The order of the hits

A double with the bases empty and two out is worth about 0.2 runs more than an average plate appearance; the same double with the bases loaded is worth about 2.5.¹ Nothing about the swing differs. What differs is who was on base, and that depends on what the hitters before it did in the same inning, before a third out could clear the bases.

That is why the same hits score more when they come together. A hit's value depends on the runners it finds, and the third out erases them, so outs in between are not neutral: they can throw away the work of the hits before them. The run expectancy table shows the stakes. In FanGraphs' example, for a league scoring 4.15 runs a game, a team with nobody on and nobody out can expect about 0.46 runs before the inning ends; with the bases loaded and nobody out, about 2.28.¹ Clustering decides which rows of that table a team gets to visit.

Measuring the order

Seeing what order is worth over a season needs an estimate of what a team's events would have produced in a neutral order. Base Runs, devised by David Smyth, provides one: it counts baserunners, how far events advance them, outs and home runs, and estimates runs as baserunners times the share of them that score, plus home runs.² FanGraphs calls it "a context-neutral run estimator" and prefers "sequencing" to "luck" for the difference between actual and estimated runs: "It's up to the reader to decide if the team is fortunate or doing something that BaseRuns isn't measuring."³

Computed from official MLB totals for every team from 2001 to 2026, leaving out the 60-game 2020 season, with Base Runs calibrated each year so that the league's estimate equals the league's runs, that difference can be measured for 750 team-seasons.⁴ It had a standard deviation of 21 runs.⁴ A third of team-seasons missed their estimate by 20 runs or more in one direction or the other, and about one in seven by 30 or more.⁴ At nine or ten runs to a win, 21 runs is about two wins.⁵ The largest gap belonged to the 2013 St. Louis Cardinals, who scored 783 runs from events worth 693 by this estimate.⁴ Ninety runs is roughly nine wins.

Does it come back?

If clustering were mostly a skill, the teams that cluster well should keep doing it. Across 690 pairs of consecutive seasons for the same club, the correlation between one year's gap and the next was 0.17.⁴ The gap kept its sign 53.5% of the time, barely more often than a coin toss.⁴ After the 101 seasons with a gap of 30 runs or more, which averaged about 41, the next season's gap in the same direction averaged about 9.⁴

Roughly three quarters of an extreme season's clustering, then, did not come back. The quarter that did could be a real skill such as hitting differently with runners on, a batting order, a ballpark, or a kind of team the estimator handles poorly. Season totals cannot tell those apart.

Within an inning, order is large and real. Over a season, the same arithmetic hands out runs according to who happened to come up when, about two wins either way for a typical team, and it hands them to the same team twice not much more often than chance would. A team that scored 40 runs more than its events were worth should expect, on this record, to keep about 9 of them next year.