On August 7, Randy Arozarena slashed a double to right. He came into second base at a trot, so evidently safe that he didn’t need to sweat it. As the camera focused on him, he turned and hyped up the dugout. There was nowhere else to look; there had been no runners on base and thus no other action to follow.
Things weren’t so sunny 10 days later. Arozarena batted with two on and two out, and a double would have been absolutely glorious. The runners would be off on contact, which meant the difference between a double and an out was two-plus runs — the two that would actually score, plus some chance of Arozarena himself scoring. But Arozarena struck out on a 1-2 slider from Bailey Falter, and the inning ended.
Advanced statistics don’t assess the value of a play in just one way. You can think about these two moments extremely differently depending on which metric you’d prefer to use. Our main offensive statistic, wRC+, ignores context on purpose. It works out the average value of a home run across all home runs hit in the majors in a given year, and uses that as the value for every home run. It does the same for every offensive outcome, in fact.
Win Probability Added zooms all the way in and focuses on how much a team’s chances of winning the game change on every single play. That double was huge at the time; the Mariners trailed by a run in the eighth inning, and Arozarena’s hit instantly put the tying run in scoring position. It was the second-most-positive offensive event the Mariners recorded all day, trailing only Cal Raleigh’s two-run blast that accounted for all the team’s scoring. Water is wet, candy is delicious, and the Mariners can’t score.
RE24 gets talked about less, but it’s an equally reasonable way of assigning offensive credit. It works off of base/out states. There are 24 of them: eight different ways runners can be arrayed on the bases across zero, one, or two outs. There’s intuitive appeal to this way of doing things. A deep fly out with one out and a man on third is really valuable, while the same ball with no one on or with two outs is just like a strikeout. Batters change behaviors based on the situation. Why wouldn’t we credit them for their ability to do that?
I’m not here to tell you which of these options you should prefer. I am here to tell you that I decided to use RE24 to power WAR and see how much our perception of hitters would change if we focused on what they did to affect the base/out state instead of treating their offense with pure context neutrality.
Using Arozarena wasn’t an idle starting point. He’s actually the hitter most affected by this switch, losing a whopping 1.67 WAR in this new way of looking at things. If you’re looking for a reason why, it’s pretty simple. With the bases empty, he’s hitting .231/.356/.426, comfortably above average. With runners on, he’s hitting .189/.296/.321. He’s batting only .125 with a runner on second base, the times when hits are most valuable.
On the other side of the ledger, Brandon Nimmo is having a solid season no matter how you look at it. His offense is down slightly from his career level, but it’s still above average, and he’s on pace to end up with 20-25 homers and an enviable OBP. If you consider the base/out context of his hits, things are much better than that. His WAR improves by 1.5 if you replace context-neutral offense with RE24. He’s the reverse Arozarena, in other words. With a runner on second, he’s batting .310. He walks quite a bit when there’s no one on base, but gets more aggressive when a hit would be most potent. He’s hitting singles when they’re most valuable.
In other words, the smoothing function performed by wRC+ specifically says that all singles are the same, but RE24 notes that they aren’t. It’s pretty clear to me that wRC+ does a good job of explaining the most elemental things about what make up a player’s offense, the things least likely to change; variation in base/out states is out of a hitter’s control, and their approach is pretty similar in many situations. Variance swamps signal; hitting a grand slam doesn’t tell us that much more about a player than hitting a solo home run, but RE24 counts them incredibly differently.
On the other hand, WAR isn’t all about understanding the stable parts of a hitter’s profile. If you want to use WAR to understand talent level, I think that wRC+ is the gold standard. You could use a different context-neutral statistic if you’d prefer – DRC+ or OPS+ or whatnot – but looking at things stripped of context does a great job of cutting through noise and focusing on key skills.
