1893 Prelim

Lots to think about, especially the relative value of pitching when seven pitchers threw 380+ innings. Dan R's WAR discounts the bulk innings but it's hard to completely ignore that kind of contribution. Lots of unearned runs, league ERA was 4.66 but RA was 6.57. Kid Nichols was a top pitcher with a K/9 of 2 and BB/9 of 2.5. Cy Young led the league in K/BB at 0.99.

1) Amos Rusie - too many innings to ignore
2) Ed Delahanty - best position player
3) Cy Young
4) Elmer "Mike" Smith - hit 23 triples!
5) Ted Breitenstein
6) Kid Nichols - there is a clear "top 4" among pitchers
7) Billy Hamilton - only played 82 games but was the best hitter by rate stats
8) Cupid Childs
9) George Davis
10) Frank Grant - he was playing for the Cubans but there are no recorded stats. Age 27 season and a Hall of Merit caliber player.

11-15) Frank Killen, Denny Lyons, Bid McPhee, Sam Thompson, Jake Beckley
16-20) Billy Nash, Roger Connor, Dan Brouthers, John McGraw, Herman Long

Hmm, it sure is interesting to think about how to weight pitching v hitting. If you pitch 40% of your team's total games, that's just a ton of accumulation for a position player to stack up against.

1) Cy Young - 3rd best bWAR (tight at the top), best fWAR (bigger lead here, relative), lots of innings, 4th on total run prevention, but they're all really quite similar there, so I have to let WAR do some of the work of sorting out how much of the differences might be defensive quality.
2) Kid Nichols - best bWAR, 2nd fWAR, lots of innings, 2nd best total run prevention
3) Amos Rusie - 4th bWAR, 3rd fWAR, 3rd on total run prevention (essential tie with Young), biggest innings total
4) Ed Delahanty - best position player pretty easily
5) Ted Breitenstein - 2nd bWAR, 5th fWAR, fewest innings, best total run prevention (essential tie with Nichols)
6) George Davis - back in these days, BABIP pretty much equaled batting average, and you needed to be about .350 to make the leaderboards for this ballot
7) Frank Killen - the 5th pitcher who I think belongs in the top of the pack pitcher conversation
8) Elmer Smith
9) Billy Hamilton
10) Denny Lyons

11-20) Cupid Childs, Sadie McMahon, Frank Grant, Jake Beckley, Roger Connor, Bid McPhee, Sam Thompson, John McGraw, Jack Glasscock, Jesse Burkett

Billy Nash would be my #21, and then after that, a mess of players who I wouldn't bother trying to differentiate.

1893 preliminary ballot

I think it is obvious that Michael Humphrey's Dra fielding numbers will less accurate at this point in baseball history, but I don't feel that any other system is any more accurate. So I will continue to use them in my calculations. The major effect of using DRA numbers is with the pitchers. With reduced strikeout numbers, DRA runs above average are usually lower than just RAA.

1. Ed Delahanty 8.93 WARR: Best offensive and defensive position player.
2. Frank Killen 7.59 WARR Pitched in a hitter'e park. Hie hitting was excellent; whereas the a number of the other pitchers (Young and Breitenstein are below average)
3. Cy Young: 7.22 WARR
4. Theodore Breitenstein 6.63 WARR
5. Kid Nichols 6.63 WARR
6. Amos Rusie 6.15 WARR
7. Billy Hamilton 5.10 WARR
8. Sadie McMahon 5.10 WARR
9. Cupid Childs 4.53 WARR
10. Elmer (Mike) Smith 4.44 WARR

Rest of top 15
11. George Davis
12. Roger Connor
13. Bid McPhee
14. Buck Ewing
15. Hermon Long

Oh, yay! 1893 will be fun. Also, I'm now on my fourth rewrite of my giant article on WAR's positional adjustments, which is what I deserve for trying to write it as I went along, instead of getting all the research done first. Anyway, I've come up with a general outline of what this all amounts to. Here's that outline:

Sigh. This is the fourth rewrite of what I thought would take a week or so. What happened was that I didn’t refrain from writing during my research process. So, nothing was in order, and I tried writing it a couple of different ways, but I’ve come down to this: I’m going to outline what the whole process is, and then, in comments to follow this one, I go into detail. Detail is important. I’m trying to tell you that a 73-year-old man, who has never once had access to a play-by-play database, has found catastrophic holes in BB-Ref’s WAR Gang’s computation of positional adjustments. I’d better provide some support for that.

