lidsky
Played with the Odds ratio - quite cool. From what I saw in the few years that I looked at it gives similar results to the n-1 Standard Deviation method I was using. (Recall (K/BFP-avgK/BVP)/SD).

My first question was is one better than the other in comparing across vastly different era. I used Pedro 1999 vs. Vance 1924. In both methods, Vance reigned supreme. I figured for them to be reasonable, Pedro had to have a shot of being "equal" to Vance wrt to the league.

Off a baseline of 313K of 835 BFP.
Pedro would have needed 26 more Ks to equal Vance using the SD method.
Pedro would have needed 30 more Ks to equal Vance using the Odds Ratio.

that to me is an indication for that case, either seems reasonable. All else being equal, the Odds Ratio is superior for it's simplicity. I'll have to look more to see if there are any meaningful differences.

In general, looking across the whole population of pitchers with enough innings, both results are monotonic, but the SD method has a steeper slope away from the averages. Not sure what to make of that yet.
6:44 PM Aug 21st

lidsky
Yeah - I saw that in my original analysis which is why I clipped it at 100 innings and asked Charles what he used for his population size originally.

Anyway, clipping at 100 got rid of the tail and the peak near zero as expected. It didn't change the irregular shape of the rise and fall which is why I stuck with the n-1 method. I'll take the time to look at more years as I only looked at 5 different years so far so my sample size isn't large enough to speak with real intelligence. (I'll also check if 120 fixes the shape).

I'll compare methods, the shape may be regular enough that it's close enough to binomial etc, but I want to see if I can find a better way to pump through the data then importing year by year into excel and manipulating the numbers.

I haven't spent much time with Access - but given it has visual basic support I'm guessing I should be able to write scripts to figure this all out year by year automatically. I'll have to teach myself Visual Basic and Access so it will probably be a couple/three weeks before I get to it and through it.
11:07 PM Aug 20th

lidsky
That last comment was on Odds Ratio.

Charles - can you explain why you think the distribution is binomial? It doesn't look that way to me. (Note - my stats background is in electrical engineering where I know from physics exactly what the distributions are, so this is my first dive into it applied to baseball - I apologize if these are basic questions)

I have to run for a while (my kid has a dog walking business to attend to, which means I do to!)
12:35 PM Aug 20th

CharlesSaeger
Lidsky: I did K/BFP, but K/(IP*3) would also be binomial. Tom using odds ratio, which I didn't think would work myself, does get at the same thing: you need to combine the fact there's a declining number of not-strikeouts as strikeouts increase.
12:27 PM Aug 20th

lidsky
Correction. I didn't look at the whole population - I restricted the population to those with 100+ innings a ps with fewer innings a stat like k/9 isn't an accurate representation.
10:13 PM Aug 19th

evanecurb
Dave:

I think the limit on ERA is 0.00. Gibson and Walter Johnson each came within 1.1 run per nine innings of that limit. Even in pitcher's eras, when the league ERA might be as low as 3.00, they were around 35% of the league ERA. So they got to 35% vs. a limit of 0%.
8:19 PM Aug 19th

DaveFleming
I'm a little wary of using standard deviation, just because I suspect that it would favor players in smaller leagues and earlier eras, where it was a little easier to stand out. Like Babe Ruth in 1920 out-homering everyone....I think it was a bit easier to be a distance away from the league back then. I suppose it would be possible to check that, but then we'd be looking at standards of deviations of standards of deviation, which would be a little like the snake swallowing it's tail.

To evanecurb's point about OPS or ERA....the big differences between those stats and strikeout rate/percentage is that a) those other metrics have fluctuated over the years, while the strikeout rate has only really risen, and b) those other metrics aren't as close to their possible ceiling as strikeouts.

That's the thing that seems crucial to me...the ceiling issue. The highest possible OPS is 5.000....Barry came within 30% of that metric's limit, but it's pretty rare to see anyone tick past 25%. No one has come close to the limit on ERA, because the limit for that is infinite.

But with strikeouts....three guys are putting in top-ten seasons of K/9 this YEAR (Sale, Scherzer and Kluber). In about five years, if the trend continues, the league leaders will tip across 50% of the absolute limit of K/9. And unlike those other metrics (which ebb and flow), the strikeout rate continues to increase steadily.

* * *

On a completely different note, if anyone wants to discuss something that isn't about high levels of math...

For a while, I think the consensus among us nerdy types is that strikeouts are one of the better ways of measuring a pitcher's ability. But if the trends of this season hold (escalating strikeouts, escalating dingers), will we hit a point where strikeouts become less indicative of a pitcher's ability than a pitcher's ability to prevent homers?

This is a Rick Reuschel thing...he wasn't a great strikeout pitcher, but he was exceptional about keep the ball in the yard. That skill had SOME value in the 1970's, especially in Wrigley, but it might have more value going forward.

