clock menu more-arrow no yes mobile

Filed under:

Sabermetrics: More Stats, New Insights

Talking Sabermetrics: The under appreciated RE24 statistic and comparing pitchers

Greg Fiume

"Dammit, Jim, I'm a blogger, not an adding machine."--title of Feb. 10, 2010 column

I realize that traditional baseball fans sometimes complain about too many statistics and numbers, particularly in the world of advanced metrics. But I think that's the wrong way to think about it. That's like saying there are too many words in our language. Maybe it would be simpler if a synonym didn't exist for almost every word. But that would wipe away a richness in our vocabulary, since each similar word conveys a different perspective or nuance. Similarly, the value of a statistic arises from the insight it provides. And different statistics can provide different insights about the game of baseball. So long as the stat is useful, isn't "more" a good thing?

With that background, let's talk about two statistical measures that add to our insight into baseball. The first topic discusses the under appreciated "RE24" statistic. The second topic addresses a unique way to compare pitchers. This takes us back to one of the original purposes of the Talking Sabermetrics column: explaining the use of particular advanced statistics in order to add to your enjoyment of the game.

RE24 Statistic

RE24 (short hand for run expectancy for 24 base-out states) is shown on Fangraphs' win value page. Frankly, I hadn't paid much attention to this component of the win value family of stats until recently. When Tango, co-author of The Book, says this is one of his favorite stats, I probably should be paying attention. RE24 won the "March Madness Tournament: Baseball Stats Edition" at Beyond the Boxscore, which swept a somewhat lessor known statistic onto the main stage.

What is RE24?

24 possible combinations of outs and base runners exist. (Examples of base-out states: bases empty no outs; runners on 1st and 3d, 1 out; bases loaded, 2 out, etc.) The value of particular hitting events (like walks, strike outs, singles, HRs, etc.) will vary for each base-out state, because the probability of a run scoring (run expectancy) after the event depends on the bases with runners and the remaining outs. For instance, a strike out has the same impact as any other out with the bases empty, but becomes much more damaging with a runner on 3d and 1 out. We could give examples for every other type of event from GIDPs to HRs. Hitting stats like wOBA and wRC calculate a player's offensive production by assigning a league average value to each type of hit or event. RE24 assigns a value to each event based on the event's impact on the actual, known base-out state. In particular, RE24 calculates the increase or decrease in the probability of run scoring based on the change in base-out state. Fangraphs explanation of RE24 is here.

You may be wondering how RE24 differs from other win probability stats like "Win Probability Added" (WPA) and "clutch." RE24 is based on run expectancy, rather than win expectancy. Significantly, RE24 is more context-neutral than WPA and clutch because it does not incorporate the inning and game score components of leverage.

Why is RE24 useful?

RE24 can be thought of as a middle ground between advanced stats and old school stats. If you like RBIs, this is the advanced stats' replacement for RBIs. Some traditionalists think that sabermetrics belittles productive outs and sacrifices. Well, RE24 gives credit for productive outs and successful sacrifices, just like it debits the player for a GIDP. In fact, RE24 takes everything into account in arriving at the change in run expectancy.

Tango goes so far as to suggest that RE24, accumulated over a sufficiently large career sample, may (arguably) be superior to wOBA and wRC in gauging the value of a hitter over his career. Tango's discussion of RE 24, here, is helpful in understanding the concept. Although I won't go into it in this article, RE24 can also be used to compare pitchers, and may be particularly useful in evaluating relief pitchers. As Tango points out in the article, above, RE24 is expressed as runs above league average.

As shown on the fangraphs leaderboard, the Top 5 career leaders in RE24 (for as long as the data is available) are: (1) Barry Bonds; (2) Frank Thomas; (3) Chipper Jones; (4) Albert Pujols and (5) Jeff Bagwell. (Hall of Fame voters, take note of the Astros first baseman.)

What does RE24 say about the current Astros?

The 2012 RE24 for Astros' hitter is shown here. Justin Maxwell led all Astros with 11.21 RE24, followed by Jed Lowrie and Jose Altuve. Obviously the 2013 stats are in small sample size territory, but the current leaderboard is here. Chris Carter, Jose Altuve, Carlos Pena, and Rick Ankiel (is Dave Cameron listening?) are the current leaders in RE24.

