Touching BasesMarch 10, 2009
Baserunning and Leverage
By Jeremy Greenhouse

Let’s set the scene.

2004 ALCS. Yankees vs. Red Sox. Game 4. Red Sox down a run, Dave Roberts on first, ninth inning, no outs.

Dave Roberts advanced on a stolen base to 2B.

2007 National League one-game playoff. Padres vs. Rockies. 13th inning tie game, Matt Holliday on third, no outs.

Jamey Carroll hit a sacrifice fly to right (Liner). Matt Holliday scored.

That’s how it looks in the box score, but those two baserunning plays might be the two most momentous swings in baseball over the last five years.

Baserunning statistics are rarely looked at, yet the difference between the best and worst individual baserunners is about 20 runs, or two wins. Pretty significant. Players like Holliday, Carlos Beltran and Ichiro Suzuki, and other efficient baserunners become underrated when this skill isn't accounted for. So is baserunning an underrated commodity in the grand scheme of things?

There are several advanced metrics for baserunning, but my choice for this analysis is Bill James Online’s “net gain,” which takes into account “basestealing, avoidance of the double play, and success at taking the extra base while avoiding being thrown out.” I tend to think of four bases as equivalent to about one run, though I could be off base there. Here's the relationship between runs scored and net bases. Each dot represents a team's single season total over the time span 2002-2008.

The r-squared between runs and net bases is .17, so it’s pretty clear that the least important part out of the four facets of the game—hitting, pitching, baserunning, and defense—is baserunning. The difference between the best and worst baserunning teams in the majors is around 50 runs. That can be compared to 125 run swings in fielding, and between 200-300 run differences in pitching and hitting, depending on the year.

As demonstrated by "The Steal" and "The Sac Fly," mentioned at the beginning of this article, baserunning can at times be the make or break factor in any given game. Tom Tango developed, and statistically quantified, the concept of a leverage index to provide context to any game state. Baserunning, defense, hitting, and pitching can all be leveraged, be it through pinch-runners, pinch-hitters, defensive substitutions, or relief pitchers. I’d like to look at whether good baserunning teams also perform better in high-leverage situations. So, using one of my favorite statistics in fangraphs “clutch” score and one of my favorite types of visual presentations in google’s motion chart, I compared a team’s baserunning to its ability to come through when it matters most. Here's a year-by-year graphic of all 14 American League teams' baserunning metrics plotted against their clutch score.

And now the National League:

The correlation coefficient between net baserunning and clutch score is .12, which isn’t significant, but it’s not zero. Furthermore, going from first to third or scoring from first has a bit of a stronger correlation than avoiding the double play and stealing bases. Strikeout percentage has an inverse relationship of similar strength to baserunning, so there are a couple variables that might weakly relate to how well teams can come through when it matters most.

The average American League team is seven bases a year better than National League teams. I still don’t know what a National League style of play means other than inferior baseball. The Phillies have been the best baserunning team over the time frame, but they have been rather unclutch. The Angels rank sixth in baserunning, right behind the Yankees ironically enough, and the Halos have been twice as clutch as any team in the time period. Meanwhile, the Ozzieball White Sox and Bowdenball Nationals lagged in basferunning, while they put up neutral clutch scores.

How about a leaderboard of the most and least clutch teams since 2002?


I find the bottom five teams on this list interesting. Well, the Tigers .265 winning percentage is interesting too. But the Astros, Cubs, Indians, and Giants were all quality teams that won in spite of bad luck, unlike the Angels and Red Sox at the top who won because of it. Anyway, it looks like the clutch teams are better baserunners, but barely.

People sometimes try to explain the difference between a team’s Pythagorean winning percentage and their true winning percentage by the strength of that team's bullpen, baserunning, and "smallball" in general. But however a team creates or prevents runs, it is accounted for in the Pythagorean record. Then again, in many situations these aspects of the game are leveraged. So I decided to look at the difference between a team’s winning percentage and its Pythagorean winning percentage and winning percentage in one-run games. The results indicated that overall baserunning can’t explain how a team fares in close games at all, despite Dusty Baker's claim that "you gotta have some speed to win close ball games."

