78 thoughts on “Cannon Fodder: The Casualties of WAR (Volume 1)”

1. mosc September 25, 2012 at 5:16 pm

   Can we just call this the hall of DWAR oddities or something? DWAR ruins all perfectly reasonable statistical discussions. As a SABR enthusiast, I refuse to acknowledge DWAR as useful. Every fiber of my statistical being has taught me to evaluate a stat’s usefulness BEFORE incorporating it, let alone stuffing it into every single player’s lifetime value like some kind of statistical panacea. Clearly any analysis of DWAR itself shows how horrible it is. Why then do people continue to rely on it??

   The pitchers look interesting though, I’ll reserve judgment on that side of the ball.

   Reply ↓
   1. John Autin Post authorSeptember 25, 2012 at 5:24 pm

      You may have rushed to judgment there, mosc. Doc Cramer? Alfredo Griffin? Just look at the OPS+ on the positional lists. A couple guys over 100 on the 1B list — that’s not a good number for an offensive position. For the other infield positions, only 3B Wigginton is at 90+, and if you don’t objectively think he’s a bad defender, you ain’t been watching. In the OF, two guys over 100 — I don’t know if you object to Bichette’s dWAR rating and I didn’t see much of him in action, but I’m damn sure that Moreland was awful out there.

      Reply ↓
   2. John Autin Post authorSeptember 25, 2012 at 5:35 pm

      Mosc, in response to your concerns, I’ve added an addendum showing the WAR Runs breakdowns for the position players. Only 4 of the 50 are in the black for Rbat.

      Reply ↓
   3. Ed September 25, 2012 at 6:09 pm

      Mosc – I’m a bit unclear on your comment. Where did John do anything that isolates dWAR? His lists were generated off of WAR, not dWAR. Can you clarify your comment? Thanks!

      Reply ↓
   4. birtelcom September 25, 2012 at 6:13 pm

      Mosc: The top career WAR fielding nuimbers of all time are held by, in order, Brooks Robinson, Mark Belanger, Ozzie Smith, Andruw Jones and Roberto Clemente. That sounds to me as if dWAR is not exactly random or blatantly contrary to intuition.

      Reply ↓
   5. mosc September 26, 2012 at 12:56 pm

      It’s probably not a good thing I had the first comment there. You guys probably prefer my negativity about dwar to be a little more buried. Thank you for discussing it though. JA, you’re very knowledgeable guy and I enjoy reading your stuff tremendously I hope my comments didn’t come across as against you, I just really don’t like when dWAR is lumped in with other statistics. On it’s own it means what it means and it’s simple enough to evaluate. Used as an adjuster in “wins above replacement” is really where I start to get upset.

      I don’t watch much NL baseball so I guess in my mind I guess Bichette is a well aged, productive DH power bat who made a lot more contact than Joe Carter. It was indeed his name that jumped out at me as absurd. Looking at his stats shows a long and generally above average career with the bat, low walks aside. His OBP without taking a lot of walks was always fairly average with above average power. The Colorado years OPS+ still reflects very positively. Not a complete player, but certainly a regular starter coors field be damned.

      birtelcom, I agree DWAR generally gets it right. The main issue is positional adjustment. A mediocre middle infielder is probably one of the 100 best gloves on the face of the earth. Put him anywhere but catcher and with an adjustment period you’re going to have a good fielder. Speaking of catchers, DWAR may have the relative ranking reasonably close but the magnitude in relation to offensive statistics is way off. I don’t like positive DWAR on first basemen, for example. They play first. If they had any range to go with their decent gloves they’d play somewhere else. DH, for example, nets you 0.0 DWAR even though you’re effectively THE WORST fielder on the team. So bad that they’d rather put out 8 other guys in the field than fit you in at ANY position. The issue though is not really these problems, it’s when the dwar stat is mixed in with other stats. The metric is not close to perfect, all would agree. The issue is when that inherent error is accounted for incorrectly by including it in a much more general stat like “total WAR”.

      JA, you’re right. A lot of those guys are out machines! I generally read a lot more than I comment on here. I’ve read all the threads and probably most of the comments. I’ll try to mimic the more objective tone that is more common here.

      Reply ↓
      1. nightfly September 26, 2012 at 1:31 pm

         WAR accounts for ease of replacement, mosc, which is why a mediocre MI can be one of the hundred best gloves on Earth, and still check in as negative in dWAR. If you like, think of it as the inverse of the bonus that hitters get at the MI positions. A 15-HR 2b is more valuable than a 15-HR LF, because it’s so much easier to find the LF who can do that, rather than the 2b; likewise, a good defensive 2b is not that hard to find – the trick is finding one who can also give you some sort of offense.

         Likewise, a DH provides 0.0 dWAR instead of taking the field and being a negative. A good 1b in the field (say Ike Davis) does, in fact, save runs over a bad one. The metric is designed to describe something that is actually happening, not necessarily to identify the eight best fielders on any given team.

         Reply ↓
      2. mosc September 26, 2012 at 2:42 pm

         I guess I would look for an offset based on position. Simply being a Left Fielder should hurt your dwar. Simply being a shortstop should help it. You have to play like Brett Gardner out in left if you want to be a replacement level defensive corner outfielder in my book. Now, this would also change the magnitudes compared to offensive statistics, which is where Bichette comes in. He’s not a good fielder, but he’s not at an important fielding position. He can’t do much damage.

