User Comments, Suggestions, or Complaints | Privacy Policy | Terms of Service | Advertising
Page rendered in 0.5706 seconds
48 querie(s) executed
You are here > Home > Baseball Newsstand > Discussion
Baseball Primer Newsblog
— The Best News Links from the Baseball Newsstand
— The Best News Links from the Baseball Newsstand
Wednesday, March 24, 2010
The Book Blog: Tango: Thank you Mr Forman
I haven’t even read the jazz yet…but I concur!
I was on Sean’s case like crazy regarding ERA+, as many of you know. Basically, while every other index stat in the world did the value of the metric divided by the “average”, ERA+ did the league average divided by the ERA of the player. In effect, instead of ER per IP, it was doing IP per ER. What made it worse is when people started to use this in calculations, using it for simple averages etc. The math did not work out.
I had proposed that he do it the consistent way, which would mean someone who gives up runs at half the league average show as 50, rather than 200. Sean was rightfully concerned that people are used to “bigger is better”, and so, that would look like a sticker shock.
Guy proposed something very simple: 2 - ERA/lgERA then times 100. This way, what would look like 50 for me would show up as 150. And the top end is 200 in the Guy method (or 0 in my method). And Sean did just that.
I was relentless in such a seemingly small thing. But it was important to show the symmetry of 50 and 150 to hold. Either my method or Guy’s method would have done that (and the original version of ERA+ did not hold to that). Kudos to Sean for being good enough for taking the brunt of my esotericness (esotericality?). For all the crap I gave him about it, I deserve to give him his kudos just as much.
Login to submit news.
You must be logged in to view your Bookmarks.
Hot Topics
Newsblog: Who is on the 2024 Baseball Hall of Fame ballot and what’s the induction process?
(273 - 10:12pm, Nov 30)
Last: Tom and Shivs couples counselor
Newsblog: Reds, RHP Nick Martinez agree to $26M deal, sources say
(6 - 9:24pm, Nov 30)
Last: Walt Davis
Newsblog: Jackson Chourio extension: Brewers closing in on historic deal with MLB's No. 7 prospect, per report
(12 - 9:17pm, Nov 30)
Last: Adam Starblind
Newsblog: OT Soccer - World Cup Final/European Leagues Start
(287 - 8:50pm, Nov 30)
Last: AuntBea odeurs de parfum de distance sociale
Newsblog: Sportsnet's Ben Wagner out as voice of Blue Jays radio broadcasts
(1 - 8:44pm, Nov 30)
Last: Walt Davis
Newsblog: Former Yankee Luis Severino agrees to 1-year, $13 million deal with Mets: reports
(20 - 8:22pm, Nov 30)
Last: Howie Menckel
Newsblog: Andre Dawson Wants His Hall of Fame Cap Changed to the Cubs
(45 - 7:52pm, Nov 30)
Last: It's regretful that PASTE was able to get out
Newsblog: OT - NBA Redux Thread for the End of 2023
(109 - 7:49pm, Nov 30)
Last: a brief article regarding 57i66135
Newsblog: Zack Britton details analytics ‘rift’ that’s plaguing Yankees
(2 - 7:18pm, Nov 30)
Last: McCoy
Newsblog: Source: Cardinals adding Sonny Gray to revamped rotation
(32 - 4:22pm, Nov 30)
Last: DCA
Newsblog: Reds add reliever Pagán on 2-year deal
(7 - 3:55pm, Nov 30)
Last: Walt Davis
Newsblog: The future of live sports TV reaches a tipping point
(52 - 2:14pm, Nov 30)
Last: Buck Coats
Newsblog: Oakland-area fans start Ballers, an independent baseball team
(15 - 12:45pm, Nov 30)
Last: Der-K's no Kliph Nesteroff.
