Certainly Mongo is correct. League quality is a huge factor in doing correctly what we are trying to do. I think it is fair to say that, rightly or wrongly, the only real league quality factors on which there is consensus of any kind are:

• Any and all 19C MLs are weak compared to any 20C ML. My evidence for this is pretty simple--the timeline--which I am guessing about half of our voters employ as a "strong" factor.

• The AA was weaker than the NL pretty much throughout its lifetime. Again, I would guess about half our voters thought so and factored it in.

• The NL of the 1910s or thereabouts was weak and I would guess that maybe one-third to one-half of our voters use that as a factor.

• Finally, anyone who relies heavily on WARP3 is undoubtedly accepting a variety of additional league quality adjustments as outlined by TomH above.

Generally, voters here might be thought of as being in three groups--those who ignore the whole issue, those who employ adjustments 1-2-3 above, and those who use WARP3.

I guess I have two questions/observations about all of this.

1. If my typology is anywhere close to correct, do items 1, 2 and 3 comprise a pretty selective use of league quality considerations? Is it fair to use these considerations on such a selective basis? I worry. And BTW I am as guilty as anyone here, having used the AA discount in the 19C. (I am among those who ignore 19C and NL 1910s quality, however.)

2. And secondly, among those who use WARP3, are you really sure the league quality (timeline) adjustments make sense?

3. And to anyone who uses any of the above discounts, is this measuring value or ability? Why is it correct to measure and reward ability rather than value?

PS. to me, the idea that the AL was weaker in '30 and '32 than it was in '44 casts this whole methodology into question. Not to diminish the work TomH has done, it sounds like a reasonable approach. But is there some reason--small sample size, the intrusion of other variables like aging or injury that cannot be properly accounted for--that this method might end up failing?

>conclusions from this data
NL was stronger thru 1936
AL caught up 1937 thru 1941
AL was weaker in 42 and 43
real 'war years' were only 44 and 45
small weakening effect during Korean conflict
AL thru mid-late 1950s was actually Weaker than it was in 1940.
NL strongest by late 1950s, but not by much over earlier years.

As for Tom's conclusions and the Cramer study, if I read it correctly, suggests that the AL never really caught up, other than an occasional one-year blip, from 1931 through 1966. Cramer also has the AL and NL in '44-'45 as weaker only than the period from '40-'43 but better than '39.

BTW Cramer also supports my long-time contention that the NL in 1878-79-80-81 was essentially as good as it was during the AA era.

Meanwhile, the FL was only about as good as the AL-NL in '01-'02-'03.

If integration had an impact on league quality, Cramer sees it as very gradual. In fact, the AL was never again as good as in 1947 until 1958, while the NL was not as good as it was in 1946 until 1952 (blip) and then 1955.

Maybe someone can explain this to me. The AL was better 1903-1919 than it would be again until 1929, Otherwise it was worse through the 1920s than in 1903. Not only that but generally the same is true of the NL though by fairly small margins. Why would the caliber of ball have declined in the '20s versus the '10s?

So anyway, this is tricky stuff. And finally, Cramer sees virtually no difference between the AL of the 1910s and the NL of the 1910s.

To pile onto Sunnyday's point about league quality questions. Anyone who uses Chris Cobb's conversions to assess NgL players is also accepting league-quality adjustments. He is adjusting downward from the NL (mostly). So this accepts two premises:
1) That there is a difference in league quality between the NgL, CWL, MxL, etc...and the white majors.
2) That the differences in league quality between the various leagus of color and the white majors do not vary, or at least very little.

Additionally, we as a group have arrived at no consensus regarding the relative league strengths of the various leagues of color compared to themselves (that is 1927 NNL vs. 1926 NNL or 1937 NAL vs. 1948 NAL), so we're also selectively operating under the assumption that, by and large, these leagues were of a consistent quality.

Whether these are accurate premises is certainly open to debate, but until we get our hands on the work of the Hall's committee, we're not likely to have a true sense of it because we'll lack the publicly available comprehensive data to do anything about it.

Doc, right on. I ovelooked those adjustments. And I will also add, what about MiLs? Some folks add value for, say, Earl Averill's or Gavy Cravath's AAA years. Those are adjusted for competition, presumably.

My rule of thumb has always been: I adjust for leagues that are understood NOT to be the best of their time--e.g. AA, MiLs, NeLs, MxL, etc.

I do not adjust over time because 1) we really don't know what the proper adjustment was and 2) a pennant is a pennant, meaning a ML pennant.

I do not adjust over time because 1) we really don't know what the proper adjustment was and 2) a pennant is a pennant, meaning a ML pennant.

Curiousity: Will you adjust for integration? Seems that a white major leaguer in 1940 is facing a very different competition set than a white National Leaguer in 1960 that has very little to do with timelining. Certainly this has to be adjusted for, right?

My rule of thumb has always been: I adjust for leagues that are understood NOT to be the best of their time--e.g. AA, MiLs, NeLs, MxL, etc.

I do not adjust over time because 1) we really don't know what the proper adjustment was and 2) a pennant is a pennant, meaning a ML pennant.

I agree - leagues that were CLEARLY less than major compared to the league/years around it - UA, some years in AA, PCL, Negro Leagues, AL 1901-1902, Federal League, MLB 1944-45, etc. should be adjusted for.

Curiousity: Will you adjust for integration? Seems that a white major leaguer in 1940 is facing a very different competition set than a white National Leaguer in 1960 that has very little to do with timelining. Certainly this has to be adjusted for, right?

As was mentioned above, the integration impact was very gradual. As such, my inclination right now is to not adjust for it.

Quick n dirty check:

One way some (such as Steve Gould) have suggested to estimate league quality is by the ability of the best players to dominate a league.

I took bb-ref's leaderboard, and found the 5th best OPS+ and ERA+ leadeers from the AL in 1931-40, and again from 1951-60. I averaged the results for each 10-year period.
lg years.. OPS+ ERA+
AL 1931-40 145 136.5
AL 1951-60 140 132

This implies the later AL was stronger, as the best hitters and pitchers stood out less from the average. This is different than the conclusion arrievd at by the BP ##s, which would yield us to believe that (partly due to Korean war yeras) the AL was actually slightly Weaker in the 1950s.