         There are lots of other little problems with DWAR. I don’t like the way it handles playing multiple positions. Specifically when a below average third basemen plays a below average right field (Bautista). That’s not a positive. There are very few position swaps I would really give extra credit for. A Catcher also playing first base? Yawn. Puhols plays a pretty decent third base though, that adds real value defensively especially compared to his regular position. I mean, how far does it go? You gunna give arod extra defensive credit because he’s got a league average dwar when he DH’s? HA! Major leaguers maybe get a little spoiled when it comes to playing the same position. In the minors, guys cover for each other all the time.

         Reply ↓
         1. nightfly September 26, 2012 at 5:01 pm

            OK, I think I see what you’re driving at. Of course, if you do it that way then you’re talking about a completely different measurement. And I think it’s important to know who’s capable of being a good fielder at an important position and who needs to be hidden where it matters less. The thing is, to measure that well, I think dWAR should stay as it is. Otherwise you’re measuring something incorrectly, IMHO – by saying Yuni Betancourt, i.e., gets a dWAR boost because he’s a shortstop, you run the risk of missing that he’s a terrible shortstop (by MLB standards). We don’t just need to know if he’s better than a Dante Bichette, but also if he can make plays a majority of his peers cannot. If not – if he gets credit merely for standing helplessly to the right of second base while balls roll past him – we may never discover that he’d be better-suited elsewhere.

            dWAR already anticipates less-important positions, because there are fewer opportunities to make plays there and affect the game positively OR negatively. If I’m at second and seeing 5 or 6 playable balls per game in my area, I’m already having my contribution measured relative to a corner outfielder who may only see that many chances in a week. If we’re both missing one of those chances, at the end of the year I’ll be 150 or so plays in the hole, and he’ll only be 30.

            So let’s say that I get moved to the corner OF instead. The team gets you to play second, because not only do you reach those chances, you make another 50 besides that; the team is now 200 defensive plays to the good. Maybe that’s worth an extra four wins per year. But in right, I’m no worse than I was before – it’s not a positive, as you observe, but it is much less of a negative. Instead of missing 150 plays, I only miss 30, just because there are fewer plays available for me not to make. As long as I hit better than the former corner OF, the team gains even if I field as badly as he did. Heck, maybe I prove to be a better OF than 2b, and the team really makes out like bandits.

            Reply ↓
2. Doug September 25, 2012 at 5:19 pm

   Casualties of WAR – love that, JA.

   Tony Cloninger had much more WAR (2.8) as a batter, despite a .192/.205/.277 with 33 OPS+.

   Other oddities
   – Rollie Hemsley and Doc Cramer were 5-time All-Stars.
   – Tommy Thevenow placed 4th in MVP voting with a 59 OPS+.
   – Dante Bichette has the fewest walks ever (22) in a 40 HR season. Next lowest total was 32 walks in Andre Dawson’s 49 HR season.
   – Steve Blass lost almost half of his career WAR in his final 93.2 IP (5% of his career IP)

   Reply ↓
   1. Richard Chester September 25, 2012 at 5:30 pm

      Tommy Thevenow holds the record for most consecutive at bats without a HR, 3347.

      Reply ↓
      1. Doug September 25, 2012 at 5:55 pm

         But, when you’re hot …
         – Thevenow hit both of his 2 career HR in the same week.
         – Similarly, Johnny Cooney (3673 career PA) hit both of his 2 career HR in the same series.

         Al Bridwell may have had a longer streak than Thevenow. Bridwell hit his first career HR in the 1913 season, after 3208 AB through the 1912 season. Certainly, Bridwell had a longer PA streak at 3745+, to Thevenow’s 3605.

         Reply ↓
         1. Ed September 25, 2012 at 10:26 pm

            Including his World Series home run, all 3 of Thevenow’s home runs came during an 11 game stretch.

            Reply ↓
   2. Jameson September 25, 2012 at 10:10 pm

      Tommy Thevenow went 10 for 24 with a 1.023 OPS for the Cards in the 1926 WS. He also had a big hit in game 7 as they beat the Yanks for the title. That definitely had something to do with his MVP consideration.

      http://www.baseball-reference.com/postseason/1926_WS.shtm

      I had never heard of the guy before this post.

      Reply ↓
3. JDGentile September 25, 2012 at 5:57 pm

   This is excellent. Dante Bichett, Kieth Moreland! What a list. This really is a thing of beauty…

   Reply ↓
4. birtelcom September 25, 2012 at 5:59 pm

   My thought had been to create an annual award for the guys with the most PAs or IP while being nothing more and nothing less than replacement level — say, no more than 0.1 WAR, no less than -0.1 WAR. I figure the Fredo Awards would be an appropriate name, in honor of both Alfredo Griffin, the classic replacement level player, and Fredo Corleone, the classic replacement level brother.

   Reply ↓
   1. nightfly September 26, 2012 at 9:49 am

      Every acceptance speech: “I’m smart! And I want respect!”

      Reply ↓
   2. Tmckelv September 26, 2012 at 12:42 pm

      Replacement Level Brother scale:
      George Bailey 10 WAR
      Barry Gibb 8 WAR
      Charlie Babbitt 5 WAR
      Alec Baldwin – 2 WAR
      Fredo Corleone – 0 WAR
      Scar – -2 WAR
      Billy Carter -5 WAR
      Cain – -8 WAR

      Reply ↓
5. Brooklyn Mick September 25, 2012 at 6:00 pm

   The first name to jump out at me was Dante Bichette with a career WAR of 3.0. It seems ludicrous that over a pretty long career, despite defensive shortcomings, he was only worth 3 wins above your average AAA call-up.