Anyway....just thinking out loud...
2:56 PM Aug 19th

CharlesSaeger
I tried it this way:

Pitcher's K/BFP - League K/BFP

divided by the League Standard Deviation of K/BFP, which is:

Square root of (LgK/BFP*(1-LgK/BFP)/Pitcher's BFP)

Tom Tango did this with a whole bunch of goalies here:

www.insidethebook.com/ee/index.php/site/comments/how_to_figure_out_how_much_talent_there_is_example_with_nhl_goalies/

Anyways, when I do this, I get Petey 1999 on top, as before, but now Ryan 1973 second. The three 2017 pitchers are still at the bottom, likely due to their lower BFP since the season isn't done yet since there will be a higher standard deviation in fewer trials. (The AL's K/BFP is slightly higher than the NL's as of this post. As K/BFP approaches 100%, it really doesn't matter if the batter is a pitcher, a DH, or a potted plant.) Pedro's 1999 is 17.3 standard deviations higher than the league average while Ryan's 1973 is 16.3 higher, giving credit to facing 62% more batters.

As for Dazzy Vance? His 1924 season beats Martínez's 1999, at 19.3 standard deviations above the mean. The rest don't.
1:33 PM Aug 19th

Mike137
yorobert wrote:
Well, Mike137, you said that unless an unanswerable question ("what if there are now more good strikeout pitchers?) can't be answered

I never said any such thing or anything that can be reasonably interpreted that way. I even suggested an approach to finding the answer to question.

yorobert wrote:
then "the question (about strikeout pitchers) cannot be answered."

That is called "logic". It is a useful skill.

yorobert wrote:
So I guess in your comparison, you're the cop who tells everybody to go home, because the keys are never going to be found anywhere.

No, I am the guy saying that you need a flashlight, because otherwise you are just wasting your time.

The claim was made in the 538 article that "If Pedro Martinez was pitching today, he’d have even more strikeouts than the Red Sox current ace." Nothing that has been said here either supports or refutes that assertion.
1:06 PM Aug 19th

yorobert
Well, Mike137, you said that unless an unanswerable question ("what if there are now more good strikeout pitchers?) can't be answered, then "the question (about strikeout pitchers) cannot be answered." So I guess in your comparison, you're the cop who tells everybody to go home, because the keys are never going to be found anywhere.
12:21 PM Aug 19th

jpc1957
Back in 2004, I had an article on this very topic appear in SABR's Baseball Research Journal. Its title was "Baseball's Most Dominant Strikeout Pitchers." If you're interested, you can check it out (and all of the other articles in that issue, too) here: https://100incheon.files.wordpress.com/2012/02/brj33-final1-69-73-400-club.pdf

I also thought that division was better than subtraction. (By the way, Baseball Digest disagreed - I originally submitted the article to BD, but they wanted me to redo it using subtraction.)

The conclusion of the article was that Vance, Waddell, Grove, Martinez, Score, Ryan, R.Johnson, Feller, and Dean all had seasons among the top 25.

The most notable absences (from a list of the top 60 seasons) were Koufax, Clemens, Seaver, Carlton, and McDowell.

I also included a similar list for relief pitchers. Not surprisingly, Gagne, Wagner, Dibble, and Lidge were at the top. I haven't updated it since 2004, but I suppose Kimbrel, Marmol, and Chapman would rank pretty high.

Anyway, I agree that a ratio is not fair to modern-day pitchers. The proper way to do it might be similar to what Bill Deane proposed when he discussed normalized winning percentage for pitchers' won-lost records (which you can find here, if you scroll down far enough:
https://www.fantasyguruelite.com/mlb-sabermetric-primer).

In any case, an interesting topic!

11:05 AM Aug 19th

MarisFan61
Of course it's easier to get strikeouts now, so of course you have to do some accounting for when it was done.
(BTW the "of course" is pretending :-) .....there are no of courses on these things, but this is pretty close to one. Because.....)

It's an of-course if for no other reason than that there has been (apparently) a decision within baseball that makes it so. And, I'd add, I think it's a decision that is largely attributable to the influence of sabermetrics.

Probably by the late '80's (give or take), it was well-known that sabermetrics was saying that strikeouts by batters were far less costly, compared to other outs, than had been believed and followed. Whether or not the change within baseball is attributable to this, it is clear that in these last 3 decades and probably with a continuing trend throughout the period, teams have cared less and less about batter strikeouts; maybe by now it has plateaued, maybe there have been blips in time where it reversed a little.

This thing (assuming it's true) has operated on at least 2 separate levels:
-- Batters' approaches, and
-- Teams' selection and development of players.

