Analysis
FIP explained: the baseball stat that ignores your fielders
FIP explained for baseball fans: the exact formula, where the 13, 3 and 2 come from, how the season constant is set, and the three things it throws away.
By CricketTaken EditorialPublished Analysis20 min read
A pitcher finishes a season with an ERA of 3.10 and a FIP of 4.30. Half the internet says he was lucky. The other half says the number is nonsense because runs are runs. FIP explained properly settles almost none of that argument, but it does something better: it shows both sides exactly what they are disagreeing about, because unlike most modern baseball statistics this one has no black box in it at all.
Here it is, in full.
FIP = ((13 × HR) + (3 × (BB + HBP)) − (2 × K)) / IP + constant
That is the entire thing. Four event counts, three whole-number coefficients, a division by innings pitched, and a constant. No model, no regression, nothing hidden. It is arithmetic anybody can do in a notebook, and almost every argument about it turns out to be an argument about the four events it counts rather than about the maths.
The original version, as it was first written down, did not include hit batters. The version in general use adds them to the walks, on the sensible grounds that a hit batter and a walk put the same runner on the same base by the same non-fielding route.
What the coefficients are actually saying
The 13, the 3 and the 2 are not arbitrary and they are not tuning parameters. They are the relative run values of the three event types, scaled so the output lands on a runs-per-nine-innings basis. That scaling is why they are whole numbers rather than decimals: the run value of a single event is a fraction, and the formula reports on a nine-inning scale, so each fraction has been multiplied by roughly nine before being rounded to something you can hold in your head.
What matters for reading the statistic is the ratios between them, and those ratios are a statement about baseball.
A home run costs a bit more than four and a third walks. A strikeout saves about two thirds of what a walk costs. Sit with those two numbers for a moment, because they explain most of what FIP does to a pitcher's reputation.
They mean that FIP is savage about home runs. One ball over the fence wipes out more than six strikeouts. A pitcher who misses bats brilliantly and gives up home runs will be punished hard, which is the correct treatment and also the reason his FIP can swing violently from year to year on a small number of events.
They also mean that strikeouts are worth less than most people assume. A strikeout is an out. Most other things that happen with a bat on the ball are also outs, and the strikeout's advantage is only the difference between a certain out and a probable one. Two thirds of a walk is about right for that, and it is a lot less than the way strikeout totals get discussed suggests.
- 13Multiplier on a home run
- 3Multiplier on a walk or hit batter
- 2Multiplier on a strikeout
- 0Balls in play in the formula
Four coefficients, four event types, and one number that never appears. FIP is not park or league adjusted either; adjusted versions are published separately.
That last tile is the whole argument in one line. Every ground ball, line drive, pop-up and hard-hit fly out that stayed in the park is worth exactly nothing to this statistic, in either direction.
The constant is a scaling term, not a fudge
The constant at the end of the formula is the part people assume is arbitrary, and it is the most precisely defined element of the lot.
FIP constant = lgERA − (((13 × lgHR) + (3 × (lgBB + lgHBP)) − (2 × lgK)) / lgIP)
Read that carefully and you can see exactly what it is doing. It computes the raw FIP expression for the entire league, subtracts it from the league's actual ERA, and calls the difference the constant. Add that constant to every individual pitcher's raw expression and the league's average FIP comes out identical to the league's average ERA by construction.
The purpose is legibility. Without the constant, FIP would produce numbers on a scale that means nothing to anybody. With it, a FIP of 3.40 sits in the same mental space as an ERA of 3.40, and the two can be compared without conversion. That is the only job the constant has.
Two consequences follow, and both are useful.
The constant is recalculated for every season, which means it silently absorbs changes in the run environment. When scoring rises across the league, the constant moves, and a pitcher's FIP stays comparable to the ERAs around it. This is a genuine strength: the statistic self-calibrates to the era it is measuring, without anybody having to decide what the adjustment should be.
And because the constant is derived from league totals rather than chosen, it cannot be used to make a pitcher look better or worse. Every pitcher in a season gets the same number added. In recent run environments the figure has tended to land somewhere near 3.10, and the exact value for any given season is published by the organisations that compute it rather than being something to guess at.
- Take four numbers off the season lineHome runs allowed, walks issued, batters hit, and strikeouts recorded. Nothing else from the line is used, and nothing that happened to a ball in play appears anywhere.
