Every half-inning of baseball is a position you can describe in two numbers: which bases are occupied, and how many outs there are. There are eight ways to arrange the runners (from empty bases to bases loaded) and three out totals, zero, one, and two. Multiply them and you get exactly twenty-four distinct situations. Baseball calls them base-out states, and the entire modern understanding of in-game value is built on a single table that puts a number next to each one.
That number is run expectancy: given a base-out state, how many runs does the average team go on to score from that point to the end of the inning? Answer it for all twenty-four states and you have the run-expectancy matrix: the quiet, unglamorous backbone underneath sacrifice-bunt math, stolen-base break-evens, and a play-by-play hitting stat called RE24. It is, in a real sense, the periodic table of run scoring.
The twenty-four base-out states
Start with the bookkeeping, because the structure is the whole point. The bases can be empty; a runner can stand on first, second, or third alone; runners can occupy any pair (first-and-second, first-and-third, second-and-third); or the bases can be loaded. That is eight configurations. Each of those can occur with nobody out, one out, or two outs. Eight times three is twenty-four, and every legal moment of an inning, before the third out ends it, lives in exactly one of those cells.
The trick is that we don’t guess what each cell is worth. We measure it. Take years of recorded play-by-play, find every time a team was in, say, a runner-on-second, one-out state, and average how many runs that team scored for the rest of that inning. Do it for all twenty-four states and the matrix fills itself in from history rather than intuition. The values shift a little with the run environment (a high-offense season pushes every cell up, a pitching-dominated one drags them down) but the shape stays remarkably stable.
Reading a representative matrix
Here is a small, deliberately rounded version of the matrix. Treat these as representative figures for a typical modern run environment: the exact decimals vary by season and by source, and you should never quote them as a specific year’s official numbers. The point is the pattern, not the third decimal place. Our own measured matrices for 2014, 2019 and 2025 follow in the next section.
Runner on first: 0 out: ~0.85 1 out: ~0.50 2 out: ~0.22
Runner on second: 0 out: ~1.10 1 out: ~0.65 2 out: ~0.32
Bases loaded: 0 out: ~2.30 1 out: ~1.55 2 out: ~0.75
Two regularities jump out, and they govern everything that follows. First, reading across any row, run expectancy falls hard as outs pile up: outs are the scarcest, most precious resource in the inning, and burning one roughly halves what you can expect to score. Second, reading down any column, advancing runners raises the number, but never by as much as an out costs. A runner on second with nobody out (about 1.10) is worth more than a runner on first with nobody out (about 0.85), but the gap is far smaller than the cliff you fall off by recording an out.
Our run-expectancy matrices, measured
We computed the matrix ourselves from Retrosheet’s play-by-play event files. For every plate appearance that began in the first eight innings, we took the base-out state when the batter stepped in and counted the runs his team scored from then to the end of the half-inning, then averaged by state. Half-innings cut short before a third out (a called game) are left out, and so is the ninth inning and later, where walk-offs cut innings short and extra innings start with a runner on second. The 2025 matrix rests on 165,364 plate appearances:
| Runners | 0 out | 1 out | 2 out |
|---|---|---|---|
| Bases empty | 0.50 | 0.27 | 0.11 |
| Runner on 1st | 0.90 | 0.52 | 0.23 |
| Runner on 2nd | 1.15 | 0.67 | 0.33 |
| Runner on 3rd | 1.24 | 0.90 | 0.34 |
| 1st & 2nd | 1.61 | 0.96 | 0.46 |
| 1st & 3rd | 1.85 | 1.24 | 0.48 |
| 2nd & 3rd | 2.01 | 1.42 | 0.62 |
| Bases loaded | 2.52 | 1.61 | 0.76 |
The two regularities described above hold in the measured table. Every row falls as outs pile up, and the out costs more than the base: a runner on first with nobody out was worth 0.90 runs in 2025, a runner on second with one out (what a successful sacrifice buys) 0.67. The same matrix moves with the run environment. 2014 was the lowest-scoring season from 2000 to 2025 (4.07 runs per team per game), 2019 had the most home runs of the span (6,776), and 2025 sits between them (4.45 runs a game):
| State or decision | 2014 | 2019 | 2025 |
|---|---|---|---|
| Bases empty, 0 out | 0.46 | 0.55 | 0.50 |
| Runner on 1st, 0 out | 0.83 | 0.96 | 0.90 |
| Runner on 2nd, 0 out | 1.07 | 1.19 | 1.15 |
| Runner on 2nd, 1 out | 0.63 | 0.74 | 0.67 |
| Runner on 3rd, 1 out | 0.93 | 1.00 | 0.90 |
| Bases loaded, 0 out | 2.37 | 2.37 | 2.52 |
| Steal of second breaks even at, 0 out | 71.6% | 74.8% | 71.6% |
| Steal of second breaks even at, 1 out | 72.7% | 73.2% | 73.4% |
| Steal of second breaks even at, 2 out | 64.8% | 69.7% | 69.0% |
| Successful sacrifice, runner on 1st, 0 out (runs) | −0.20 | −0.23 | −0.23 |
The start of an inning (bases empty, nobody out) was worth 0.46 runs in 2014 and 0.55 in 2019, and the scale shift runs through every cell. The verdicts it drives barely move. The successful sacrifice cost between 0.20 and 0.23 of a run in all three seasons, and a steal of second with fewer than two outs needed to work about 71.6% to 74.8% of the time to break even.