   Then there’s Jesus Alou with a -1.0 WAR. Was he really that bad? I don’t think so.

   And Moose Dropo, who led the league with 144 RBI as a rookie ended up with a career 1.3 WAR.

   I think the defensive components of WAR are very vague, too heavily weighted, and ultimately obscure the overall “picture” of the player.

   Reply ↓
   1. birtelcom September 25, 2012 at 6:24 pm

      Dante Bichette was a corner outfielder with a career OPS in away games of .730, and he played much his career in an era when run scoring league-wide was very high. His mediocrity on offense was heavily masked by his home fields and the era he played in. The fact that he wasn’t very good on defense is just gravy.

      Reply ↓
   2. bstar September 25, 2012 at 6:30 pm

      B Mick, you’re not the first to question Dante Bichette’s low WAR total. In fact, his 1999 season where he hit 34 HR, 133 RBI and slashed .298/.354/.541/.895 might be the most controversial WAR season in recent memory. Despite those offensive numbers, Bichette logged -2.6 WAR in 1999.

      How is this possible? Just look at the fielding and baserunning components of his WAR total and consider that despite his high power numbers, he only had a 102 OPS+ because of the steroid era and Coors Field. So he only had +3 batting runs that year, -34 fielding runs, and -5 baserunning runs. It’s one of the worst fielding years in history.

      Here’s a good article about that infamous season:

      http://www.fangraphs.com/blogs/index.php/dante-bichettes-unbelievable-1999-season/

      Reply ↓
      1. bstar September 25, 2012 at 6:33 pm

         Oops, you were slightly ahead of me, birtelcom. Sorry.

         Reply ↓
         1. birtelcom September 25, 2012 at 7:37 pm

            The more the merrier, bstar — Dante is the sabermetric fan’s party pinata

            Reply ↓
            1. Brooklyn Mick September 26, 2012 at 10:18 am

               birtelcom, is it fair to say that Bichette, who led the league in hits, home runs, RBI, SLG, and total bases in 1995, while finishing 2nd in the MVP voting, is the best bad player of all time?

               bstar, thanks for the link to the FG article…good read.
            2. birtelcom September 26, 2012 at 11:38 am

               Brook-M: I think it’s fair to say that Bichette had the best raw career numbers, by a considerable margin, of any poalyer whose overall career value was basically that of a replacement level player. His career represents a perfect storm for this purpose — a slow corner outfielder who happened to play in a very high-run scoring era and for much of his carer on a team with an extremely high run scoring home park. All these factors combined to create a huge spread between his raw hitting numbers and his real overall value.
   3. Richard Chester September 25, 2012 at 7:28 pm

      Concerning Dante Bichette there is something about his stats for the 1999 season that bothers me, especially when compared to teammate Ben Petrick. I have previously submitted a similar comment.
      Here are their stats for that year
      Player………..1B……..2B……..3B…….HR……BB……HBP……Outs
      Bichette…..103…..38……2……34…….54………2………426
      Petrick……..13………..3……..0……..4…….…10………0………42

      The Rbat totals were 3 for Bichette and 2 for Petrick which looked sort of out of whack to me. Using approximate linear weight coefficients obtained from Dr. Doom I made my own calculations for their Rbat totals. The coefficients I used were 0 .3 for BB, 0.3 for HBP, 0.5 for 1B, 0.7 for 2B, 1.0 for 3B, 1.4 for HR, and -0.25 for outs. My results were 38 for Bichette’s Rbat and for 7 for Petrick’s. They both have the same park factors. Am I missing something?

      Reply ↓
      1. topper009 September 25, 2012 at 7:38 pm

         Those are the raw run values, but Rbat is relative to average, not total. Bichette had a 102 OPS+, so a little above average.

         His Rbat of 3 is saying WAR expects an average player in his park to produce 35 runs given his PAs, so Bichette produced 38 and was 3 above average. All the WAR breakdowns are relative to average until the replacement level factor is put in there.

         Reply ↓
         1. birtelcom September 25, 2012 at 7:51 pm

            Exactly — the Rbat stats are telling you that Bichette was essentially an average level hitter in 1999, and that Petrick was above average but only had 79 PAs so only generated a couple of runs above average.

            Reply ↓
      2. bstar September 25, 2012 at 7:40 pm

         Richard, I read that comment where Dr. Doom gave you that quick-and-dirty for calculating Rbat. My suggestion to you would be to email Sean Forman directly about it. He unilaterally has responded back personally to me whenever I find an error on a player’s stats page, so he may just have a more detailed answer for you.

         I personally think OPS+ is the driving force of Rbat; Petrick had a 124 OPS+ in 1999 in only 72 at-bats. Bichette, as I mentioned above, was very close to league-average at 102, so, looked at from that perspective, is it really a big surprise that a 102 OPS+ yields an Rbat close to zero?

         Reply ↓
         1. Richard Chester September 26, 2012 at 11:16 am

            Thanks Topper, birtelcom and bstar. I guess the calculation for Rbat is not as simple as I thought it was.

            Reply ↓
   4. Hartvig September 26, 2012 at 11:25 am

      Bichette’s defensive short-comings were well known to both his former employers and to much of the public in general either from first-hand experience or via the accounts of various sportswriters long before he ever set foot in Colorado.