I should add that the reason this took so long was that I kept surfing BB-Ref, over and over, trying to find something, ANYTHING, that would explain what the heck the WAR Gang thought they were doing. That did worse than fail; it uncovered MORE problems, convincing me beyond any doubt that WAR’s positional adjustments are entirely subjective and can never be trusted. Of course, if you can’t trust the positional adjustments, then you can’t very well trust the fielding numbers, and if you can’t trust those, then you can’t trust the pitching or batting numbers, either, because they all have to add up somewhere. That’s why I’ve been using just a list of Win Shares to make up my MMP ballots. I have lost all faith in WAR’s numbers.

What I’m going to do here is just outline things. I understand the approach the WAR Gang tried, and I’ve produced an outline of how that approach might work, if you decided to try it. The WAR Gang’s method is almost a school course in how to NOT do this right. There will be details.

In the process of this, I will refer to two different pages in BB-Ref. Those two pages contain most of the catastrophic problems. I labelled one of them the “kcgard2” page, and the other the “Brock” page. The kdgard2 page is the one to which kcgard2 sent me when I was first dealing with this nonsense. The Brock page is one I discovered during my wanderings through BB-Ref. Each of these pages contains a huge chart of positional adjustments for all playing positions for each season for each league of MLB. The charts do not agree. I THINK the one that actually shows up on BB-Ref player pages is the kcgard2 one, but the whole thing is such a mess that I can’t even be sure of that.

Here is the url for the kcgard page:
“https://www.baseball-reference.com/about/war_explained_position.shtml”

Here is the url for the Brock page:
https://www.baseball-reference.com/about/war_explained_position.shtml

OK. Enough introduction. Time to outline the process, as it might look if you were trying to do it right. Fair Warning: The BB-Ref process is actually based on Bill James’ work 20 years ago, in the New Historical and the book Win Shares. It starts with the article “Why did the Defensive Spectrum jump?” on pages 182-184 of Win Shares. I know how you guys love Bill; it’s not my fault that the WAR Gang tried to appropriate a lot of his 20-year-old stuff and failed. More of Bill’s 20-year-old stuff will show up in the detail comments. WAR won’t look any better.

So, here’s the outline:

A. Find a stat that you think measures offense well. I call this the Foundation Stat. The stat that BB-Ref uses, tOPS+, does NOT measure offense well. It is, in fact, very bad at that. This is the First Catastrophic Failure of the WAR Gang. Bill used Runs Created, which is much better than tOPS+. Surely, after 20 years, something better than RC, much less tOPS+, has emerged – something that does NOT include anything from WAR itself, since that would be circular reasoning.

B. Use the Foundation Stat to evaluate the offense of every position in every league, in every season of MLB. This is a GROSS expansion of what Bill was doing. Bill was trying to document a general thing over spans of decades, having to do with 2B and 3B only. He CANNOT be held to account for someone using his method to evaluate every playing position in every league in every year of MLB. The failure of the WAR Gang to understand this is the Second Catastrophic Failure. And we’re on Step B.

C. Make a very SHORT list of MINOR adjustments that you might make to the results of Step B. Make this IN ADVANCE of doing Step D! The WAR Gang clearly did not do this, which is the Third Catastrophic Failure. We’re on Step C.

D. Translate the info from Step B so that it is expressed like OPS+ is expressed, with or without the t. That is, call the average 100, and then compute relative values. 107 means 7% bettter than average; 94 means 6% less than average. This produces what I’m calling a Quantified Offensive Spectrum. The WAR Gang did do this, at least.

E. Try making the MINOR adjustments from step C. Pick the version which best matches your expectations; they won’t differ by much, because the adjustments you are making are MINOR. The WAR Gang applied MASSIVE adjustments, which is the Fourth Catastrophic Failure. We’re on Step E.

F. a) - If the offensive Spectrum that your Foundation Stat produces is nowhere near anything you might have expected, do NOT make massive adjustments. If you do that, you will end up so far from your Foundation Stat that your numbers will become unmoored, and you will end up with a purely SUBjective number. (This is what happened to the WAR Gang.) Start over, at Step A, by finding a better Foundation Stat. Repeat: Do NOT make massive adjustments to get yourself out of this.

b) - If your Foundation Stat’s Offensive Spectrum differs from expectation more than usual in one place (say, DH NOT being on the far end of the OS), take some time to consider whether your OS is trying to tell you something (like, that there are playing positions that have more offense than DH). Repeat: Do NOT throw massive adjustments at your OS until the difference goes away, or you will end up with WAR’s unmoored Catastrophic Failures.