Newsblog: OT: Wrestling Thread November 2014
(3018 - 10:18am, Nov 30)
Last: a brief article regarding 57i66135
Newsblog: Update on Yankees’ Juan Soto trade talks: Teams talking players, but not close on agreement
(18 - 6:47pm, Nov 29)
Last: Howie Menckel
(273 - 10:12pm, Nov 30)
Last: Tom and Shivs couples counselor
Newsblog: Reds, RHP Nick Martinez agree to $26M deal, sources say
(6 - 9:24pm, Nov 30)
Last: Walt Davis
Newsblog: Jackson Chourio extension: Brewers closing in on historic deal with MLB's No. 7 prospect, per report
(12 - 9:17pm, Nov 30)
Last: Adam Starblind
Newsblog: OT Soccer - World Cup Final/European Leagues Start
(287 - 8:50pm, Nov 30)
Last: AuntBea odeurs de parfum de distance sociale
Newsblog: Sportsnet's Ben Wagner out as voice of Blue Jays radio broadcasts
(1 - 8:44pm, Nov 30)
Last: Walt Davis
Newsblog: Former Yankee Luis Severino agrees to 1-year, $13 million deal with Mets: reports
(20 - 8:22pm, Nov 30)
Last: Howie Menckel
Newsblog: Andre Dawson Wants His Hall of Fame Cap Changed to the Cubs
(45 - 7:52pm, Nov 30)
Last: It's regretful that PASTE was able to get out
Newsblog: OT - NBA Redux Thread for the End of 2023
(109 - 7:49pm, Nov 30)
Last: a brief article regarding 57i66135
Newsblog: Zack Britton details analytics ‘rift’ that’s plaguing Yankees
(2 - 7:18pm, Nov 30)
Last: McCoy
Newsblog: Source: Cardinals adding Sonny Gray to revamped rotation
(32 - 4:22pm, Nov 30)
Last: DCA
Newsblog: Reds add reliever Pagán on 2-year deal
(7 - 3:55pm, Nov 30)
Last: Walt Davis
Newsblog: The future of live sports TV reaches a tipping point
(52 - 2:14pm, Nov 30)
Last: Buck Coats
Newsblog: Oakland-area fans start Ballers, an independent baseball team
(15 - 12:45pm, Nov 30)
Last: Der-K's no Kliph Nesteroff.
Newsblog: OT: Wrestling Thread November 2014
(3018 - 10:18am, Nov 30)
Last: a brief article regarding 57i66135
Newsblog: Update on Yankees’ Juan Soto trade talks: Teams talking players, but not close on agreement
(18 - 6:47pm, Nov 29)
Last: Howie Menckel
About Baseball Think Factory | Write for Us | Copyright © 1996-2021 Baseball Think Factory
User Comments, Suggestions, or Complaints | Privacy Policy | Terms of Service | Advertising
User Comments, Suggestions, or Complaints | Privacy Policy | Terms of Service | Advertising
Reader Comments and Retorts
Go to end of page
Statements posted here are those of our readers and do not represent the BaseballThinkFactory. Names are provided by the poster and are not verified. We ask that posters follow our submission policy. Please report any inappropriate comments.
I liked it better when Mo was #1 on this list with a 194 and Pedro was 2nd at around 154. Now it looks like they are much closer.
ERA++ is too silly, but...
actually OPS+ scales up pretty nicely with runs
very simple "study", the baseline is .260/.327/.420, 550 PAs, so any .327/.420 line will generate a 100 OPS+
Ab hit tb bb avg obp slg ops+ rc rc/27 OPS+ % rc %500 130 210 50 .260 .327 .420 100 68.7 5.02
500 140 220 50 .280 .345 .440 110 76.0 5.70 10.3% 10.6%
490 130 235 60 .265 .345 .480 120 81.2 6.09 19.7% 18.1%
480 140 242 50 .292 .358 .504 130 86.8 6.89 29.6% 26.2%
the top line is the base line
going from a 100 to a 110 OPS+ meant the player went from 68.7 RC to 76.0
a 103 % increase in OPS+ matches a 10.6% increase in RC
granted RC isn't the best run estimator, plus while OPS+ scales u[p nicely with runs- it doesn't scale up nicely with Runs per OUTS
same chart, last figure is % change in RC/27:
Ab hit tb bb avg obp slg ops+ rc rc/27 OPS+ % rc %500 130 210 50 .260 .327 .420 100 68.7 5.02
500 140 220 50 .280 .345 .440 110 76.0 5.70 10.3% 13.7%
490 130 235 60 .265 .345 .480 120 81.2 6.09 19.7% 21.4%
480 140 242 50 .292 .358 .504 130 86.8 6.89 29.6% 37.4%
520 160 250 30 .308 .345 .481 120 86.4 6.48 20.0% 29.2%
480 150 197 70 .313 .400 .410 120 78.8 6.45 19.9% 28.6%
OPS+ at the low end goes negative too
Huh? Not true at all. Lots and lots of middle infielders, even regulars, have OPS+ figures in the 60's or 70's. For example, in 2009 Edgar Renteria, the starting SS for the Giants, had 510 PAs, and his 250/307/328 line yielded an OPS+ of 66.
Agreed.
Also, Nu-ERA+ is less granular. Old ERA+ differentiated between 1913 Walter Johnson (259), 1968 Bob Gibson (258), and 1906 Mordecai Brown (253). Now they're all tied at 161. This could be avoided by going to a single decimal place, but its weird to talk about a 161.4 ERA+.
The fact that Brian Sabean can't recognize major league talent, or a lack there of, should not be used as an argument against me.
