# Is speed really that important?

August 3, 2007 6 Comments

90 feet. Go.

Baseball is a game of hitting and running (and pitching, but that’s something else). In order to score, a player must actually run the 360 feet around the basepaths (in 90 foot increments) to end up back at home plate, and the faster a runner he is, the more able he will be to accomplish this. Right? Speed is one of those things in baseball that everyone seems to think is important, but they’re not entirely sure why that is. After all, it’s not entirely raw speed that influences whether or not someone is good on the bases. For example, in stolen bases, some amount of ability in reading the pitcher must be involved, and for what it’s worth, the pitcher’s abilities in holding the runner and the catcher’s arm must be taken into account. But then again, speed can influence categories commonly thought of as being hitting-related. Beating out an infield hit still raises a player’s OBP and AVG.

But how much does speed really matter on the basepaths? Runner on third, one out and the batter hits a high fly ball to left field. You’re the third base coach. Do you send the runner? How will you make that decision? Well, in previous work, I found that you’re not thinking about how fast the runner is. You’re looking at how far away the ball is. Even more, it doesn’t really seem to matter how fast the runner is as to whether or not he makes it. Not only do 97% of all runners who attempt to score on a fly ball to the outfield make it home safely, but speed doesn’t seem to be much of a determining factor as to whether or not the runner will make it home safely. In fact, most runners who attempt to take an extra base generally make it.

Time to break out one of my favorite techniques: binary logistic regression. This is a type of analysis that answers the question “What are the odds?” and allows us to see how different factors influence those odds. How do I know that speed isn’t much of a determining factor in telling us whether the runner will make it home safely on the potential sac fly? Well, binary logit tells us whether or not something significantly influences the odds, and if so, by how much (for you stat geeks, I look at Nagelkerke R-squared as my measure of variance explained).

One quick question to answer: How to measure speed? I’m using the good old Bill James speed scores method. In an article to be published in the upcoming edition of *By The Numbers**,* I actually developed my own speed measure from the ground up using some fairly high-level methodology (which took forever to calculate). James’ method and mine correlated at .81. Mine was a slightly stronger measure in terms of scale properties, but his is much easier to calculate (and the scale properties are still pretty good). After going through all the analyses that I did, it turns out the James method works pretty well. I stuck with it.

So, when would speed come in handy? It sure would come in handy trying to go from first to home on a double. I coded for all times from 2003-2006 in which a runner was standing on first, his teammate at bat doubled, and the runner did not stop at third. Either the runner in question was slapping hands with the on-deck guy after scoring (success!) or he went back to the dugout having been thrown out (failure). I only looked at those runners who made the attempt and ignored all the guys who just stopped at third and called it a day. I entered his speed score in as a predictor into the logistic regression. Sure enough, speed significantly predicted the odds that he would be safe in the expected direction (faster runners were more likely to be safe), but the Nagelkerke R-squared was…. 1%. That’s it. One percent of the “recipe” for the odds of whether or not a runner will make it is how fast he runs. I tried restricting the sample to situations with less than two out (still 1%) and situations with two out (a little south of 1%.) I looked at fly balls with less than two outs figuring that the runner might hesitate to see if the ball would be caught (all the way up to 1.5%). I made sure that the ball went through the infield (1.1%).

Second to home on a single? R-squared was 1.2%.

First to third on a single? R-squared was 0.2%.

What about the most obviously visible time when speed comes into play: the stolen base. I isolated all SB attempts of second base. The R-squared for speed did make an impact here: 4.2%. Not anything to sneeze at, but probably less than you expected. Previously, I’ve found that something as simple as whether the pitcher throws over is good for explaining even more of the variance (7.6%).

But there is one extra thing that speed helps with. I looked at situations in which there was a runner on first and the batter hit a ground ball to one of the infielders (i.e. a double-play ball). Did the batter’s speed help him to stay out of the double play? Yes, and it explained 5.5% of the variance.

Why are these number so low? Even on something like a stolen base, only 10% of the variance in success rates has to do with speed? Well, the other determining factors of whether the runner will make it are how big a lead he gets, how good the pitcher is at holding him on (those two will be correlated), how well the catcher throws, what sort of pitch the pitcher throws, and a few other factors. With some of the extra base advances on hits, there’s the issue of where the runner is when the ball is picked up and how far away that is from the base he’s trying to reach. Plus, you try throwing a ball 350 feet and hit a target to within a foot or so. The point is that while a speedy runner will have an easier time about things than a slower runner, the contribution of speed is not all that big. I’d guess that a lot more has to do with how well the fielders react.