- Weight themMultiply home runs by 13, walks plus hit batters by 3, and strikeouts by 2. These are relative run values scaled to a nine-inning basis, not tuning parameters.
- Add the costs, subtract the savingHome runs and free passes go in as positives. Strikeouts come out as a negative. The result is a raw number with no natural units.
- Divide by innings pitchedThis converts the total into a rate. Two pitchers with the same events over different workloads are now comparable, which the raw total could not do.
- Work out the season constantRun the same expression across the whole league, subtract it from the league's ERA, and keep the difference. It is the same for every pitcher in that season.
- Add the constantThe output now sits on the ERA scale by construction, so that the league's average FIP and the league's average ERA are the same number.
- Read it against his own leagueFIP is neither park nor league adjusted. A pitcher in a park that suppresses home runs will show a lower FIP for reasons that have nothing to do with him.
- Compare with his actual ERAThe gap between the two is everything the formula threw away: the fielders, the park, the order events happened in, and chance. It is a question to investigate, not an answer.
Every step is arithmetic on published counting statistics. The only input that is not on a standard box score is the season constant, which is derived from league totals.
A worked example, with invented pitchers
Two pitchers, both constructed for this article, both with 180 innings, chosen round so the arithmetic can be checked by hand. Neither is real and neither corresponds to anybody.
Pitcher A allows 20 home runs, issues 50 walks, hits 5 batters and strikes out 200.
Pitcher B allows 12 home runs, issues 80 walks, hits 5 batters and strikes out 120.
Run A through the formula. Home runs: 13 × 20 = 260. Free passes: 3 × (50 + 5) = 165. Strikeouts: 2 × 200 = 400. So (260 + 165 − 400) / 180 = 25 / 180, which is 0.14. Add a constant, and for this example take a round 3.10, close to where the published figure has tended to sit. A finishes at 3.24.
Now B. Home runs: 13 × 12 = 156. Free passes: 3 × 85 = 255. Strikeouts: 2 × 120 = 240. So (156 + 255 − 240) / 180 = 171 / 180, which is 0.95. Add the same 3.10 and B finishes at 4.05.
The two are eight tenths of a run apart, which is an enormous gap, and the chart shows precisely where it came from. A gave up eight more home runs, which cost him 0.57. He also struck out 80 more men, which saved him 0.89, and issued 30 fewer free passes, which saved him another 0.50. The strikeouts and the control together more than paid for the home runs.
Notice what did not enter the calculation at any point. Neither pitcher's defence. Neither pitcher's park. Neither pitcher's tendency to give up his home runs with the bases empty or loaded. Neither pitcher's ground-ball rate. None of it is in there, by design.
Why anyone thought this was a good idea
The intellectual foundation was laid by Voros McCracken, who posted his findings on the rec.sport.baseball newsgroup in 1999 and published the argument properly in a January 2001 Baseball Prospectus article asking how much control pitchers actually have.
His finding, which was received at the time as close to heretical, was that a pitcher's rates of striking batters out and preventing home runs held up well from one season to the next, while his rate of preventing hits on balls put into play did not. The same pitcher, throwing the same way, would post very different results on balls in play in consecutive years. Whatever was driving those results, it was not something the pitcher was reliably doing.
Tom Tango later derived FIP as a practical implementation of that idea: if the outcomes that persist are home runs, walks and strikeouts, build a statistic out of only those, put it on the ERA scale, and you have an estimate of a pitcher's performance stripped of the things he does not control.
The finding did not survive entirely intact, and the amendments are worth knowing. McCracken himself flagged pitchers with unusual deliveries, knuckleballers in particular, as behaving differently. Tom Tippett's work at Diamond Mind in 2003 found meaningful career-long differences in balls-in-play results between pitchers that correlated with how good they were. The modern position is a compromise rather than a victory for either camp: pitchers exert some genuine influence over what happens to balls in play, that influence is much smaller than the raw year-to-year variation suggests, and it takes a very long time to establish.
How long is a number worth carrying. Batting average on balls in play is the statistic that measures this directly, and the published guidance is that a hitter's figure starts to mean something after around 800 balls in play, roughly two seasons, while a pitcher's needs something like 2,000, or about three full seasons. Three seasons is a career phase. That is the actual scale of the problem FIP was built to route around.