-----

2nd subject: While integration WAS gradual, I can't see how we cannot conclude that by 1959 the effects of having blacks play had raised the average level of play over 1946. We may debate whether the expansions of 61/62/69 give it all back I'm sure. I will most certainly adjust for this, in the sense that I might take the 8th best candidate of the 1950s era over the 7th best of the 1930s. If that makes me an enemy of the 'pennant is a pennant' crowd, so be it.

To me, league adjustments over time are pretty much impossible because you've got two different variables going.

One is the pool. As the size of the population pool from which players are drawn increases, in theory so does the average skill level among your sub-pool--i.e. major league players. This is a theory which (in theory) the Cramer study and others verifies, though every study has come to vastly different conclusions about the size of the differentials.

Two is the individual player. There is no logical reason why the individual player should be discounted the same as the average player.

Or to put this in the terms Tom used above in #118:

>AL 1931-40 145 136.5
>AL 1951-60 140 132

>This implies the later AL was stronger, as the best hitters and pitchers stood out less from the average.

But the degree to which the best hitters and pitchers stand out from the rest is a function of both. How do we know that "the rest" in the '50s were 2-2.5% better than "the rest" in the '30s, rather than that "the best" were 2-2.5% better? We don't.

Four conclusions:

1. The pool argument is fairly compelling, especially when combined with what we know to be better health and training, in that the average player today IS better than before, and in fact the average has probably increased on a fairly level curve over (a long) time.

2. But the increase has been so gradual that we sometimes see league's being worse than they used to be. In the micro, the curve is in fact not a level curve at all.

3. But great players are outliers and their greatness is not particularly illuminated by comparison to the average, especially when the average at a micro level is so uneven and unpredictable. I mean, do we know that Joe Medwick faced better opposition than George Sisler (etc. etc.). Not with much certainty at all.

4. Timeline adjustments generally vastly overstate the change and ignore the fact that great players are not dependent upon the average in the first place.

For all of these reasons, if there is any science to rating and ranking ballplayers, timelining and league adjustments are not part of it but are purely art. They belong in what Bill James calls the "bullshirt dump." I don't mean to say they are "bullshirt," but that they are art and not in my view something you can build into a statistical evaluation with anything close to the kind of precision that is claimed for any statistical evaluation system worth using.

Sunnyday, this point is something I've wondered about often

. The pool argument is fairly compelling, especially when combined with what we know to be better health and training, in that the average player today IS better than before, and in fact the average has probably increased on a fairly level curve over (a long) time.

I see two points of view on this.

1) For all the reasons you mention and many more, in absolute terms, today's players are better than their yesteryear peers.

2) But in relative terms, the average player is, by definition, no better than the average player of yesteryear.

We often hear it said: would So and So be as dominant today as back in the day if they got put into a time machine to today. Reframing this question: Would the average player from 1896 or 1928 or 1952 be an average player today.

My answer is emphatically yes, assuming that he had the same access to training, healthcare, etc... that any player today does.

To put this another way, I imagine that the talent of average players is essentially always the same (assuming a large enough pool to draw from to make it statistically likely), it's the surrounding environment and their access to ideas and technology that make them "better" than their historical counterparts.

I'm a strong anti-timeliner, so you will be surprised when I say--today's average play is clearly better than he used to be, probably as recently as the 1970s and surely anytime before that.

1) He is better in absolute terms because he comes from a larger pool (today, that means including Latin America and Japan) and because of better health, training, diet, etc. etc. So, no, the average player from earlier eras could not step off the time machine and compete with these guys.

2) He is even better in relative terms because the superstars do not dominate him as much as they used to.

There are two problems applying these concepts. First, we are not electing average players, we are electing star players and we cannot assume that star players of any given era are better than star players from earlier eras by the same percentage that average players are (even setting aside the fact that we don't know with any certainty what that percentage is).

Second, we are meant to be "fair to all eras," and I have always taken that to mean that we are electing value and not ability.

Of course, the average player has more ability today, but more importantly he has more value because the star player does not dominate to the same degree.

But what of the star player? Well, maybe he has more ability than star players from earlier eras, but it is my contention that we do not know this for certain. The truth is, however, that he is less valuable today because he does not dominate. This is the conceptual problem for those of us who eschew the timeline. I've been accused of using a reverse timeline because I don't blanket-adjust for the fact that it was easier to dominate in the old days.

My PHoM includes some 19C players not in the HoM--Childs, Jennings, Williamson, C. Jones, H. Wright and Bond--but the HoM also has 19C players not in my PHoM--Sutton, Stovey, Kelley, Keeler and Galvin. I have only 1 more 19C player in my PHoM than the HoM has.

The fact is I do adjust for the fact that it was easier to dominate in the 19C. I just do it on a case-by-case basis, not by applying a blanket adjustment that derives from the performance of the average player rather than from the performance of the candidate himself. Like I said, it is art, not science.

Sorry for all those italics everyone, I missed a </em> somewhere.

One very important feature of the WARP-3 league quality analysis is missing here.

A regression line is calculated from the numbers for 1947-2004. All measurements are then made with respect to that regression line.

What this means is that there will be no significant change in overall quality post-1950, just measurements of how expansion impacted the "local" quality (temporally speaking).

This also means the "timeline" pre-1950 is not really a "timeline" per se, but a measurement of how the local evolutionary rate differed from the modern (post-war) rate. The showing that 1930 was significantly weaker than 1940 really means that baseball quality evolved faster during the 1930's than it has in the modern era. I assume that this is due to the rise of the farm systems and the harvesting of the last independent top-minor league, the PCL.

JimD,

Can this statement really be accurate in light of integration?

The showing that 1930 was significantly weaker than 1940 really means that baseball quality evolved faster during the 1930's than it has in the modern era. I assume that this is due to the rise of the farm systems and the harvesting of the last independent top-minor league, the PCL.

