Saturday, 7 September 2013

Wx - Lx Records: Part 2

In this post, which in retrospect I will call Part 1, I introduced the idea of using RE24 and REW to get an idea of what pitchers' "true" win-loss records should be. This is an attempt to reduce or eliminate the effect of team offence on what is deserving of a win, a loss, or a no decision, without removing the start-to-start luck experienced by the pitcher.

If a pitcher gets lucky and has a great season despite allowing an alarming number of base runners and warning track fly balls, that pitcher is deserving of wins, as they currently stand, despite the fact that you may not want that pitcher to lead your staff next year. If that same pitcher ends up with a lot of losses and no decisions because his team's offence only averages 2 runs per game- those are the kind of W-L fluctuations I am trying to see past.

One of my sharp readers commented that in the hitters post, I was comparing Wx-Lx for hitters with W-L for pitchers. Good point.

In this post I want to put some big tables of Wx-Lx for pitchers, then:
a) see if it still makes sense
b) see whose W-L records may not be as great as they seem

a) and b) are related - part of the Wx-Lx making sense is who would benefit and who would not from this alternative scoring system.

I'm also limiting the data to post-1950 because REW doesn't exist much before that.

So, here are some lists. I'd love to give a big sortable table, but I don't know how.

The not-obvious columns are:
Rk - Rank of most number of wins since 1950
Rk x - Rank of most number of xWins since 1950
REW - base out wins added
Wx - xWins
Lx -xLosses
dWx - difference between Wins and xWins. A positive number means more xWins. A negative number means more Wins.
dLx - difference between Losses and xLosses. A positive number means more xLosses. A negaitve number means more Losses.
diff - the difference between the Win-Loss spread and the xWin-xLoss spread. By spread, I essentially mean the difference in games over (or under) .500: wins minus losses. A positive diff means that the Wx-Lx record is better than W-L. A negative diff means that W-L is better than Wx-Lx.

1. Most (actual) Wins since 1950

2. Most xWins since 1950

The top 8 pitchers are the same for both. Actually, the top 14 pitchers are the same for both. The order is shifted slightly. Here are some things from these tables that stand out for me:

A. Jim Kaat has only 3.9 REW. 4530 innings is a lot for a guy who may have been, overall, an average pitcher. I guess 16 gold gloves go a long way towards keeping a guy on the mound.
B. Greg Maddux is 1 W ahead of Roger Clemens, but 15 Wx behind. I will get more into why I think this is later, but it's interesting that it's not as close.
C. Three players are different between the two lists. The players removed from the W list are Andy Pettite, Jack Morris, and Juan Marichal. The players added are Jim Bunning, Jerry Koosman and John Smoltz.
D. The only way to get to 300+ wins is to pitch a LOT of innings. A lot of these guys are over 5000 innings. That is twenty years of 250IP per year, or twenty-five years of 200IP. If someone wants you to pitch for that long, you have a great chance of getting to 300 wins, of either kind. Since 1901, there have been 9 pitchers who have thrown 5000 innings, and all of them have 300 wins.
E. I'm angry that Roger Clemens took PEDs and upset that we don't get to see where he would have been without them, because he had a spectacular start to his career.

3. Of the 200 pitchers who have the most wins since 1950, these are the pitchers with the highest REW:

Things that stand out to me:
A. How ridiculously good was Pedro Martinez? Really, really, really ridiculously good. His 56.5 REW was put up in just 2827 innings, compared to the other guys on this list. For comparison's sake, Roy Halladay has been terrific over the last 10 years and has been considered one of the "best of the era" kind of guys, but in similar innings he is a full 20 REW behind Pedro. 20 REW is a good career. Pedro's true W-L gives him full and accurate credit for his brilliance, by the way.

B. I think that Kevin Brown was much better than he has been given credit for. I think that some veterans committee will put him in the hall of fame, eventually.

C. By this measure, Johan Santana has not been so different from Sandy Koufax.

D. Who is Billy Pierce? Looks like he was a bit of longevity away from the hall. He twice led MLB in REW for bad White Sox teams, going 18-12 and 15-10 (that one with 200 ERA+).