Let’s put it another way. I looked at the 2022 and 2023 seasons and took every hitter who batted at least 400 times in both years. I converted RE24 to a rate statistic (it’s a counting statistic by nature) by dividing by the number of plate appearances, then compared how internally consistent each statistic was. Of the variation in year two wRC+, 28.6% could be explained by year one variation. Only 15.9% of the variation in year two RE24/PA could be explained by year one variation. In other words, wRC+ is far more predictive of itself in the future.
This makes good sense, for the reasons I outlined above. It also explains why we use it as our marquee offensive statistic: It does a good job of showing which hitters are the best in a stable way. Context-neutral statistics have been part of baseball since the very beginning. Every single part of a slash line is context-neutral, and home run and stolen base totals are too. No one would ask how many home runs a player hit in each base/out state and try to use that to project their future home run rate in varying base/out states; they’d ask how many bombs the guy hit, period. The argument for wRC+ is pretty obvious.
But if you’re concerned not with talent level but with what happened in the past, the arguments for RE24 get better. Yes, in the long run, singles are worth about 0.71 runs more than making an out, but with a runner on second and two outs, they’re pretty obviously worth more than that. To evaluate what actually happened, which plays were of value to a team not for their predictive power of future outcomes but for what they did in the past, you probably have to consider context.
Another way of thinking about it is that at the team level, RE24 does a much better job of predicting run scoring than wRC+ (or Off, our measure of offensive value, because for inscrutable database reasons RE24 includes stolen bases and times caught stealing). More specifically, team RE24 has a 0.89 r-squared to team runs scored; almost all of the variation is explained by variation in RE24. Offensive runs still does a good job, at 0.79, but obviously including the context helps sharpen the correlation.
What does this all matter? In some sense, it doesn’t matter at all. You can say whatever you want statistically; it doesn’t change how the games are actually played. What statistic you prefer doesn’t make the outcome different. A lot of what we do here at FanGraphs is about predicting the future, whether it’s prospect rankings, trade value, or our analysis of player breakouts and breakdowns. For things like that, context-neutral statistics just provide more predictive power.
When you’re analyzing past value, though, I think you can make an argument for subbing in RE24 (or WPA if you’d prefer — I’ve actually made that argument before). Sure, in the long run Arozarena’s general level of offense will help increase run scoring, but in 2024, his actual results haven’t done so. From RE24’s perspective, he’s actually been below replacement level, if you account for how much of his positive production has come when it’s less important and how many bad outcomes have happened with runners aboard. Nimmo’s having a down year in terms of his true talent, but in terms of actual delivered value to the Mets offense this season, he’s been exceptional.
One of the cool things about WAR is that it’s just a skeleton that you can modify how you’d like. Different measure of offensive runs produced? Throw it in. Different defensive system? Sure, it can handle it. New positional adjustments? I promise you, WAR still works in a broad sense even if you think the first base adjustment is wrong.
This is a great example of that. If you want to use WAR to say who the best players are, our calculation of it is well adjusted for just that. You could improve it! You could overweight outcomes that are sticky and indicative of batter skill, and down-weight things where variance is king, like BABIP and, to some extent, defense. For the most part, though, our calculation of WAR is built around answering the question of who’s the best.
“Who’s helped their team the most” is a different question, but you can make WAR answer that too. You just need to swap in some new metrics, and again, it’s pretty easy to do so. RE24 is a good one, which is why I’m using it as an example today, but the key part is that you should make your version of WAR do what you want it to do, because it really can do pretty much anything.
I don’t want to leave you with that preachy paragraph, so let’s throw some lists in to close things out. Here are the 10 hitters whose WAR would increase by the most using RE24 in place of wRC+:
And here are the 10 whose WAR would decline by the most:
And here are the top 10 overall players in terms of RE24-WAR:
Lastly, as befits one of my janky statistical looks, here’s a complete leaderboard as of the games of August 25. You can do this on your own with a bit of Excel manipulation if you’d prefer to learn to fish, but hey, I might as well provide it for you in any case. Whatever your opinion on RE24 or any particular offensive statistic, I think that understanding how they work, and seeing how that theory looks when applied to the current season, is always worthwhile.