The top minor league talent was at least available for purchase prior to and during the development of the farm system, but no players of color were allowed across the line. Were integration instant, it would have had a remarkable effect on replacement level, at the major league level, but even at its own pace, it should have had a simultaneous effect on the minor leagues where blacks were effectively trapped by MLB quotas and whatnot. This should have, therefore, had a bottlenecking effect on talent flowing through the minors, allowing a smaller percentage of filler (read: MLB replacment level) players to rise that high. If this postulate is correct, then integration at the minor league level should still have raised the level of major-league play.

So I guess I'm wondering if the great progress in the 1930s reported by BP's system is actually some kind of recognition that pitching found a little more equilibrium with hitting after the bat-tastic twenties (esp. in the NL).

I think people underestimate two effects here:

1) the power of the minor league farm systems which improved the quality of the average and replacement level players in the majors. The top stars practically always found their way into the majors eventually; average, sub-average, and bench players were much improved by the systematic organization of talent, which replaced the "star search" system of pre-1925.

2) economic concerns. Was it Gadfly that argued that more young blacks became players in the 1930's then the 1960's due to lack of alternative opportunities? (Frankly, I think he's underestimating how slowly doors opened post-1960, but that's irrelevant to my argument.) The same argument applies to the white population during the 1930's and the Great Depression. Between the growing popularity of baseball during and after the Babe Ruth era, and the lack of alternative employment opportunities, I'd bet that more kids tried to become baseball players than at any time previously (and possibly since).

If we wanted to create a comprehensive list of all the factors that have influenced the quality of play by the hypothetical average player, or we could express it as the total quality of the pool... What all would be on the list?

1. The size of the pool, which itself consists of a lot of things including the population of all the various relevant ethnic, racial, national, etc., groups that have comprised the pool at various times.

2. The quantity of playing experience available at X age and the quality of coaching and/or peer instruction available from birth right up through attainment of ML status. All the things that enhance tangible skills and knowledge of the game.

3. Health, diet, training, etc., all the things that optimize performance over and above raw "skill" and knowledge of the ins and outs of the game.

4. As jimd says, the economic attractiveness of baseball as something more than a youth activity or hobby.

5. And there are all the other things that make the game attractive, like media coverage, groupies...??? Or less attractive, like the counter from other sports like basketball, football, maybe boxing in earlier days, not to mention the opportunities or lack thereof for various racial and ethnic groups to earn a living in all the other areas outside of sports.

6. And all the things other than racial segregation that enhance access to the game, whatever that means.

7. Not sure if it's fair to include things like improvements to facilities and equipment, but maybe....

8. And of course, as jimd says again, though maybe this is related to #2 or even #6, the farm systems and/or all the various factors that create an efficient market for the evaluation and acquisition of talent.

9. And of course you've also got the size of the employee market to be filled up which is one of the few factors that militates against more or less constant improvement.

I'm probably missing another 9.

But I guess my point is that this is a massively complex social problem and I don't pretend to be even in the ballpark, pun intended, of having a way to really deal with it.

The other question is whether this is all just hypothetical, as in a hypothesis, or whether we think that the improvement has in fact been measured, really and truly, by Cramer and others. If so, then who cares about cause and effect. But if we don't really trust the measurement and we fall back to cause and effect arguments, then we're really only advancing hypotheses. Plausible ones, but hypothese nonetheless. And I'd guess we understand evolution a lot better in terms of cause and effect.

And I'd guess we understand evolution a lot better in terms of cause and effect.

Except in Kansas.

[Runs for cover before anyone can comment...]

don't even go to Kansas, toto....

I found the Cramer study. It is in units called Batter Win Average, which are in runs per plate appearancs. Since one out into a single equals about .73 runs, I will divide by that factor to turn them into EqA.

The study did not publish a table, only a difficult-to-interpolate chart. :(

Beginning in 1931 thru about 1950, the NL was stonrger than the AL by .015 in EqA most years. The gap rose slightly toward the end.

League quality overall steadily increased throughout this time, excpet for a lull in the late 1930s.

1944 and 1945 are seen as significant 'war years', where avg skill was down by about .015. 1943 had no effect. The NL in 1945 was stronger than the AL in 1942!

From 1950 to 1960, the NL increased its lead even more, to about .022, after which the AL would catch back up until it pulled even in about 1975.

The two metrics agree on NL superiority in the 30s and 50s, and on the fact that the 'war discount' is not huge (maybe 8 htis in a year for a batter).

They disagree on league qual in the 1940s, and the rise in level of play over time.

Well, there you have it. Interpretation will truly be in the eyes of the beholder.

Random unrelated question that I didn't know what thread to put in:

Has anyone studied (or is anyone even aware of) the California Winter League? I had never heard of it, but it was an integrated league that had lots of star players appearing in their respective off seasons. The individual teams weren't integrated, but there was always at least one or two black team playing against a group of three or more white teams.

In the 1926-7 season, for example, the Philadelphia Royal Giants put on a team with Bullet Rogan, Andy Cooper, Will Foster, Turkey Stearns, Biz Mackey, Rap Dixon, Crash Holloway, Willie Wells, and Newt Allen.

Obviously -- none of the opponents being the 1927 Yankees -- they won by a lot. But the opponents were not all career minor leaguers, either. They were all at least half major leaguers and the rest top PCL guys. One opposing team in 26-27, for example, had both Meusels, Fred Haney, Johnny Rawlings, Chicken Hawks, and Ping Bodie. Not pushovers.

Looks like an interesting place to look for Negro League comps, in a league that I really didn't know anyting about.

Mongo - thanks for getting this rolling again.

The big issue from what I've heard of the Cramer study is that with each comparison, you are introducing a minor error to the study - nothing is perfect. So from year to year this isn't an issue. However, as the study grows, the errors compound, so while looking at 1910-11 isn't an issue, looking at 1910-1998 is an issue. You'll have to find a way to avoid this for your study to work.

Tom - Regarding the Gould thing - Chris Dial was going to do a presentation this year on the 'domination' idea at SABR, but his proposal was lost and he wasn't notified until it was too late to resubmit. But essentially, he's convinced that it's easier to dominate now than it has been in awhile. Look at Greg Maddux 1994, or Pedro 1999-2000, or Barry Bonds, etc.. Everybody seems to think the players now are the best ever, but if so, why are we having these extreme deviations from average?