Question to anyone how might know: will replacement level ERA+ still be around 80?
How do the new numbers scale? From the posts so far it seems like that the great majority of MLB players will see little change in their ERA+ but at the extremes the change will be um...more extreme. True?
Has this resulted in a change of how the calculation is being done? In other words, if I took a list of every MLB pitcher with the old method and compared it with the new method, would it be the same order or would there be significant differences?
I don't like this new method. It isn't descriptive and and I can't easily tell how much better the guy did then league average, which is the whole point. For example, a guy who posted a 1.00 ERA with a league average of 5.00 would have an old ERA+ of 500 and a new one of 180. A guy who posted an ERA+ of 1.00 with a league average ERA of 10.00 would have an old ERA+ of 1000 and a new era+ of 190. In this case, the old one seems to do a much better job of showing the quality difference in performance.
Extreme numbers are probably not the best way to determine the quality of a stat, though.
Okay. But can the fact that in 2009, White Sox 2B Chris Getz had an OPS+ of 74, Cubs' 2B Ryan Fontenot had an OPS+ of 72, Tigers' SS Adam Everett had an OPS+ of 59, and Marlins' 3B Emilio Bonifacio had an OPS+ of 61 be used as arguments against you? Just checking.
I called it Relative ERA myself. I think Thorn and Palmer called it Normalized ERA. My guess is that calling it ERA+ saved precious bytes back in the days of transistor tubes.
It's still based on ERA, so the order will still be the same. As has been mentioned, the new method seems to lose some granularity at the high end so you will have situations where say the old method had two guys at 130 and 131 and the new method has them both with the same number.
For the new formula, everything boils down to a percent (once you get past the zero point being 100, that is).
So a 110 ERA+ means that a pitcher was 10% better than the league average. 150 means 50% better, and so on. This was not true of the previous version.
EDIT: Notably the same IS in fact true of OPS+, if you look at it as a hitter's contributions to team runs, not OPS itself. OPS+ has the "slope" right; it's OPS itself that has the wrong slope. An OPS 10% higher does not correspond with 10% higher run scoring; an OPS+ 10 higher does.
ERA old new
1.00 400 175
1.50 267 163
2.00 200 150
2.50 160 138
3.00 133 125
3.50 114 113
4.00 100 100
4.50 89 88
5.00 80 75
5.50 73 63
6.00 67 50
6.50 62 38
7.00 57 25
The point is that the guy with the 3.00 ERA is as far from 4.00 as the 5.00 ERA guy is from 4.00. They are both 1 run away. The new ERA+ scale shows that properly. The old one did not.
Thanks. That scale illustrated the point well.
As in,
Barry 2002: 268 OPS+
Pedro, 2000 (old ERA+): 282
Pedro, 2000 (new ERA+): 166 ---whoop dee doo!
I have been toying around with looking at Z-Scores to compare ERA across seasons. I am interested in how extreme a performance was, in terms of standard deviations from the mean. IMO comparing ERA+ doesn't tell you as much, because you are comparing unscaled multiples of averages.
<u>Pitcher Year...ERA...IP...Subset Lg ERA...Subset StdDev...Z Score</u>
Gibson 1968 1.12 305 3.04 0.87 2.20
Johnson 1913 1.14 346 3.10 0.93 2.10
Brown 1906 1.04 277 2.78 0.86 2.02
So, all three are certainly extreme seasons, <u>maybe</u> with Gibson's being the most extreme of the three, depending on your view of the precision of the inputs (see below). Johnson's is arguably the most valuable, given the fact that he pitched far and away the most innings.
I set a minimum threshhold for the pitching seasons before I calculate the standard deviation, which is unstable when you include pitchers with small number of IP. I use the same minimum for calculating league average, for the sake of consistency. The league standard deviation seems pretty stable for threshhold levels from 80-150 outs pitched per year. This is a low enough minimum so that the standard deviation is stable and the subset league average I calculate doesn't differ <u>too</u> greatly from the actual league average. Below I used 105 Outs (or 35 IP) as the threshhold. (One could possibly argue that the difference in Z-Scores here is less than the precision of the method used.)
Thoughtful opinions would be greatly appreciated.
EDITED for format
Feel free to rip me a new ####### - I'm interested to hear how I'm wrong. But right now, I wish it was at least a different stat. Because, fundamentally, it IS a different stat, isn't it? It requires a totally different approach, and its results are demonstrated in a fundamentally different way.
ERA++ is too silly, but...
Then the next version can be ERA#
I liked it better when Mo was #1 on this list with a 194 and Pedro was 2nd at around 154. Now it looks like they are much closer.
