PNC Park Fit
Savant park factors by batting side, shrunk for noise, applied to each player’s outcome rates over half his games, with a spray-aware home-run factor.
What it predicts
Runs per 600 plate appearances (hitters) or batters faced (pitchers) at PNC against an average park, and against the parks he played in.
Inputs
- Baseball Savant park factors, 2024–26 rolling, by batting side
- Statcast event counts at each park (the noise in each index) and every batted ball’s exit velocity, launch angle and spray angle
- The Lineup Lab’s outcome rates by pitcher hand; outcome rates allowed by batting side for pitchers
- Each player’s home parks, 2024–26 weighted 3/4/5; the Winter Plan’s 2027 roster and branches (decisions feed)
- Outside players: the Roster Builder’s candidates (hitters and pitchers) and every Free-Agent Board player, at his projected 2027 playing time
Method
- Each index shrunk toward 100: noise from its event count at the park and the road comparison sample; true spread across parks = observed spread minus noise
- Neutral rate = observed ÷ (½ × home factor + ½); PNC effect = ½ × neutral × (PNC factor − 1) × wOBA weight, by batting side against each pitcher hand
- Home runs: park home-run probability by exit velocity × launch angle × pull angle, with a structured prior on the log-odds scale (each park’s region, pull direction × launch angle, toward the league; each bin toward its region), prior strengths chosen on held-out 2026 balls. Shrinking every bin straight toward the league erased most of PNC’s own level (0.94 against 0.78 unshrunk for right-handed hitters)
- A hitter’s adjustment is how his balls fare relative to the average same-side hitter’s, shrunk toward 1 by its split-half reliability (his own noise) and by the share of its consistent spread that is not park noise (a parametric bootstrap of every park’s balls)
- Team: hitters’ effect at plan plate appearances minus the staff’s at plan innings; an average team nets about zero because opponents play the same park. Plate appearances and innings the plan leaves to replacement players count as average (about zero)
- Park wins for one player: runs against an average park at his projected 2027 plate appearances or innings ÷ the depth chart’s runs per win (for a pitcher, the runs he allows count against the team), shown beside him on the Park page, the Roster Builder and the Free-Agent Board; not added to any page’s wins, which use park-adjusted WAR
- Test: hitters who changed home parks, predicted vs. actual change in home-minus-road wOBA, side-level and spray-aware against a no-park guess
How well it works
Computed by the pipeline from the data on this site
- Backtest: 38 hitters who changed home parks (2024→25, 2025→26; 150+ PA at home and on the road both seasons), predicting the change in his home-minus-road wOBA
- Error (RMSE) no park adjustment .074 · side-level factors .070 · spray-aware fit .070 Correlation with the actual change side-level r = .41 · spray-aware r = .42 The actual changes spread .074 (SD) and the predictions only .012: the park explains a small, real part of a noisy home-road gap. The spray-aware fit and the side-level factors are 0.2 points of wOBA apart, far inside what 38 hitters can detect, so this test can’t judge the spray model. The factors cover the same seasons, so it is not fully out of sample.
- Spray-aware park fit (where each batted ball went, by park and batter side)
- Fit gain log-likelihood +208 over no park effect, against +96 for the earlier model that shrank every bin straight toward the league (prior: 10 expected home runs toward the league’s level by region, 40 balls toward the region by bin) A hitter’s PNC adjustment, odd vs. even games r = .66 (391 hitter-sides); 89% of its consistent spread is not park noise PNC home-run rate vs. league (spray model) left-handed batters ×0.91 · right-handed batters ×0.82 (×1.00 = average); Savant’s home-run index LHB 98, RHB 70
Data: Baseball Savant park factors, 2024–26 rolling, by batting side; every Statcast batted ball 2024–26 with its spray angle.
What it can't see
- The factors and the park-change test cover the same seasons, so the test is not fully out of sample.
- FanGraphs WAR adjusts with one factor for all hitters; the add-on is first-order.
- Strikeouts, outs and weather are not modeled; pitchers get side-level factors only.
- Tampa Bay’s 2025 season and the Athletics’ parks have no Savant factor and count as average.
How we'd measure impact
- Home-minus-road wOBA of hitters who join or leave Pittsburgh
- Whether park fit ever changes a Winter Plan call