MATCHDAY LEARN · BASEBALL

Tactics & decisions: baseball

MLB — what actually decides these games, and how the model reads it, including where it honestly falls short.

General education + model notesAnalytics only · not betting advice

The long season

162 games is a lot of baseball. It's long enough that randomness washes out over a full season in a way it doesn't in a 16-game NFL slate — but it also means a team's April record can be almost meaningless. A hot two-week stretch to open the season looks identical in the standings to a hot two-week stretch in July, but it carries far less information about who the team actually is.

How the model reads thisMatchday explicitly flags prediction confidence as "early" versus "established" based on how many games a team has actually played, damping confidence during small-sample stretches rather than treating six games the same as sixty.

Run differential over time

A team's win-loss record can lag its true quality — a club that's 24-20 but has been outscored on the season is usually a worse bet going forward than a 22-22 team that's outscoring opponents comfortably. Run differential, tracked over enough games, tends to predict future performance better than the record alone, because close one-run games have a real element of luck that evens out slower than blowouts do.

How the model reads thisScoring margin — runs for minus runs against, per game — feeds the model directly, and the model's own opponent-adjusted rating layers division and interleague schedule strength on top, since not every team's 60 games so far have come against the same quality of opposition.

What the model honestly doesn't see

Baseball is a pitching matchup sport in a way few others are — a team's record against left-handed starters can look completely different from its record against right-handers, and a single ace on the mound can flip a game's expected outcome more than almost any other single factor in team sports. That's genuinely hard to model well without individual pitcher data.

How the model reads thisMatchday's MLB model does not currently track individual starting pitchers — it reads team-level results, not who's on the mound that day. That's a real, honest limitation worth knowing about, not something the model quietly papers over. Team-level scoring margin and the opponent-adjusted rating are still meaningful signals; they just don't know it's Tuesday's fifth starter instead of the ace.

Home advantage is smaller here

Home-field advantage in baseball is real — travel, sleeping in your own bed, a groundskeeper who knows how to set up the infield dirt for your team's style — but it's smaller than in most other major sports. There's no crowd noise disrupting play calls the way there is in football, and pitching mostly neutralizes ballpark familiarity except in a handful of parks with unusual dimensions.

How the model reads thisMLB's home-field weight is set lower than football's or basketball's, reflecting how much less the venue itself tends to swing baseball outcomes.

See also: the Q&A page for what Elo, edge, and confidence mean generally, and the Content hub for weekly model recaps.