How Teams’ Initial Closers Performed
Over the past week, I have collected information on how spring training closers battles have worked out from 2013 to 2016. Today, I go over the results. It’s now time to release the tables.
The first set of data shows how the team’s initial closer fared.
| Season | Count | % |
|---|---|---|
| Closer from beginning to end | 47 | 39% |
| Lost to injury | 26 | 22% |
| Poor performance | 29 | 24% |
| Traded away | 9 | 8% |
| Traded for | 3 | 3% |
| Suspension | 2 | 2% |
| Replacement returned | 4 | 3% |
Just 40% of closers were able to make to the season’s end. About the same percentage lost their jobs to injury and poor performance.
The next table shows how these closer groups were projected to perform according to their Steamer projected ERAs.
| ERA | Average | Median |
|---|---|---|
| Closer from beginning to end | 3.03 | 3.08 |
| Lost from injury | 3.17 | 3.16 |
| Poor performance | 3.37 | 3.33 |
| Trade away | 3.47 | 3.51 |
| Traded for | 3.37 | 3.43 |
| Suspension | 3.29 | 3.29 |
| Replacement returned | 3.21 | 3.24 |
| Overall | 3.19 | 3.24 |
The closers who kept their jobs were about 0.15 runs better than the average and 0.30 runs better than those who lost their job because of performance. Additionally, the pitchers who got injured were close to the overall average. Finally, many of the closers moved at the trade deadline were below average. It seems like teams want to add a closer but not pay for a good one.
Now, I will examine the same data points but group by ERA.
| ERA Range | Whole Season | Poor performance | Injury | Other | Count |
|---|---|---|---|---|---|
| < 2.50 | 70% | 10% | 10% | 10% | 10 |
| 2.50 to 3.00 | 48% | 9% | 39% | 4% | 23 |
| 3.00 to 3.50 | 35% | 31% | 18% | 16% | 55 |
| >3.50 | 31% | 28% | 19% | 22% | 32 |
First, closers with an ERA over 3.00 are three times more likely to be replaced for poor performance than those with an ERA under 3.00. Additionally, there is a nice steady drop in pitchers making it as a full season closer as their ERA increases. I found the average ERA of each group and determine two best-fit equations from the data.
Equation #1 (r-squared = .93): Full season closer chances = -0.2454*Projected ERA + 1.1888
Equation #2 (r-squared = .98): Full season closer chances = 2.137*Projected ERA^(-1.488)
Instead of using the formula, here is a chart with some simple rates.
| ERA | Equation #1 | Equation #2 |
|---|---|---|
| 2.00 | 70% | 76% |
| 2.50 | 58% | 55% |
| 3.00 | 45% | 42% |
| 3.50 | 33% | 33% |
| 4.00 | 21% | 27% |
| 4.50 | 8% | 23% |
Time to put these numbers to use. Here are the chances a pitcher (10 or more projected Saves) will keep the closer’s role if he starts with the job.
| Name | Team | SV | ERA | Equation #1 | Equation #2 |
|---|---|---|---|---|---|
| Andrew Miller | Indians | 12 | 2.04 | 69% | 74% |
| Aroldis Chapman | Yankees | 35 | 2.33 | 62% | 61% |
| Zach Britton | Orioles | 34 | 2.38 | 60% | 59% |
| Kenley Jansen | Dodgers | 33 | 2.38 | 60% | 59% |
| Wade Davis | Cubs | 30 | 2.48 | 58% | 55% |
| Mark Melancon | Giants | 34 | 2.63 | 54% | 51% |
| Edwin Diaz | Mariners | 33 | 2.76 | 51% | 47% |
| Seung Hwan Oh | Cardinals | 32 | 2.92 | 47% | 43% |
| Kelvin Herrera | Royals | 33 | 3.00 | 45% | 42% |
| Craig Kimbrel | Red Sox | 32 | 3.01 | 45% | 41% |
| Ken Giles | Astros | 30 | 3.03 | 45% | 41% |
| Jeurys Familia | Mets | 25 | 3.07 | 44% | 40% |
| Cody Allen | Indians | 16 | 3.10 | 43% | 40% |
| Roberto Osuna | Blue Jays | 34 | 3.13 | 42% | 39% |
| Alex Colome | Rays | 31 | 3.13 | 42% | 39% |
| Shawn Kelley | Nationals | 30 | 3.20 | 40% | 38% |
| Cam Bedrosian | Angels | 27 | 3.21 | 40% | 38% |