So what does this mean for teams? Some teams keep a guy around whose only real purpose is to pinch run (and have a few at-bats in garbage time). Sure, there are some situations where every little bit helps, but the contribution of speed in most situations is much less important than I believe is generally thought. Even the strategy of pinch-running for a big-hitting (but slow running) player with a banjo-hitting but faster guy late in the game seems to have its drawbacks. Why lift your best hitter for a pinch runner, especially in a tie game, when the hitter might need to come up in the extra innings? All things being equal, a faster runner is a better player, but remember: you can walk home on a home run.

Dan Fox posted this on Baseball Prospectus:

http://www.baseballprospectus.com/unfiltered/?p=463

Look at each instance where speed can help you, it doesn’t seem like a huge deal, but add them all up and we’re looking at an extra win per year from the best players. Two of our favorites, Grady Sizemore and Chone Figgins, are tops this year so far. On a team level, the best baserunning teams are about 1 win above average, and the worst about a win below, at the 2/3 mark of the year.

Funny how the range of the best and worst teams is not much more than the best and worst individuals. While Figgins, Willits and Cabrera are busy adding runs on the bases, we’ve got Casey Kotchman showing his slowness and Vlad Guerrero running into too many outs.

Certainly, it can add up over time, and Dan’s previous work (and the column you link) makes it clear that this is the case. I don’t want to come off as saying “speed doesn’t matter.” It does. A little bit. I suppose that my point is that a good deal of baserunning outcomes (on a case-by-case basis) have less to do with speed and more to do with the fielders (and park effects!) than is generally believed. Over time, the signal of baserunning ability/speed can be detected, but every time I see Eric Wedge pinch run Mike Rouse for Travis Hafner late in a game, I shake my head. Assuming that Rouse even gets a chance to try to take an extra base, the contribution of his speed (and he’s no Vince Coleman… he’s just not… well, Travis Hafner) isn’t huge.

Over 90 feet, the difference between the World Record holder in the 100m dash and me (let’s just say Hafner might beat me in a race) is about 2 seconds, so even the most ideal pinch running scenario (let’s say Rickey Henderson for Cecil Fielder) buys you a second or so. Sure a lot can happen in a second, but I’m guessing that the standard deviation for how long it takes the a fielder to track down a ball in the gap and then make the appropriate throw (and relay throw) is a lot wider than that. That’s probably why the R-squared is so low.

Perhaps instead of pinch running, managers should consider sending the batboy out at the appropriate time to club the right fielder in the knee.

I won’t dispute your point, the location of where the ball is hit is going to be much more important than the speed of a runner. But in this case, a second is a huge deal. I’ve never timed anyone running from 3rd to home, but its the same 90 feet as home to first, and a typical speedy runner gets down the line in 4 seconds. Almost every player, when running hard, can get there in 5 seconds. Hafner, Cecil, Bengie Molina maybe a bit over 5.

Therefore if a speedy runner just beats the throw, a slow guy on the same play is going to be out by 18 feet!

Is the 97% stat (for runners scoring on flyballs) for real? That’s astounding.

Failing to send runners home from third on flyballs strikes me as one of the great inefficiencies in modern baseball. With a runner on third and two outs, for example, you should really send the runner if you think they have even a 40% chance of scoring. Since the next batter up can be expected to drive him in no more than 35% of the time (and accounting for a few wild pitchers, etc.).

Presumably, this is a function of third base coaches and managers avoiding being second guessed when 60% of the runners they send are thrown out–even though you’d pick up a bunch of extra runs each season.

Here is something I posted on the SABR list a few years ago. I found that speed probably does not explain much of a team’s winning percentage.

I wondered how important speed and

defense are to winning. Could we set an upper bound on how important they were?

What if we knew what percent of winning percentage was due to non-speed factors

were? Then maybe we could say that the percent that was not explained by them

is an upper bound on the share of winning attributable to speed and defense.

So I ran a regression in which team winning percentage depended only on

variables that had nothing to do with speed. They were frequency of walks,

strikeouts, homeruns by the offense and walks, strkeouts, and homeruns per

inning by

the pitching staff. Here is the regression equation.

WPCT = .553 + 4.77*HR% + 2.11*BB% – 1.17*SO% -1.17*HRA/IP – .486*BBA/IP +

.198*SOA/IP

The variables were all statistically significant. The r-squared was .739.

This means that 73.9% of the variation in winning percentage across teams is

explained by these variables.

This would set an upper bound of about 26% for the role of speed and defense

in winning. Their combined role could be much less.

Some of these variables may be highly associated or correlated with some

other variables that also affect winning percentage. For example, if a team

does

not strikeout much it might be because they are good hitters and therefore get

alot of singles, doubles and triples. But that would have nothing to do with

speed (unless there is some correlation between speed and how often you

strikeout).

I did not put in variables like overall hit frequency because some hits may

be the result of speed. I also did not include non-HR hits allowed by pitchers

since those are affected by defense.

I used walks + at-bats as my numerator for the hitting frequencies. I also

ran a regression with the strikeout variables taken out of both the offense

and pitching.

Cyril Morong