Diving into Statcast Sprint Speed

In April, Statcast quietly introduced a new speed metric dubbed Sprint Speed. It wasn’t until late June that Baseball Savant made the leaderboards publicly available and we now have data going back to 2015. I have been meaning to dive into the data to find any incremental value, and finally the day has come. From the leaderboard page, the metric is described as thus:

Sprint Speed is Statcast’s foot speed metric, defined as “feet per second in a player’s fastest one-second window.” The Major League average on a “max effort” play is 27 ft/sec, and the max effort range is roughly from 23 ft/sec (poor) to 30 ft/sec (elite). A player must have at least 10 max effort runs to qualify for this leaderboard.

You might recall that we already have a speed related metric on our player pages. It’s the Spd, or Speed Score, metric which was originally developed by Bill James and its components include stolen base percentage, frequency of stolen base attempts, percentage of triples, and runs scored percentage. So my initial curiosity stemmed from whether Statcast’s Sprint Speed correlated better with BABIP than Spd. My most recent xBABIP equation accounted for a hitter’s speed using Spd, but would Sprint Speed be an even better metric to use?

My assumption was that yes, of course it would be, as Sprint Speed is straight speed and ignores a hitter’s supporting cast (Spd is slightly team dependent given the “runs scored percentage” component, and also managerial philosophy dependent given its reliance on stolen base attempts). But rather than just run the correlation between Sprint Speed, Spd, and BABIP, I decided to compare the correlations between those speed metrics and a host of speed related stats. I only ran correlations for 2015 and 2016, ignoring 2017. So let’s get to it…

I highlighted which speed metric correlated better with each speed-related stat. I was rather surprised to find that Spd “won” significantly in four of the six stats, while one of the two Sprint Speed wins was by a minor amount. Furthermore, I also expected Spd and Sprint Speed to correlate more strongly with each other. But it just goes to show you how context affects Spd.

At first, I wondered how Spd, which includes four components, could possibly correlate more strongly with triples per ball in play than Sprint Speed. Triples are essentially doubles for speedsters, so the faster you are, you figure will lead to more triples. But then I realized that triple percentage is a component of Spd, so of course it’s going to correlate highly with the metric. The same is true of the stolen base related stats I tested the correlations on.

BABIP, though, was not a component of Spd, so comparing the correlation here is a fair fight. Unfortunately, Sprint Speed barely wins, which is disappointing. I was so sure that Sprint Speed would prove to be far more correlated, which would help me improve my xBABIP equation, but that hasn’t proven to be the case. However, just because it doesn’t correlate much better with BABIP doesn’t necessarily mean it’s not the better metric to use. When the 2017 season ends and I have three years of data, I will run my xBABIP equation with the Sprint Speed metric instead of Spd and see if it makes much of a difference. Fingers crossed.

Interestingly, Sprint Speed won IFH% by a significant margin. Infield hit percentage is not a Spd component, so again it’s a fair fight. This might be telling that Sprint Speed is indeed a better metric to look at because the stats where Spd won were included in the calculation of the metric itself.

We still don’t have enough data to perform any heavy analysis. But what I would like to see is if we could use Sprint Speed as a stolen base breakout indicator. If a player hasn’t attempted many steals in the Majors, which likely resulted in a mediocre Spd score, but Sprint Speed confirms the player has excellent speed, does he beat his stolen base attempt projections the following year?