Right. They should be much closer; the old way had distortions at both ends, that tended to vastly overstate the differences between high-end pitchers.
ERA old new
1.00 400 175
2.00 200 150
3.00 133 125
4.00 100 100
5.00 80 75
6.00 67 50
7.00 57 25
So, yeah, the old ERA+ would cause severe distortions. The 1.00 ERA guy is as far from 4.00 as the 7.00 is. The average of the two (1.00 and 7.00) is league average.
The new scale, with 175 and 25, gives you an average of 100 and they are symmetrical.
In the old scale, how is it apparent that 400 and 57 together average out to 100?
It seems like a cosmetic change that's intended to make it work, percentage-wise, in a way that's similar to OPS+, but that's misleading in that hitters and pitchers have different goals - it's impossible for a pitcher to have an ERA+ greater than 200, because 100 % "better" than average is 0.00. I understand what ERA+ does - I understand that, statistically, a 200 ERA+ isn't literally 100 % better than average. But in terms of pitching, it's twice as good as average, because a pitcher's goal is to prevent runs rather than create them. Being literally analogous to OPS+ is irrelevant, because pitchers and hitters have different goals. Besides, OPS+ and ERA+ don't even measure remotely the same things. Does that not make sense? Am I wrong?
This strikes me as a different stat - it takes a different route to its results and its results are demonstrated differently. It has its merits, definitely, but the old ERA+ makes more sense to me, conceptually, and I don't see why it can't be included in the discussion as well.
But thanks for responding to me respectfully.
It'll take a bit of getting used to the new scale, of course. But the real advantage is that you can average ERA+ weighted by innings. Or if a guy has a few seasons with roughly the same innings but widely-varying ERAs, you can now just do a straight average of them to get the ERA+. With the old scale, people did that all the time in posts here, but it was incorrect.
For example, in a 4.50 league, a guy throws two 200-IP seasons at 3.00 and 4.50. By the old scale, that's 150 and 100, but the average is 120. By the new scale, that's 133 and 100, and the average is 117 (just like you'd expect).
I apologize for my short response. I hope I did a better job explaining it in #39 (and I see Tango's #33 now).
I understand that, statistically, a 200 ERA+ isn't literally 100 % better than average. But in terms of pitching, it's twice as good as average, because a pitcher's goal is to prevent runs rather than create them.
See, this is what I disagree with. It's not twice as valuable as average (in terms of wins added or anything like that), and "twice as good" is a subjective term.
Tango's formula (which would call that performance an ERA+ of 50) does make it look "twice as good", which is one advantage it has over Guy's formula (but then it has the lower-is-better "problem"). I don't really think lower-is-better is a deal-breaker, and it would address this issue you bring up (while also still averaging properly).
There was never a single formula for ERA+. The concept has been around for thirty years, and it's been called "ERA+" for over twenty years, but there have been a number of different implementations, differing in details like application of park factors.
Because many people have been introduced to stats like ERA+ and OPS+ via B-R, there is an assumption that B-R's formulas are THE formulas for these stats, but that's not true. This isn't the first time Sean has adjusted the formulas, but it is admittedly a far bigger change than any previous ones.
That would be my concern. I understand that for Tango and Colin and others who work with these numbers, the change is welcome. But for the baseball fan who is slowing coming to accept the new metrics, which always seemed to be one of the key goals of the stathead movement, this seems like a nice way of turning him off.
Concur. If I was not a primate, I would not have seen Tango's entry, and would not know what had happened. So I would have gone over there, seen, for example, that Mo's career-leading ERA+ had suddenly "collapsed" from around 200 to 151, and be very confused.
I would wager that neither this site nor Tango's site gets the traffic that B-R does, so there could be a lot of confusion. I also agree that changing something that we had been working to gain acceptance for in this manner may be counterproductive.
I guess that even with something as recent as ERA+, it is possible to be a hidebound traditionalist.
I am also surprised that Sean didn't mention this anywhere (even the BR blog doesn't get nearly the traffic as the stats there, I'd guess). I prefer the new formula to the old one (though it's still imperfect), and understand the sentiment here that it should've been given a new name.
I honestly wonder what's better:
* Dramatically changing the formula for ERA+ to one that is better
* Adding the better formula as "ERA#" (or something) to the stat line, and saying, "ERA+ is still here, but it's flawed, so you should use ERA# now instead."
It's cool. You did - thank you.
I have some questions then. I was using "twice as good" to mean "twice as valuable." I didn't know that a 200 ERA+ wasn't twice as valuable - but I guess it makes sense, when I think about it, considering the probabilities of winning with different runs-allowed totals and everything. But does OPS+ demonstrate "twice as valuable" more closely than the old or new ERA+ does (from a purely hitting/pitching standpoint - not taking into account baserunning or fielding)? Does the new ERA+ scale in a way that more properly accounts for overall value with pitchers? Was Felix Hernandez closer to 23% more valuable than John Danks last year, or closer to the 50% or so that the old method showed?