      And while Bichette may well be the best known qualifier for the title of the most over-rated as viewed by the SABR crowd I do think that Doc Cramer may actually be the all time champion. Even though he’s largely forgotten now in his day he was a 5 time AllStar even though they didn’t start playing the AllStar game until he was in his 5th season as a major leaguer. Still he was named as an All-Star at times over the likes of Goose Goslin, Earl Averill, Bob Johnson and in 1939 over his teammate Ted Williams.

      It’s hard to imagine that there has ever been a player who’s perceived value has ever been more at odds with his actual worth than Cramer.

      Reply ↓
6. topper009 September 25, 2012 at 7:15 pm

   Off topic, Brewers pitcher Miek Fiers has been struggling recently so he shaved his beard into a mustache that makes Wade Boggs look like a little boy. After throwing 1 inning tonight he will instantly lead the league in MAR (mustache above replacement, with Craig Counsell being a replacement level mustache player).

   If he makes 2 more starts with it he will ascend into Dick Tidrow/Rollie Fingers territory (remember you have to adjust for the league mustache index, in the 70s they were much more common than today)

   Reply ↓
   1. bstar September 25, 2012 at 7:45 pm

      topper, this should help the Brewers recover some lost MAR when John Axford sadly shaved his draping scraggle earlier this year. As bad as Axford was struggling, I probably would have shaved it off too.

      Reply ↓
      1. topper009 September 25, 2012 at 7:50 pm

         That like a pitcher taking a loss without allowing an earned run, he (Axford’s mustache in this case) gets the blame but it wasnt really his fault.

         Reply ↓
         1. bstar September 25, 2012 at 8:11 pm

            A cleanly-shorn John Axford is a sad sight to these eyes. It just looks so wrong.

            Reply ↓
            1. topper009 September 25, 2012 at 8:18 pm

               Now he doesn’t look like he could get into a PG-13 movie. Just sad
            2. bstar September 25, 2012 at 10:57 pm

               Speaking of the Brewers and hair abnormalities, what is up with Yovani Gallardo and his chia-pet Mullet Gone Wild? That thing’s taken on a life of its own; I’m worried it’s going to grow around his ears and eat his face, right when Yovani’s game has risen to a new level.

               Or perhaps that shock of hair has a Samson-like quality to it…
            3. nightfly September 26, 2012 at 9:53 am

               Last week when Gallardo popped up on a live-look in on MLB network, I immediately called out, “Hey, look, it’s Pascual Perez!”

               You’d think a bunch of old rec-league hockey guys would have recognized the reference.
            4. bstar September 26, 2012 at 7:50 pm

               Ha, I-285 Pascual! I-285 is a big circle encapsulating metro Atlanta; dude drove around the circle three or four times before he found the exit to the old Launching Pad, arriving at a game he was supposed to start 20 minutes before the first pitch. He was one-of-a-kind.

               He even used to bunt for base hits because of his raw speed.
            5. Jason Z September 27, 2012 at 12:56 pm

               bstar-I remember going to a game in Atlanta
               shortly after the incident you described with
               Pascual.

               They gave everyone a T-shirt that had a zany
               caricature of Pascual with a confused look as
               the long ago loops around Atlanta encircled
               his head.

               How I wish I still had that shirt and could post a pic here.
7. BryanM September 25, 2012 at 7:34 pm

   Is there a good source for the confidence interval of WAR? What I mean by that is how confident are we that a player producing 2 WAR playing every day had a ” better ” year than his teammate. Who produced 1 WAR? I read somewhere on BRef that. Team WAR was accurate to about 15 runs but I can’t find the reference. Putting defense aside for a moment , are we 70 percent confident that Jaime Navarro was better than Ross Grimsley? 90? 55? Is a 3 WAR Difference big or little?

   Reply ↓
   1. Ed September 26, 2012 at 8:48 am

      Okay BryanM’s question re: confidence intervals for WAR. I’ll write what I know though if others know more or differently, please chime in.

      1) I don’t think the concept of confidence interval really applies to WAR. Confidence intervals are normally constructed when you have a set of observations that are drawn from a larger set of possible observations. Think of the typical news story showing that candidate x is leading candidate y in the polls. Obviously they can’t interview every potential voter so they draw a sample, normally around 1,000 people. In this case, the sample serves as a proxy for the larger population of all voters. Or think of measuring your height. If we measured you height 100 times, we could construct a confidence interval around those measurements (we could theoretically measure your height an infinite number of times so the 100 measurements we took is just a subset of all possible measurements).

      2) What we really want to see for WAR is what is known as a Sensitivity Analysis. There are definitely assumptions that go into constructing WAR and the question is “how sensitive is WAR to changes in those assumptions?” For example, check out this chart that compares the various versions of WAR.

      http://www.baseball-reference.com/about/war_explained_comparison.shtml

      If you look at the 3rd and 4th columns (Fangraphs WAR and Updated B-R WAR), you’ll see lots of instances where Fangraphs and B-R made different assumptions/decisions. So the question is: what would happen if we changed those assumptions? What would happen if you took a case where Fangraphs said “No” and B-R said “Yes” and changed the B-R assumption so that it matched the Fangraphs assumption (or vice versa). Ideally you’d want to see this sort of analysis on each single component that goes into WAR.