E. What took so long to get Bert Blyleven in the hall. Seriously. I know he became the poster boy for stathead hall of fame cases, but it was pretty much just his poor W-L record maintaining this aura of mediocrity around him. More on that later.

Now, I want to see whose records are improved or hurt by a transition from W-L to Wx-Lx.

4. Here are the top 25 most improved by W-L to Wx-Lx:

Things that stand out to me:

A. Mike Morgan seemed like he was bad, but maybe he was just average. Oh.

B. Bob Friend pitched almost exclusively for the Pirates, with brief stints with both the Yankees and Mets coming in his age 35 1966 season before retiring. In his first 7 seasons, the Pirates finished 7th or 8th out of 8 teams. He went 73-94 with a 103 ERA+. Then the pirates finished 2nd in 1958, and Friend 22-14, leading the league in wins. The Bucs slipped to 4th in 1959 (Friend 8-19), then back to 1st in their world series-winning 1960 season (Friend 18-12). The Pirates finished in the middle of the pack for the rest of Friend's career, and he went 70-79 for them with a 116 ERA+. My point is that, overall, he was a slightly above average pitcher. His pitching deserved a fairly mediocre 207-194 record more than it deserved the stinky 197-230 record he ended up with. But he spent nearly his whole career with a terrible team, and his record mirrored that of his team almost the entire time*. He was pretty average, so he fared how his team fared. Since the whole point of this was to try to remove the pitcher from team context, I would say that this is an example of its effectiveness.

*The lone exception may have been 1955, when he went 14-9 and led the league with a 2.83 ERA. 

C. Bert Blyleven. If Blyleven had ended up with a 316-237 record instead of 287-250, he would have been a first ballot hall of famer and all time legend. Instead, he lingered on the ballot and finally got in to the hall based on the emergence of the internet and the education of the majority of voters that wins and losses are not the whole story. His 39.5 REW is the 12th highest since 1950. Although in-depth cases of his situation have surely been done all over the internet, I will take a look at his stats and his team stats.
Team stats are in blue. RS/G and RA/G are runs scored and runs against per game, respectively. Those values in white are for the games Blyleven pitched (RS) or the amount of runs (total) he allowed (runs against).

Okay, so I've been looking at this for a while, and I can't really figure it out. This time, the team performances aren't that bad.Except for a couple of cases (1976, 1980), the runs his team scored for him are not bad. What I think is interesting is that this difference basically is concentrated over the first half of his career. From 1981 onwards, his Wx-Lx record (129-107) matches his W-L (131-109) almost exactly. I don't really feel like getting into the game logs today, but there has to be something strange about the way runs were scored in his starts during the first half of his career. Maybe he was one of those guys who was either dominant or average? In any case, I am getting away from things. Back to #4D, things I notice about the list of players who receive an improvement from W-L to Wx-Lx.

D. In almost all of these cases, it is subtracting losses more than receiving lots of extra wins that is the main factor in this difference. That says to me that these pitchers were saddled with losses when their bad teams struggled to score during their starts more than they received no decisions when their teams blew bad leads. Something that I find very interesting is that this model seems to work despite me using 9.0 innings pitched per start as a metric to determine the number of overall decisions a pitcher should be awarded*. This seems to suggest that starting pitchers overall earn decisions at the rate of about 1 decision per 9 innings pitched. A pitcher who starts 9 games and averages 7 IP per start will average about 7 decisions. A pitcher who starts 36 games at 7 IPS will average about 28 decisions (24-4 / 20-8 / 18-10 / 14-14 / etc)

*It's not perfect - in this sample of the 200 pitchers with the highest W totals since 1950, the pitcher averaged 7.2 fewer Wx-Lx decisions than W-L decisions - but it's close, and it is a round number. Adjusting this parameter to get the number of decisions to match resulted in an innings per start of about 8.75. 

