Stealing Success Against Pitch Speeds and Pitch Heights

Sometimes there just isn’t a way to sex up a headline. The other day I tried to sate my own curiosity by looking at what happens to the called strike zone when there’s a runner on the move. The results supported what I expected to be the case, but the data’s also incomplete, so it’s not like anything could be proven one way or another. Ultimately it turned out to be half study and half idea-introduction. There’s not a lot I can do about it now.

The post was powered by the searchable Baseball Savant, which somewhat recently added a “stolen base attempt” check box. This time around, I want to do something a little more obvious with the data, since it’s data I’ve never played with before. There’s information for more than 14,000 stolen-base attempts in the past four seasons, which doesn’t cover all the stolen-base attempts, but does cover most of them. Let’s assume, for the moment, the data that’s available is accurate. How do stolen-base rates change by pitch velocity? How do stolen-base rates change by pitch height? Do the trends follow the patterns we’d expect?

I probably shouldn’t need to tell you the patterns we’d expect. In theory, success rate goes down the faster the pitch. In theory, success rate goes down the higher the pitch (to a point). We might as well just dive into the numbers. First, the velocity table, broken into somewhat arbitrary groups:

Speed	SB	CS	Attempts	SB%	Avg. Speed
95+	609	175	784	77.7%	96.2
90-94	3699	1319	5018	73.7%	92.2
85-89	2815	1050	3865	72.8%	87.7
80-84	2128	623	2751	77.4%	82.6
75-79	1026	308	1334	76.9%	78.0
Under 75	308	74	382	80.6%	71.9

Immediately, something interesting stands out. Yes, the highest success rate comes against the slowest pitches, which is one of the things we’re looking for. But there’s no clear trend, and the second-highest success rate comes against the fastest pitches. Runners had a better-than-average success rate trying to steal against pitches that averaged 96.2 miles per hour. The sample is small, but not so small it can be dismissed.

Here’s what I suspect: We could be observing the results of selection bias. The pitcher on the mound doesn’t have any secrets. If that guy has a big fastball, everybody knows about it. The guy who’s reached base knows about it. The guy who’s reached base knows that, because of the velocity, the ball will arrive to the catcher faster. So runners might take off only when they feel more certain. More significantly, this could be selecting for better runners overall, where worse runners don’t even try to take the chance. It could be that mostly only premium runners try to steal against the hardest-throwing pitchers. This is one potential explanation of several.

It’s worth noting we don’t have a breakdown of attempts at second and attempts at third. It’s worth noting that, between groups, we’re talking about differences of only some hundredths of a second that the pitch is in flight. It’s worth noting that the hardest-throwing pitchers might just be below-average at holding runners on, perhaps because they haven’t had to worry about it; perhaps because they throw so hard. All groups here aren’t even, so it’s interesting to think about why we might see the things we see in these numbers.

Similarly, runners are a little more successful in group No. 2 than they are in group No. 3. Also, they’re a little more successful in group No. 4 than they are in group No. 5. I don’t think the message is pitch velocity is irrelevant. I think the message is pitch velocity is only a small factor. A steal takes place in three to four seconds. The difference between a fast pitch and a slow pitch might be between 2% and 3% of that.

Now let’s look at steal success rates against pitch height, again broken into somewhat arbitrary groups:

Height	SB	CS	Attempts	SB%
4.5+	282	154	436	64.7%
4.0-4.4	369	210	579	63.7%
3.5-3.9	799	339	1138	70.2%
3.0-3.4	1284	518	1802	71.3%
2.5-2.9	1722	639	2361	72.9%
2.0-2.4	1940	636	2576	75.3%
1.5-1.9	1750	507	2257	77.5%
1.0-1.4	1347	337	1684	80.0%
0.0-0.9	962	189	1151	83.6%
Under 0	130	20	150	86.7%

Height is feet above the ground at the front of home plate, as determined by PITCHf/x. A negative height would refer to a pitch that bounced in the dirt in front of the plate. Apparently some runners have been thrown out stealing even after a dirt ball. It’s uncommon, but it’s happened, and it’s probably embarrassing.

Here, I’d say we see the trend we’d expect. Generally, a throw to a base is made from a standing position. A standing position is a higher position, so the higher a pitch, the easier it is for a catcher to receive it and get rid of the ball quickly. If a pitch is down, a catcher either has to wait to get up, he has to drop back down again or he has to suck it up and throw from his knees. The trend is nice and consistent, and the success rate changes from below- to above-average between groups No. 5 and No. 6. With the highest pitches, runners aren’t even successful 70% of the time.

Yet it’s probably worth adjusting to remove designated pitchouts from the sample. Pitchouts are successful at nailing runners on the basepaths, but they’re also sort of their own beast, and they kind of skew the data. Pitchouts are mostly thrown high, and removing them only meaningfully changes the data for the first two groups. The changes:

Group 1: 64.7% –> 75.1%
Group 2: 63.7% –> 68.4%

Here, runners start to look successful against the highest pitches. Some of these pitches were simply too high, getting by the catcher or at least causing him to jump and end up out of position. From the catcher’s perspective, pitchouts excluded, the sweet spot is more between 3.5 feet and 4.4 feet above the ground. These pitches will drop a little more as they go from the front plane of the plate to the glove, and they’ll allow the catcher to almost immediately transfer the ball to his bare hand. At that point, all else being equal, the catcher has about a three-in-10 chance.

There are probably more things that could be done with this data. I’ll note there’s no real difference at all by horizontal pitch location. You’d expect there would be a difference by horizontal pitch location against intended horizontal pitch location, but we can’t measure that so we can’t speak to it. And honestly, I’m content for now. This is data I’ve never investigated before, so it’s nice just to have it, even if it doesn’t reveal anything groundbreaking. And it might still do that, given further study.