      3) We could do a sensitivity analysis for WAR at the aggregate level (i.e., for all underlying assumptions as opposed to each individual one) by comparing Fangraphs WAR to BR WAR. The problem with doing so is that the measures are set on different scales, based on what they assume the level of a replacement player to be. That being said, let’s take a look at Miguel Cabrera. Fangraphs shows Cabrera to have 50.5 career WAR; BR shows him to have 44.2. That’s to be expected because of the scaling issue. But when we start looking at individual seasons, we see some interesting (strange?) results. For example, in 2004, BR actually gives Cabrera a higher WAR than Fangraphs (3.2 vs 2.8). On the other hand, Fangraphs has Cabrera much higher in 2007 than BR, almost twice as high (5.6 vs. 3.0). On the third hand, for each of the past 3 years, Fangraphs and BR have almost the exact same WAR for Cabrera (all years within 0.2 of each other).

      What to make of this? I don’t know but it does, in my mind, raise the question of sensitivity. Why was BR higher than Fangraphs in 2004 (14.3% higher) when typically it’s lower (12.4% lower across all years)? What the heck is going in in 2007? I don’t begin to have the data (or the knowledge) to answer these questions, but it is the sort of thing that BR, Fangraphs and others SHOULD be looking at (though I suspect they’re not).

      4) All that being said, we do know that a team’s aggregate WAR correlates fairly strongly with Team Wins. Fangraphs most recent version of WAR has been shown to correlate about .88-.91 with teams wins. Not sure about B-R’s WAR.

      Reply ↓
      1. Jim Bouldin September 26, 2012 at 9:52 am

         Ed,

         I think you are 1 for 2 🙂

         Your explanation of what sensitivity analysis is all about, and how it applies to WAR calculation is right on the money.

         However, it’s also true that confidence intervals and standard errors can be computed (and likely are, just never presented), because the computation of WAR requires first computing (1) “player runs” (and average player runs and replacement player runs as well), and (2) “player wins”, the latter being computed from the former). The only way that either of these metrics can be computed is via some type of optimization algorithm, which is to say, some type of regression equation. This automatically means there is uncertainty in those equations, which means there must be confidence intervals and standard errors. Furthermore, the uncertainty is additive between the two steps (the so-called “error propagation”), because the first step is input to the second step. Whether the uncertainty arising in the first step is included when fitting the regression equation of the second step (WAR from RAR) is unknown to me (haven’t checked the algorithms), but it *should* be if one wants to make a full accounting of the uncertainty. However, I have my doubts that it really is, because that requires a stochastic modeling approach that is more complex than just fitting two consecutive regression models to two sets of data. Somebody who really knows the algorithm in detail can correct me if I’m wrong on any of these points.

         So, both confidence intervals *and* sensitivity analyses are necessary if you want to fully evaluate the merits of WAR as a statistic. Not to mention of course, fundamental conceptual issues, such as “what exactly is a ‘replacement player’ and how do we define that?”

         At least that’s my undertanding and hopefully it’s understandable.

         Reply ↓
         1. BryanM September 26, 2012 at 10:50 am

            Ed, Jim. Thank you both for your very well written answers to my question. I initially had in mind the point Jim addressed. ( fitting regressions to event data to estimate runs, and then using estimated runs to fit to an estimate of wins) . As Jim says the data is freely available to those who calculate WAR and could usefully be presented at the same time. As it is, when B-R says that Andrew McCutchen is leading Buster Posey 7.0 to 6.6 in the WAR race it says he is ” probably ” having a better year. What is their estimate of how probably? Of course the whole concept of a “replacement” player contains estimation errors of its own…
            That said , I was blind to the issue Ed raises, namely the basis of the assumptions necessary to construct the metric, and sensitivity to them. That is another whole area for further gaps to arise between calculated WAR and effect on team performance . The. .88 to . 91 number for team WAR Is not particularly impressive. Bill James’ s Original Pythagorean formula from 30 years ago does as well, and it is a complete heuristic.

            Am I the only one who worries about this ,? Why isn’t the data-hungry community more curious.
            Thanks again, Ed and Jim
            .

            Reply ↓
            1. Ed September 26, 2012 at 11:13 am

               Jim – Thanks for chiming in since I think you know more about stats than anyone else here. (sadly all my advanced stats classes are 20 years in the past and I haven’t had the chance to keep up).

               Anyway, one question/comment…while I’m sure it’s true that there are regression equations that feed into some of the components that make up WAR, it’s not clear to me that the final WAR number “pops out” of a regression equation. Would you still be able to calculate confidence intervals if that’s the case?

               Anyway, here’s an interesting article by HHS own Adam Darowski, comparing Fangraphs and BR WAR for the top 200 players in career WAR. He finds that in most cases Fangraphs WAR is higher (the scale issue I mentioned) but there are 11 players who are higher on BR WAR. Several are 19th century players. But both Albert Pujols and Ichiro also show up on the list, two players with basically the complete opposite skill set. That should raise some red flags for anyone who thinks about it for even a half-second.

               http://www.beyondtheboxscore.com/2010/11/29/1839730/are-fwar-and-rwar-on-different-scales

               And yet, I’m not aware of these issues being discussed. Though there’s obviously a lot I’m not aware of. But I do agree with BryanM that we should worry/think/ask more about this sort of stuff. We, the consumers of WAR, are being asked to accept WAR at face value without really knowing how it’s being calculated. I do applaud B-R for being much more open about there methodology than in the past and yet it still seems like there’s a lot we don’t know. And when both Ichiro and Albert show up on a list, we ought to say, wait a second, what the heck is going on here???
            2. Jim Bouldin September 26, 2012 at 11:57 am

               No Bryan, you’re not the only one who worries about it. I have numerous concerns about the validity or usefulness of WAR, some serious, others probably (but not certainly) minor.

