Syndergaard’s Stolen-Base Problem and the Postseason

Noah Syndergaard has reached a point of excellence this season that finds him capable (to the extent that anyone is capable) of challenging Clayton Kershaw for the title of baseball’s most dominant starter. If compelled to pinpoint the most glaring difference between the two Cy Young candidates, however, it would be this: whereas Kershaw is historically masterful at stopping the running game, Syndergaard is historically poor. The Mets’ ace gave up a whopping 48 steals this year, one of the 10 worst seasons for steals allowed since 1974, when Retrosheet’s full records begin.

The reason for Syndergaard’s struggles is clear: the 6-foot-6 righty is really slow to the plate. This has been a problem all year, making it a popular talking point for the New York media. That included speculation from John Harper a couple of weeks ago, as the Daily News writer made the case that these struggles would make Syndergaard an unwise choice to start the wild card game.

That might even raise the question of who should start the wild-card game. As dominant as Noah Syndergaard can be, his problems in controlling the running game are a consideration in a win-or-go-home scenario, where a couple of stolen bases could prove crucial.

“That would be a factor for me,” an NL scout said Friday. “Everybody says stolen bases aren’t important anymore, but then you get to the playoffs, and they can be the difference in a ballgame.

This argument that the Mets should sit the exceptional Syndergaard is suspect. But the scout’s theory is worth testing. Maybe, given the magnitude of postseason games, runners attempt more steals when it counts, and contribute more towards team wins. Let’s check.

Below are probability density functions (PDFs) — in which the area under each curve sums to one — showing the win probability added on stolen-base attempts. The chart shown below was built with PITCHf/x-era data, spanning from 2008 onward, and compares the regular season against the postseason.

We can see that both PDFs mirror one another. For the scout’s claim to hold true, the playoff curve would need to be shifted farther rightward than the regular season curve. Moreover, both curves peak sharply at ~1%, showing that the impact that stolen-base attempts make in games isn’t much. More than half of attempted steals change win expectancy by an amount under 2.5%, and 82% of attempts result in a WPA change under 5%. Little gains like these can add up in bundles, but this result is still a far cry from the “X factor” that Harper and the scout posited. Big playoff steals may endure in our memories, but they aren’t the norm.

Another way that stolen bases could “be the difference” in playoff games is by occurring more often. With teams’ seasons hanging in the balance, base-runners may try to manufacture more runs. An already vulnerable pitcher like Syndergaard could, hypothetically, become further exploited in the playoffs.

To test this, I turned to a boosted decision-tree model to predict whether or not a steal of second base will be attempted. I plugged seven factors into the model:

• Inning
• Score (lead or deficit, in runs)
• Outs
• Home-run rate of the batter at the plate (min. 75 PA)
• Runner stolen-base-attempt rate (min. 25 opportunities)
• Joint pitcher/catcher stolen-base-attempt rate (min. 25 opportunities)
• Whether opportunity occurred in regular season or postseason

For an “opportunity” to count, second base had to be open in front of the runner at first, who couldn’t have advanced by errant means (like a wild pitch). Only matchups that met the sample-size benchmarks above were included, which still left over 221,000 opportunities in the PITCHf/x era to analyze. That’s enough to handle factors such as whether the runner is reluctant to steal when the batter at the plate has shown power, or whether the particular pitcher/catcher combination is good at deterring steal attempts.

The model did a nice job in its predictions, posting a strong cross-validated AUC of 0.84. Below is its output — the percentage influence of each variable in predicting attempts.

The decision to go or not is ultimately up to the runner, so it’s no surprise that the runner’s aggressiveness is the most vital factor. The vulnerability of the pitcher/catcher tandem matters about half as much, but is still a big piece of the puzzle. Little impact is made by the other variables — the least significant of which is whether the game took place in the regular season or playoffs. Controlling for all other factors, the model shows that when big base-stealers match up against susceptible pitchers and catchers, stolen-base attempts rise by 1% to 3% in the postseason. This year, Syndergaard averaged 5.3 stolen-base opportunities per game, so it would typically take at least six playoff starts to see any change. In practice, this playoff bump means very little.

And fortunately for Syndergaard tonight, the Giants do not present prominent base-stealing threats. With the aggressive Eduardo Nunez sidelined, San Francisco has just two starting position players who attempt steals of second at an above-average rate: Angel Pagan and Denard Span. But even then, the model predicts that Span will attempt steals in about 23% of opportunities off Syndergaard, while Pagan will attempt slightly more at 30%. Between the pair, it would take four opportunities on first base until we’d expect a stolen-base attempt.

Now, it is possible that the Giants push the envelope against Syndergaard, attempting a bunch of steals like they did back in May. But this research shows that it’s unlikely, and it’s also unlikely that the running game leads to Syndergaard’s downfall tonight. By FiveThirtyEight’s game predictions, the Mets, with Syndergaard as their starter, have a 55% chance of topping Madison Bumgarner and the Giants in their wild-card showdown. Even with his stolen-base vulnerability, Syndergaard is a uniquely dominant pitcher, and that dominance should persist all through the playoffs.