The extra-inning runner is worth slightly less
Since 2023 every extra half-inning has started with a runner on second and nobody out, the same cell as a runner on second with nobody out in a regular inning. They do not play the same. In 880 visitors’ half-innings from the 10th on in 2023–2025 that started with the automatic runner, teams scored 1.12 runs on average and scored at least once 56.7% of the time. When a team reached the same state in innings 1–8 of those seasons (6,625 plate appearances), it scored 1.14 runs on average and at least once 61.4% of the time. The average runs differ by less than the sampling error (±0.05), but the 4.7-point gap in scoring at least once is about 2.6 standard errors. This measurement does not say why; the states are reached differently (a leadoff double against the starter, versus a free runner against a late-inning reliever), and that is the obvious place to look. Home half-innings are left out because a walk-off ends them early.
How we measured it
Retrosheet’s regular-season event files for 2014, 2019 and 2023–2025 (www.retrosheet.org/events/), read with our own play-by-play parser, which tracks the bases, outs and runs through every play. As a check, the parser’s output for 2010–2025 matches Retrosheet’s separate game logs exactly: every final score and the season totals of at-bats, hits, doubles, triples, home runs, walks, intentional walks, hit batters, sacrifices, strikeouts and catcher’s interference. The information used here was obtained free of charge from and is copyrighted by Retrosheet. Interested parties may contact Retrosheet at www.retrosheet.org.
RE24: crediting a play by what it changed
Once you have the matrix, you can value any single play with one subtraction. RE24 (the “24” is a nod to the twenty-four states) scores a play as the run expectancy of the state it created, minus the run expectancy of the state it started in, plus any runs that actually crossed the plate during the play. In words: a play is worth how much it improved your run-scoring outlook, plus whatever it banked immediately.
The beauty of the construction is that it credits the things batting average and even wOBA ignore: a walk with the bases loaded forces in a run and is worth a great deal; the same walk with two outs and nobody on is worth a sliver. RE24 sees the difference because it knows the state. Summed over a season, a hitter’s RE24 tells you how many runs better or worse than average he was, given the exact situations he actually batted in.
A worked example
Make it concrete with a clearly hypothetical play, using the representative numbers above. Suppose a hitter comes up with a runner on first and nobody out: a state worth roughly 0.85 expected runs. He doubles into the gap; the runner on first comes around to score, and the batter ends up on second. The new state is a runner on second, nobody out, worth about 1.10, and one run has already scored.
Run the arithmetic: RE24 equals (1.10 − 0.85) plus the 1 run that scored, which lands at about +1.25 runs of credit for that one swing. That is an enormous number for a single play, which is exactly right, because a run-scoring double with nobody out is one of the best things a hitter can do. Now flip it. Suppose instead the hitter, with that same runner on first and nobody out, lays down a textbook sacrifice bunt: the runner moves to second, the batter is out, and the state becomes a runner on second with one out, worth about 0.65. RE24 is (0.65 − 0.85) plus zero runs, roughly −0.20. The “successful” sacrifice handed back two-tenths of a run.
Why the sac bunt usually loses
That negative number is not a fluke of the example; it is the general rule, and it is the whole reason the sacrifice bunt nearly went extinct. The bunt trades a high-expectancy state (runner on first, nobody out) for a lower one (runner on second, one out) because the out it spends costs more than the base it buys. The run-expectancy matrix turned a century of accepted strategy into an arithmetic mistake, visible the moment you put the two cells side by side. Stolen bases get the same treatment: a steal raises expectancy modestly, a caught stealing torches it, and the break-even success rate falls straight out of the matrix.
RE24 is what happens when you apply that logic to every event in a game rather than just bunts, building a player’s seasonal value one base-out subtraction at a time.
RE24 versus context-neutral stats and WPA
It helps to place RE24 between two neighbors. On one side sit context-neutral stats like wOBA, which assign every double the same value no matter when it happened: cleaner for measuring a hitter’s underlying skill, precisely because they strip out the situation. RE24 deliberately keeps the situation: it rewards the bases-loaded double over the bases-empty one. That makes it a better account of what a player’s plate appearances were actually worth, and a worse predictor of future performance, since being handed high-leverage spots is not a repeatable skill.
On the other side sits Win Probability Added. RE24 values plays in the currency of runs; WPA values them in the currency of wins. A two-run double in a 12–1 blowout has a healthy positive RE24 (runs are runs) but a trivial WPA, because the game was already decided. The matrix doesn’t know the score or the inning; it only knows the base-out state. That is its strength for measuring run production and its blind spot for measuring drama.
The bottom line
The run-expectancy matrix is twenty-four numbers that quietly answer most of baseball’s strategic questions. RE24 is what you get when you let those numbers grade the game one play at a time: every event is worth the change in expectancy it produced, plus the runs it scored. It is why the bunt lost its job, why a bases-loaded walk is a real event and a meaningless-spot walk nearly isn’t, and why “how many runs is this worth?” finally has a defensible answer. Learn to read the matrix and you are reading the language the rest of the sport’s strategy is written in.
Sources & Further Reading
- Retrosheet: regular-season event files for 2014, 2019 and 2023–2025, the source of the matrices above. The information used here was obtained free of charge from and is copyrighted by Retrosheet. Interested parties may contact Retrosheet at www.retrosheet.org.
- FanGraphs: RE24 leaderboards and the run-expectancy framework behind them.
- Tom Tango, Mitchel Lichtman & Andrew Dolphin, The Book: Playing the Percentages in Baseball, the canonical treatment of base-out states and run values.
- SABR: historical and methodological writing on run expectancy.