I think maybe I'm starting to get into the territory of WAR and other cumulative stuff, though, aren't I.
Regardless, I still think that the old method was conceptually perfectly clear, and changing around like this without any fanfare whatsoever will be confusing and counterproductive.
For the vast majority of pitchers, their career numbers are barely going to change:
3.00 133 125
3.50 114 113
4.00 100 100
4.50 89 88
5.00 80 75
***
When you see Eck have a 0.70 ERA compared to a league average 4.20, and you see ERA+ of 600, how do you say that? That his ERA is... 6 times better? 500% better? No, you have to say that the league gives up runs 6 times as much as Eck. That is, the metric describes the league to Eck, and not Eck to the league.
Eck's ERA is 17% of the league average. Or, it's 83% less than the league average.
While I prefer the relative scale (meaning showing 17), the Guy method says 200-17 = 183. And so, you would look at the 183, take away the 100 and say that Eck's ERA is 83% less than the league average.
Phew. I feel bad when I irk somebody here, so I'm glad to see your response.
I have some questions then. I was using "twice as good" to mean "twice as valuable." I didn't know that a 200 ERA+ wasn't twice as valuable - but I guess it makes sense, when I think about it, considering the probabilities of winning with different runs-allowed totals and everything.
Well, that's part of it, but the key is really the non-linearity of the old scale. Say we have a 4.00 league, and two teammates who throw 200 innings, one with a 3.00 ERA and one with a 5.00 ERA. Their old ERA+ marks are 133 and 80. So they're each 22 runs away from average (one better, one worse), but old ERA+ puts one guy 33 points above average and the other 20 below. That's the distortion I'm talking about.
As you move further away from league average (like when talking about Cy Young contenders, or ace relievers, or what have you), the distortion just grows.
I'll address the "twice as valuable" in my next post.
But does OPS+ demonstrate "twice as valuable" more closely than the old or new ERA+ does (from a purely hitting/pitching standpoint - not taking into account baserunning or fielding)?
Leaving the "valuable" part alone for a minute, OPS+ has a linear scale; if the difference between 120 and 100 is 15 runs, so is the difference between 160 and 140. This is because the denominator is the *league's* OPS. The old ERA+ used the pitcher's ERA+ as the denominator, meaning it was different for every pitcher; the new ERA+ is similar to OPS+ in that it uses the league average as the denominator, so a 20-point difference means roughly the same difference in run value between 180 and 160 as it does between 110 and 90 (for both OPS+ and new ERA+).
Yep. You can say "various forms have been around for 30 years," but 99.9% of the people associate it with what BB-ref has been peddling, and now that's up and changed. Troglodytes don't like change.
I don't think we can value players in a vacuum; we can compare them relative to each other, or individually to a baseline.
In other words, I don't think we can say that a pitcher who pitched 200 innings at 2.25 in a 4.50 league is worth X; we can say he's worth Y runs over an average pitcher in the same number of innings, or Z wins over an average starter in the same number of starts. If we do this with multiple pitchers, we can then compare them.
So I don't think it really makes sense to say a 2.25 guy is twice as valuable as average (old ERA+ of 200), or 50% more valuable (new ERA+ of 150).
We've been discussing this ERA+ change in the lounge, comparing it to PythagenPat winning percentage as a "truer" measure of value, since it's expressed in wins. (Though I still have the same concern about measuring value in a vacuum; when talking about PythagenPat winning percentage, at least we can look at it in terms of wins above .500 or .400 or something). This chart shows how a team's ERA converts into old ERA+, new ERA+ and PythagenPat W% at various levels (I doubled the PythagenPat W% to get it onto the same scale). The distortion in old ERA+ is very apparent in that chart, especially at the high end (low ERA).
At the low end (poor ERA), PythagenPat curves back up, somewhat like Old ERA+. But PythagenPat assumes 9 IP/GS -- it works for a team, but for individuals, guys with poor ERAs don't pitch as many innings, and the real value of individuals is flatter than that curve.
But that separation shouldn't have been there. The fact that it no longer exists is a correction, not a bug.
I agree that Sean should have announced this, but I disagree with those saying he should have made this with a new name. Gosh, we have enough stats already don't we? It would just get confusing with 9 different versions of ERA+. I don't want BB-Ref to be BPro.
PS: The hover-glossary on ERA+ still shows the old calculation. Also I am not sure he is using the formula we think he is -- he has Roy Halladay's 2000 at -9, for example, and I can't replicate that.
