Measuring the Impact of Stealing Bases Using Win Probability

FanGraphs' Weighted Stolen Base Runs (wSB) is likely the most frequently used advanced metric for measuring the value of stolen bases and caught stealing. It feeds into FanGraphs' WAR calculations as part of base running value, but also operates as a standalone metric (unlike Baseball Reference's base running metric which is not broken out by the different constituent parts like steals, extra bases taken, etc.). wSB is intended to measure the number of runs a player contributes to their team through stolen bases and provides a relatively easy-to-understand value for understanding whether a player adds value through stealing bases. However, it has one characteristic that I thought might make it less useful as a measure of how well a team uses stealing bases to win. wSB assigns a value of 0.2 runs to every successful stolen base attempt, agnostic of game state. This means that a stolen base in the fifth inning of a blowout and the ninth inning of a tied game would be treated equally in terms of value under the wSB formula. This means it cannot tell us whether specific teams are particularly good or bad at using stolen bases in high leverage situations, rather than just being good or bad at stealing bases in general. Of course the two are connected, but I was interested in finding an approach that exposes the former.

Win Probability Added (WPA) stood out to me as a good tool to measure how good teams are at using stolen bases in leverage. Win probability is pretty self-explanatory, being a team's statistical chance of winning a game at any exact moment, expressed as a percentage. WPA is the difference between a team's win probabilities before and after the occurrence of some intervening event. WPA is often used to measure reliever performance as it tells us a lot about how good a reliever or team's bullpen was at performing well in high-leverage situations.

I thought the same could be applied to stolen bases, as the impact of stolen bases and outs, particularly in leverage, is clearly reflected in win probability. For example, in a tied game in the bottom of the ninth with no outs, Statcast's win expectancy for the home team is 71% with a runner on first. This jumps to 80.5% if that runner steals second with no outs, but drops to 58.2% if the runner is caught stealing and there is one out with nobody on.

The Build

In order to measure this I pulled every stolen base and caught stealing from the 2026 season and the Statcast win probability from before and after the stolen base. I exclude pickoffs, balks, wild pitches, passed balls, and defensive indifference. I also attempted to control for the outcome of the plate appearance (if there was a batter event on the same pitch as the stolen base/caught stealing) by estimating the impact of that batter event and determining which part of the change in win probability was attributable to the running event. I assessed this purely at the team level which means that double steals are treated as one play.

Results

WPA for Stolen Base Attempts by Team (2026)

Analysis

A few things immediately jump out from the results. First, every team had a positive WPA from stolen bases in 2026. This is because the average success rate for stolen bases is high.

Second, the range of WPAs does not appear large, but is potentially meaningful. Relivers with a WPA for 2026 around the 2.391 level of Washington's stolen base WPA include Tanner Scott, Jhoan Duran and Grant Taylor. Batting WPA at this level includes Shohei Ohtani, Jonathan Aranda, and Bobby Witt Jr. Relievers with a WPA around the 0.182 level of the Astros' stolen base WPA include Jack Dreyer and Ryne Stanek. Batters at this level include Jeff McNeil and Andres Giminez. The clear gap in quality between these players illustrates how strong a difference of two WPA can be.

Third, the stolen base WPA results are quite similar to the wSB results for the year, with a few notable exceptions that result from the fact that wSB is agnostic of when a steal attempt occurs. The Padres were the third best team by wSB this year but 15th by stolen base WPA, indicating that their success at stealing was greater in low leverage situations than in high leverage situations. Conversely, the Athletics were 23rd in wSB but 11th in stolen base WPA, indicating that: (i) they were more willing to take risk on stolen bases in low leverage spots (leading to a lower overall success rate) but had relatively high success in high leverage situations; and/or (ii) they don't attempt steals in low-leverage situations and concentrate their steals in high-leverage situations. The Rays (10th in wSB and 4th in stolen base WPA) and Miami (7th in wSB and 3rd in stolen base WPA) also have this characteristic.

Another interesting finding comes from the WPA/Attempt column. This generally decreases in line with WPA overall, but the Blue Jays, Dodgers, and Rangers all stand out as teams with bottom half stolen base WPA values but top half WPA/Attempt values, meaning they attempt stolen bases relatively infrequently but generate relatively large WPA when they do run. Toronto is particularly notable, being 16th in stolen base WPA but 6th in WPA/Attempt. One read of this finding is that these teams are too conservative in attempting stolen bases, but it could also be attributable to roster construction (e.g. their rosters are generally not good at stealing with the exception of certain players who are very good at it).

All in all, I think the WPA associated with stolen base attempts provides an interesting lens to look at a team's running strategy (high vs. low leverage) and at the value associated with being good at stealing bases, particularly in high-leverage situations.