The three things FIP throws away
The people who publish it are explicit about this: FIP assumes average luck on balls in play, average defence, and average sequencing. Each of those is worth understanding separately, because they fail in different ways and for different pitchers.
The fielders. A pitcher with an outstanding defence behind him will allow fewer runs than his FIP suggests, every year, for as long as that defence is behind him. This is not a distortion in the pitcher's favour, it is the statistic doing exactly its job: it is telling you what he did, not what his team did. The problem is that people read the gap as luck and it is not luck. A defence is a durable property of a club, chosen deliberately and paid for, and a pitcher who plays in front of one for six years gets six years of it.
Sequencing. FIP treats a season's events as a bag with no order in it. Three singles and a walk in the same inning score runs. The same four events spread across four innings score none. ERA sees that difference because it counts runs; FIP cannot, because it never looks at when anything happened. Over a full season the effect is usually modest and it does not reliably repeat, which is why sequencing is the component of the gap most defensibly filed under chance.
Contact quality. This is the largest hole and it is the one the original framework had no way to see. FIP charges a pitcher the full 13 for a home run and precisely nothing for a 110 mph line drive that hits the top of the wall. Two pitchers who allow identical numbers of home runs, walks and strikeouts get identical FIPs even if one of them was hit hard all season and the other was not. Contact tracking has changed that: how hard and at what angle the ball leaves the bat is now measured on every batted ball, and suppressing hard contact turns out to be a real if modest skill. FIP was built before any of that could be measured and it has never been retrofitted, because doing so would stop it being the simple, checkable thing that made it useful.
There are smaller omissions that add up. FIP gives no credit for holding runners, for inducing double plays, or for a pickoff move that erases a leadoff walk. It does not know whether a home run was a solo shot in a 7-1 game or a three-run shot in a tie. And it is neither park nor league adjusted, so a pitcher in a venue that suppresses home runs carries a structurally lower FIP than an identical pitcher elsewhere. That is what park factors exist to correct, and adjusted versions of FIP are published for exactly this reason.
What a good FIP looks like, and why the answer keeps moving
The commonly published rule-of-thumb scale runs roughly like this.
| Rating | FIP |
|---|---|
| Excellent | 3.20 |
| Great | 3.50 |
| Above average | 3.80 |
| Average | 4.20 |
| Below average | 4.40 |
| Poor | 4.70 |
| Awful | 5.00 and up |
Use it with one caution attached, which is that any such table is calibrated to a run environment. The scale above was set against a particular season's scoring level. When league scoring rises, every one of those bands should slide with it, and a table copied out of an old article will quietly mislabel an entire generation of pitchers.
The FIP constant already handles this internally, which is the elegant part: because it forces league FIP to equal league ERA every year, a pitcher's FIP is automatically on the right scale for his own season. The rule-of-thumb table is not, because it is a fixed set of numbers. Read a pitcher against his own league and the problem disappears.
The gap between FIP and ERA is a question, not a verdict
This is where the statistic gets misused most often, and the fix is a habit rather than a formula. When a pitcher's ERA sits well below his FIP, there are four candidate explanations and only one of them is chance.
His defence was good. Check who was behind him. This is the first thing to eliminate and the one most often skipped.
His park helped. A venue that suppresses home runs lowers ERA and FIP together, but a venue with a large outfield that turns doubles into outs lowers ERA alone, because FIP never counted those balls.
His sequencing broke well. Real, and the least likely of the four to repeat.
He genuinely suppresses contact. The people who publish FIP acknowledge that certain pitchers consistently beat it, and describe them as reasonably rare. Rare is not the same as non-existent. A pitcher who lives in the bottom of the zone, gets ground balls, holds runners and induces double plays is doing valuable work that this statistic is structurally blind to. If a gap persists over several seasons, across different defences and different parks, the honest conclusion is that the statistic is missing something about him rather than that he has been lucky for five years running.
The same discipline applies in reverse. A pitcher whose ERA sits well above his FIP is often described as unlucky, and might instead be pitching in front of a poor defence, in a park that punishes him, or giving up his damage in clusters because he loses the strike zone when he is behind.
There is a rough order of operations here that saves a lot of arguing. Eliminate the park first, because it is a fixed property you can look up in a minute and it explains more gaps than anything else. Eliminate the defence second, for the same reason. What remains is a mixture of sequencing and genuine contact management, and the way to separate those two is time: sequencing does not repeat and contact management does. A single season cannot tell you which one you are looking at, and pretending otherwise is how most bad pitcher takes get made.