               As for “Why isn’t the data-hungry community more curious?”, my experience is that there is a pronounced tendency by people in general to treat statistics with more confidence than is warranted, and the procedures by which those stats were computed as a black box in which they are obliged to put their faith unless they can open the box and figure out exactly what’s going on inside it. Which many cannot.

               I should emphasize that this issue is decidedly *not* just a baseball sabermetric issue. Not at all. It occurs in any area where numerical analysis involving lots of different types of numbers in a complex system are involved.

               Probably 90% of the science work I’ve done over the last five years has involved the evaluation of whether certain types of metrics are computed and/or applied, correctly for the question or hypothesis being evaluated. And I can *guarantee* you that even top level scientists working on really important topics, can make serious blunders in their statistical/numerical analysis. The reason: quantitative analyses of complex systems require *extremely* careful consideration so as not to get one’s fundamental conclusions wrong in one way or another.

               I hope to have more to say later if I get time.
            3. Jim Bouldin September 26, 2012 at 12:18 pm

               “…it’s not clear to me that the final WAR number “pops out” of a regression equation. Would you still be able to calculate confidence intervals if that’s the case?”

               Ed, agreed that WAR definitely doesn’t result from a single regression equation. If it did, the answer is yes. In fact, it would be most simple to do so in that situation. Whenever you have a regression equation, then by definition, you are going to have standard errors and hence, confidence intervals. [Except in the trivial case where you have a fully determined equation with no random component, which is never going to happen.]
            4. birtelcom September 26, 2012 at 12:22 pm

               It is no coincidence that b-ref’s WAR correlates to team wins to the same degree as the pythagorean projection correlates to team wins. B-ref’s WAR was built to do just that. WAR is intended to reflect a player’s value in contributing to team wins by contributing to run scoring and run prevention. The pythagorean projection shows what team wins would be expected based on the team’s run scoring and run prevention. The gap between the pythag projection and actual team wins is the result of the (at least currently) unpredictable allocation or spread of run scoring and run prevention to each of the finite number of individual games in a season. That gap either represents random luck or some hard-to-identify run-timing skills or some combination of those two. WAR does not try to allocate to the players that foggy gap between run scoring/run prevention and actual team wins. So when you correlate WAR to team wins you will always find a cap on that correlation that is set at the correlation between pythag projection and team wins.

               The predecessor to WAR, Bill James’ Win Shares, did actually allocate the gap between between run scoring/run prevention and team wins to the players on the team, so that Win Shares correlates perfectly with team wins. But James just assigned the gap proportionally to all the players on the team — he didn’t have any better idea than the rest of us how to actually allocate the gap to particular players. But his theory was that player wins ought to add up to team wins, so he allocated the gap as best he could across all the players on the team. The result is that in Win Shares, the players on this year’s Orioles, a team that is heavily outperforming its pythag expectation, will get bonus Win Shares not ascribable to their run scoring and run prevention performance. WAR doesn’t try to do that kind of bonus, so the WAR numbers of the polayers on the 2012 Orioles will add up only to the team’s pythag projected wins not to the team’s actaul wins. http://joeposnanski.blogspot.com/2012/08/tango-on-war.html
            5. mosc September 26, 2012 at 1:08 pm

               WAR should still total up for all players on all teams to 30 teams times the quantity of 81 – whatever a replacement level team would win (say 30?). So I’d expect total WAR to be something on the order of 50*30 or 1500 for the league total.
            6. birtelcom September 26, 2012 at 3:51 pm

               mosc @61: B-ref uses a .320 winning percentage as replacement level. That’s 52 wins in a 162 game season. So for the majors as a whole there are 30 x 52 replacement level wins (that’s 1,560) and 30 x 81 total wins (that’s 2,430). That means MLB-wide wins above replacement should add up to 2,430 minus 1,560, which comes out to 870. In 2011, if you add up all the WAR (pitching WAR and everyday player WAR, combined) for all the players who played in 2011, you get 875. The tiny variance from 870 is probably the result of the fact that the WAR numbers for all the individual players are rounded to one decimal place, so you are bound to get some rounding error. But the numbers essentially balance — there should be almost exactly 870 total WAR across the majors each full season, assuming 162 game schedules for each of the 30 teams.
            7. bstar September 26, 2012 at 8:16 pm

               Glenn DuPaul wrote this article using graphs of selected years to compare rWAR team totals to actual wins:

               http://www.hardballtimes.com/main/article/what-is-war-good-for/

               Tango replied, saying basically the first half of the article is irrelevant because, as birtelcom says, rWAR is DESIGNED so that r=1 at the same-season runs level, while the old Win Shares forces the fit so that r=1 at the team wins level:

               http://www.insidethebook.com/ee/index.php/site/correlating_war_to_same_season_wins/

               I’m not sure why Sean Forman tried to touch on this subject recently, because all he did was muddy the water in his article about this subject on the front page of B-Ref.

