[Update: The concept of W/WAR has been shown to be fundamentally flawed; see comments by kds @11 and me @25. But I’m leaving the post open for any further comments in the threads.]
I’ve been playing around with ways to reflect the role of luck in starting pitchers’ W-L records, using data that can be pulled up with the Play Index. After muddling about with run support, defensive support and general team quality — none of which can be summoned by the Play Index — it occurred to me that those things are already factored into Wins Above Replacement (“WAR,” using the Baseball-Reference formula).
A simple ratio of Wins per WAR, then, might provide a snapshot of Wins luck for starting pitchers. The higher the ratio, the luckier the pitcher; the lower the ratio, the tougher the luck.
Some caveats:
(1) This is a junk stat, with no saber-value. My only purpose in running W/WAR* is to provide another point of reference in discussing pitcher Wins. If you’re not interested in pitcher Wins, you shouldn’t be interested in this. (Hint, hint.)
(2) Relief notes: Since I’m using career numbers for all pitchers, including those like John Smoltz with significant Save totals, I made a rough adjustment to WAR by subtracting 0.05 WAR per Save from the career WAR total, before dividing by Wins. That’s why I’m calling this W/WAR*, with an asterisk. The 0.05 value is the combined ratio of WAR to Saves for all pitcher-seasons of 20+ Saves in the past 20 years. Since modern closers rarely amass significant Win totals, I made no effort to remove relief wins.
(3) W/WAR* also reflects the luck of being allowed to pitch for many years despite poor results. (“Paging Mr. Russ Ortiz … Your rotation spot is ready….”)
(4) I would not use this method to compare pitchers from different eras, due to changes in SP usage patterns that (presumably) would cause significant difference in the average W/WAR* across eras.
The following table is centered on the past 20 years, 1992-2011. These are the career W/WAR* figures for every pitcher with:
- 100+ career wins; and
- 50+ starts in the period 1992-2011; and
- at least 50% of his career games as a Starting Pitcher.
In other words, the career totals of Roger Clemens, Greg Maddux and others who began before 1992 are included here, because they had 50+ starts within the specified period. Also, the 50% SP requirement weeds out the likes of Dennis Eckersley.
The table is currently sorted from worst luck to best luck.
I welcome any suggestions for improving this concept, or for losing belly fat by following this one simple rule.
(Aside to HHS authors: If you know how to get more separation between the column headers and the “sort” icons, please let me know!)
[table id=36 /]
I don’t have much to to analyze right now, but this list seems to correlate highly with ERA+
I believe I already stumbled across and mentioned that correlation in a blog of mine a little while back but I forgot with which post in particular. It was probably on the old B-R blog.
He doesn’t qualify for John’s criteria but if he did Matt Cain would be tied with Clemens at 2.76. Cain’s career isn’t over but right now he’s the only pitcher in history with 200+ starts, ERA+ above 120, and a losing record.
I love Matt Cain. Seven years of pitching outstandingly for a team with an anaemic offence, and not a word of complaint. When his luck was at its most horrendous around 07/08 even Barry Zito, with an ERA pushing 5, was picking up more wins than him for crying out loud.
Ed:
Not to pick nits, but I believe Jim “Death Valley” Scott might fit your Matt Cain criteria – 226 GS, 107-114 W-L (.484), and a 121 ERA+ prior to WWI. By the same token, some of Scott’s decisions may have come in relief. After returning from Wilson’s “war to end all wars”, Scott won another 125 games in the minors pitching into his late 30’s.
For all his hard luck, Cain is going to get the big payday…..
I forgot to mention that I also used the criterion of 75% of appearances being starts. Scott was at 71% which is why he didn’t appear in my search.
If Felix Hernandez qualified, he would be fourth on the list at 2.92, and Tim Lincecum would be fifth at 2.99.
This seems to give us more a measure of quality than luck. I think part of the problem may be that Wins starts from zero while WAR starts from a higher level. If you put a replacement level starter in for 180 innings you would expect him to get some wins. It is the wins over that that we are interested in.
Observation tells us that starting pitchers get about 1 decision for every 9 IP. Tom Tango, who had a lot of influence in the construction of both rWAR and fWAR, tells us that a replacement level starting pitcher is about .380 W/L%. We can use these to find out how many replacement wins we would expect for the IP, add the WAR and see how this compares to actual Wins. This should take much of the luck out it.