High run enviroments are very influential on extreme deviations, perhaps even moreso than league quality issues. So I'm not convinced that that's the right way to go, even though most seem to think it is.

"A regression line is calculated from the numbers for 1947-2004. All measurements are then made with respect to that regression line.

What this means is that there will be no significant change in overall quality post-1950, just measurements of how expansion impacted the "local" quality (temporally speaking)."

Jim I thought 1947-2004 were treated on their own merits (meaning it is NOT a straight line adjustment year by year), but the straight line those points form was used to project back BEFORE 1947 (for overall MLB quality, the difference between the leagues is also a factor).

Am I misunderstanding your post?

"Regarding the Gould thing - Chris Dial was going to do a presentation this year on the 'domination' idea at SABR......he's convinced that it's easier to dominate now than it has been in awhile. Look at....Pedro, Barry Bonds, etc.. Everybody seems to think the players now are the best ever, but if so, why are we having these extreme deviations from average?"
--
Too bad. I have been waiting for someone (I know, Tom, why don't you do it yourself ya lazy bum) to address this on a good platform. It's Very tempting to put Barry (and Rocket) at the top of my theroetical all-time list, but the number of guys putting up surreal numbers gives me pause; it would be good if we had a few more years of data (= perspective) before annointing the current superstars status in Valhalla.

This is extremely provocative, as there are two very very deeply ingrained theories in obvious conflict.

1. Quality has increased and continues to increase.

2. The degree to which individual (superstar) players dominate is the inverse of quality.

My gut feeling is that 1 remains true and there is a problem with 2. IOW, the reason individual players could dominate in the old days was indeed quality (low, as posited in theory #1). The reason certain players dominate today is something different.

What could that something different be? Two possibilities:

1. Physical training, diet, steroids, etc. etc.

2. Mental ability. The corollary to this of course would be provocative, and that is that in the old days individual players lacked the mental X to challenge the status quo in terms of playing and training and other accepted wisdom, Babe Ruth being the exception that proves the rule. Today, as our society has become more individualistic, certain individuals have found new paradigms for preparing and playing and conceptualizing the game in its mental aspects.

The other explanation is that Bonds and Clemens represent too small of a sample to be significant. That once upon a time "dominance" was something that was shared by, say, one or two or even three dozen players in each league, whereas today dominance to a similar degree (in SDs?) is only shown by a very very few players.

Or maybe the old-timers really were better, as was often claimed when I was a kid!?

The atypically superior ERA+ figures put out in the last 12 years I think can be attributed to more 'below-average' pitchers getting more mound time - the 11th and 12th hurlers on the staff.

The five-man rotation surely has had an effect as well. Tom Seaver, by consensus the last 4-man rotation era pitcher with the greatest career value, only has an ERA+ of 127, yet he won four ERA+ 'titles', once with 'only' a 142. It does seem it was much more difficult to put up eye-popping numbers in 1975 than in 2000.

If anyone still checks this thread, I'm having a real problem applying a timeline. Basically, I can see two ways to do it:

1. Simply multiply a player's wins above replacement player (WARP) by the league difficulty. In a league 10% tougher than average, 10 WARP become 11, while in a league 10% easier than average, 10 WARP become 9. The obvious problem with this is when you get to players below replacement level. If, say, Cristian Guzman was (hypothetically, I don't remember the real number) minus 3 WARP in 2005, which was 15% more difficult than 1960, it's absurd to multiply that by 1.15 and give him negative 3.45 WARP. On the contrary, his -3 WARP in 2005 (or whatever) probably would have been like -2 WARP in 1960, not -3.45.

2. Raise or lower the league replacement level in relation to difficulty. This solves the problem of players like Guzman, but totally biases your results towards career guys over peak guys in hard leagues and peak guys over career guys in weak leagues. If you lower the replacement level 15% (or whatever it is), a guy who gets 1 WARP will suddenly be worth 3 (a 200% increase), while a guy who gets 10 WARP will be worth 12 (a 20% increase); conversely, if you raise it 15%, guys who were a steady 4 WARP a season for 20 years will lose 50% of their value, while someone who had 8 years at 10 WARP will only lose 20% of his value.

Both of these seem really suboptimal. Is there maybe a way to combine the approaches? Or just a different way to do it entirely? Let me know if you have any ideas.

Thanks,

Dan

I'm not a fan of timelining. At all.

That being said if you must . . . and jimd is the better to one to ask than me on this . . . you need to do both.

You need to raise/lower the replacement level - and if that drops the player out of the league, then he gets nothing. If it moves him into the league he gets more PT. Then with what is left after that, you apply your percentage gain/loss.

Kind of tough to do. Short of that, I think you should go with adjusting replacement level.

"If you lower the replacement level 15% (or whatever it is), a guy who gets 1 WARP will suddenly be worth 3 (a 200% increase), while a guy who gets 10 WARP will be worth 12 (a 20% increase); conversely, if you raise it 15%, guys who were a steady 4 WARP a season for 20 years will lose 50% of their value, while someone who had 8 years at 10 WARP will only lose 20% of his value."

That's the cost of doing business. If a guy was a steady 4 WARP in a weak league, he may deserve to lose half his value. That 4 wasn't right in your mind anyway, that's why you are adjusting it. So don't get married to it, or worry if you have to change it significantly.

But again, don't timeline. A pennant is a pennant.

I would add that while I'm against timelining, I would (and do) apply the above approach to things like adjusting for the AA being weaker than the NL in a given year, war years, etc..

By the way, Italics Man is my #####.

Forget I said timelining, then. But yeah, what about adjusting for the Federal League, for example? Let's take, say, Eddie Plank's 1915 in the Federal League. Just to keep it simple, Plank gave up 75 runs that year in a 3.85 R/G league. Using the BP formula replacement RA that year would be 4.61, or 137 runs in Plank's innings, making Plank 62 pitching runs above replacement. Now let's say I have the Federal League as 85% the difficulty of the AL and NL (assuming they were equal then, which they weren't) that year. How would I implement the adjustment? I just can't seem to make the math work myself.

Say "competition adjustment." (Wink wink nudge nudge say no more say no more)

Here are the various approaches that occur to me.