E. Relief pitchers break my model. Hoyt Wilhelm, Dennis Eckersley and John Smoltz are all on this list, and they the few pitchers with enough wins to make the list who spend a significant portion of their careers in relief. In the case of Smoltz and Eckersley, they were called upon to finish the game mostly with their team ahead. When closers earn wins, it is usually vulture wins - blown saves with their team coming out ahead before they come out of the game. Anyways, I think the rate of wins and losses earned per inning is different for relief pitchers - closers in particular - so I think that this model is applicable to starting pitchers only.

5. Here are the 25 pitchers who were most hurt by a change from W-L to Wx-Lx.

Things that stand out to me:

A. I could have guessed - Andy Pettite at the top of the list. No pitcher has gotten more mileage from being merely good than Pettite, who was fortunate enough to break in with the Yankees at about the same time they got their mojo back. 255-151 is great, amazing stuff. 208-157 is a nice career, but there is no need to mention Pettite and the HoF in the same sentence. That's the difference here.

B. A lot of the pitchers on this list are very good pitchers, or are regarded as very good pitchers, anyways. On the previous list there were a lot more guys I had never heard of, bad pitchers improved to .500, etc. Here, there are some pitchers who I consider to be great who Wx-Lx shows not to be bad, but to be maybe a little less great. That's interesting. Do I consider them to be great because they are? Or because they had the extra advantage of a pretty W-L record, which often led to the hall of fame?

C. Jack Morris is on the list. If Bert Blyleven is the internet poster boy for Wins obscuring a brilliant career, then Jack Morris is at the other end of the spectrum, with an internet reputation for W-L record that looks way better than it ought to have been. Morris's (254-186) is transformed to a less impressive (226-199). As for his reputation for pitching to the score, well, that might be for another post. But for a quick summary: it's easy to create Wx-Lx with WPA instead of REW. WPA cares about the game situation - extra points are added late in close games, and actions matter far less early in games or when the score is not close. Morris's WPA Wx-Lx is (227-198), about the same as REW, and far off his regular W-L total.

D. Aaron Sele seemed like a .500 pitcher when I was following his career. His Wx-Lx makes me happy.

E. Mike Mussina was a great pitcher - his 41.9 REW in just over 3500 innings is terrific and 10th since 1950. Even so, his win totals were apparently inflated - he pitched for very good Orioles and then Yankees teams. I still think he should sail into the hall of fame. After his 4-5 rookie season, Mussina finished under .500 just once - 11-15 for Baltimore in 2000. Despite an ERA+ of 125, the Orioles managed just 3.4 runs per game in his starts. His Wx-Lx that year was 17-10. Everyone wins some and loses some (except maybe Andy Pettite).

F. Similar to Mussina, Wx-Lx reduces Whitey Ford from one of the greatest winning percentages ever to merely an outstanding record. The 50s and early 60s were a good time to be a Yankees pitcher.Tim Hudson is another one of these - an outstanding record now seems like more of a product of being a good to great pitcher on outstanding A's and Braves teams.

G. I could go on, but the names I recognize in this list all have the same kind of feeling to me. Kind of an "Oh, yeah he didn't seem to be as good as that W-L record" feeling. Kenny Rogers. David Wells. CC Sabathia. Jamie Moyer. Bob Welch.

6. Okay, last one. Of those top 200 in Wins, here are the players best predicted by the model, that is, who have the smallest difference between their actual W-L records and their Wx-Lx records:

Things that stand out to me:

A. Well, the message here is that, if you remember these guys, xWins and xLosses shouldn't change too much about what you think of their win loss records.

B. Don Sutton earned a bit of a reputation as a compiler because he did not necessarily have a brilliant peak, but maybe there is something brilliant about being a 5000 IP compiler.

C. I still have a huge baseball crush on Pedro.

I don't know how to put a spreadsheet up here, so message me if you want to sort out and look at the numbers yourself - although it's very easy to calculate for any career or season:

Wx = IP / 9.0 / 2 + REW
Lx = IP / 9.0 / 2 - REW

No comments:

Post a Comment