Clemens,
kds, good point about the “wins over that.”
I haven’t yet managed to focus in on what I’m looking for out of this. I started out thinking about Tom Seaver. As any Mets fan of a certain age will tell you (long after you’ve stopped listening – and I say this as a Mets fan of a certain age), Seaver got poor run support during his Mets tenure, even by Mets standards of the day.
This is broadly true, but the exact degree of it, and how much it affected his W-L record, is not that easy to determine. Seaver’s career .603 W% (IIRC) seems pretty well aligned with his 128 ERA+. So I was thinking, well, maybe Seaver didn’t really lose many wins to poor support, after all; maybe whatever he lost with the Mets, he got back in his time with the Reds and/or White Sox.
But then I decided to calculate W/WAR, and Seaver is near the bottom of the list of live-ball 250-winners.
Anyway, I’m kind of all over the dish, but thanks for helping me think it through.
I didn’t think I had posted this early version. #11 is more complete. Tango did say .38 but I think that goes with Fangraphs slightly lower replacement level.
Nobody thinks that highly of Jason Marquis, but this chart makes him seem downright terrible. He’s only been worth 3.4 WAR over 12 seasons???? Is that the lowest career WAR total for any pitcher with at least 100 wins?
Good instinct there, Josh. Marquis actually has the 2nd-lowest WAR of any 100-winner. Tony Cloninger underdid him, with 2.9 WAR and a record of 113-97, 88 ERA+.
Just would like to nitpick that Marquis does have 2.6 extra WAR tacked onto his resume due to his hitting.
Is Luck the term you really want to use to describe this phenomenon? Clemens, at least, with his .658 win %, seems an anomaly, unless the stat is saying that with just a little more Luck, or a lot, he would have been invincible on the mound.
The other paradox is that the fewer wins you have, the better your Luck is from the perspective of what’s supposed to count here, not getting your due. I’d rather my team had the wins, however they came.
Luck may not be the precise word I’m after.
However, it’s not a stretch to say that Clemens could have had even more wins and better W% than his already amazing numbers. He does have the 3rd-best ERA+ of any modern pitcher with 3,000+ IP. Lefty Grove, whose 148 ERA+ is just 5 points better than Clemens, had a .680 W%, and won 59% of his starts; Clemens won an even 50% of his starts.
Lastly, I may be brain-cramping, but I’m not following the paradox you describe. Holding WAR constant, more wins produces a higher ratio of W/WAR, which I have associated with more luck. Am I misunderstanding you?
Thanks for responding. My second point is convoluted, but—
Example: Clemens is at the top of the list because he is the unluckiest, which is “good” from the perspective of not getting his due. If he had been awarded more wins, WAR being the same, he might have fallen from the top spot. Being Lucky would be unlucky to his ranking. Sounds like two physicians to me.
My real point is that wins aren’t simply a statistic. From a team perspective they are the goal of playing the game.
This is interesting…but somehow “unlucky” seems like the wrong word to use to describe the people at the top of this list.
They obviously just didn’t pitch to the score as well as you-know-who, selfishly piling up WAR points and shutout innings in games their team was obviously going to win.
🙂
This seems to give us more a measure of quality than luck. I think part of the problem may be that Wins starts from zero while WAR starts from a higher level. If you put a replacement level starter in for 180 innings you would expect him to get some wins. It is the wins over that that we are interested in.
Observation tells us that starting pitchers get about 1 decision for every 9 IP. A replacement level starting pitcher is a bit under .40 W/L. We can use these to find out how many replacement wins we would expect for the IP, add the WAR and see how this compares to actual Wins. This should take much of the luck out it.
eWINS = (IP/9*.39)+WAR. eWINS is expected, earned or estimated Wins.
Clemens eWINS = 341
Jack Morris = 205. Suggests he was very lucky in his choice of teammates.
Roy Halladay = 171
Very interesting stuff, kds.
I would use 8.5 rather than 9 for the IP per SP decision, based on an every-10-years snapshot check of the available splits on B-R (and it’s pretty consistent across eras).