I attempted to break the data down further by looking at pinch-runners and performance in different situations, but unfortunately the only data readily available were stolen base and caught stealing scores.

  • The Athletics last year had the two most steals from substitute players of any team since 2002, thanks to pinch-runner extraordinaire Rajai Davis. Davis had 42 plate appearances as a sub, picking up 11 singles and one walk, but he pinch-ran so often that he had more stolen base attempts than times he reached first base. Oddly, Davis was a better hitter than basestealer as a sub on the A’s, as he hit to a tune of .341/.357/.561, while he was successful in just 11 of 16 theft attempts. It didn’t really matter for the A’s, who showed unremarkable splits in clutch situations. However, I wouldn't dismiss the idea of keeping a 25th-man on the roster as a specialist pinch runner.
  • The Phillies, the best baserunning team in the league each of the last two years, have topped the league in contributions from substitutions on the bases as well. Their sub-baserunners have put up 28 steals compared to a single caught stealing, while in the ninth inning the entire team has recorded 31 steals to one caught. But again, it seemingly makes no difference in the team’s record in tight games.
  • The incredibly unclutch Indians of 2005 were 3 for 11 stealing bases in situations with a leverage index above 1.5, and it probably did take them out of a game or two.

The sample sizes in these situations are small, so it’s hard to make conclusions using this data. But I think that the small sample size is a decent conclusion. While baserunning might be under-appreciated in today's game in a macro sense, it might be over-valued in explaining how an individual game is won and lost. Teams can leverage their baserunning to add a few runs over the course of a season, if that. Teams hold constant true-talent levels for baserunning, and it doesn't appear that the better clubs are able to achieve greater success by leveraging the ability at opportune times. Over 162 games, the difference between a team's offensive performance in high-leverage situations relative to their normal run production levels can't be explained by their baserunning.


Nice work Jeremy! Did you look at this using any of Dan Fox's baserunning metrics from BPro?

Jeremy - the A's have used the idea of the designated runner before: the famed Herb Washington, who finished his career with 31 steals (in 48 attempts), 33 runs, and exactly ZERO plate appearances. They did win the Series in '74, but check those postseason numbers... in a short series, where swinging even one game can be huge (as the Roberts steal was), the impact of such a player should be greatest.

Matt, I do like the baserunning metric on BP, but they currently only have years dating to 2007, so I decided to forego it.

Nightfly, I think I heard of a sprinter being used as a designated runner before. He was picked off his first time on base, right? I wonder whether having a baserunning specialist, fielding specialist, or LOOGY is most beneficial.

it's worth noting that Matt Holliday was in fact OUT AT THE PLATE, but the blown call allowed the Rockies to advance (and continue their unprecedented streak)

Right, but if I recall correctly he wasn't tagged out either.

And was that throw by Giles that bad? He might have the worst arm of any right-fielder in the game, and he got a lot of heat for not gunning down Holliday, but the throw was so accurate I kind of think it wasn't that bad. Definitely one of the most exciting games I've ever seen.

You have to remember that the value of good baserunning is already covered - albeit indirectly and accidentally by most overall offensive measures (ones that don't take situations into account) because the singles and doubles of good baserunners are not as valuable as the singles and doubles of bad baserunners because the latter do more to advance baserunners.

This obviously isn't a great way to measure anything, but you can't just say that a good baserunner is undervalued by his runs created,
equivalent runs, etc. without taking this into account.

I did a study a long time ago to see if good baserunning teams were undervalued by sabermetric stats, and the answer was that they don't appear to be. So the value of baserunning is already baked into these stats, although it may not be apportioned correctly.


I'm not sure I follow why base hits by good baserunners are not as valuable as base hits by bad baserunners. Why would one do more to advance baserunners than the other?

If runs created, equivalent runs, etc. don't account for baserunning, then they are not capturing a player's total offense. It's perfectly possible that a player can be overrated while being a solid baserunner, but that's beside the point.

Baserunning isn't a big enough part of the game to make a significant difference in the difference between's a teams expected runs and runs created, as random statistical variance would almost certainly outweigh it. But conceptually, I don't quite understand how it's possible that a statistic that doesn't include baserunning would be able to properly value a good baserunning team.