| Raisel Iglesias | Reds | 18 | 3.23 | 40% | 37% |
| A.J. Ramos | Marlins | 28 | 3.33 | 37% | 36% |
| Tony Watson | Pirates | 30 | 3.35 | 37% | 35% |
| Addison Reed | Mets | 12 | 3.36 | 36% | 35% |
| Joaquin Benoit | Phillies | 10 | 3.38 | 36% | 35% |
| David Robertson | White Sox | 29 | 3.46 | 34% | 34% |
| Sam Dyson | Rangers | 30 | 3.54 | 32% | 33% |
| Greg Holland | Rockies | 23 | 3.56 | 32% | 32% |
| Arodys Vizcaino | Braves | 22 | 3.58 | 31% | 32% |
| Francisco Rodriguez | Tigers | 33 | 3.59 | 31% | 32% |
| Drew Storen | Reds | 16 | 3.68 | 29% | 31% |
| Jim Johnson | Braves | 14 | 3.71 | 28% | 30% |
| Adam Ottavino | Rockies | 14 | 3.73 | 27% | 30% |
| Ryan Madson | Athletics | 25 | 3.78 | 26% | 30% |
| Fernando Rodney | Diamondbacks | 28 | 3.85 | 24% | 29% |
| Brandon Maurer | Padres | 26 | 3.90 | 23% | 28% |
| Brandon Kintzler | Twins | 24 | 4.00 | 21% | 27% |
| Neftali Feliz | Brewers | 27 | 4.12 | 18% | 26% |
| Jeanmar Gomez | Phillies | 20 | 4.32 | 13% | 24% |
Now that the chances are known, the ‘when’ can be examined
Here are the average and median date for a pitcher losing his role
| Change that happened | Average Date | Median Date |
|---|---|---|
| Poor Performance | 06-04 | 05-18 |
| Injury | 06-18 | 06-05 |
| Trade involved | 07-23 | 07-27 |
| All | 06-16 | 06-11 |
On average, owners can expect a non-trade move to happen around June 1st with trades happening at the trade deadline (rocket science). Besides the average and median dates, here are the accumulated rates by month.
| Month | Injury | Change | Poor Performance | Change | All | Change |
|---|---|---|---|---|---|---|
| March/April | 23% | 23% | 31% | 31% | 25% | 25% |
| May | 42% | 19% | 52% | 21% | 42% | 18% |
| June | 58% | 15% | 79% | 28% | 60% | 18% |
| July | 81% | 23% | 83% | 3% | 84% | 23% |
| August | 88% | 8% | 86% | 3% | 89% | 5% |
| September. | 100% | 12% | 100% | 14% | 100% | 11% |
Changes happen early and stay steady for the first four months. Then for some reason in August, no one gets hurt or performs badly. I am still wrapping my head around this finding. I wouldn’t be surprised if the drop involved teams being stuck with their closers after the trade deadline.
One final piece of information for today, here are some numbers for teams with a closer competition going into spring training
| Stat | Value |
|---|---|
| Average ERA | 3.43 |
| Median ERA | 3.57 |
| Full season role | 31% |
| Average date for change | 06-12 |
| Median date for change | 06-07 |
The reasons teams have a closer competition is their talent sucks. Remember, these are the pitchers who initially won the closer’s role. The other canidates were probably worse. Additionally, the 31% survival rate is right in line with the formulas which project a 31% to 35% rate of holding the job for a full season.
I am sure I can tease some more tidbits out of the data, but I am done for today. Please let me know if there are any additional pieces of information you would like to have available. For now, I am just going to let the information stew around in my brain and will look at the data at a later date.
Jeff, one of the authors of the fantasy baseball guide,The Process, writes for RotoGraphs, The Hardball Times, Rotowire, Baseball America, and BaseballHQ. He has been nominated for two SABR Analytics Research Award for Contemporary Analysis and won it in 2013 in tandem with Bill Petti. He has won four FSWA Awards including on for his Mining the News series. He's won Tout Wars three times, LABR twice, and got his first NFBC Main Event win in 2021. Follow him on Twitter @jeffwzimmerman.
Do you have any data on the incumbent closers and how they performed? Did they keep the job?
