How Managers Were Fooled by the Home-Run Spike

If you follow baseball at any level, pitch counts are a part of your life. Some people hate them, some people think they need to be more heavily enforced for amateurs. They impact our thinking about pitcher health, durability, and effectiveness. Every broadcast tracks them.

The interest in pitch counts isn’t simply a media/outsider-driven affair, either. Teams have significant financial and competitive incentives to keep pitchers healthy and effective, and it certainly seems like they’ve stopped pushing their pitchers as much within individual games over the last decade.

But even amid this general trend, something significant happened over the last two seasons. Instead of following the long-term trends, pitch counts fell sharply starting last year and have fallen again in 2016. While we can’t offer a definitive explanation without spending time inside the heads of MLB’s managers, the evidence seems to suggest that the culprit is something other than a newfound appreciation for protecting arms.

Let’s get right into the data (through Wednesday’s action for 2016). Here are the percentage of starts which reached a certain pitch count for every season dating back to 1995. There’s quite a bit to discuss in this table, but I’ll call your attention to a few of its most important points.

First, there’s a clear adjustment over the course of 1997 and 1998, likely as the league adapted to expansion from 28 to 30 teams. With a diluted talent pool, starting pitchers threw more pitches to cover the innings that would otherwise have gone to player who wouldn’t have been on MLB rosters in 1996.

Beyond that, high-pitch-count starts have been trending downward for quite some time. In 2002, 0.56% of all starts went 130 pitches or more. Over the last three seasons, that number has been more like 0.04% of all starts. The 130-pitch outing is virtually dead, basically only occurring when a pitcher is chasing a no-hitter.

But the trend exists at lesser thresholds as well, such as 115-plus-pitch starts. This graph does the trick:

However, while it’s been easy to see the steady vanishing of high pitch counts over the years, there’s a less obvious trend occurring, as well. Over the last two seasons, teams have begun to back off even on the 90/95/100-pitch thresholds. Not only are teams staying away from those really long 125-pitch outings, they’re cutting back on how often they let their pitchers throw 95-plus pitches in a start.

It probably won’t shock you to learn that pitchers are not becoming any more or less efficient on a per-inning or per-batter basis in recent years. Over the last 15 years, pitches per batter faced have been trending up for starters, but there hasn’t been a dramatic change over the last two seasons. Pitches per inning have held relatively steady over the entire sample as well.

What we do observe, however, is that, over the last two years, there has been a marked drop in pitches per start and innings per start. Given that pitchers are throwing fewer innings and fewer pitches in tandem, we know that manages are pulling starters earlier, but it’s not yet clear if this is inning driven or pitch-count driven just from looking at this initial data.

Let’s confirm this by looking at the percentage of starts to complete a certain number of innings. If you look at the table below, you see a clear drop over the last two seasons at the five-, six-, and seven-inning thresholds.

The question we want to answer is this: is the league getting more cautious with their arms or are teams responding to what’s happening on the field? You could imagine a scenario in which teams are collectively being more cautious with workloads, deciding to shave pitches or outs whenever possible. But it’s also possible that this is the product of something else and that the lower pitch counts are a symptom rather than a cause.

There isn’t a clean way to test the idea that teams are being more cautious. We could devise a few strategies, but given the multitude of confounding variables, getting a clear picture would be difficult. However, it does seem rather straightforward to study one “happening on the field” explanation.

Over the last 13 months, scoring has been up thanks to an uptick in home runs. It seems possible that, as offense has returned over the last two seasons, that teams have pulled their starters more quickly because they have allowed more runs. There’s a relatively easy test of this, considering that offense spiked right after the break last year. If we compare 2014 and 2015 by half, we can see if this trend started in the second half of last year, right alongside the home-run spike.

When home runs started flying out of ballparks in the second half of 2015, managers started to pull their pitchers earlier. At an individual level, this makes plenty of sense: if you have a pitcher who is allowing more runs, especially more home runs, you’re likely to think he’s having an off night and the other team has his number. No reason to keep him in the game any longer, even if he has a few more pitches left before hitting his usually threshold.

But at the macro level, pulling him is not particularly logical because the individual pitcher isn’t performing any worse relative to his peers, he’s just performing worse relative to what you would have expected before the 2015 All-Star Game.

What’s fascinating about this is that managers are responding to the home-run spike. If you look at earlier seasons, there’s not a clear relationship between scoring and pitches per start or innings per start. And you wouldn’t expect there to be in the aggregate. The decision to pull the starter should be based on his effectiveness relative to the other pitching options, not his un-adjusted in-game performance. But that’s not how it happened. Managers didn’t realize the league was scoring more runs; they all acted as if their starters in particular were losing effectiveness.

Put another way, the home-run increase disrupted managers’ ability to evaluate their own starting pitchers. Instead of recognizing that run-scoring was up and that expectations should be adjusted as a result, managers started pulling their starters in response to the runs they were allowing.

It makes sense from their perspective. How would they know last August that the game was fundamentally changing? It wasn’t an easy change to spot and, at first, it could easily have been brushed aside as a blip. Outside analysts are now confident that the home-run spike wasn’t random, but we still haven’t definitively established the cause. It seems likely that the ball is juiced, but you can’t blame managers for failing to see that at the time.

This gets right to the heart of an idea that fascinates me about human cognition. As a species, we’re extremely good at adapting to changing conditions. We can get used to almost anything. Even things that seem unimaginable now will seem normal after a little exposure. But when it comes to evaluating numeric performance, our minds get anchored to certain numbers we consider meaningful and we have a really difficult time replacing those benchmark statistical levels when the world changes around us. We can adapt extremely well, but we have a hard time adapting baselines that are numeric. Numbers are sticky, it seems.

The interesting question is what happens next. Will managers readjust once they find their footing and get a sense of the new normal? Or will this change be permanent and lead to earlier pulls for starters despite the fact that individual pitchers are no less effective relative to the league than they were two years ago?

This is one of the consequences of a meaningful change to the baseball, intentional or not. If it doesn’t happen out in the open, decision makers aren’t going to be able to respond to the changes in the right way, and by the time they figure out where they went wrong, it might be too late to go back.