Explore hits per pitch as a team metric

[Mattsface]

Goal

Explore whether hits, at-bats, and pitches seen can produce a useful descriptive team metric.

Starting point

Consider:

Hits per 100 pitches = hits / pitches_seen * 100

Also compare it with batting average and pitches per at-bat so we understand what information is actually added.

Questions

• Is the metric meaningful enough to chart?
• What does it measure well, and what does it miss?
• Does pitch count need plate appearances or walks to become more useful?
• Should it stay descriptive rather than be called an efficiency or quality metric?

Out of scope

No implementation yet. This issue is for defining and validating the statistic first.

[Mattsface]

Hits-per-pitch metric exploration (issue #32)

Goal: decide whether hits / pitches_seen * 100 ("hits per 100 pitches") is
worth adding as a team chart, per issue #32. This is a validation-only spike —
no implementation — answering the issue's questions with real data before any
schema or chart work is proposed.

All numbers below come from running python-mlb-statsapi 0.8.0 against the
live MLB Stats API for the completed 2025 regular season (three teams, 162
games each): Seattle Mariners (136), New York Yankees (147), Los Angeles
Angels (108). Per-game hits, at_bats, number_of_pitches, and
plate_appearances all come from the same gameLog/hitting split already
used by app/services/team_game_logs.py — no new MLB endpoint is needed to
compute this metric.

1. The data isn't persisted today

TeamGameBattingLineRecord (app/database/models.py) stores only hits,
runs, and strikeouts. _normalize_batting_line in
team_game_logs.py never reads split.stat.at_bats or
split.stat.number_of_pitches, even though both are present on
SimpleHittingSplit. Computing this metric for real would require a schema
migration (two new columns) and a change to normalization — not just a new
app/analytics/*.py module. The numbers in this doc were pulled with a
one-off script calling mlbstatsapi directly, bypassing the app entirely.

2. Hits-per-100-pitches is not new information — it's AVG rescaled

Algebraically:

hits / pitches * 100 = (hits / at_bats) * (at_bats / pitches) * 100
                      = AVG * (100 / pitches_per_at_bat)

Measured season totals confirm this exactly:

Team	AVG	Pitches/AB	Hits/100 pitches	AVG / (Pitches/AB) × 100
Mariners	.244	4.43	5.52	5.52
Yankees	.251	4.52	5.54	5.54
Angels	.225	4.42	5.09	5.09

The two right columns match to the decimal in every case, because they are
the same computation. Charting hits-per-100-pitches next to AVG would show
two lines that move together almost in lockstep — confirmed at the game
level too: corr(hits/100p, AVG) across a season's games is 0.94–0.95
for all three teams. Pitches-per-AB is the only source of any daylight
between the two, and it barely moves (game-level stdev ~0.51 pitches, on a
mean of ~4.4 — about 11% relative variation, versus AVG's own ~30% relative
game-to-game variation). AVG dominates the ratio.

3. It doesn't separate teams any better than AVG already does

Team	AVG	Hits/100 pitches
Angels (worst)	.225	5.09
Mariners (mid)	.244	5.52
Yankees (best)	.251	5.54

Same ranking, similar relative spread (AVG: .225→.251 is +11.6%; hits/100p:
5.09→5.54 is +8.8%, slightly compressed relative to AVG because
pitches-per-AB is mildly anti-correlated with hit rate at the game level
(corr(hits/100p, pitches/AB) is -0.30 to -0.48). So beyond a small
damping effect, a chart of this metric across teams would reorder nothing
and reveal nothing AVG doesn't already show.

4. Where it actively misleads: it counts plate discipline as waste

number_of_pitches is pitches across every plate appearance, including ones
that end in a walk or hit-by-pitch. at_bats (and therefore hits)
excludes those outcomes by design — the same reason AVG excludes walks from
its denominator, so that working a full count and drawing a walk isn't
scored as a failure to hit.

hits / pitches re-introduces exactly the problem AVG's definition avoids:
a team that takes more pitches to draw walks sees its pitch count rise with
no matching rise in hits, dragging the ratio down — even though drawing a
walk is a good outcome, not an inefficiency. A metric that punishes plate
discipline shouldn't be labeled "efficiency" or "quality"; doing so would
contradict the standard sabermetric view (why OBP is preferred over AVG, and
why pitches-per-PA is tracked as its own plate-discipline signal rather than
blended into a contact-rate stat).

5. Answers to the issue's questions

• Is the metric meaningful enough to chart? No. It's a deterministic
  rescaling of AVG by pitches-per-AB (exact algebraic identity, confirmed
  against real 2025 data), correlates 0.94+ with AVG at the game level, and
  produces the same team ranking AVG already gives.
• What does it measure well, and what does it miss? It measures almost
  nothing AVG doesn't. What it adds — sensitivity to pitches-per-AB — is a
  small damping effect, and it adds that by penalizing plate discipline
  (walks, long counts) rather than isolating contact quality.
• Does pitch count need PA or walks to become more useful? Yes. If the
  goal is a genuine new plate-discipline signal, pitches-per-plate-appearance
  (not hits-per-pitch) is the metric to chart: it's computed from the same
  per-game fields, is not a rescaling of AVG (across these three teams it
  shows no consistent relationship with AVG — the Angels, the worst hitting
  team, has essentially the same pitches/PA as the Yankees, the best), and
  is a real, independently-tracked sabermetric concept on its own.
• Should it stay descriptive rather than be called an efficiency or
  quality metric? If it's built at all, yes — but the stronger
  recommendation is not to chart hits-per-100-pitches as proposed. If there's
  appetite for a new metric from pitches_seen, chart pitches-per-PA
  instead, framed explicitly as a plate-discipline stat, not an efficiency
  or contact-quality stat.

Recommendation

Don't implement hits-per-100-pitches as a team chart. It duplicates AVG
under a different label and mislabels plate discipline as inefficiency.
Pitches-per-plate-appearance is the metric worth a follow-up issue if a
plate-discipline chart is wanted — but that requires the same schema work
noted in §1 (persisting at_bats and number_of_pitches/plate_appearances
per game), which today's model doesn't carry.

[Mattsface]

Okay closing this