With that adjustment, the Est. Wins for Clemens comes out to 354, his exact actual wins. Seaver comes out to 325 (311 actual), Blyleven 318 (287 actual). Perry & Niekro each gain about 27 est. wins by this method.
kds, it’s pretty clear now that you’re right about why W/WAR is distorted. Here’s the example that makes it clear to me:
Consider two long-career pitchers with 400 decisions each. And for the sake of argument, imagine that their W-L records perfectly reflect the quality of their pitching.
Able’s record is 260-140, an outstanding .650 W%.
Baker’s record is 200-200, an even .500.
Their WAR values might be something like 80 for Abel and 40 for Baker.
So the ratio of Able’s wins to Baker’s wins (the numerator in W/WAR) is 1.3, but the ratio of their WAR (the denominator) is 2. That will tend to raise the W/WAR of the weaker pitcher, relative to the better one.
(I think I didn’t realize this distortion before because I started out looking at the W/WAR only for the top winners.)
kds, Bill James did a study a while back using estimated wins, or something very close to it, to compare the careers of Jack Morris and Rick Reuschel. He took a much closer look at it and compared them year by year and how their win totals would look if Reuschel had pitched for the teams that Morris(winning%-wise) did and vice versa. He definitively proved that Reuschel and Morris’ wins totals(214 vs. 254) could essentially be reversed if they had pitched for the others’ teams throughout their career.
Hate to ask a really dumb question, but why not include a pythagorean calculation of total runs allowed against league-average scoring?
Seems like a perfectly reasonable question, Mike. But since the league scoring data are only available by year, I would have to set up a fairly sophisticated spreadsheet in order to compare a pitcher’s career totals against the league average for that span of years.
I think I’ve done something like that before, and it might be worth doing here. But I’d still prefer an easier route. 🙂
I have a sort of “estimated wins” system using annual ERA+ to determine league-average scoring and then negative binomial distribution to arrive at “team neutral” win-loss records. And then I jump completely off the cliff by applying Bill James’s Fibonacci method to assess the value of that new win-loss record. . . .
Jack Morris is essentially a 232-208 pitcher, while Rick Reuschel should have ended up 232-173.
The results are really interesting. I wonder if, John, you have proven that quality that the outermost, stictly-saberminded people refuse to acknowledge: that elite pitchers are actually able to squeeze more wins out of iffy win situations than pitchers of less quality. They give up their hits at the right time, don’t give up the big hits at the wrong time, tend to strand just enough runners to get the job done, pitch to the score, etc. This has been refuted almost unanimously as not true as there is no real evidence that one pitcher is better at it than another, but I wonder if that is what this study is proving. Am I totally off here? The quality at the top of the list really strongly suggests that talent is what is driving a low W/WAR, not luck.
bstar, I wouldn’t draw any conclusions from those data just yet. It’s probably true that they don’t really make up any kind of luck index, which is what I was shooting for. But what they do comprise, if anything, has yet to emerge.
Yeah, it’s really got my head spinning as to what it is proving. That was kind of a wild stab and I had already talked myself out of it by the time you responded. Here’s another thought: are the great pitchers’ nonwinning starts, which typically are adding more WAR to their total than lesser pitchers, creating what’s driving them to the top of the chart? They tend to pitch better overall when they lose or get a ND than lesser pitchers, so they are accumulating more WAR when they don’t win, lessening their W/WAR. But now I’ve talked myself out of that one, too, because the great ones are also accumulating more WAR per win than lesser pitchers, so the ratio shouldn’t really change. Color me confused.
bstar, consider what I just posted @25, following up on kds’s critique.
I think the whole concept of W/WAR is fundamentally flawed. I’m now thinking about kds’s formula for Estimated Wins.
The guys at the top weren’t “unlucky”
They simply didn’t waste any years racking up wins while not racking up WAR. Clemens is at the top of the list because he was ALWAYS getting WAR, every season.
This is why the pitchers at the top of the list, are pretty close to being a list of the best pitchers of all time.
Stottlemyer and others are at the bottom of the list, because they spent many years racking up WAR, but not wins.
Greg Maddux, for example, doesn’t make it that high on the list because at the end of his career he racked up wins every year but not that much WAR.