At the low end (poor ERA), PythagenPat curves back up, somewhat like Old ERA+. But PythagenPat assumes 9 IP/GS -- it works for a team, but for individuals, guys with poor ERAs don't pitch as many innings, and the real value of individuals is flatter than that curve.
OK, I tried to estimate what the PythPat curve should look like for starting pitchers, by correcting for IP/GS. I took all pitchers who have exclusively started for a season over the last three years, and calculated average IP/GS for each ERA bucket. Then I calculated the PythPat W% based on those IP and league-average bullpen support, and converted that back into wins added per 9 IP.
So here is the new chart (the green line represents my best guess so far at starting pitcher "value", in terms of PythPat wins added per 9 IP). Note that we still see the same upwards curve at the high end, and a less-pronounced upwards curve at the low end (though still pronounced enough to show that runs-based methods and new ERA+ underrate poor starting pitchers -- especially if we think that there's value in pitching bulk innings, even at low quality).
More importantly, ERA+ was defined, at least at Baseball-Reference, by the way it was calculated and not by what it was trying to measure. The definition at the site read (and still reads), simply "100*[lg ERA/ERA] Adjusted to the pitcher's ballpark(s)". It was not defined as "ERA relative to league average, adjusted to the pitcher's ballpark(s)", at least not at Baseball-Reference.
I don't like this new method. It isn't descriptive and and I can't easily tell how much better the guy did then league average, which is the whole point. For example, a guy who posted a 1.00 ERA with a league average of 5.00 would have an old ERA+ of 500 and a new one of 180. A guy who posted an ERA+ of 1.00 with a league average ERA of 10.00 would have an old ERA+ of 1000 and a new era+ of 190. In this case, the old one seems to do a much better job of showing the quality difference in performance.
But consider the example where two pitchers are in a league with a 5.00 ERA:
Pitcher A: 1.00 ERA
Pitcher B: 2.00 ERA
Do you think Pitcher A was 100% better (twice as good at) preventing runs as pitcher B, or 25% better than Pitcher B, relative to league average? Because here are the old and new ERA+ stats.
500 old ERA+, 180 new ERA+
250 old ERA+, 160 new ERA+
The new ones make more sense to me. But that's an argument for a new stat, not a new calculation for an old one, IMHO.
You will still have different versions of ERA+, because the historical record isn't going away. Those who advocate a new name are the ones proposing not having two different versions of ERA+.
Ideally, if you had introduced a new stat and left the old one, people would eventually buy into the utility of the new one and basically abandon the old one. Trying to force that change rapidly is bad, however. IMO.
The variable runs-to-wins conversion was my main concern about this change, and this does an excellent job of addressing that concern. Nice work.
I'm not saying there aren't advantages to calculating it this way, but they seem pretty marginal to me and definitely not worth completely changing the referent point for a lot of people.
As people have pointed out, the fact that there is now a major divergence in what counts as great/good/average/terrible OPS+ and ERA+ numbers is a big issue for me. It's disconcerting to look at that list and see that the second best rate pitcher in baseball history gets a 135. Jason Bay had a 134 OPS+ last year. So there's that.
I do think Tango has driven changes single-handedly at Hardball Times, FanGraphs, Baseball Prospectus and Baseball Reference. That's pretty powerful by the standards of our little universe. It's fair to say he's the #2 guy in sabermetrics.
This is possibly a function of the dispersion of the OBP and SLG and ERA data themselves. The 100th best OBP or SLG season is probably 2.5 standard deviations above that season's league average. IIRC Lefty Grove's most extreme ERA season probably approached but didn't reach 2.5 standard deviations above that season's league average.
From now on, every time you are reading a discussion using ERA+ you will have to find out if it's old-style or new-style ERA+. Awesome as they are, Sean Forman and Tom Tango can't convert all the stuff using ERA+ that is already on the internet.
It would be simple to signal that it is different by using another name for this rescaling. ERA+ isn't that great a name anyway.
Thanks. Honestly the new ERA+ doesn't address the runs-to-win conversion. The green and orange lines on the chart do that, and I think what that shows is that the misses by not using the runs-per-win converter are much smaller than the distortions that the old ERA+ formula introduced. I still think it's worth going through the trouble of looking at IP/GS and doing PythagenPat when looking at the extremes -- Cy Young votes, how much to value ace FA starters, etc.
Here is the spreadsheet. I hope you guys can see the formulas.
Yes, this is a real problem. I think it's great that Sean considered the new formula, but I do worry about how it was rolled out.
On a side note, an ERA of 0.00 in a hitters home park would generate an ERA+ over 200, right? That would be quite funny to see...