Those are separate, and they add together.
The first of those, of course. If players know or are told that it doesn't matter that much if you strike out, just take your best swings, they'll be more likely to strike out.

And if teams believe in this, they won't mind a young hitter's strikeout tendency that much when they select him, they won't be so inclined to pick players because of their not striking out, and in the development of players they won't be so inclined to emphasize contact over power.

One reason that this "of course" isn't an of-course is that other things also change over time, and I don't know that there haven't been other factors, unrelated to this, that work in the other direction. Like (just making this up; I don't mean I think it's true), if batters' eyesight or hand-eye coordination has gotten better, that would counter the above. But it would seem (IMO) that the above is such a distinct and strong thing that it would be hard for other things to cancel it out.
10:02 AM Aug 19th

Mike137
steve161 wrote:
if I understand you correctly, your point is the same one Stephen Jay Gould made about .400 hitters: it has to do with a change in the level of competition. Is that a fair statement?

I don't know, since I don't know what point Gould made.

My point is that if you want to normalize for something that has changed, you have to know what has changed. What has changed to cause strikeouts to be more common than they were 20 years ago? I can't answer that question. Without being able to answer that question, we can't make a meaningful comparison of Sale to Pedro.
8:03 AM Aug 19th

lidsky
So for a few seasons I calculated the leaders in the following way: (K/9-LGAVGK/9)/STDDEV(LGK/9)

Don't know how well that will read, so in words. I found the distribution for each year of interest in terms of Standard deviation. I then took the players K/9 subtracted the league average K/9 and divided it by the Standard Deviation for that year.

Thus I get how many stddev above average somebody is. It is cumbersome so I haven't done many yet.

Pedro 1999 was 4.8 std deviations above the league average of 6.19
Gooden 1984 was 4.12 std deviations above the league average of 5.26

I'll do the rest of the list and add to Readers Posts, but it will take me a bit. I looked at a few and the stat gives similar leaders as in the article.

Then I decided to check the issue when the strike out rate is low, so I looked at Vance's 1924 season:

5.32 Std dev over the average of 2.6

So at first I thought - okay not a good method for those early years. But i noticed nobody else was near Vance, where in terms of Std Dev, there were a couple of players closer to Pedro and Gooden. So I thought, maybe Vance really was comparably that good. To check the small number issue, I checked 1922. Interesting result in that the top pitcher, Pruett didn't blow away the league even though the numbers were low:

2.83 Std dev over and average of 2.79

I'll look at a lot more when I have time, but the metric seems to pass the sniff test of able to work for both eras in that Pedro had struck out 21 more batters he would equal Vance's 5.32 std dev above average.

Perhaps, in terms of K/9, Vance was better than Pedro. (Banish the thought says this Sox fan!)


1:26 AM Aug 19th

SteveN
It even sounds dumb to me when I think it but would calculating the number of batters not struck out be a better indicator? Its late and I am not thinking too clearly right now so maybe this is extra dumb.
10:54 PM Aug 18th

Mike137
A police officer came upon a drunk anxiously searching the sidewalk under a streetlight. The office asked the man what he was looking for. "
My house keys", the drunk replied.
The cop asked "are you sure you lost them here?"
"No I lost them in that alley."
"Then why are you looking for them here?"
"The light is better".

yorobert wrote: "all we can do is compare players to their contemporaries to see which players were the best of their peers, and by how much. Thus, I think this article is on the right track."

Only if the drunk is on the right track.

This article, and the 538 article it referred to, starts by making an assumption that is not only unproven, it is unexamined. That is no way to make progress in understanding.



10:10 PM Aug 18th

yorobert
Mike137,
Your objection is true, but it is true for every comparison between eras, so all we can do is compare players to their contemporaries to see which players were the best of their peers, and by how much. Thus, I think this article is on the right track.

I'm guessing that cross-generational comparisons begin to falter when the numbers/statistics are so drastically different between eras, however, resulting in p-values that prevent any sort of accurate conclusions. I wish I knew enough about math to know how to do this myself.
8:03 PM Aug 18th

MarisFan61
Good and interesting and thorough work.

But, I'd argue with one of your points:
It's far from clear that it's better to "divide" than "subtract." In fact, I think it's better how 538 did it -- i.e. subtract. I agree that your eventual lists are better than theirs, but I don't think that the "divide/subtract" thing is a reason.

And funnily, the main reason I think 'subtract' is better is this thing that you said, in a different context:

"The possible strikeout percentage has an upward limit. As the overall percentage of strikeouts rises, the game gets closer and closer to that upward limit."

This makes it hard to compare pitchers from different times in any straightforward way, but I think it causes more of a problem if you use division than if you use subtraction, because when the league rate is very low, you can get very high results, even without exceeding the league rate by that much -- far higher than it's possible to get in a higher-strikeout time.


4:45 PM Aug 18th