               I guess I don’t really get why people need to see how well team WAR correlates to team wins when it’s obviously designed to do that.
            8. Ed September 26, 2012 at 8:53 pm

               FWIW, Dave Cameron also thought it was worthwhile to correlate team WAR with teams wins. So I dunno.

               http://www.fangraphs.com/blogs/index.php/war-it-works/
            9. bstar September 26, 2012 at 9:21 pm

               It makes a little more sense for fWAR since it isn’t forced to fit exactly to team runs or wins.
      2. Ed September 26, 2012 at 10:00 pm

         Ah now I see where I was confused. I had skimmed the DuPaul article before my post #37. He starts off talking about the Cameron article and I thought he was updating Cameron’s analysis. In my skimming, I missed that he was looking at B-R WAR. Thanks for straightening me out Bstar!

         Reply ↓
8. topper009 September 25, 2012 at 7:56 pm

   I remember going to this game during the 1998 HR chase at Old County Stadium where John Mabry started for Mark McGwire just give big mac a rest. Well as you can image no went to that game to watch Mabry or the Brewers, as McGwire would later say on the Simpsons, they wanted to see dingers.

   In his first AB Mabry was booed mercilessly by the crowd for seemingly preventing them from witnessing dingers, but he cranked one anyways.

   Reply ↓
   1. oneblankspace September 26, 2012 at 9:23 pm

      I was at a Midwest League game that night… and it was rained out in the 2nd or 3rd inning. Looked like a good pitcher’s duel, too.

      Reply ↓
9. no statistician but September 25, 2012 at 11:35 pm

   JA:

   Late to the discussion, but I think Keith Moreland and Bill Dietrich should get passes, since they were made butts of public, if comic, humiliation years ago by Steve Goodman and Gene Shepherd respectively. Isn’t there a penalty for piling on?

   Reply ↓
   1. John Autin Post authorSeptember 25, 2012 at 11:57 pm

      You’ll have to clue me in, amigo.

      Reply ↓
      1. tag September 26, 2012 at 6:32 am

         John,

         We’ll excuse your non-Windy City-centric worldview. Check out Steve Goodman’s “A Dying Cubs Fan’s Last Request” on YouTube for his remarks on Moreland.

         And I think Gene Shepherd’s remarks about Bill Dietrich, who even I’m pretty sure pitched a no-hitter (nsb?), are in “In God We Trust, All Others Pay Cash.”

         Reply ↓
         1. no statistician but September 26, 2012 at 9:28 am

            tag:

            Correct. Gene Shepherd grew up out your way in sunny Steelmill Land, I believe, where the air is fresh and pure. Just as an aside, I’m pretty sure Moreland and Goodman were friends, and Moreland’s voice can be heard in the chorus of “Go, Cubs, Go,” if they still play it on WGN. I’m no longer in Chi and have cease to follow the Cubs.

            Gene Shepherd’s remarks about Bill Dietrich, from what I remember of the routine, are kind of belittling—he wore glasses, is probably now working as a gas jockey in a service station, etc.—but his description of Dietrich’s no hitter in front of a crowd of a few hundred is kind of poignant.

            Reply ↓
            1. tag September 26, 2012 at 3:07 pm

               nsb,

               They still play “Go, Cubs, Go” in the ballpark after every win. Which of course is not very often these days.
   2. Hartvig September 26, 2012 at 10:38 am

      Another reason I love this site.

      As I type this, the Steve Goodman CD I just popped in my stereo has just begun to play- I had the pleasure of seeing him play live when I was in college in the mid- to late 70’s- and 4 books of collected stories of Jean Shepherd sit on my bookshelf just a short stretch from me.

      Great minds think alike, I guess.

      Reply ↓
10. John Autin Post authorSeptember 26, 2012 at 12:00 am

    Off topic: Tuesday night against Seattle, Zack Greinke racked up 13 Ks in 5 innings, but was done after 110 pitches. It’s the first known instance of as many as 12 strikeouts in an outing of 5 IP or less.

    There were 2 groundouts and a pickoff. One of the strikeouts reached on a wild pitch. One HR, 3 doubles and 3 singles; 2 walks, both among the last 3 batters Greinke faced.

    Reply ↓
11. Mike Felber September 26, 2012 at 5:39 am

    Thoughts on Belanger? He was excellent in the field, but his #s per game are otherworldly. Jimmy Piersall too: Rtot/year of 19 & 13, respectively! Belanger I understand had an otherworldly instinct for positioning. He had dWAR years of 4.0, 4.4, & 4.9, always missing some games! Though his range factor per game is actually .42 UNDER league average!

    Anyone have a good idea whether certain quirks of the pitching staff & other fielders exaggerated HOW good these guys were?

    Reply ↓
    1. Ed September 26, 2012 at 6:19 am

       Mike Felber – A few responses re:Belanger.

       1) dWAR is basically uninterpretable because it also includes the position adjustment. I really wish BR would stop publishing it because it seems to lead to a lot of confusion.

       2) The better number to look at is Rfield which shows the number of runs saved relative to the league average. Of course that’s a comparison to “league average” rather than “replacement” but it’s really the best summary number for defensive value.

       3) You cite Belanger’s range factor per game as being under league average. But BR also has range factor per 9 innings which seems like the more relevant statistic. In that case, Belanger is not only above league average, he has the 4th highest range factor per 9 innings all time. His fielding percentage was also above league average.

       Reply ↓
       1. Mike Felber September 27, 2012 at 2:26 am

          Thanks Ed. Where were you when I asked 2x for a justification for a 70.5% SB break even point, when elsewhere I have not heard above 2/3 argued a historical average? 😉

          I did not know Belanger was so high in RF/9, nor how that could be so very different from RF/G. But Rtot I think seems to sum up total defense well.