What does ERA+ of 400 mean on the old scale?
Maybe the next one can be "ERA?"
Seriously though, as a user this makes intuitive sense to me, especially thanks to tango's 33 . Will probably get a lot of crap from msm about changing the data, or moving the goalposts, or something.
One other thing about the rollout. I actually rolled it out unintentionally. I wanted to see how things changed, so I ran my sandbox site with the new formula and forgot to change it back before launching the uniform numbers, so it slipped into the wild without my blessing. I'm working out what to do next. I may change it back at least temporarily. I apologize for the confusion and it certainly doesn't say much for the QA process here at Sports Reference that this happened.
My blog post on this issue
That's a good point. At least that's comparing career to single season. To put it another way, the best single season ERA+ (which is 166) is equal to Adrian Gonzalez's 2009 OPS+.
I really don't care one way or the other about the change, which I suppose qualifies as a tepid complaint against the change. (If I don't think it makes much difference, it ain't worth the hassle). When I interviewed Sean for THT a few years ago, I asked him how he tries to balance mass appeal with sabermetrics, and he says he errs on the side of the former, because the size of the latter is limited. By that standard, I'm not sure it's a good idea to have the ERA+ refining. Then again, the only people who might care about this are us statheads, which would help justify it.
I definately disagree with the idea that this change should've been an addition, not a replacement. That's just too confusing, especially since the numbers are rarely changing that much.
It's not so much that it means anything in its raw form, but it means something to the reader/user of the data, that the pitcher was ridiculously good. 32 degrees farenheit doesn't mean anything particularly meaningful to 99.9% of us (I'm sure it means something scientifically) but I know that's when my water will freeze.
As logical as this change seems to be it appears that it will be more useful to the very small set of people who do various calculations than it will be to the majority of us who simply like it as a tool to figure out if someone had a good year. Now we have to adjust our mental scale.
EDIT - Never mind, just read your blog post and obviously you've done something pretty similar with the leaderboard. That's helpful.
Yes, the formula is not intuitive. But ignore that. The metric is very intuitive: 140 means a pitcher is 40% above average (i.e. gives up 40% fewer runs than average), an 80 means a pitcher is 20% worse than average. And best of all, 180 really is "twice as good" as 140, relative to average.
Under the old system, 20 "points" of ERA+ could mean an improvement of .70 in ERA (120 vs. 100 ERA+). But it could also mean a reduction of just .20 in ERA (180 vs. 200 ERA+). How does that make any sense?
The main argument against the new ERA+ seems to be that it doesn't give really high scores to amazing performances. This is really an aesthetic argument, and I see where it's coming from. It's cooler, I suppose, to see Pedro at 250 than a 150. But we could do the same thing to BA by counting hits from 150 to 200 as 1.5 hits, and then hits after 200 as 2 hits each. Then Ichiro would hit .430, and that would be really cool ("my amplifier goes up to 11"). But, you know, it would be wrong. And the old ERA+ was also wrong.
What does ERA+ of 400 mean on the old scale?
That his ERA was 1/4 of league average maybe? That's why I'd prefer it without the "two minus" part. To me, it's not intuitive that 160 means 40% of league average. Yeah it's easy to calculate, if you know how. But I can garantee you that I'll be converting it back each time, at least at the startm, which to me kinda defeats the point...
Edit:And best of all, 180 really is "twice as good" as 140, relative to average.
No, this is the opposite of intuitive, "twice as good" as 140 should be 280. That wuold be intuitive. If you ask somebody who has no idea about ERA+ "what is twice as good as 140" he will guess 280.
Right. My concern was that the change would increase the distortion, which is clearly untrue.
It's like if we all of a sudden started using ERA as earned runs allowed per inning instead of earned runs allowed per 9 innings.
As in, if I say that the league ERA in 2007 was 4.37 and in 2009 it was 4.23, that means something to you.
If you expressed it on a per inning basis, 2007 ERA was 0.497 and in '09 it was .480.
In this particular case, they both mean the same thing and they're both "right", only that the "longer/wider" scale helps as a visual. Then again, we would probably get used to the new scale in due time. Life goes on...
Put me in the "new formula is better, but it's unfortunate that this happened the way it did and it probably would be better as a separate column (space permitting)" camp.
You are correct. But does anyone use it that way? When you see Eck with a 621 ERA+, is someone thinking that it's 100/621 of the league average?
But why not do what I say and think that 160 means that it's 60% below the league average?
The first column is the average ERA+ of the group of pitchers in that class. The second column is their actual win%. The third colum is the new ERA+ divided by 2. The last column is the number of pitchers in that class.
