Why one closer over another?

image
Joel Hanrahan might not be a household name right now, but he could be just as valuable as a guy like Mariano Rivera and come at a fraction of the cost. (Icon/SMI)

Around this time last year, I showed that saves are unpredictable and explained why I don’t like paying much for them on Draft Day. Does that mean, however, that we should treat all closers as precisely equal commodities and flip a coin to decide which one to take in any given situation? Of course not. If Mariano Rivera is sitting next to Fernando Rodney in Round 20, is it a toss-up? No, sir. I’m taking Rivera, and so are you. But why?

The answer deals largely with probability. While both Rivera and Rodney are expected to start the year closing games (assuming the Tigers don’t sign someone like Brandon Lyon), what’s the probability that each will end the year closing games? I’m sure most would agree that Rivera is a much surer bet than Rodney to still be closing games in September. But how do we quantify the difference?

I’ll be the first to admit that the process I’m about to outline is a bit subjective, but that’s the only way we can do this. This will change based on your own judgments of situations, but hopefully the process I lay out will prove useful.

When looking at how long a closer will last, there are two primary probabilities we much come up with: the probability of injury and the probability that poor performance or managerial whim will lead to removal. Also, if you’re drafting early and the team’s closer hasn’t been announced yet, the probability of winning the job is important too.

To elaborate further, the injury probability isn’t the probability that a player will get injured, because that begs the question “how long?” Instead, the percentage of time we expect the player to miss due to injury. As a guideline, one month is roughly 17 percent of the season.

The save system

Now let’s take a look at how you might go about putting this all together. For our purposes today, keep in mind that these figures aren’t based on any real measurements. They are quick estimates simply to show you how this should be done.

To keep things simple, I’ll ignore the impact of team quality on save opportunities and the impact of closer skill on save conversions, since the effects aren’t that large. If we were to do this in a more scientific manner, both would need to be considered, although there would be a heavy regression to the mean component since there is so much random variation in these things (check out the article I linked in the first sentence to see more precise figures).

In this vein, I assigned every closer 42 save opportunities (which I got by taking the average number of opportunities for all pitchers who closed the entire year in either 2007 or 2008) and a conversion percentage of 88 percent (the aggregate rate for this same set of pitchers).

+-----+-----------------+-----+---------+--------+---------+-----+----+
| ADP | CLOSER          | SVO | GET JOB | INJURY | REMOVAL | SV% | SV |
+-----+-----------------+-----+---------+--------+---------+-----+----+
|   9 | Joakim Soria    |  42 |    100% |    11% |      3% | 88% | 32 |
|   7 | Mariano Rivera  |  42 |    100% |    15% |      6% | 88% | 30 |
|  18 | Joel Hanrahan   |  42 |     97% |    12% |     10% | 88% | 28 |
|  23 | Fernando Rodney |  42 |     65% |    22% |     60% | 88% |  7 |
+-----+-----------------+-----+---------+--------+---------+-----+----+

Note: I didn’t include it in the table above, but for some closers, you could add another column with the percentage chance that the pitcher is traded to a team who will only use him as a setup man or that the team will trade for a closer to supplant him. In 2009, this could apply to a guy like Huston Street or Jonathan Broxton.

Again, while these are based on some quick, subjective judgments on my part, you can see that — strictly in terms of saves — it is completely unnecessary to take a closer in the early portion of a mixed league draft. Under this method, early round options like Joakim Soria and Rivera would be just as good of bets as Joel Hanrahan, who could come a full ten rounds later.

As long as you’re making reasonable assessments and you pick the right late-round closer options, you will be making the correct percentage play. Of course, if you pick the wrong option, you could wind up with Rodney’s seven projected saves (ADP: Round 23).

But remember…

It’s very important to keep in mind that this method will not parallel real world results, and using it will severely decrease the value of closers in comparison to the rest of the player pool. Even if we assign a closer who is 100% to win the job a 0 percent injury score and 0 percent removal score, he would still only project out to 37 saves. The save leader next year will have far more than 37 saves.

The problem is that whoever this is will get there through a lot of good luck, something we simply can’t project. Therefore, when making your projections (or looking at someone else’s), it would be imprudent to assign any closer more than 40 or so saves, and certainly no more than 45. If you’re using a set of projections that have several closers above 40 or 42 saves, I would definitely consider looking elsewhere for save projections.

Further implications

While a closer projected to save 45 or 50 games (or 35 or 40 games with poor skills, ala Todd Jones)—as some systems will project—might have 60 or 70 percent of his value tied up in saves, a closer projected to save 30 games might only receive 40 percent or so of his value from saves.

This is notable because, when evaluating closers using this method, it makes it more important to identify the closers with good skills. Not only will good skills decrease a closer’s “Removal percentage”, those good skills will translate to a better ERA, WHIP, and strikeout total, which now make up a greater portion of the pitcher’s value.