1. Straight multiplier. 62 pitching runs *.85 = 53 pitching runs.
2. Straight replacement level. 137 replacement runs *.85 = 116 pitching runs, making Plank 116-75 = 41 pitching runs.
3. Replacement level jiggered to look like a multiplier. If we want Plank to be 53 pitching runs (as if it were a multiplier), then his replacement must have allowed 75 + 53 = 128 runs, which makes a replacement RA of 4.31. Use 4.31 as replacement RA for the whole league.
4. Combination of straight multiplier and straight replacement. Repeat step 2 to get Plank's 41 runs over the higher replacement level, then multiply THAT by .85, making him just 35 runs above replacement (ouch!).
5. Combination of straight multiplier and jiggered replacement. Repeat step 3 to get 53 runs above replacement, then multiply by .85, for 45 runs above replacement.
6. Half-and-half. We want Plank to have the same 53 pitching runs he'd have using a straight multiplier, but with half the adjustment coming from changing the replacement level. Thus we use a 4.46 RA replacement level, which is 132 runs in Plank's innings, and multiply that by .925, giving us 53 runs above replacement. Use 4.46 and .925 across the league.

It seems to me the best thing to do would be to actually look at the data used to derive the league difficulty factors. For example, in Plank's case, I have him at 7.3 wins above replacement in 1911, 6.4 in 1912, 6.4 in 1913, 3.2 in 1914, a Federal League spike to 8.2 in 1915, and back to 5.2 in 1916. In this one case, that seems to me to be decent circumstantial evidence that Plank was probably a 5.6 win pitcher or so in 1915, or 46 runs above replacement, which would anecdotally suggest that Option 5 is the way to go.

If a guy was a steady 4 WARP in a weak league, he may deserve to lose half his value. That 4 wasn't right in your mind anyway, that's why you are adjusting it. So don't get married to it, or worry if you have to change it significantly.

Yes.

1. Straight multiplier. 62 pitching runs *.85 = 53 pitching runs.

No. Like Joe said. This would mean no adjustment for someone at league average. Your own misgivings show that leagues do not differ in this way. If you arrived at 85% by asking what is the best uniform multiplier for runs above league average then you took the wrong road and it would be a miracle if you arrived at the right place.

Phillybooster 10 months ago:
Has anyone studied (or is anyone even aware of) the California Winter League? I had never heard of it, but it was an integrated league that had lots of star players appearing in their respective off seasons. The individual teams weren't integrated, but there was always at least one or two black team playing against a group of three or more white teams.

One of William F. McNeil's books features the California Winter League. I don't know it but there is a copy in a nearby used book store.

Mongo #107:
I don't know what relationships you are measuring --something about league, year, OPS+, age at the player level.

I clearly know that a straight multiplier isn't the way to go, but it's not true that there would be no adjustment for a league average player. A league average pitcher would be around 22 runs above replacement in the 1915 FL, so with a straight multiplier that would drop to about 19.

With Pete Browning and Charley Jones receiving strong support in recent HoM elections, now seems like a good time to revisit the issue of the quality of the American Association. Like many voters, I joined the project after the main discussion of this issue (even though I’ve been voting since the 1931 election). Although I was aware of the general conclusions on AA league quality, for me the issue was tangled up with other, more controversial issues (such as the WARP “for all time” adjustments), and I didn’t have a good sense of the quality of the evidence supporting the conclusions.

Past studies (such as a 1980 Dick Cramer study and BP’s “for all time” adjustments used for WARP2 and WARP3) have attempted to measure both (a) the relative quality of the leagues at a point in time, and (b) the changes in overall quality over the years. These studies have been controversial, especially with regards to the second issue—many people (including me) are skeptical that these studies can appropriately measure changes in quality of play over time. However, it seems to me that comparing quality of two leagues at a point in time should be simpler and less controversial than measuring the changes over time (provided, of course, that there was an adequate number of players moving between leagues). So I decided to undertake a study comparing batting performance of players who switched between the AA and the NL, covering the period when Browning played in the AA (1882-89). As my measure of performance, I’ll stick with simple conversion factors for the components of plain vanilla OPS+, OBP+ and SLG+, thereby hopefully avoiding the uberstat wars.

My comparison sets consist of players with at least 300 PA in the AA for year X and at least 300 PA in the NL during the period X – 3 to X + 3. For example, for 1882 I looked at all players with at least 300 PA in the 1882 AA and at least 300 PA in the NL during 1879-85. I found six players who met the criteria. (Fortunately, for the other years there were 16 to 30 such players.) For each player, I calculated his OBP+ and SLG+ for both leagues. I then averaged across players by taking the geometric mean (which has nice mathematical properties for averages of ratios). If the leagues were of equal quality, I’d expect the ratio of the players’ mean OBP+ and SLG+ in the two leagues to be about 1. On the other hand, if the AA has lower quality, I’d expect the performance to go up on average when a player moves from the NL to the AA and to decline when he moves the other direction. Thus, the ratios of mean OBP+ and SLG+ will be my estimates of the conversion factors.

I made two exceptions to the rules outlined in the last paragraph. For 1890, the Players’ League is generally considered the premier league in terms of quality, so I substituted the PL for the NL for that year. Also, for the 1889 AA, I set the comparison period as 1886-91 rather than ending with 1892, since the NL’s quality of play is generally considered to have risen in 1892 with the contraction to one major league. Other than these two exceptions, I am ignoring other ups and downs in the quality of NL play and simply treating the NL as the benchmark against which the AA is to be compared.

Probably the most important caveat for this study is that it doesn’t adjust for player age. Fortunately, a lot of players moved back and forth between the AA and the NL, and they included young, mid-career, and older players. Nevertheless, when you see a pickup in performance that’s bigger than other players in the group, chances are it’s a young player, and when you see an unusually large decline, more often than not it’s an aging veteran. I’m not going to try to adjust for age (I’ve spent way too much time on this study as it is), but I’m listing all of the comparisons in case one of you wants to do more analysis with this dataset. I’ll also include a few historical notes about the AA, mostly drawn from Total Ballclubs.

