# The Slider: A Visual Analysis

Let’s take a visual stroll across Statistical Boulevard and meander past Analytics Way as we dissect the slider (as classified by PITCHf/x) in various ways. The intent of this article is to show some interesting visuals on the aforementioned slider and perhaps bring to the surface an interesting tidbit or two.

We begin with a brisk walk on Top Season Street and take a peek at the top slider-throwing seasons since 2009. I’ve excluded 2008 since it was the first season to have PITCHf/x data, and it showed some weird inconsistencies with the rest of the data set. 2008 will be excluded from all data in this article. The graph below is courtesy of FanGraphs’ leaderboards feature.

We pause our brisk walk to catch our breath and marvel at how dominant Yu Darvish’s slider was in 2013 (almost 10 runs better than any other slider since 2009). Questions about whether or not the 37 percent slider usage was a contributing factor in his subsequent injury problems bubble up, but they are moved aside as we notice four of Clayton Kershaw’s last five seasons show up in the top 27 and the fact that Tyson Ross shows up for his last two seasons despite his slider being worth fewer than two runs/100, an indicator of just how much he throws his slider. Scroll down to the bottom and marvel at just how bad Rick Porcello’s slider was in 2012.

What do most of these top slider seasons have in common? They appear to be dominated by pitchers who get a lot of swings and misses, as opposed to pitchers who generate a ton of ground balls. Is it possible to have a slider that is both a dominant swing-and-miss pitch as well as a dominant groundball-inducing pitch?

Here we see sliders with swinging strike percentage on the vertical axis and groundball percentage on the horizontal axis. Each circle represents a distinct pitcher season in which at least 100 sliders have been thrown by that pitcher. You can mouse over any circle to see which data point it relates to.

Red circles indicate either a SwStr% above 28 percent or a GB% above 14 percent. These are purely arbitrary but serve to highlight that the high-GB% pitchers are clustered below the average SwStr% and the high-SwStr% pitchers are clustered to the left (below average) GB%. This is natural, since if you get a ground ball, you necessarily did not get a swing and miss.

Click on the blue circle that’s at 24 percent SwStr% and 13 percent GB%, which represents Joba Chamberlain’s 2011; Tableau will now highlight the other seasons we have data for Chamberlain. Clearly, the 2011 season in which he was well above average in both metrics was an outlier compared to his other seasons.

Click on the circle again to clear the selection and go back to the default view. You can find Kershaw right on the average GB% line at the 30 percent SwStr% line. Clicking on that circle will show you his five most recent seasons, all with very consistent profiles compared to the other two circles, which represent 2009 and 2010. If you can find Aroldis Chapman, you’ll see that his slider isn’t very good at getting ground balls.

Let’s hypothesize that it’s far easier for an elite pitcher to leverage the slider as an elite swing-and-miss pitch than as a groundball pitch. By browsing through the top GB% slider pitchers (i.e. by mousing over the red circles to the right of the graph), the only potentially elite pitcher is Marcus Stroman, for whom Jeff Sullivan waxed eloquent about his pitch comps.

While the top of this graph is dominated by relief pitchers, you’ll see that pitchers who are above-average SwStr% slider pitchers typically will be consistently above or below the SwStr% average line and fluctuate between positive/negative on the GB% axis (and the same can be said for high-GB% pitchers).

We take a detour on the Road to Swinging Strikes and switch to a jog as we flip through three charts, all showing little to no correlation between SwStr% and the potential predicting variable.

The first chart plots velocity against SwStr%, and we see almost no relationship. The second plots absolute horizontal movement against SwStr%, and again there is no relationship. The third charts vertical movement against SwStr%, and while we see a non-zero relationship, it is negligible and probably has more to do with an increased probability of throwing lower in the zone (more on this soon).

What then (of the elements we can measure in PitchFx) are predictive factors for inducing swings and misses? There are two variables I could find, the first one being the number of strikes in the count, the second being the median vertical location of the slider (lower being better).

We can see pretty clearly that the number of strikes in the count is a pretty solid indicator of the probability of getting a swing and a miss. The number of balls is only relevant in a 3-0 count, when the batter is extremely unlikely to swing, especially at a slider.

This brings us to the second component, median vertical location. I chose median over average since it does a better job of getting rid of the noise of really bad pitches skewing the results. This leads us to our next chart:

Red indicates a high SwStr%, yellow an average one, and green a low one. Here we see a fairly strong relationship between median zF (final location of the pitch as it crosses the front of the plate, measured in feet). The R-Squared value of 0.27 implies 27 percent of the variation in swinging strikes can be explained solely by the vertical location of the pitch.

This model estimates that, for every foot lower you can throw your slider, you’ll gain about 10 points of SwStr%. This then begs the question: If throwing a slider lower is better, is this a repeatable skill that a pitcher can demonstrate year to year?

Glad you asked.

What we see here is a surprisingly strong relationship, with an R-squared in the neighborhood of 0.5. What’s even more interesting is that this relationship holds up year to year if you flip through the “year” filter at the top, always in the 0.5 ballpark. This would suggest it is a real skill to be able to throw a slider consistently low and one that will translate to increased swings and misses. One note: The year-to-year graph uses average instead of median since Tableau can only compute using the average for some reason.

I’m going to leave you with one last chart, with which you can peruse through all the top slider seasons by runs prevented per 100 pitches, graphed against slider usage.

Great article, easy read. Powerful graphs. I wonder what the SwStr% vs. median zF chart would look like at three different strike counts – 0,1,2. Maybe all with 1 ball or something to control it even further.

Thanks for the feedback. Here are the correlations in the following strike counts (all ball counts)

0 Strikes: 0.17

1 Strike: 0.22

2 Strikes: 0.17

It looks like there is a lot more SSS noise as you slice and dice and when you compare it to the aggregate correlation of 0.27, it would indicate that it is likely to be a count-neutral effect, which is surprising to me, since I would have guessed that the 2-strike correlation would be significantly higher (as you suggested).

What is interesting is that the average height of a slider is very strike-count dependent.

0 Strikes: 2.2 feet above the plate

1 Strike: 2.0

2 Strikes: 1.8

Zack Greinke has a dominant ability to throw low sliders with two strikes (2 of the 3 lowest zF scores for 2 strike sliders).

Thanks for the comment!

This was great and informative! Thank you Eli