          Reply ↓
          1. Ed September 27, 2012 at 9:44 am

             Glad to help Mike Felber! The difference between Belanger’s RF/9 vs. RF/G stems from the fact that he was used as a late-inning defensive replacement a lot, particularly late in his career. Plus he was pulled a lot for pinch hitters. Even though he played in 1,971 games, he only started 1,691 of them and only played a complete game 1,467 times.

             As for the stolen base issue, a quick google search of “stolen base break even point” shows that most experts thing the break even point is 75%. That being said, the 2/3 you cited does sound vaguely familiar, perhaps some old work Bill James did?

             Reply ↓
12. Timmy Pea September 26, 2012 at 8:13 am

    Kieth Moreland played for Detroit as well, that’s not listed.

    Reply ↓
13. nightfly September 26, 2012 at 10:00 am

    I would like an honorable mention for Vinny “Cash-Stealer” Castilla, who racked up nearly all his career WAR value (15.9 out of 16.7) in Coors, where the air is rarefied. Just check the home/road splits. Somehow, he was even a worse base stealer on the road, one area where you’d imagine such a split wouldn’t hold.

    Reply ↓
    1. John Autin Post authorSeptember 26, 2012 at 10:08 am

       I think Volume 2 will focus on overrated guys like Castilla.

       Reply ↓
       1. Brooklyn Mick September 26, 2012 at 2:27 pm

          Wonder if Big Daddy Fielder will make the cut. Two home run crowns, three RBI crowns, back-to-back runner-up MVP, one of the most feared hitters in baseball in the early 90’s.

          In 1990 and 91 he accounted for 9.9 of his career WAR of 14.7. Take away those 2 years and he amassed a measly 4.8 WAR over the remainder of his career.

          Reply ↓
14. Brent September 26, 2012 at 10:11 am

    How much does lefthandedness (as a pitcher or hitter, not as a fielder) have to do with a long career without much WAR? The names Lenny Harris and Mark Sweeney (as well as Jesse Orosco) jumped out at me on this list. I am sure there are others.

    Reply ↓
    1. Brent September 26, 2012 at 10:20 am

       I hate to reply to my own post, but I thought of something else, since the list was made in respect to games played, as opposed to plate appearances or innings pitched, it probably contains more pinch hitters and Loogies than it would otherwise. Kind of hard to amass a lot of WAR when you have one AB/game or pitch to one batter/game.

       Reply ↓
       1. Brent September 26, 2012 at 10:25 am

          OH, geeze, never mind about the Loogies, but the point about players is still valid. (head in hands)

          Reply ↓
          1. nightfly September 26, 2012 at 10:47 am

             For the really long-tenured, it would probably be helpful to also look in detail at when they compiled WAR, and when they started giving it away. In the thread above, for example, someone mentioned Steve Blass, who undid half his entire overall career value in his final 90+ innings. Before then, he was usually an effective pitcher.

             A lot of these guys didn’t get chances despite being bad, they kept getting chances once they stopped being decent. Sidney Ponson, through age 27, was worth 12.6 WAR – not amazing, but a fair number for a middle-of-the-rotation guy. Then he fell off a cliff, and played for six different teams in five years trying to regain his form.

             Incidentally, JA, your listing of Jesse Orosco above is for his HITTING. For any reliever to be worth positive WAR with the bat is actually darned impressive. On the mount he was worth 21.7 WAR in his long career.

             Reply ↓
    2. Voomo Zanzibar September 26, 2012 at 11:01 am

       Lenny Harris pulled to the right, yes, but his value was as the perceived “super utility guy”

       485 games at third base,
       300 at second base,
       161 in right field,
       157 in left field,
       87 at first base,
       52 at shortstop,
       12 at DH,
       3 in center field,
       one scoreless inning as a pitcher

       And, of course, he was a prolific pinch hitter.
       Prolific in plate appearances, though not in results.

       His career slashes:
       .269 .318 .349 .667

       And cold off the bench 883 times:
       .264 .317 .337 .654

       Reply ↓
       1. Brendan Bingham September 26, 2012 at 8:13 pm

          It’s cool that Harris falls near the top of the list in Games and near the bottom in PA — consistent with his role as a pinch hitter.
          I would love to see the list of all-time leaders in WAR (or some other such metric) for pinch hitting appearances. It’s probably safe to say that Harris is not at the top of the list. Is Manny Mota, perhaps?

          Reply ↓
15. MikeD September 27, 2012 at 9:02 pm

    Just by the question, “Who is the all-time SABR whipping boy,” I thought of two names immedatiately: Dante Bichette and Joe Carter. I was approaching it, however, from the persepctive of which player the SABR community has devoted the most ink to discussing in an effort to balance perception vs. reality. Carter rightfully doesn’t make any of the above lists, but his value was a bit inflated.

    Now that said, while I do look at WAR as a solid guideline, I still don’t have enough faith in it to view it as anywhere close to an absolute. Between defensive data input and positional adjustments, it’s quite possible that a player showing a -2 could in reality be more valuable than a player showing as a +2. So maybe Bichette wasn’t really a -2.6 player in 1999, but even if he’s a plus 1 or 2 or so, it does shows that his value is nowhere near what his triple-slash stats would indicate.

    Reply ↓