Splits, and When They Are Real

A full season is already a modest sample. Cut it by handedness, then by home and road, then by month, and each cell holds a fraction of the original.

A hitter with 500 plate appearances who is described as "hitting .340 against left-handed pitching" may have 120 plate appearances in that split. From What Sample Size Does to a Batting Average, you already know how wide the interval around that is — wide enough to contain both stardom and mediocrity.

Not all splits are equally suspect. A split is more likely to be real when there is a physical mechanism behind it, and when it stabilises quickly:

• Platoon splits (batter versus opposite-handed pitching) have a genuine mechanism — pitch movement away from versus into the hitter — and are among the more reliable splits, though individual estimates still need heavy regression toward the league-wide platoon effect.
• Ground-ball and fly-ball tendencies stabilise relatively quickly and describe how a player actually plays.
• Home and road splits mostly reflect the park, which you can look up directly instead.
• Month-by-month, day-versus-night, and versus-a-specific-team splits are usually noise. There is rarely a mechanism, and the samples are tiny.

Before believing a split, ask:

• Is there a mechanism? Can you say, in one sentence, why this would be true of this player specifically? "Left-handed sliders move away from him" is a mechanism. "He is a good June hitter" is not.
• How many chances are in the split? Not the season total — the cell.

If a split has a mechanism and a decent sample, take it seriously and still regress it toward the league average for that split type. If it has neither, it is a coincidence that got a name.