A Way to Measure Speed and Quantify Its Effects: DP-

Speed is listed as one of the five tools, but metrics for it are either lacking or impenetrable. Stolen bases have been around forever, and in tracking them we’ve learned a lot about speed on the base paths and who is the greatest of all time. Stolen bases start and dominate the conversation, but they don’t represent speed as much as they are a result of multiple factors, speed among them, of course, but also aggressiveness, timing, preparation (studying tape to learn pitchers’ pickoff moves), and opportunity. Some managers enforce conservatism on the base paths, and some players can slug just as well as they can run, so they bat in the heart of the order, more likely to reach base with a player ahead of them (can’t steal unless it’s a double steal). Also less likely to risk getting caught stealing when the next guy in the lineup is a good power hitter. Because Dingers and Ribbies are best enjoyed together, some players with good speed and the inclination to steal will miss out on opportunities.

We could spend all day noting the imperfections of stolen bases, the pappy of all speed statistics, but that’s unproductive, about as unproductive as writing down all the flaws in your real pappy. My goal is to find another statistic that complements stolen bases to help us better understand which players have the best speed, or whose speed impacts the game the most. (Those might be the same question.) To start, I looked at double plays.


Data from Fangraphs

Above you see how the percent of ground balls that became double plays has fluctuated over the last decade. They seem to have rebounded back to normal last season. You can also see that for the last eight years the AL has been more conducive to ground ball double plays (GDP). The difference is only a few tenths of a percentage point, but it is interesting, if you have an uneventful life like mine. I would guess the DH/pitcher disparity is the difference here. If pitchers are batting with less than two outs (i.e. the only time there could be a double play) and a man on first or multiple men on base, that pitcher is probably going to bunt. Meanwhile, a DH is usually selected for his power. If he was speedy they’d put him in the outfield. So a DH is probably more likely to come up in double play situations and ground into double plays during those situations.

So, with the league average rate of ground balls going for double plays, I compared the rates of individual players. In 2013, 6.5% of ground balls resulted in double plays (GDP/GB = DP%). Matt Holliday, in an extreme case, grounded into 31 double plays, out of 200 total ground balls hit. 31/200 = 15.5%. By multiplying Holliday’s total of 200 by the league average rate of 6.5%, we can say the league average hitter would have made only 13 double plays (xDP, x for expected). With the same amount of ground balls, Holliday made 18 more double plays than we would expect of an average hitter, thus “costing” his team 18 outs.

On the other hand, Norichika Aoki hit into only 9 double plays out of 328 ground balls, good for a 2.7% DP%. A league average hitter would be expected to hit 21.3 double plays, so Aoki “saved” his team 12.3 outs. Aoki is a fast player, Holliday notoriously slow and gimpy and getting old. These extreme cases jive with what we’d expect. In the narrow view speed doesn’t always break up a double play, but in the wide view it should, so let’s look at the players alongside Aoki and Holliday at the ends of the spectrum.


2013 leaderboard using the MLB average (i.e., not split between AL and NL). All data from Fangraphs. What the hell is Adam Dunn doing here? Click to embiggen.

Again, xDP is the expected number of double plays based on the league average DP% of 6.5. DP- is the difference between a player’s actual and expected double plays (GDP – xDP = DP-). DP- gives good players negative scores, the lower the better, as in ERA- or FIP-.



The leaderboard was filled with speedy guys, the anti-leaderboard with catchers and portly players. It does seem like speed is the driving issue behind the differences at the extremes. Of course, guys at the top of the second graph had a lot of opportunities to bat with runners on base, which may have inflated their DP- a little. Even so, Occam’s Razor wins out: The reason fast guys populated the first board and slow guys the second is that DP- measures pretty accurately the difference between fast and slow players in terms of double plays. In general, slow guys hit into more double plays and fast guys fewer, regardless of external circumstances; those always even out in the long run.

Stolen bases are choices. Not everyone decides to steal, but everyone must run to first as quickly as possible when they hit a ground ball. In ground ball double plays, a few tenths of a second make all the difference–fast players get to first base before the ball does, slow players can’t. Double plays above or below average are therefore a good index of speed; and therefore DP- is a good index of speed, and a purer one than stolen bases, for there is no element of choice. Over the course of a season, a individual’s speed when running against a potential double play can make a difference of 15 outs or so. This is just one small way speed shows up, the start of an inquiry.


N of players: 385

Fifteen extra outs may not sound like a lot, especially because that number represents the extremes, but I’ve started to mess with Tom Tango’s linear weights in order to convert the outs gained and lost to runs gained and lost. With what I have so far, Matt Holliday cost the Cardinals 3 runs and Norichika Aoki saved the Brewers 2. From there, it’s easy to get to -0.3 wins for Holliday and +0.2 wins for Aoki. So only fractions of a win, with admittedly crude calculations, but again, this is just a first look of one kind of situation in which speed is manifest.