Jeremy - Because dosingles and doubles by good baserunners don't advance baserunners as much as singles and doubles by bad baserunners. Ichiro will make a double out of some hits into the outfield that Adam Dunn only gets a single out of, but the runner on first is going to likely advance the same number of bases in both cases. This is most true for players who get a lot of infield/bunt hits, which rarely advance baserunners more than one base. Mike Piazza's average single advances baserunners a lot more than Luis Castillo's average single. But both are counted as having the same value in RC, EQA, LW, etc.

So while these stats don't include the value of good baserunners' baserunning, they overvalue good baserunners' hits.



I'm not sure I follow. Baserunning is an independent skill from hitting. Arguing that Mike Piazza's singles are more valuable than Luis Castillo's singles is an entirely logical argument, and the case that power hitters are systematically undervalued by a linear weights approach is very thought-provoking and intuitively correct, I don't see the connection to baserunning. The three examples of terrific baserunners I used were Carlos Beltran, Matt Holliday, and Ichiro Suzuki. That's a very diverse set of players, and I don't see how each one of their run production could be systematically overvalued. I think your syllogism
that good baserunners are light hitters, light hitters have singles that are overvalued, so good baserunners' singles are overvalued is not appropriate, since there's no causation between baserunning and being a light/power hitter.

Also, the two measures for team performance I used were runs and clutch score, which account for any type of run production and accurately measure any type of hits that lead to runs.

Jeremy - I wasn't referring to the measures you used later in the article, just the general statement from earlier in the article.

Anyway, a common thread of players who get infield hits is that they are generally good baserunners. You don't get many infield hits if you are a slow baserunner. There are other factors, of course - right-handed batters get more infield hits, batters who hit a lot more ground balls get more infield hits - but speed, which is obviously a huge factor in baserunning, is probably the most important factor. The same thing goes for doubles that advance a baserunner on first to third instead of home. A good baserunner will have a greater percentage of doubles that advance a runner on first only to third instead of home than a bad baserunner. because a good baserunner is more likely to get to second on the same batted ball that gets a bad baserunner to first.

I don't ordinarily post simple "me too" remarks, but I want to strenuously agree with Greg's position. Analyses of decades' worth of data by analytic means shows--to me, at least--that the advantages of speed, such as they may be, are already "built into" any decent run-production equation, notably, as Greg says, in the extra bases. You can see 1138 team-seasons, MLB for 1954 through 2001, graphed by one methodology at; there is no explicit "speed" component in the equation, and in fact overall accuracy hardly matters if we omit SB/CS data.

Fair enough. I didn't know that you were referring to cases of individual players, but now I see what you're saying. Pretty much an "empty" RC/EQA. That makes sense. I might look into that. Compare something like fangraphs' RE24 to wRAA and see whether theres a systematic difference between the types of players based on speed or other traits. Sorry I didn't understand your point before.

Eric, I'm having a tough time understanding how a system that doesn't account for baserunning would properly weigh it. The regression model will create a nice line of best fit, since getting on base and hitting for power are by far more important than baserunning, but if baserunning is properly accounted for, wouldn't that be incidental? There would still be omitted variable bias.

My plan is to regress baserunning and a predicted measure of runs (Can the linear weights method of wRC be accepted?) against runs. I believe that my hypothesis is that good baserunning will add to the predicted runs, as it is not accounted for in the model, while you say that it will subtract from predicted runs, as it has already been adjusted for?

I hope I'm representing the argument correctly here.

agreed that Jamie Carroll's fly ball to right was huge...Holliday scoring that run off of Hoffman on the extra day of the regular season was just enough to bump my opposition's ERA above mine and I won my Fantasy League on that call at the plate...

Jeremy - sorry for the late reply. Herb Washington is the guy you're thinking of, I'm almost certain of it.

Alright, I did some rough tests. Adding the baserunning metric does indeed increase the accuracy of run estimators of wRC, though a change of an r-squared from .9 to .91 isn't really significant. For each base, a quarter of a run can added or subtracted from the predicted model. The correlation coefficient between baserunning and the difference in wRC and R is .19.

So not including baserunning will at times create misestimations of players and teams, and baserunning does help explain some errors in run estimators.