contrast, Rondon entered Saturday night’s game against Boston with a one-run lead in the ninth. His pitches painted the corners as he struck out two. Rondon preserved the lead for his first save of the season.

After the game, manager Brad Ausmus said the following about Rondon:

“Rondon’s just got a different look, a different look in his eye. He looks confident. He’s attacking the hitters with his fastball. He’s not trying to trick them. He’s going right at them. There was an intelligence in his approach, too.”

It’s hard to think Rondon has changed so much in a short period of time. Perhaps he has been a solid major league relief pitcher the entire season. He left some pitches out over the plate in his early appearances, which allowed major league hitters to get some hits.

But even major league pitchers can’t pinpoint the location of their pitches down to the inch. There’s variability in where a pitch crosses the plate, and this randomness has shifted in Rondon’s favor recently.

Small sample size

For relievers like Rondon, the small sample size of innings pitched makes it impossible to get an accurate assessment from either numbers or watching his performance.

On the numbers side, I’ve written about the randomness on batting average on balls in play (BABIP). Pitchers tend to allow a BABIP of .300, and deviations result from factors out of the pitcher’s control, such as defense and randomness.

This season, Rondon has allowed BABIP of .400, well over the expected average. He might have a lower average if he threw underhand lobs toward the plate. This quantity already has regressed to the mean, as his BABIP allowed was .455 on July 29.

However, we has humans make the curse of small sample size even worse. Let me explain.

Human obsession with patterns

In his book “The Ravenous Brain,” neuroscientist Daniel Bor discusses how humans are obsessed with finding patterns in information. This tendency runs deep in our culture.

“And we really are a decidedly strange species for actively seeking out games with patterns in them, when such activities seem to serve no biological function whatsoever, at least not in any direct way. It’s as if we were addicted to searching for and spotting structures of information, and if we do not exercise this yearning in our normal daily lives, we then experience a deep pleasure in artificially finding them.”

Sounds like he’s talking about baseball, right?

This search for patterns has resulted in technological marvels such as putting a man on the moon and a supercomputer in your pocket. However, this same search for patterns can cause problems when looking at randomness.

In a previous article, I made some visuals of random sequences to show how these sequences have streaks. But the human obsession with patterns makes us tell stories about these streaks.

When Rondon leaves pitches out over the plate, he’s the fireballer with no control over his pitches. When he found the corners against Boston on Saturday night, he looks confident.

As humans, we want to tell stories about what we see in life. However, it’s important to check these stories with the underlying numbers, a difficult task for relievers in baseball.

Other Tigers relievers

Let’s look at two other examples of how sample size statistics might mislead you into telling stories about randomness.

Before closer Joakim Soria was traded away, he had allowed home runs on 19 percent of fly balls, much higher than his career average of 8.4 percent. This statistic depends on which batters a pitcher faces, but randomness plays a big role. On average, 10 percent of fly balls leave the park in the majors.

These small sample sizes can also work in a reliever’s favor. Blaine Hardy has become one of the better relievers for the Tigers this season, as he has reduced his walk rate from last season. However, he has not allowed a single home run in almost 50 innings pitched this season, an unsustainable pace.

Ed Feng has a Ph.D. in chemical engineering from Stanford and runs the sports analytics site The Power Rank. Have a question about the Tigers you want addressed in this column? Email Ed Feng here.