1882.
The AA’s first season, it consisted of six teams (Baltimore, Cincinnati, Louisville, Philadelphia, Pittsburgh, and St. Louis), at least four of which had existed earlier as independent teams playing outside any formal league. Our comparison set suggests that for the first season, relatively little recruiting was done from NL ranks, as only six position players had significant, recent NL experience. All six hit better in the AA than in the NL, most of them hitting much better. Their differences in batting performance between the two leagues are consistent with the modern difference between the majors and the Double-A level. The AA that first season clearly should not be regarded as playing at a major-league level.

AA quality (part 2)

1883.
The AA expanded to eight teams (adding an existing independent team from New York and a new franchise in Columbus). They aggressively recruited players until the signing of the Tripartite Agreement with the NL and the Northwestern League, in which the leagues agreed to respect each others contracts and allowed each team to reserve eleven players. The data suggest that the AA’s quality of play was significantly improved compared to 1882, but remained lower, relative to the NL of that era, than the quality of a modern Triple-A league relative to the majors.



1884.
In response to the organization of the rival Union Association, the AA expanded to 12 teams, adding franchises in Brooklyn, Indianapolis, Toledo, and Washington. Predictably, the quality of play slipped a little relative to the contemporary NL; it remained between what would now be considered a double-A and triple A level.



1885.
With the threat from the UA past, the AA contracted back to eight teams, dropping Columbus and the new teams that had been added in 1884, except for Brooklyn. The data indicate that the AA’s quality of play picked up. In 1885 the AA stood, relative to the contemporary NL, about where modern triple-A leagues stand relative to the majors.

AA quality (part 3)

1886.
There were no changes in teams, but the data show a continued improvement in batting quality relative to the NL. This is the first season for which the AA’s quality of play was clearly “major league,” in the sense that the gap between the AA and the contemporary NL was smaller than the gap between a modern triple-A league and the majors. It appears that at this point the improvements were primarily coming from recruiting and developing young players, since there was not a significant flow of quality players from the NL to the AA.



1887.
The Pittsburgh franchise moved from the AA to the NL—they had finished second in the AA in 1886 and would finish sixth in the NL in 1887—and was replaced by a new team in Cleveland. The data indicate that the AA’s quality of play reached its apex, only slightly lower than the NL.



1888.
The New York AA franchise, which had been cannibalized by the NL New York Giants when both teams were owned by the same ownership group, folded and sold its remaining good players to Brooklyn; the league awarded the franchise to weak team in Kansas City. The data indicate that AA quality of play slipped this year.



1889.
The Cleveland team left the AA and joined the NL (finishing in sixth place in both leagues). It was replaced by a new franchise in Columbus (which also finished sixth). The data show the quality of play rebounding slightly—the AA continued to be slightly below the contemporary NL, but still well above the triple-A level we now think of as minor league quality.



<Summary</b>
The AA began life in 1882 with quality of play equivalent to a double-A minor league. The league’s quality gradually increased, but not until 1886 did it reach “major league” levels (that is, demonstrably better than a modern triple-A league). Its quality peaked in 1887 and remained a little below the NL throughout 1886-89.

Quick Q on your methodology: how can you include, for example, Tip O'Neill in your set of players used to determine translations, since he never played in the NL from 1884-1889? Are you just using his 1890 PL and 1892 NL seasons to determine his "true" talent level relative to the NL and comparing those to his AA performance? Isn't it possible that he actually just peaked in 1887, and was a much worse player by 1890 and 92?

Sorry, I should have read your description of your approach before posting, which answers the question. Nonetheless, I think maybe the cost of a smaller sample size might be worth the benefit of removing guys like O'Neill from your calculation.

Brent, this is excellent work! Thank you!

I disagree with Dan on removing any players from the study: a larger sample, and a full representation of the acutal data, is preferable. When I was doing the NeL conversions, I found that adjusting for aging patterns did make a difference of a three percentage points in the outcome, but that was with a much smaller data set. I am pretty confident, therefore, that yours would not change more than 2 or 3 percent, unless there are years in which, by chance, the players are all old or all young.

I'll be curious to see what Pete Browning's and Charlie Jones's OPS+ scores look like with these modifiers applied.

Nice job, Brent. Your study corroborates the less precise discounts that were used here in the early elections when evaluating Stovey, et al. Except for 1885, which appears to be somewhat weaker than was earlier assumed (more like 1888 than 1883).

For comparison, here are the Davenport translation differences in EqA between Browning (AA) and Brouthers (NL) from 1882 to 1889. 20 pts in EqA is about 30 pts in OPS (Tom's rough estimator).

yr PBrow DBrou
1882 -75 -18
1883 -50 -18
1884 -51 -29
1885 -40 -25
1886 -32 -24
1887 -29 -16
1888 -23 -17
1889 -21 -12

avg.. -40 -20 shows Browning would lose about 30 pts of OPS, relative to legue/park avgs of course, (15 each of OBA and SLG) if he had been in the NL.

My argument, put more explicitly, is that since the sample size isn't big enough for aging differences to just cancel each other out, the players in the sample should be carefully selected to make sure they are representative of the gap between leagues we are trying to measure. I don't think the fact that Tip O'Neill had a 106 OPS+ in the 1890 Player's League says anything about what OPS+ he would have generated had he played in the 1887 NL instead of the 1887 AA. Nonetheless, Brent, your conclusions do seem to line up with the consensus view on what the appropriate discount is.

The AA’s first season, it consisted of six teams (Baltimore, Cincinnati, Louisville, Philadelphia, Pittsburgh, and St. Louis), at least four of which had existed earlier as independent teams playing outside any formal league.

Cin, Lou, StL and ??
From memory [I won't keep saying that. I am not checking anything.] I think Philadelphia was a new Athletics club.

<i>Our comparison set suggests that for the first season, relatively little recruiting was done from NL ranks, as only six position players had significant, recent NL experience

At the last minute (how late?) the AA overturned its plan to ignore the NL blacklist.

--
What is the "modern AAA" and "modern AA" level?

--
A big question (elephant), why do the results differ so greatly from Cramer's? Because OPS+ differs so greatly from his BWA? Hard to believe and a concern if true.

Is Cramer's plot (average Batter Win Average) available by internet? What about his table (separate corrections for Batting and Slugging averages)?

