Funnel plot, year-to-year correlation
ERA minus FIP: luck, skill, or the fielders?
The gap between what a pitcher allowed and what his strikeouts, walks and home runs say he should have. A funnel shows how much of it is noise; last season shows how much carries over.
Computed on request from the record book (Oct 13, 2026) and 720 finals in the game log.
The finding
Among the 185 pitchers with 20 or more innings this season, ERA minus FIP ranges from −3.64 to +4.50 runs per nine. The funnel below shows the spread narrowing as innings pile up, at about the rate chance alone would produce — the widest gaps belong to pitchers with the fewest innings.
Last season is the sharper test. For the 67 pitchers who threw 40+ innings in both years, last year's gap predicts this year's at r = 0.14 (95% −0.10–0.37) — weak. A pitcher who beat his FIP last year is not reliably any more likely to do it again. Last year's FIP predicts this year's ERA at r = 0.25, last year's ERA at r = 0.17; re-centred for the change in league scoring, FIP misses this year's ERA by 1.27 runs per nine on average and ERA by 1.49. FIP is the better forecast, as the theory says, though with this many pitchers the two intervals overlap.
Gap repeats at
0.14
−0.10–0.37
FIP → next ERA
0.25
0.02–0.47
ERA → next ERA
0.17
−0.08–0.39
Pitchers compared
67
40+ IP both years
Small sample. 67 pitchers in both seasons. A correlation from this many has an interval about ±0.24 wide near zero.
The funnel: ERA minus FIP against innings
If the gap were pure noise, its spread would shrink with the square root of innings. The shaded bands are that shape (one and two standard deviations), scaled to the league as a whole.
What carries from one season to the next
Correlation between last season's number and this season's, for the same pitchers, with 95% intervals.
Method
FIP = (13·HR + 3·(BB + HBP) − 2·K) ÷ IP + C
FIP is computed the same way as on the player pages, with the constant C solved from each season's own totals so that league FIP equals league ERA. The funnel's band assumes the gap's variance falls as c ÷ IP and sets c = 69.6 from all 185 pitchers, so it shows the shape noise would take rather than testing against an outside standard. The carry-over correlations match pitchers by their record-centre ID across the 2025 and 2026 snapshots; intervals use Fisher's z transformation, and the gap's correlation has p = 0.25.
ERA minus FIP is not all luck. It holds the defence behind the pitcher, his park, how the bullpen handled the runners he left on base, and any real skill at limiting hard contact — none of which the KBO's public data separates. The year-to-year comparison only includes pitchers who kept a job for two seasons, which leans toward the better ones.