Run expectancy and RE24

At any moment in an inning, exactly two things define the situation: which bases are occupied and how many outs there are. There are eight possible base configurations — empty, first, second, third, first-and-second, first-and-third, second-and-third, loaded — and three out counts, zero through two. Eight times three gives 24 base-out states, and every moment of every inning is one of them.

Run expectancy is the average number of runs that score from that state to the end of the inning, measured across thousands of real innings. It doesn’t predict what will happen next; it tells you what the situation has historically been worth.

Lay all 24 values in a grid and you get the run expectancy matrix — one of the most quietly useful tables in baseball.

Read the numbers and several things about baseball fall out at once.

Outs are the scarcest resource. With the bases empty, expectancy falls from about 0.48 runs with nobody out to 0.10 with two out — an out costs more than most fans intuit. An inning has only three of them, and every one spent is a fifth of a run gone.

Advancing a runner is worth less than keeping the out. A runner on first with nobody out (0.86) is worth more than a runner on second with one out (0.66). That single comparison is the analytical case against the sacrifice bunt: trading an out to move a runner usually lowers the expected runs.

Bases loaded with nobody out is the peak at roughly 2.29 runs — and it drops to about 0.75 with two out. The same base configuration is worth three times as much depending only on the out count.

Once you can price every state, you can price every play. That statistic is RE24, sometimes called run expectancy 24 or REW.

The calculation is one line: take the run expectancy of the state after the play, add any runs that scored on it, and subtract the run expectancy of the state before.

RE24 = (RE after + runs scored) − RE before

A double with a runner on first and nobody out moves the state from 0.86 to about 1.96 (second and third, nobody out), for roughly +1.10. A strikeout with the bases empty and nobody out moves 0.48 to 0.25, for about −0.23. Sum a player’s RE24 across a season and you have a context-aware measure of how many runs he actually created or cost.

Unlike batting average or OPS, RE24 knows the difference between a single with the bases empty and a single with two on. That context is precisely what it was built to capture.

RE24 has an obvious sibling in win probability added, and the difference between them is worth being clear about.

RE24 measures runs. WPA measures wins. RE24 asks how much a play changed the expected runs in the inning; WPA asks how much it changed the team’s chance of winning the game.

The consequence is that RE24 is context-neutral about the game situation. A grand slam in the second inning of a blowout scores the same RE24 as one in a tie game in the ninth — both created about the same number of runs. WPA would value them completely differently, because only one changed who was likely to win.

Neither is more correct; they answer different questions. RE24 is the better measure of production. WPA is the better measure of impact on the outcome.

TwentySeven is built on the win-probability side of this family, but the same logic drives its play-by-play view: open any game and you see player impact rankings and the five highest-leverage plays — the moments where the base-out state and the score combined to matter most.

If run expectancy is the grammar of an inning, win probability is the grammar of a game. Start with how win probability works to see how the same historical-frequency method scales up from one inning to nine.