I spoke to Dick Cramer at SABR34 about continuing interest in his classic study. He planned to redo it taking advantage of modern computing speed (25 years of quickening!). By email a year or two ago, I learned that the old findings held up remarkably well. I wonder whether he published.

Cramer used all the data. Can anyone explain why using it all, weighted by playing time, would yield findings systematically different from using only the players with 300 AA plate appearances in one season?

--
Somewhere here I posted Pete Palmer's findings for UA1884 and FL1914-1915. He finds those leagues remarkably close to the contemporary NL, remarkable if we accept Brent's estimates and the conventional wisdom about rival leagues. Eg, Palmer finds UA1884 relatively stronger than Brent finds AA1882. Cramer finds the former as far below AA1882 as that league below the NL.

. . . Actually, Brent may be finding differences much greater than anyone has found for any league. For 1885, the first season when it is generally agreed that the two leagues were similar in quality:

1885 AA:NL quality ratios estimated by Brent Lastname
Player OBP+ SLG+
Ratio 89.2% 85.3%</pre>
Does this means OPS+ "ratio" ~75? (90, 85-->75)
(Evidently, my brain doesn't working today.)
That is about double Palmer's finding for UA1884, universally considered the relative-weakest rival major league.

Estimating League Quality - Part 1 (the concept)

I may have posted findings by Palmer or someone else, or contributed relevant discussion, to Part 2 or Part N.

"Part 1 (the concept)" was a good idea, I'm sure, but the group didn't come close to maintaining the discipline necessary for that. So I am copying here in this "part" (because it is now active) what I posted to "Ron Guidry" a day or two ago.
[hr]
:: Dick Cramer's classic study is explicitly limited to batting skill.

: True, Paul, but there were others who used Cramer's concept for fielding and pitching and came up with a similar pattern.

True. But
(a) [incomprehensible. Similar pattern or similar magnitude? If similar magnitude then Pierce, Bunning, and Wynn are silly selections --I guess.]

(b) Others have not generally replicated Cramer's finding regarding batting skill. For the 1950s-60s, maybe so (and maybe that is what you mean [John Murphy]), but here is Cramer's finding.

batting skill relative to contemporary NL

American League 1901-1980
AL>NL: 1973-1980 (dh
AL=NL: never
AL<NL: 1902, 1907, 1917, 1924-1925 (plotted gap less than two quanta, I think)
AL<<NL: 01 03 05-06 08-11 16 18-23
AL<<<NL: 04, 12-15, 26-72 (that's right, 1926 to 1972)

AL<<<NL roughly comparable to AA<<<NL1884: 1901 and 1942-1968!
(1884 is the best of the five clearly inferior AA league-seasons, comparable to Federal League inferiority)

American Association 1882-1891
AA>NL: 1886
AA<<NL: 85, 87-89

Here is another qualitative representation of the gap according to Cramer, omitting the designated hitter seasons 1973-1980.

batting skill relative to contemporary NL [##] = number of MLB league-seasons in this class
+2: [01] PL 90
+1: [01] AA 86
==: none
-1: [04] AL 02 07 24-25
-2: [19] AA 85 87-89 ; AL 01 03 05-06 08-11 16 18-23
-3: [25] AL 04 12-15 26-41 69-72
really big gap:
-4: [31] AA 83-84 ; FL 15 ; AL 01 42-68
-5: [04] AA 82 90-91 ; FL 14
off the map:
-9: [01] UA 84

Now reread line "-4".

If the true difference in average pitching skill is half so great as that, then Pierce, Bunning and Wynn may be the HOM's biggest mistakes.
Fortunately or unfortunately, the size of the Cramer-measured difference casts doubt on the method. Where others broadly using the same method estimate a notably smaller gap, we really need to know why.

Thanks, everyone, for your comments.

Dan Rosenheck wrote:

since the sample size isn't big enough for aging differences to just cancel each other out, the players in the sample should be carefully selected to make sure they are representative of the gap between leagues we are trying to measure. I don't think the fact that Tip O'Neill had a 106 OPS+ in the 1890 Player's League says anything about what OPS+ he would have generated had he played in the 1887 NL instead of the 1887 AA.

While I concur that player aging can be a problem, I'll note that the problem goes both directions. Tip O'Neill may have been near the end of the line in the 1890 PL, but Billy Hamilton and Herman Long were just getting started in the 1889 AA. All that says is that increases and decreases in player performances are influenced by other factors as well as league quality. I also agree with Chris Cobb's comment that it is unwise to cut the sample size by removing selected players. I can think of a couple of approaches that seem preferable. (a) I could set up a control group of players who remained in the NL, attempting to match the players who switched leagues in age and approximate baseball ability. Since I don't have computer software set up to select control groups, and such a comparison would take many hours, I don't think I'll try this approach. (b) I could run a multiple regression analysis that controls for player ages. I won't have time to try this approach in the next week or so, either, but at least it seems do-able. Either approach would adjust for the effects of aging without throwing out data from the sample.

Chris Cobb wrote:

I'll be curious to see what Pete Browning's and Charlie Jones's OPS+ scores look like with these modifiers applied.

For future reference, the modified (or National League-equivalent) OPS+ scores are posted on the Browning and Jones threads.

Jimd noted a difference for 1885 from "the less precise discounts that were used here in the early elections." What were those discounts? BP's WARP2/3? (I wasn't around for the early elections, and as you know, many of the threads for pre-1925 elections were gutted when the HoM switched Web sites.)

Paul Wendt asked:

What is the "modern AAA" and "modern AA" level?

Working from memory, I believe that the modern AAA level is about 90-92% for OBP and about 86-88% for SLG. For AA level, I think we're talking about 82-84% for OBP, and about 75-77% for SLG. If someone can find a source that provides different conversion factors, please correct me.

A big question (elephant), why do the results differ so greatly from Cramer's?