The coefficients are frozen and the game is not
Here is a wrinkle that rarely gets aired, and it follows from something already established: the constant floats every season, and the coefficients never move.
The constant is recalculated annually, which keeps the league's average FIP pinned to the league's average ERA no matter what happens to scoring. That is the self-calibration described above and it works. But the 13, the 3 and the 2 are fixed whole numbers, set once, and they encode a particular relationship between the run value of a home run and the run value of a walk.
That relationship is not actually constant across eras. In a high-scoring environment with runners on base more often, a home run is worth more runs than in a low-scoring one, because more of them clear the bases rather than travelling alone. The formula does not know this. It charges 13 in every environment.
The effect on the league average is nil, because the constant absorbs whatever the aggregate error is. The effect on individual pitchers is not nil, and it goes in a specific direction: when home run rates rise across the sport, the fixed coefficient means the spread of FIP values widens, and pitchers separate on home run prevention more than the underlying run values justify. When home run rates fall, the reverse happens and FIP compresses.
None of that makes the statistic wrong. It makes it slightly less precise at the extremes in unusual environments, which is a reasonable price for a formula you can compute by hand. It is worth knowing about mainly because it is the sort of thing people assume a widely used statistic must have solved, and this one has not, deliberately. Every attempt to solve it produces a number nobody can check on the back of an envelope, and the checkability is a substantial part of why FIP won.
FIP changed what pitchers are asked to do
A statistic that becomes the default way of valuing a job eventually changes the job. FIP has been the default lens on pitching for long enough that its fingerprints are visible on how pitchers are developed, and this is the part of the story that gets left out of the explainers.
The formula says three things clearly. Strikeouts are good. Walks are bad. Home runs are extremely bad. It says nothing whatsoever about anything else a pitcher might do with a baseball.
Follow those incentives and you arrive at the modern pitcher. Maximise swing and miss. Live at the edges of the zone but do not miss so far that you walk people. Keep the ball out of the middle of the plate at the top of the zone, where it leaves the park. Everything else, the weak ground ball, the broken bat, the popped-up sinker, the double play, is worth nothing on the sheet and therefore worth little in a development meeting.
That is not a caricature. It is a fair description of the direction pitch development has taken, and the tools that made it possible arrived alongside it. Pitch design is fundamentally an exercise in producing swings and misses and avoiding damage contact, and the metrics used to evaluate a new pitch in a lab, spin and its actual effect on the flight of the ball included, are chosen because they predict exactly the outcomes FIP counts.
There is an argument that this has cost the sport something. Whether it has is a matter of taste. What is not a matter of taste is that a metric which values only the three fielder-independent outcomes will, if used as the primary evaluation tool, produce pitchers optimised for those three outcomes and indifferent to the rest. Contact-quality data has begun to push back on that, because it can finally measure the thing FIP was blind to. But it arrived two decades after the incentives were set, and incentives set for twenty years are not undone quickly.
What came next, and the trade every successor makes
FIP has descendants, and each one buys prediction with description.
xFIP replaces the home runs a pitcher actually allowed with an estimate: his fly balls multiplied by the league's average home-run-per-fly-ball rate. The logic is that the share of fly balls leaving the park bounces around from year to year far more than skill can explain, so replacing it with the league rate removes noise. It works. xFIP correlates about as well with future ERA as anything this simple does. It also means the statistic is no longer describing the season that happened, because home runs a pitcher genuinely gave up have been deleted and replaced with home runs he did not give up.
SIERA goes further, modelling the interactions FIP treats as independent: that high-strikeout pitchers produce weaker contact, that ground-ball pitchers get more double plays when there are runners on, that pitchers with very high fly-ball rates tend to give up home runs on a smaller share of them. It is only slightly more predictive than xFIP, and it is considerably better at explaining why a pitcher succeeds, which is a different and often more useful thing.
The pattern is consistent and worth stating plainly, because it is the axis every one of these arguments actually runs along. Descriptive statistics tell you what happened. Predictive statistics tell you what is likely to happen next. ERA is the most descriptive and the least predictive of this family. SIERA is the most predictive and the least descriptive. FIP sits in the middle, which is exactly why it became the default: it is predictive enough to be useful and descriptive enough that you can still see the season in it.