The Astros' high strike out numbers were a narrative early in the season. Nobody really knows what strike out rate is "too high." One of the criticisms of the three true outcome players on this Astros team is that the Ks take away the opportunity for productive outs or other positive things that come with contact. The other side of the argument points to the HRs that typically accompany strike outs. This debate is unprovable. Or is it? Perhaps RE24 can cut through the polemics and determine how much the hitters are contributing to run expectancy. What I like about this approach is that it focuses on the individual player, rather than making generalizations. And so far, hitters with 3 of the 4 highest K rates on the team lead the team in RE24, and produced higher than anticipated run expectancy. Despite a 35% K rate so far, Chris Carter has managed to be an American League Top 25 player in RE24 in the last week.

How is RE24 used to determine "added value?"

Runs Above Average (wRAA) reflects the average runs expected from a player's offensive actions compared to the average hitter. Comparing RE24 to wRAA shows whether the hitter produced more or fewer runs than expected by wRAA. For example, Justin Maxwell's 2012 wRAA was 4.1, which means that Maxwell's actual contribution to run expectancy (11.4) as measured by RE24 was 170% high than the runs expected based on his wRAA--which is an excellent ratio.

Remember the debate when Jim Rice was voted into the Hall of Fame? His supporters said his RBI totals made him a feared hitter. Some of the stat heads responded by pointing to his propensity to ground into double plays (GIDP), as well as his low walk rates. Fangraphs shows Rice with a total RE24 of 247, almost 40% less than his total wRAA of 394, suggesting that perhaps the double plays significantly reduced his run production below expectations. (Rice led his league in GIDPs for 4 consecutive years, and was among the league leaders in GIDPs in other years.)

Below I have identified the highest RE24 contributors to the Astros since 1990, and calculated the difference between the player's RE24 and wRAA. The Value Added Percent is the potential percentage increase or reduction in the hitter's runs above average if the player's performance during the 24 base-out states is taken into account.


Net of wRAA

Value Added


Jeff Bagwell




Lance Berkman




Craig Biggio




Moises Alou




Carlos Lee




Hunter Pence




Derek Bell




Carl Everett




Bill Spiers








Luis Gonzalez




Ken Caminiti




Mike Lamb




Jeff Kent




I doubt that you are surprised that Bagwell, Berkman, and Biggio are the leaders in RE24 by a wide margin. Among those three Astros' icons, Biggio has a significantly higher margin between RE24 and wRAA. This makes sense, given that Biggio was a tremendous situational hitter and virtually impossible to double up during his peak seasons. The differences between wRAA and RE24 may be more problematic as we move down the chart and sample sizes become less reliable. In particular, many of the part time players produced large percentage differences between RE24 and wRAA, which may reflect sample size issues. In addition, part time players may have been more likely to enter games with runners on base. Bill Spiers, a super utility player, accumulated a RE24 far higher than the run production predicted by his wRAA. Undoubtedly this is why Spiers was prized as a pinch hitter. The extreme margin between Derek Bell's RE24 and wRAA is not easily explained. Apparently Bell was only somewhat above average as a hitter---except for hitting in base-out states when he could do the most damage.

Pitcher Similarity Scores

Just a short piece on pitcher similarity scores. Stephen Loftus at Beyond the Boxscore has developed pitcher similarity scores to compare pitchers based on their pitching arsenal. He uses pitch f/x data to identify comparable pitchers based on the following parameters: pitch velocity, pitch break (horizontal and vertical), pitch location, and pitch release points. He applies his method to a sample of pitchers in 2012, and arrives at most similar and least similar pitchers.

I have waded through his spreadsheet to find the most comparable pitchers linked by similarity score to current Astros pitchers. So, without further ado, here are the most similar pitchers:

Travis Blakely: Bartolo Colon

Lucas Harrell: John Lester, Zach McAllister, Luis Mendoza.

Phil Humber: Jeremy Hefner, Chris Volstad, Wade Miley, Dillon Gee.

Bud Norris: Christian Frederich, Tyson Ross, Jarrod Parker.

Jordan Lyles: Gavin Floyd, Matt Cain, Ivan Nova.

Dallas Keuchel: Jason Vargas.

All of these scores meet Loftus' criteria as "very similar." Lyles' similarity to Gavin Floyd was the highest score among these pitchers. And, how about Matt Cain as a comparable to Lyles? We can only hope that Lyles will be similar to Cain. Hopefully Lyles gets it together in Oklahoma City and returns to the Astros in the near future.

Bartolo Colon as a comparable to Blackley is slightly surprising. Perhaps it's more indicative of Colon's current state.

I can see Norris and Parker as similar pitchers. Frederich and Ross? I don't know enough about them to say.

Are you surprised by the results of these similarity scores? Is this a good way to compare pitchers' "stuff?" Do you see any good uses for this type of metric?