Unless you’re playing in a league of full owners who pay attention to peripheral stats (as many of you do), you can gain a bit of an advantage here. Closers are often known for great “stuff” and blazing fastballs, but they certainly don’t need these things to succeed. Trevor Hoffman is a change-up specialist who throws his fastball just 86 MPH, but he is still a good pitcher, far better than a guy like Matt Lindstrom who can touch 100 MPH. Still, they are both being drafted around the 18th and 19th round.

Closing thoughts

Hopefully this provides you with a new, better way of evaluating a closer’s save potential. Again, it’s absolutely subjective, but many of these things simply can’t be objectified, and this doesn’t have to be a precise exercise. You can gain value from it simply by separating the Hanrahan’s from the Rodney’s. It doesn’t matter if your judgments put Hanrahan at 25 saves and Rodney at 15; that’s still a large enough gap to differentiate them for drafting purposes.


15 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments
Robert Dudek
15 years ago

There is one more calculation that can be made in this regard: expected (team) save opportunities.

There are some teams that generate more save opportunities than others. Good teams generally generate more than bad teams – and teams that play low scoring games get in more close games than teams that score and/or give up a bunch of runs.

Even though I have not done the work to yield a save opportunity predictor formula, in principle it would not be difficult to fit one with past data to determine expected team save opportunities.

One would expect, for example, that the Angels closer would get more save opportunities than the Orioles’ closer, because the Angels are expected to be a better team and are also a team that tends to play lower scoring games.

Nick
15 years ago

Speaking of closers and Brandon Lyon in particular, I noticed that you took him late in the mock draft a few days ago.  Why is Lyon being drafted but not Chad Qualls?  Qualls has better skills, and had the job at the end of last season.  Is there something I’m missing?

BCBarney
15 years ago

Robert, while your thinking sounds good it is actually false.  Good teams do not generate more save opportunities than bad teams.  I implore you to take a look at the data or maybe at the very least do a google search on the subject.  I don’t think anyone has come up with a reliable statistic that correlates well with save opportunities.

BCBarney
15 years ago

Even better, I found a great article for you in the archives of this very site:

http://www.hardballtimes.com/main/fantasy/article/are-saves-predictable/

Nate
15 years ago

Well, the Giants pretty much stunk last year (and were predicted to), yet Wilson got a ton of Saves.

Derek Carty
15 years ago

Nick,
Qualls actually went in Round 17 of that draft.  If he lasts longer than that in any of your leagues, he makes an excellent pick.
i
As far as Lyon goes, he’s actually a free agent (though MDC has him lasted on his previous team, the D’Backs).  There’s talk of the Tigers signing him, in which case he would be a heavy favorite to start the year as their closer.  In round 23, I was perfectly fine taking that chance, especially since there are some other teams with shaky closer situations who could be wlling to sign him.

Robert Dudek,
BCBarney’s got it right.  That’s actually my article he linked to (which I actually mentioned twice in this article, sorry if it went overlooked).  While team quality has *some* bearing on save opportunities, it isn’t as large as you would think.  Here’s how I addressed this in today’s article, in case you missed it:

“To keep things simple, I’ll ignore the impact of team quality on save opportunities and the impact of closer skill on save conversions, since the effects aren’t that large. If we were to do this in a more scientific manner, both would need to be considered, although there would be a heavy regression to the mean component since there is so much random variation in these things (check out the article I linked in the first sentence to see more precise figures).”

Doom, PHD
15 years ago

I understand the concept of “not paying for saves”, but if you draft second or third rate closers aren’t you paying for inflated WHIP and ERA?

Derek Carty
15 years ago

Doom,
Not necessarily.  You could very easily not select a single closer until after Round 16 and still get guys with elite (or close to elite) skills.  Heath Bell, Joey Devine, and Chad Qualls seem to fit this bill.

To address your question further, though, I’d definitely read this article (http://www.hardballtimes.com/main/fantasy/article/draft-strategy-closers-are-overrated/).  It talks about why the drop-off isn’t as big as you might imagine.

To summarize, though, the combination of a R5 Hitter/R18 Closer is more valuable than a R5 Closer/R18 Hitter.

Looking at the 2009 THT projections, Mariano Rivera has the highest value – $25.  He has a 2.52 ERA and 31 saves.  Trevor Hoffman has a 3.55 ERA and 31 saves and is worth $15.  That may seem like a big gap, but keep reading.

A R5/6 OF, for instance (let’s say a guy like Alex Rios), would be worth $15 or so.  A R18/19 OF would only be worth about $1.  That’s a $14 difference, larger than the gap for the closers.  And that’s if we’re looking at the extremes.  The difference between Rivera and Heath Bell (adjusted to 31 saves) is just $6.  A 31-save Joey Devine is just $3 different.

If you combined Rios and Bell you get roughly $34 of value.  Combining Rivera and, say, a $1 Pat Burrell gets you just $26 of value.