I didn’t do any in-season work. It was a pain to pull together the preseason data. Maybe if I have a week of my life to completely waste, I will look into it.
Is there a correlation between ADP and keeping the role for whole season?
No idea. And someone would have to have a nice dataset of previous ADP data.
Jeff, great stuff as always. This is actually pretty surprising (at least to me). I knew the rates at which teams changed closers were high, but did not think they would be this high.
On an unrelated topic, I recently learned that, in regard to the missing Statcast data, a significant portion of the missing data comes from a small number of dates where we don’t have data for entire games, as opposed to where the data is missing due to Statcast failing to record certain weakly hit balls. Would it be possible to separate the Missed% column in your data into the percentage of batted balls missed due to having occurred in one of these games that Statcast missed altogether vs. the percentage of batted balls that were missed in a game in which we have data but where Statcast failed to record it for whatever reason?
Thanks for this article.
Basically why? I insert equivalent data depending on fielding position. Why does it matter how it was lost.
A batted ball fielded by the third basemen where we have no idea about the exit velocity and only a general idea about the vertical launch angle (i.e. those occurring in a game for which we have no data) is very different than a batted ball ball fielded by the third basemen where we have a very good idea that it was hit weakly (i.e. the rest of the missing data) and should be treated as different.
The deal is that I will use the same replacement values, which includes all tracked batted balls, for either one. There is just no way to know otherwise. You are pretty much split pubic hair at that point and not worth the time.
This is the kind of stuff that really makes me tick. Well done.
When you say “(l)ost to injury”, Jeff, do you mean job lost for good? That is, if Chapman goes down for the month of June and then comes back to the job, which slot are you putting that in?
I didn’t look for return time chances. I had to draw the line at some point collecting the data.
So you’re saying that would go into the ‘lost to injury’ slot, Jeff?
Regarding Equations #1+2, I don’t know that this is the “(t)ime to put these numbers to use”. Disregarding injury means they see Jansen and Davis as equally likely to lose the job via that way. Which is fine, in fact laudatory, for examining how job security links to the one variable of Projected ERA. But way less good for actually positing that ‘so-and-so closer does have this X% chance of holding onto the job’.
OK
Would be interesting to see how start-of-season closers fared when bucketed into ranges of saves projections by Steamer. E.g.,what pct. of closers with 15-20 projected saves actually made it through the whole year? Then possibly add the saves projection as an input to your equation.
I thought of this, but I don’t like projected Save totals as the role is in so much flux. Depending on when a draft or auction happens, the projected number of Saves changes.
For the closers displaced due to poor performance, was there an identifiable “closer in waiting” with perhaps a better projected ERA, at the start of the seasons?
I didn’t collect that data but feel free. It’s all yours.
This might be the best research and column of the offseason. Great stuff, Jeff.
Thanks. I have been wanting to do it for a while but needed to back check the data.
So what I take from the ERA projection correlation is that 75%+ of season starting closers are more likely to not finish the season in the role than to finish the season in the role.
I suppose that doesn’t necessarily mean you should punt SVs as you can handcuff your closer to have some additional certainty of retaining the closer role for a team over the course of a season, but it sure doesn’t point to any ability to be comfortable drafting closers with an expectation of being set for SVs.
This seems to advocate for an approach of punting SVs and focusing more than drafting relievers that help you across other statistical categories and hunting for SVs throughout the season.
I think an owner has to do both. They need some Saves to start with possible handcuffs. But an owner needs to be on top of the situation. So 60% change hands once during the season. Only 25% happen in the first month so 60%*25%*30=4.5 changes then. In a 12-team league, I think owners can hunt with .38 new closers per team. Moving to a 12-team AL-only league, the number is half at .19 and most of these setup men will be own.
This comment is basically turning into an article, so I will try to answer in more detail later next week. I have another cool toy to bring out today.
How were cases like Andrew Miller and Cody Allen handled in your historical analysis? It seems like neither of them should qualify as an “initial closer” at the moment.
They would fit into competition and the only case I can remember is when Betances and Miller were in the same situation a few years back but Miller was given the role right before the regular season.
I wonder if Neris should be added to the chart, considering many sources guess that he will take over for Jeanmar at some point. The chart certainly backs up the thought that Jeanmar will lose his closer role eventually.
Edit: Perhaps Carter Capps as well, considering Maurer’s relatively low standing too?