What makes a good MLB bullpen?

The answer here, on the surface, is obvious: don’t allow runs. But what are the factors which go into a bullpen that’s good at run prevention? There are some obvious metrics which clearly have an impact. A high strikeout rate, for example: you can’t score, if you are carrying your bat back to the dugout. A low walk rate, similarly: free passes are never a good thing. Not allowing home-runs, too. These really don’t deserve looking into much. But there are other numbers where the correlation is less apparent. Ground-ball rate, say. Does dialing up a lot of ground balls help a bullpen? Or are fly-balls, which typically turn into outs more often, better?

To find out, I pulled numbers from Fangraphs for the last five years, for each of the thirty MLB teams. That’s 150 data points in total. I chose eight metrics and worked out the correlation of each against both ERA and FIP. Correlation is a number between -1 and +1: the closer to the extreme, the greater the connection between the two sets of numbers. +1 is a positive correlation: as X goes up, so does Y. -1 means as X goes up, Y goes down. 0 means the two data sets appear not to be linked. I’d expect the number of runs the D-backs score and my calorie intake at breakfast that day to be a zero correlation.

Below, you’ll find the metrics ranked in increasing order of average correlation.

#8. Fastball velocity. Average correlation: -0.088

Well, this was a surprise. The least significant metric to both bullpen ERA and FIP was pure power. Now, there is a known and logical link between velocity and K-rate. But for bullpens as a whole, that doesn’t seem to apply much. I think that may be because the best relief corps don’t just blow pitches past people. They have a variety of arms, approaches and slot angles, which let them keep batters off-balance. If all you’re seeing is 98-99 mph, hitters will eventually settle in. Variety makes that harder. If you look at the 2017 D-backs bullpen (perhaps our last truly good one), yes, they had Archie Bradley averaging 96.4. But they also had T.J. McFarland.

#7. Ground ball rate. Average correlation: -0.216

The negative number means the higher the ground ball rate, the lower the ERA and FIP. Which is what you want. While short of a definitive link, anyone who remembers Brad Ziegler will know it helps, even though ground balls are more likely to become hits. That factor is countered by a couple of things. Firstly, ground balls are much more likely to lead to double-plays, which is the best possible outcome. They also won’t become home-runs (without significant help!), the worst possible outcome. All told the OPS on ground balls last year across MLB was .514, while on fly balls it was .840, and that doesn’t take into account the sweet, sweet joy of a well-turned double-play.

#6. Line-drive rate. Average correlation: 0.290

Line drives are bad, m’kay? They are far more likely to become hits, with a BABIP of .616, compared to .245 for ground balls and just .091 for fly-balls (though the last have the best home-run chance). That helps lead to a meaty triple-slash line of .629/.622/.871 on line drives, a 1.492 OPS. So, more important for pitching than ground balls or fly balls, is simply avoiding line drives. While that is a skill, it’s one which is subject to random variation. In theory, we could look to buy low on a pitcher with a typically good LD rate which has spiked. But there’s always the risk the spike is not random.

[Random aside. Bunts last year had a .482 batting average! You may be wondering why hitters aren’t trying it more often. That’s partly a false number, because if you bunt with a runner on-base, you’ll typically get credit for a sacrifice even if you fail, and that doesn’t count against batting average. Include those back in, and the average goes down to .269. With close to no chance of an extra-base hit, that’s why you’re usually better to swing away. We now return to your regularly scheduled article.]

#5. Exit velocity. Average correlation: 0.382

You might be beginning to detect a theme here, although this one might almost fall into the category of obvious enough not to need confirmation. If you give up a lot of hard-hit balls, bad things happen. Yeah. This is my unsurprised face. I am, however, gratified that my number is close to that from a Fangraphs study, which found a similar correlation at the individual player (rather than the team) level: “an r of .33 for ERA and an r of .40 for FIP”. The study notes that pitchers with lower exit velocities also tend to be better at limiting home-runs, which is logical – and also better at striking out batters. That seems a little less obvious.

#4. Hard hit %. Average correlation: 0.398

You’d think “Hard hit” would be easily defined, but note there are two different metrics here: Hard% and HardHit%. The former comes from Statcast, and is simply balls in play with an exit velocity of 95+ mph. But Fangraphs also has Hard%, which is those “classified as hit with hard speed by Baseball Info Solutions”. Nobody seems to know specifically what this means – it’s proprietary – but seems to involve trajectory as well. We will get to that in a bit, but the Fangraphs version here is only a few points better correlated than the simple average of exit velocity in the paragraph above.

#3. Barrel%. Average correlation: 0.492

Barrels are the best kind of hard-hit balls, the highest quality of contact when combining power and direction. They need to be at least 98 mph, and hit at a certain launch angle. That latter component varies depending on the velocity: the harder the ball, the bigger the range of acceptable launch angles. They aren’t common: only a half-dozen batters passed 70 barrels over all of last year. But when they happen, they are kryptonite for pitchers: the season they were initially defined in 2016, barrels resulted in a batting average of .822, and a 2.386 slugging percentage. Give up very many of those, and you’ll have a bad day on the mound.

#2. Hard%. Average correlation: 0.590

As noted above, this is best described as the BIS version of a barrel. Although nobody seems to know exactly what this means – it’s proprietary – but seems to involve trajectory as well as velocity. One comment on that post suggests its just BIS interns eyeballing things. But that is on Reddit, where anyone can claim anything (and frequently does). However, it clearly does a better job at correlating with bullpen results than the related metrics we discussed above. On the other hand, it may not be skills: studies suggest “hitters just have much more control over how hard their batted balls are hit than pitchers.”

#1. Left on base%. Average correlation: -0.666

Turns out nothing is better for a bullpen than stranding base-runners. However, this was the metric where there was the biggest gap in correlation between ERA and FIP. For the latter, the correlation was down at -0.515, but for ERA is was all the way up at -0.818. Why the difference? I suspect because LOB% is quite heavily dependent on defense. The problem is, pitchers don’t tend to have much control over this, unless you think “clutch” pitching i.e. with RISP is a thing. High K pitchers tend to be better there. But as we’ve seen, they tend to be better anyway. However, free agents with a lower than usual LOB% might represent good bounceback candidates for 2026.

What have we learned. Probably less than would justify me writing fifteen hundred words on the topic. Though I did gain some knowledge regarding these metrics, and it beat staring out the window and waiting for the D-backs to do anything. To build a good bullpen, you want pitchers that are adept at avoiding hard contact (however you want to measure it), and are particularly proficient at leaving runners on base. Good luck with that. Ground ball pitchers are somewhat preferred preferred, but you don’t necessarily simply want to load up on flamethrowers, because that will allow hitters to get comfortable.

However, I want to point out again that consistently creating a good bullpen is hard. Indeed, it may be close to impossible. In 2018, Jeff Sullivan plotted bullpen Win Probability against the same number the following year. The results are the amorphous blob shown above, and show almost no correlation (0.04). Having a good bullpen one season does little or nothing to help you the next season – for comparison, starting pitching and hitting WP both had a 0.14 correlation. So it’s possible Mike Hazen has simply had the equivalent of coming up tails ten times in a row. Maybe 11th time will be the charm?

Read the full article here