G. Flip your Offensive Spectrum to generate a Quantified Defensive Spectrum.

H. Calculate all this for every playing position in every league in every season of MLB. (Remembering that you are using Bill’s method in MUCH more detail than Bill did, so you cannot blame Bill if this goes wrong.)

I. Apply a few Plausibility Tests to your results, in order to find anyplace where your Defensive Spectrum fails. If you find that this happens, return to Step A. (The WAR Gang apparently did not do this. I say this because their Positional Adjustments, subjective though they are, failed the very first Plausibility Test I applied. The failure to do this is the Fifth Catastrophic Failure.)

J. Make up a Big Ol’ Ass Chart of the adjustments for every position in every league in every season of MLB. Oh, and By The Way. If you are only going to use one decimal place in your numbers, do NOT add three trailing zeroes to the end of each number when you publish the Chart. (Both of BB-Ref’s charts, on the kcgard2 and Brock pages, are polluted so badly with trailing zeroes that they are really hard to read. This is not a Catastrophic Failure, but is a Big Damn Nuisance, BB-Ref.)

Well, that’s the outline. The remaining comments are documentation of all this. I want to add here that you may find my writing tone cranky. I can’t help it. My biggest problem, all the way through here, is an inability to understand How The WAR Gang Could Possibly Have Missed This? I figured out all of the problems in the first week or so; some of them in a few minutes. How did this group of Sabermetric Analysts manage to miss things that I saw so obviously? I don’t know. And the inability to figure out where the WAR Gang went wrong is what drove me to keep looking and looking at BB-Ref, trying to find some sort of explanation. I have failed at that, leaving me still wondering, “How Could They Have Missed All This?”

Oh, and just to try to stop the Win Shares pitcher complaining, I think that the ballots I have seen above are very realistic. In 1893, pitchING was probably not any more important than it is now, but individual pitchERS pitched so many innings that the good ones will regularly end up in the Top Ten, and will often lead the league. I suspect that my ballot will look very much like the ones above.

This is the second comment in my series of comments on WAR’s Defensive Positonal Adjustments. Before I start in, I want to insert the two BB-Ref urls that were most important to me in doing this, so you can fact-check me without going to another comment. Here they are:

Here is the url for the kcgard page:
“https://www.baseball-reference.com/about/war_explained_position.shtml”

Here is the url for the Brock page:
https://www.baseball-reference.com/about/war_explained_position.shtml

I’m going to start with Step A of the outline that I posted earlier. That step involves choosing a Foundation Stat for the whole process of deriving Defensive Positional Adjustments from a Foundation of an Offensive stat. The stat the WAR Gang chose as their Foundation Stat is tOPS+.

Here’s a chart that I will use a lot. It lists the Offensive Spectrum that tOPS+ generates, sorted in order of greatest offense. The numbers are the tOPS+ quantification of this Offensive Spectrum, normalized to 100 = average. Then, after the slash, it lists the Defensive Posiitonal Adjustments that the WAR Gang ended up with, sorted in order of worst defense. The numbers after the positions are the actual Positional Adjustments themselves, normalized to zero.

1B: 121 / DH:-15
RF: 117 / 1B: -9.5
DH: 112 / LF: -7
CF: 104 / RF: -7
LF: 102 / 3B: +2
2B: 97 / CF: +2.5
3B: 96 / 2B: +3
C: 95 / SS +7
SS: 94 / C: +9

So, according to tOPS+, 1B is the position with the highest offense, and that offense is 21% over average. The list of Positional Adjsutments has DH as the worst “defensive” spot, and that the Final Posiitonal Adjustment for DH will be -15 runs.

There are just a few problems. First, not even one playing position occupies the same slot in both lists. So tOPS+ did not generate anything like the eventual Adjustments.

Second, DH is not at the far end of the tOPS+ spectrum. It is at the far end in the Adjustment list, where almost everyone thinks that it should be, if you include DH in the Spectrum at all (I don’t inlcude DH or Pitchers, but then, I can find Defensive Win Shares, which don’t require Defensive Positional Adjsutments.).

Third, there is an enormous difference, according to tOPS+, between LF and RF (117-102 = 15). The Final Adjustments for the two spots are identical at -7.

Fourth, CF is far too high in the tOPS+ list, as compared to the Adjustment list.

Fifth, tOPS+ thinks that Catchers hit better than Shortstops.