It's been a number of years since I looked at Cramer's study, so I really can't answer. My guess is that a big part of the difference is that I am treating the NL as if it were constant quality, whereas his results allowed quality of both leagues to vary. For example, to compare the 1884 AA with the NL, I'm really looking at comparisons of 1884 AA to 1882 NL or 1884 AA to 1887 NL. Because few, if any, players switched leagues during the season, I don't have any comparisons of 1884 AA directly with 1884 NL. In order to compare the leagues for the same season, you really have to link two (or more) comparisons--for example, the 1884 AA with the 1882 NL, and the 1882 NL with the 1884 NL. I'm only making the first of those comparisons, so that could be a source of our differences.

The 1884 NL is an interesting case, because its quality probably dropped due to competition from the UA. The BP data cited by TomH (# 141) indeed show a decline in NL quality for 1884. What I don't understand, however, is why the BP NL data don't rebound more in 1885. Didn't most of the quality players who'd defected to the UA return to the NL in 1885?

I've tried to fully document my own study and make it as transparent as possible. Other than that, I can't pretend to explain differences from the Cramer, Palmer, or BP studies.

By quality I mean relative quality of contemporary leagues throughout. Same for related terms such as gap, inferior, etc.

Brent finds huge gaps in league batting quality, huge in context of what others* have estimated, or conventional wisdom. (*unless Except Cramer on the 1940s-60s, who finds a persistent huge gap.) AA1883 according to Brent is comparable to UA1884 according to Palmer (OPS+ ~75% of contemporary NL). AA inferiority in 1882 and 1884 (and presumably 1890-1891) is notably greater than UA inferiority according to Palmer. Only in its best years, according to Brent's estimate, the AA was about as good as the Federal League according to Palmer.

I don't know what's wrong but I can't believe it any more than I can believe Cramer on the American League inferiority of 1942-1968 --comparable to AA1883-84 and FL1915, stronger than only five leagues in major league history.

What is the "modern AAA" and "modern AA" level?

Working from memory, I believe that the modern AAA level is about 90-92% for OBP and about 86-88% for SLG. For AA level, I think we're talking about 82-84% for OBP, and about 75-77% for SLG. If someone can find a source that provides different conversion factors, please correct me.

Unfortuantely I'm at work, because I did a little figuring on this a couple weeks ago. There's a couple different points of view, the BP POV and the BJ POV.

In the 1985 Abstract, in the introductory essay, James said that the QoP discount for AAA was 18%, which is assessed to the run environment. He offered nothing for any other classifications or for international play.

So I went hunting on BP to find out what Clay Davenport's figures say, and to estimate how James' MLE system might compare. A clutch of articles helped me figure it out.

-In the article "Japanese Baseball" he writes:
The Triple-A/majors multiplier is .860; if the transitive property holds, then Japanese EqA is worth about .948 of a major-league EqA, which conveniently enough is almost identical to what we got from major leaguers.

-From "Evaluating the Olypmians":
This is still using the Olympic average=.260 EQA rule. Given the difficulty level we worked out before, an average major-league team on this scale would have an EQA of .318. Stepping through the levels, an average Triple-A team should have produced at about a .289 EQA, a Double-A team would be .271, a high-A team would be .251, a low-A (South Atlantic, Midwest) would be .232, a short-season A team (New York-Penn, Northwest) would be .212, and a rookie league team (Appalachian, Pioneer) would be .201.

-From "Winter and Fall League Translations"
The 2001 AFL rates as being 5.3% easier than the International League, in terms of EqA. The IL, in turn, is rated at 13% below the AL (again, in EqA terms), making the net assessment of the AFL to be 18% below the AL.

and

Given that the Triple-A leagues rate between .850 and .870, and the Double-A leagues come in between .790-.800,...

-In "Julio Franco":
Unfortunately, for Mexican pride and Braves' fans hopes, the level of play in Mexico is a lot closer to the Midwest League than the Pacific Coast. That means Franco's .400 EqA should translate to about .270 or so in the National League.

-In "Translating Cuban Performance" Davenport says this:
The difficulty rating is the ratio between what a run is worth in this league and what it is worth in the major leagues. A player who produced 100 runs in Cuba, even after allowing for the offensive level of the league, would only be expected to produce 45.6 runs in the majors. The closest American league to that level of play is the New York-Penn League, which over the last four years has averaged a .436 rating.

OK, here’s what I get out of this. I plotted it all on a spread sheet, then used the one known value for James’ system to extrapolate the remainder.



It’s not perfect because EQA is not runs. And there are examples within the stuff above that may be contradictory. And perhaps a straight-line extrapolation ain’t quite the right way to go. But it at least offers a little bit of a sense of this all.

I like working in runs when possible, Chris Cobb has shown us information like this in terms of batting average and SLG, which present as higher for reasons to do with how runs are built. Anyway, the Mexican League of the 1940s probably falls somewhere between AA and AAA (so .8-.85 by EQA or .76-.82 by R), I’m still studying the matter and I hope to have a full report soon.

Meanwhile, the NgLs are likely falling in the realm of AAA, maybe a little lower than Japan. So .85-.95 by EQA or .82-.90 by runs.

By conventional wisdom, the PCL of the 1900-1950 era likely shows up as higher than the .86 or .82 shown above due to the fact that it wasn’t a farm league and held reserved contracts on lots of top-flight west coast talent during that time, as well as because it became a haven for top unaffiliated black players during the dissolution period of the NgLs. It also tended to have a very strong core of recently retired MLB players at any given moment. The IL and AA were probably similarly situated until around 1930ish.

No idea yet on the Candian Provincial League (also a haven for NgLers as well as for MxL jumpers), but I’ve got a little bit of info on those leagues that I’ll someday tease useful conversion figures from.

Anyway that’s just my take on it, I hope someone else has a better take on it all.

See #55-90 on the preceding page.

In #55, DonF says "The Cramer spreadsheet is now uploaded" to the Yahoo egroup.

After I have my say on Cramer, Palmer, Davenport (mainly C & P) through #78,
Cblau #79 reports on Davenport and
jimd through #90 makes some good points about this family of studies.

I am not comfortable with EqA but I think the Davenport numbers reported by cblau #79 show a larger gap than my remarks imply, closer to Brent's estimates.

I have not assessed the contributions by Mongo and following (continuing on this page).

--
Reprinting Paul Wendt #72
Pete Palmer utilized only 1913-1916 data in his assessment of