Catcher framing is a model output
Pitch framing is the great success story of public baseball research. A skill that scouts had described in vague terms for a century turned out to be measurable, large, and badly mispriced by the market — the best framers were worth wins that no existing statistic captured, and front offices that noticed early bought catchers cheaply for several seasons. All of that is true. It is also true that framing runs are not a measurement in the way that a home run or a stolen base is a measurement. They are the output of a model, and the number a catcher carries depends on choices made by whoever built it.
What the model actually does
The core method is straightforward. Take every taken pitch, record its location as it crosses the plate, and build a probability surface: given this location, how often is this pitch called a strike? Then compare each catcher’s actual called strikes to what the surface predicted, and convert the difference into runs using the run value of a ball versus a strike. The catcher who converts borderline pitches into strikes above expectation is credited; the one who loses them is debited. Nothing about that is observed directly. Every framing run is a residual.
The controls decide the answer
Early public models used location alone. Better ones add pitch type, handedness of batter and pitcher, and the count, because all of them move the strike probability substantially — the zone an umpire calls on 3–0 is not the zone he calls on 0–2. Add the umpire and you are controlling for the single largest source of variation. Add the pitcher and you start stripping out the credit that belongs to command. Each control is defensible, each changes the leaderboard, and different public and private models make different choices. Two framing figures for the same catcher in the same season can differ by a win.
The run-value conversion carries its own assumptions. Turning extra strikes into runs requires aa value for a ball-to-strike swing, and that value depends on the count it occurred in, the run environment of the season, and the base-out state. Models that apply a flat average per call will diverge from models that price each situation, and the gap compounds across the thousands of borderline pitches a full-time catcher receives.
Credit the catcher is not owed
A framing model attributes the residual to the receiver by construction, but at least three other parties shaped it. The pitcher who hits his spots gives his catcher easy receptions; the one who misses by a foot forces a reach that no technique saves. The umpire brings a personal zone that is stable enough to be modeled. And the batter influences the call by his own reaction, since umpires are measurably affected by whether a hitter flinches. The catcher is the only one of the four who gets the number.
Sample size behaves unusually here, and in the catcher’s favor. A full-time receiver sees thousands of taken pitches a season, which is why framing stabilizes faster than almost any other defensive measurement in baseball — one of the genuine arguments for trusting it. But the stability is in the estimate, not in the modeling choices behind it. A number can be precisely estimated and still be answering a question the model specified slightly wrong.
Measurement error in the location itself
Tracking systems report a pitch location with error, and the error is not uniform across the zone or across parks. Since framing value is concentrated almost entirely in a thin shell of borderline pitches, small location errors land exactly where the model is most sensitive. System changes between eras of tracking hardware produce discontinuities in framing leaderboards that have nothing to do with catchers changing their technique.
Pitcher attribution is the thorniest of the four. A staff with exceptional command will make every catcher who works with it look competent, and framing models that do not control for the pitcher hand that credit entirely to the receiver. Models that do control for the pitcher face the opposite problem: catchers and pitchers are not randomly paired, so controlling for one quietly removes real catcher influence on where the pitcher aims. There is no clean solution, only a choice about which bias to accept.
The skill is real but it is shrinking
Two forces have compressed framing value since its discovery. Teams priced it, so bad framers stopped catching — the worst receivers left the position, which narrows the spread between best and worst. And the strike zone has been increasingly scrutinized, with umpire evaluation systems rewarding accuracy, which reduces the room a catcher has to move a call. A skill worth two wins when nobody was looking is worth less once everyone is, and that is the normal life cycle of a market inefficiency rather than evidence the research was wrong.
Catchers have adapted in ways that complicate year-over-year comparison as well. The one-knee receiving stance spread rapidly across the sport because it improves presentation at the bottom of the zone, at some cost to blocking and throwing. A catcher whose framing figure jumped after a stance change did something real, but the model cannot distinguish that improvement from a season spent catching a staff that happened to live at the knees.
Automated zones end the argument
The logical terminus is a strike zone called by a machine, in whatever hybrid or challenge-based form it eventually takes. To the extent the zone is automated, framing value goes to zero by definition, because there is no longer a human judgment to influence. That makes framing the rare statistic with a scheduled expiration date, and it makes long-term contracts justified primarily by framing a genuinely risky bet on a rulebook.
The framing debate also crowded out the rest of catcher defense, which is awkward because the other components are easier to observe. Blocking pitches in the dirt, controlling the running game, and handling a staff are all real duties with measurable consequences, and they have smaller run values than framing in most published systems. That ordering is itself a modeling result, not an observation, and it deserves the same skepticism as the framing figure that produced it.
How to read it on StatLine
StatLine’s MLB tables carry the offensive and pitching lines rather than catcher defense, which is the right place to start a framing conversation anyway. Check what a catcher’s staff does with and without him in terms of plain run prevention, and compare his bat to the position baseline, because framing estimates are frequently used to argue that a weak-hitting catcher is secretly valuable. That argument may be right. It rests on a model, and the model should be named rather than assumed.
The honest read
Framing is a real skill, it was genuinely undervalued, and the research that uncovered it is some of the best public work baseball analysis has produced. It is still an estimate produced by a model whose controls are contested, built on location data with error concentrated exactly where the estimate is most sensitive, attributing to the catcher a residual that the pitcher, hitter, and umpire all helped create. Treat framing runs as a well-supported inference rather than a count, expect different sources to disagree, and remember that the skill is being legislated out of existence.