Sixth, the tOPS+ percents don’t add up. It’s a list that claims that it is normalized to 100 at the average. Let’s add up the percents that were higher than 100. 21+17+12+4+2 = 56. Now, let’s add up the numbers less than 100, expressed as negative numbers (97 - 100 = -3) because they are less than the average. -3+-4+-5+-6 = -18. 56-(-18) = 38. The difference should be zero, not 38,, because the list is normalized. So, what happened? Well, I can’t be absolutely sure, but the logical conclusion would be that Pitchers rank at -38, which means 62 on th tOPS+ list. That may well be accurate. However, when I tried to figure out just what the WAR Gang was doing with Pitchers, I found a lot of ajdustments and no flat-out numbers for pitchers.

I want to add that I’m not totally sure that I understand the “t” in tOPS+. What BB-Ref says about it, in multiple mentions, is that it involves comparing players’ numbers while playing their primary positon to their numbers at other positions. That makes no sense to me. Let’s consider Catchers. Starting Catchers play a huge percentage of their games actually at Catcher. Yadier Molina, for example, has played only two games this year where he was NOT the Cardinal Catcher. He has pinch hit twice. Trying to compare those numbers is futile, because the sample size of “not at Catcher” is far too small.

It doesn’t really help things if you compare those splits for a league as a whole. This is because BACKUP Catchers play many more games at spots other than Catcher than Starting Catchers do. So, trying to make the comparison over a whole league just results in a VERY strong bias in favor of Backup Catchers. And the number of innings played “not at catcher” will still be too small to stand up to the huge number of IP actually at Catcher.

In short, the approach of comparing “at starting position” to the “not at the starting position” will fail. I have no idea what the WAR Gang was thinking, but if this is really what “t” means, then it’s a disaster.

That deals with Problem #6. The other five problems may possibly be due to one issue – Running Runs. This is a term I came up with to collect BB-Ref’s Stolen Base Runs, Taking Extra Bases Runs, and Avoiding Double Play Runs. Running Runs are a very serious issue here. I’ve always known that OPS was a garbage stat, catering to sportwriters and fans whose math abilities top out at adding two three-digit numbers together with the help of a calculator. But I had no idea of the magnitude of this until I did this research. It’s gigantic. And it’s also going to wait until my next comment, because it takes a while to work through.

Oh, hey. I'm actually here and able to post. So, here's my ballot for 1893. A fun year.

Hi. This is Brock Hanke’s ballot for 1893. It’s a list of Win Shares. One interesting feature is that, although 1893 was the first year with the 60’ 6” pitching distance, the pitchers do not seem to have been much affected. Seven of the players on my list were pitchers, including the Top Six. The three who weren’t pitchers were Hugh Duffy, Ed Delahanty and Herman Long. I assume that everybody here recognizes those three names. Well, enough of this; here’s the list:

1. Frank Killen (42)
2. Amos Rusie (41)
3. Kid Nichols (40)
4. Cy Young (35)
5. Ted Breitenstein (30)
6. Brickyard Kennedy (29)
7. Hugh Duffy (28)
8. Ed Delahanty (28)
9. Herman Long (26)
10. Sadie McMahon (26)

1. Brock says: ... I can find Defensive Win Shares, which don’t require Defensive Positional Adjustments."

Question: Are not the inherent weights, which are part of the win shares defensive number calculation, actually a way of providing a position adjustment in win shares?

2. The main point that I would make about this discussion is that the positional adjustment is not an adjustment based on batting runs, but an adjustment made to take into account the relative fielding skill levels required to to play each position. (based on fielding runs)

For example: most people would agree that a !B and a SS, both with +5 defensive runs (5 runs better than the average at each position), are not equal in defensive skill. So applying the 1B positional adjustment of -9.5 and the SS positional adjustment of +9 is an an attempt to quantify the ir relative value as fielders. Tom Tango and others have discussed the procedure used to reach these values on their blogs (but I am not knowledgeable or articulate enough to try to give an explanation).

I think the problem with baseball reference is that, on their website, they do not attempt to explain the finer points of how the position adjustments are calculated. FanGraphs does a better job of that.

I'll take a stab at a bit of this.