New winPercent newDivide2 n
159 76.0% 79.6 9
149 71.9% 74.6 29
139 68.2% 69.4 115
129 63.3% 64.6 385
120 60.2% 59.8 695
110 54.9% 54.9 1034
100 50.3% 50.1 983
90 45.7% 45.2 771
80 42.2% 40.2 518
71 39.0% 35.3 272
61 36.6% 30.3 142
52 32.4% 25.8 56
40 28.1% 19.8 16
From a newERA+ of 80 to 140, their actual win% and their newERA+ track each other very closely. Even at the top-end, it's pretty close. And, we know that it's going to converge at a newERA+ of 200 which would obviously be a top-end of 100 win%.
So, there are two ways to think of seeing a pitcher with a 140 ERA+: his ERA is 40% below the league average (140-100), or his equivalent win% is .700 (140/2).
The main issue seems to be not that this metric is better, but that there's inertia.
Unless there is a reason for 1000 IP and 3000 PA that isn't just "Hey, a round number that feels good."
This doesn't bug me at all, since OPS and ERA are measuring completely different things. If you want to compare the value of pitchers and hitters you need something like WAR anyway.
An OPS+ of 140 corresponds to 118 runs created (given 700 PA, and the average is 84 RC). That's +34 runs above average.
OPS+ and ERA+ are now comparable.
NewERA+ EquWin% IP WAA OPS+ EquRC RAA200 1.000 162 9.0 200 168 84
190 0.950 162 8.1 190 160 76
180 0.900 162 7.2 180 151 67
170 0.850 162 6.3 170 143 59
160 0.800 162 5.4 160 134 50
150 0.750 162 4.5 150 126 42
140 0.700 162 3.6 140 118 34
130 0.650 162 2.7 130 109 25
120 0.600 162 1.8 120 101 17
110 0.550 162 0.9 110 92 8
100 0.500 162 0.0 100 84 0
90 0.450 162 -0.9 90 76 -8
80 0.400 162 -1.8 80 67 -17
Anyway, does the change impact the vexed argument over whether it's easier or harder to put up extreme ERA+ numbers in a high-offense era? None of the numbers are as extreme as they used to be, so I guess it just got harder than it was, in the sense that the numbers up around 200 for Pedro and his recent colleagues have dropped quite a bit. Bob Gibson's 1968 is now tied for 6th on the single-season list: was it tied for sixth before? Is it closer to the top than it used to be? Have relative positions on the leaderboards changed, or just the absolute scale?
I believe the ordinal rankings, both for seasons and careers, should be unaffected by this change.
I assume your proof is all math?
@84: The ordering is (within rounding error) the same. The scale has changed though. The issue is described in my blog post on the subject.
What about the "Over ERA of 7" issue? See yesterday's Lounge for discussion and graphs and stuff.
I quite like these, although I wasn't sure if they were new, or if I just hadn't noticed them before.
As Harold mentioned in 39, I think how ERA+ was consistently used incorrectly was enough to warrant the new stat or a revamping.
I'm sure people will be used to the new stat well enough. I don't think a new stat was necessary, though it may have been nice - stats don't have to be intuitive for them to be mainstream. People compare two different quarterbacks by QB rating and just about nobody actually knows what that means (besides the basics that completion percentage, YPA, and TD/attempt are main factors).
I realize that this is about as mismanaged as something can be, but I'm going to roll it back. I actually have a pretty major update coming out today or tomorrow, so I really don't want this to detract from that, though the horse may be WAY out of the barn on that one.
Actually, OPS+ and ERA+ are now measuring pretty much the same thing: the rate at which a hitter or pitcher creates/prevents runs, compared to league average. And each unit = 1% of the average performance.
This does result in higher OPS+ for great hitters than great pitchers. But it should: Bonds2002 (268 OPOS+) created more than 120 extra runs for his team in 612 PA. Pedro2000 (ERA+ 166) saved his team about 80 runs over 817 opposing PAs. Bonds created runs at about twice the rate Pedro did, relative to league average. And that created about 4 more wins for his team (over fewer PAs). Why should Pedro have a higher "+" rating? (as he did under the old system)
I haven't read that yet, but I think it is easier to get a low ERA in a high run environment.
RELATIVE low ERA
Here's something Harold posted
Flip back a few pages, as the discussion is interspersed amongst #### jokes.
Harold responds
It's going to be the same issue with relativeOBP + relativeSLG minus 1. The negative in OPS+ and the negative in newERA+ are going to look weird and really hard to explain.
So, it's not something that Sean has to fix to his code. His code is correct in that this is a design limitation.
Obviously, simply doing ERA/lgERA has no such issues, but that's outside the scope of what Sean wants to do.
You must be Registered and Logged In to post comments.
<< Back to main