The Radar Gun Knows Before the ERA Does. Some Numbers Predict; Most Just Revert.
A companion issue argued that Paul Skenes isn’t really declining — his ugly ERA and .500 record are mostly noise. A sharp reader pushed back with one word: velocity. It is the perfect objection, because it opens the single most useful door in statistics. Some numbers become trustworthy almost the moment you begin counting. Others never do. Telling them apart is the whole difference between seeing a decline coming and being fooled by a mirage.
The question: can you see a decline coming before the results show it?
The answer: yes, but only from the top of the ladder. Strikeout rate settles after about 70 batters. Walk rate needs about 170. ERA needs a season and a half. Fastball velocity is trustworthy in a single start.
Why that is surprising. The numbers we quote most — ERA, wins, batting average — are the slowest to mean anything. The ones that warn us early are the ones nobody puts on the broadcast.
Every statistic is a mixture of two things: the player’s real skill, and the luck of the particular sample we happened to watch.
The ladder of trust
But the number of hits that fall in against him — his batting average on balls in play — does not settle for years. Notice what that means: the stat we quote fastest is the one we should trust last.
Why velocity wins the argument
Now put the reader’s objection where it belongs on that ladder, right at the top. Watch what velocity does that ERA cannot.
So we predict from the top of the ladder, using the numbers already made mostly of skill. We do not predict from the bottom — not from ERA, not from won-lost, not from the hits that happened to fall. Those are mostly a record of what already happened to us.
The numbers that revert can’t warn you, and the numbers that warn you don’t revert. To see a decline coming, read only the second kind.
The ProfessorWhat to take home: the rule that travels
Strip the baseball off and we are left with a habit of mind that works on any number anyone waves at us. Before you believe a statistic, ask how many chances built it.
Notes & sources
The stabilization points are from the sabermetric work of Russell Carleton, popularized by FanGraphs and Baseball Prospectus: the sample at which a statistic correlates about 0.7 with a separate, equally-sized sample of the same player — roughly the point at which half its variation is repeatable skill rather than luck. For pitchers (measured in batters faced or balls in play): strikeout rate about 70, walk rate about 170, batting average against about 630, home-run rate about 1,320, and BABIP against about 2,000 — the last effectively unreachable within a season. Exact values vary slightly by study and metric definition.
Velocity is not one of Carleton’s plate-appearance stabilization stats because it is measured per pitch, not per outcome; it is shown here at roughly one start to place it on the same axis, but its reliability is higher still — a pitcher’s average fastball velocity is trustworthy within a few dozen pitches, faster than any results-based number. The Skenes velocity figures (four-seam averages of about 98.8, 98.2 and 96.9 mph across 2024–2026) are from MLB Stats API pitch-arsenal data, and were the subject of the companion issue on his 2026 season.
This is a follow-up to the previous issue on Paul Skenes, prompted by a reader’s question. It rests on the primers for Sample Size (Concept No. 6) and Regression to the Mean (Concept No. 3). The archive is at thesportspage.net.