Choosing between them is choosing a question, not choosing a better number. If you want to know how a pitcher's season went, ERA. If you want to know whether to sign him for four years, one of the estimators. Quoting an estimator to settle an argument about what already happened is the error, and it is committed daily.
Why FIP is why the two public WAR figures disagree about pitchers
There is a practical reason this matters beyond arguments about individual seasons. One of the two public implementations of wins above replacement builds its pitcher values on FIP. The other builds them on runs actually allowed, adjusted for park, opposition and the quality of the fielders behind the pitcher.
That single choice is the largest source of disagreement between the two systems for pitchers, and it is not a bug in either. They are answering different questions. The FIP-based version asks what the pitcher himself did. The runs-based version asks what happened while he was on the mound, and then tries to strip out the parts that were not his doing. A pitcher with an unusual profile can differ by a couple of wins between the two, and neither figure is wrong.
Anyone quoting one pitcher's WAR against another's without knowing which version they are holding is making an argument they have not understood.
When not to reach for FIP at all
Four situations where the statistic should be left alone.
Single games. FIP over six innings is arithmetic on a handful of events and it means nothing. A pitcher who allowed two home runs in a complete-game win has a terrible FIP for that game. So what.
Small reliever workloads. A reliever throwing a modest number of innings can have his FIP swung most of a run by one home run. The formula's home run coefficient, which is correct over a season, is a liability over 40 innings.
Unusual deliveries. The original research flagged this and it has never been resolved: pitchers whose stuff is genuinely strange appear to influence balls in play differently from everybody else, and a statistic that assumes league-average results on those balls will systematically misread them.
Extreme parks and extreme defences. FIP is not adjusted for either. In a normal environment the distortion is small. In an extreme one it is not, and the adjusted versions exist precisely so that you do not have to do the correction in your head.
Four things to check when the numbers disagree
The point of understanding any of this sport's pitching statistics is to be able to do something with them, so here is the routine.
Compute the raw expression yourself. Take the four counts off the season line, weight them 13, 3 and 2, divide by innings. Doing it once tells you more about the statistic than reading about it ten times, and it takes about a minute.
Look at the home run count before anything else. It carries more than four times the weight of a walk, which means it is the single event most capable of moving the figure, and it is also the least stable of the four from year to year.
Find out who was fielding behind him. The FIP-to-ERA gap is the first place a good or bad defence shows up, and it is the explanation people skip most often on their way to calling it luck.
Check whether the gap persists. One season of a pitcher beating his FIP is a coin landing heads. Four seasons across two clubs, two defences and two parks is a pitcher doing something the formula was never built to see, which is a much more interesting finding than luck and a much rarer one.
Common questions
What is FIP in baseball?
FIP stands for fielding independent pitching. It estimates what a pitcher's earned run average would have been if he had received league-average results on every ball put in play behind him, by counting only home runs, walks, hit batters and strikeouts. It is reported on the same scale as ERA so the two can be read against each other directly, which is the whole point of the constant added at the end of the formula.
How is FIP calculated?
Multiply home runs by 13, walks and hit batters by 3, and strikeouts by 2. Add the first two, subtract the third, divide the result by innings pitched, then add that season's FIP constant. The constant is set so that the league's FIP equals the league's ERA, which is what puts the output on a familiar scale rather than leaving it as an abstract rate.
What is a good FIP?
On the published rule-of-thumb scale, a figure in the low threes is excellent and a figure around the low fours is roughly average, though that scale was calibrated to a particular run environment and drifts as scoring changes. The safer habit is to read a pitcher's FIP against his own league and season rather than against a fixed table. Park-adjusted versions exist for exactly this reason, because FIP itself is neither park nor league adjusted.
Why is a pitcher's FIP different from his ERA?
Because the two count different things. ERA counts the runs that actually scored, which depends on the fielders behind the pitcher, the park, and the order the events happened in. FIP counts only the four outcomes that involve no fielder at all and assumes league-average results on everything else, so a gap between them is a measure of what the fielders, the park, the sequencing and chance did on top of the pitching.
Is FIP better than ERA?
They answer different questions, so neither is better in general. ERA is the better description of what happened, because runs actually scored and the pitcher was on the mound while they did. FIP is the better predictor of what will happen next, because the components it uses are far more stable from season to season than the results on balls in play that ERA depends on.
Filed under Baseball·baseball · mlb · baseball statistics · analytics · pitching · sabermetrics