You can replace names as you wish, but I can assure you the premise is very sound… IF you take the right late round closers.

Robert Dudek
15 years ago

The referenced article is a decent starting point, but I think the data sample is far too small. We need to look at a more complicated predictive algorithm and a larger data sample before we deem save opportunities as “non-predictable”.

One thing that bothers me is whether 7th inning save opportunities that end up being blown are counted as true opportunities in these studies. These are not true opportunities in the sense that the closer isn’t going to be brought into a game in the 7th.

If there are other studies available I’d be more than happy to look at them. In the mean time, I think I’ll do a bit of work in this area and see what I come up with.

The first thing I’m going to look at is runs scored minus runs allowed versus actual team saves. This will allow me to build a larger data sample and not have to worry about the effect of whatever definition of “opportunity” is in use.

My initial hypothesis is that there will be a “sweet spot” in team run differential at which team saves peaks. If a team is really great (i.e. has a massive edge in runs scored, the save opportunities should plateau or diminish slightly because there will be more blow out wins.

Thanks for the quick responses!

Robert Dudek
15 years ago

So far I’ve looked at 1993-2008 teams (what I like to call the post Eckersley single inning save era).

Run differential per game ((Scored-Allowed)/G)is correlated to team saves per game at 0.498

Rampage
15 years ago

Would you say the same in a 10 team league??

Derek Carty
15 years ago

That’s very interesting, Robert.  That comes out to an R-squared of .248, which isn’t super high but definitely worth noting.

I checked run differential when I wrote that article last year but only got an R-squared of 0.06.  Of course, that was using fewer years (although they were more recent years, I’m not sure if there’s any kind of bias with using more years from the 1990s) and it was run onto save opportunities, not saves totals.

We also need to consider that your findings indicate that 25% or so of a team’s total saves variance can be explained given that we have their exact runs scored and allowed.  Before the season starts, all we have are projections, which will definitely reduce our accuracy.  Once we take the inaccuracy of projections into account, we might only be able to explain 10 or 15% of the variance before the season begins.  That means maybe starting the Angels closer at 45 save opps and the Nats at 39 (rough estimates on my part which could actually be too large even).  Worth noting, but I doubt it makes up the huge ADP gap.

While I definitely think that this is an avenue worth pursuing to see if anything comes of it, I don’t think that this changes the overall point too terribly much.

Team quality needs to be considered (as I’ve maintained), but there will still be a heavy regression to the mean component (though perhaps not quite as heavy as I originally thought).

I still stand by this as a general rule of thumb, but I definitely might look into this more if I get time.  I’d love to see this done with 9th inning save opportunities, as that would definitely improve our data set, although I don’t currently have that kind of data.  Perhaps sifting through Retrosheet would allow us to get it.

Again, thanks for running that, Robert.  Let me know if you ever do anything else on the subject.

Derek Carty
15 years ago

Rampage,
Yes, I would feel the same for a 10-team league.  I would actually probably feel stronger about it in a 10-team league since the player pool is shallower, since you’ll be able to get quality closers even later in the draft, and since you’ll be more likely to acquire them via the waiver wire throughout the season given more setup man in the free agent pool and fewer opponents vying for them when they step into the closer’s role (of course dependent on your league setup).

Beanster
15 years ago

Derek – I love the fresh thinking on this. Last year this approach (high job stability, great peripherals) would have yielded Joakim Soria in the 13th or 14th round.  I’m still looking for a comparable value this year.

Revisiting the percentages may be even more valuable during the season as more data is available.

Robert Dudek
15 years ago

Just for fun, I developed a basic team saves estimator using just RunsPerGame (i.e. run environment) and TeamWinningPercentage as inputs.

On page 223 of this year’s THT Annual Mitchel Lichtman has a table of 2009 team projections, which are based on assuming that personnel and usage in 2009 will be exactly the same as they were in 2008 for all teams. I adjusted the winning percentage so that MLB come out to .500 and the RunsPerGame so that they match last season’s league rate.

In lieu of a more realistic and sophisticated 2009 WPCT and RPG projection, here’s the list for the AL using Lichtman’s data. Keep in mind that aRPG is not park adjusted:

2009     aWPCT   aRPG   SAVES
BOS     0.5791   9.385   46.5
NYY     0.5600   9.673   44.7
CLE     0.5473   9.366   44.5
TOR     0.5218   8.860   43.7
CWS     0.5282   9.172   43.6
TBR     0.5346   9.447   43.5
LAA     0.5155   9.110   42.8
MIN     0.5028   9.322   41.6
OAK     0.4964   9.235   41.3
DET     0.5155   10.098   41.0
KCR     0.4582   9.116   39.0
TEX     0.4773   10.185   38.3
BAL     0.4646   9.998   37.8
SEA     0.4455   9.473   37.5

When more accurate 2009 projections come out, I’ll plug in those numbers. The top to bottom range should be 10-12 saves for a typical league.