In the chart at the page you linked, which you've referenced, noting tOPS+. tOPS+ is not created by comparing, for an example, how Yadier Molina bats when starting as a catcher and when starting at some other position, noting the delta, and then doing the same thing for all the players in the league. That would indeed be a terrible method, I agree with you. It is done at the aggregate level. The chart notes that 1B have a 121 tOPS+. This means that all the PAs in the league taken by first basement resulted in batting production 21% better than the overall average. If Yadi Molina took 1 PA as a 1B, then he would count 1 PA out of the 15,000 PA (making up a number) that all 1B in the league took that year. Rightly, his contribution to the metric for 1B would be nothing, because his playing time was nothing. In this chart, tOPS+ means the performance for the given split (the defensive position) relative to the overall average (the league as a whole). At BB-Ref, tOPS (or tANYSTAT) means performance for the given split relative to overall performance. Personally, I think this makes sense: you find how the pool of all PAs taken by each position compare to the performance of the league as a whole, to find out how much batting is expected from the position. The question, I suppose philosophic in nature, is whether batting contribution is necessarily inversely related to the demands of fielding that position. How to quantify the fielding demands of the different positions is a question that must be answered for any value stat created. Until someone finds a better way to determine it, the batting performance of the positions in aggregate is probably the best proxy we have, IMO.

BTW, the chart you reference is for a given year. In any given year, of course, the batting contribution from a given position will not always match what is assigned on the defensive spectrum. In the given year, CF hit better than LF. That will happen sometimes, even understanding that the defensive requirements of CF are greater than those of LF. In the longer run, CF will not hit better than LF, which is why CF has a higher position adjustment. I assume you are OK with using longer timeframes than a single year to determine what the defensive adjustment value for the positions should be.

Last item: the adjustments do not equal to 100 (the good hitting positions hit +56 and the bad -18). Did you notice that pitchers have a tOPS+ of 0 in the chart? For a total negative value when including them of -118. Which, granted, still doesn't equal to 100 when you sum them. Why not? Because the different positions do not get equal plate appearances. If you weighted each position by its number of plate appearances in the season, the tOPS+ splits would equal 100. If we knew which specific season the chart is showing, we could prove this. Catchers and pitchers get far fewer plate appearances in any given season than 1B and RF do. Because they are worse hitters and go in the 8 and 9 slots a lot. However, the chart is showing a seasonal line for each position "prorated" to 650 PAs. Notice it takes pitchers 283 prorated games to get there, catchers 165, and everyone else in the neighborhood of 152. I bet if we simply weighted the tOPS+ values by the (inverse) prorated games (because more games indicates fewer actual PAs for that position - it is the number of games required for the average player of that position to achieve 650 PAs), it should give something quite close to 100.

C: 1/165*95 = 0.576
1B: 1/152*121 = 0.796
2B: 1/151*97 = 0.642
3B: 1/155*96 = 0.619
SS: 1/152*94 = 0.618
LF: 1/152*102 = 0.671
CF: 1/148*104 = 0.703
RF: 1/152*117 = 0.770
DH: 1/151*112 = 0.742
P: 1/283*0 = 0
========================
0.06252 6.137

6.137/.06252 = 98.2 Give me some grace for rounding errors on this total, but it's close to 100.

If we could have just weighted by PAs directly, it would equal 100.

I haven't forgotten about this, I will start the election at the end of this month.

1893 Prelim Ballot

1. Amos Rusie, P, New York Giants: 482 IP laps the competition while 143 ERA+ comes in 3rd
2. Ed Delahanty, LF, Philadelphia Phhillies: 2nd in OPS+ with 164, leads in RC with 144, and adds +17 fielding
3. Cy Young, P, Cleveland Spiders: 144 ERA+ is 2nd, 422 IP is 3rd
4. Kid Nichols, P, Boston Beaneaters: 425 IP is 2nd, 139 ERA+ is 3rd
5. George Davis, 3B, New York Giants: 4th in MLB with 155 OPS+ while adding +2 fielding from a prime defensive position
6. Ted Breitenstein, P, St. Louis Browns: 148 ERA+ leads the league but 382 IP is well behind the top hurlers
7. Frank Killen, P, Pittsburgh Pirates: 124 ERA+ and 415 IP for a strong combo of quality and quantity
8. Elmer Smith, LF, Pittsburgh Pirates: 158 OPS+ is 3rd and he adds +6 fielding from left field
9. Sam Thompson, RF, Philadelphia Phillies: 151 OPS+ is top five while 133 RC is second only to his teammate Ed
10. Billy Hamilton, CF, Philadelphia Phillies: 167 OPS+ leads the league though Slidin' Billy missed a lot of time this season

11. Brickyard Kennedy, P, Brooklyn Grooms
12. Ed McKean, SS, Cleveland Spiders
13. Cupid Childs, 2B, Cleveland Spiders
14. Jesse Burkett, LF, Cleveland Spiders
15. Denny Lyons, 3B, Pittsburgh Pirates