The Cardinals’ Señor Octubre (aka Carlos Beltrán) is out of the game with an injury, said to be his knee – will have to wait and see how serious it is. Losing Beltran didn’t hurt too much today as his replacement, Matt Carpenter, provided all the runs the Cards would need with a two-run homer in the 3rd. The two bullpens were stingy again, allowing a combined zero runs on only 2 hits and a walk in 4.2 innings of work. The Giants cranked 9 hits, but only one for extra bases, leaving 11 on base with an oh-fer in 7 RISP opportunities. In comparison, the Cards went 2 for 4 in those situations.
Yanks and Tigers go tonight, and it’s do or die time for the pinstripers. They have their ace on the hill, so they are set up to at least prolong the series, and maybe start a comeback to get the ALCS back to the Bronx.
If you were trying to think of the last left-hander before Phil Coke to save consecutive LCS games, well, it’s been a while. Hasn’t happened since Randy Myers in games 2 and 3 of the 1990 NLCS. Coke is the first to do it in the ALCS. Last time in the World Series – Tippy Martinez of the Orioles in games 3 and 4 of the 1983 classic.
With Raúl Ibañez striking out to end game 3 with the tying and go-ahead runs on base, his legend, like mighty Casey’s (not that Casey), has now taken a back seat to … wait for it, courtesy of mlb.com, the Delmon Young legend. Seriously, the man’s a terror with 7 HR in his last 17 post-season games. Of course, in that span, he also has just 3 walks, an OBP under .300, and 15 Ks. But, hey, let’s not let a few inconvenient details spoil a good legend.
Enjoy the games, and tell us what’s on your mind.
A-Rod is benched again. Yes we know about his difficulties with righties lately but I seriously can’t see how Chavez is an UPGRADE over A-Rod right now.
If you just look at the full 2012 regular season numbers, and ignore past years, Eric Chavez actually does look like an upgrade over A-Rod. And a major upgrade if you take platoon differential into account.
It’s about to pour in St. Louis:
http://www.weather.com/weather/map/interactive/St.+Louis+MO+USMO1202:1:US
As a Tiger’s fan I’ll be more than happy to live with the memory of the Delmon Young legend, just so they don’t resign him for another year in the process.
Prior to this season, the Yankees have participated in 48 best-of-seven post-season series. They’ve been swept in three of those: the 1976, 1963 and 1922 World Series.
Three times being swept in 48 chances happens to be exactly what would be expected from a team that had a 50-50 chance to win every game.
Can we call 1922 a sweep? Isn’t there a tie in there?
Birtelcom, the link you posted about everything you ever wanted to know about 7 game series was very interesting.
On the flip side, how many times have the Yankees swept 4 – 0? The prediction would be 1/2^4 or the same 1/16 of 48 = 3 as was the case for being swept. Off-hand I’d say the Yanks have at least 2 to 3 times that number of 4-0 sweeps – how come it doesn’t square away as well?
Why have the Yankees been able to sweep more 7-game series (9 out of 48 tries) than would have been expected from a set of coin flips? Probably because the Yanks have so frequently been the more talented team on the field, even in the post-season (most of the Yankee sweeps have come in the peak eras of the franchise, which in the context of the Yankees franchise are very high peaks indeed). So the results are not random, but rather tilted in the Yankees direction.
Among the reasons the Giants have turned 9 hits and 5 walks into just 1 run through 7 innings:
– 4th inning, 1 out, men on 3rd & 1st, down by a run: Matt Cain sacrificed. Pagan then flied out.
So what if he’s a .127 career hitter? Take a shot! Lohse is a fly-ball pitcher, and Cain’s never hit into a DP. And he did single off Lohse in his next AB. BTW, Cain’s career BA against finesse pitchers like Lohse is .186 (37 for 203). Oh, and he has 6 career HRs in 457 ABs.
Drives me crazy when they won’t even try.
Hey, JA. I agree there wholeheartedly. Rooting for your Tigers except for tonight’s game.
Completely agree that he should’ve swung away, but the 0 DP statistic is silly. He has 85 career PA with less than 2 out and 1st base occupied (includes 1–, 12-, 1-3, 123). In those 85 PAs he has 51 SH, 14 SO and 1 BB. I would assume that some of those SO were PAs where he attempted to bunt and that some of the remaining 19 were situations where he bunted into a FC. So he really hasn’t had that many opportunities to hit into a DP.
Good point, Evan. Thanks for checking that.
As I suspected, the Yanks have had 9 sweeps of 4-0 in 48 such series for one in 5.33 instead if one in 16 (predicated on a 50/50 chance of winning each game. So (x/100)^4 times 48 = 9. Solving for x would indicate the odds the Yanks had of winning each game.
X = 65.8 or pretty close to a 2/3 instead of a 1/2 chance of winning each game. I’m too lazy to figure out what the Pythagorean Winning Percentage would be for those 9 series, but it’s probably pretty close to that .658 figure
You can’t do that Larry. You can only assess the likelihood that a specific, per game win likelihood (e.g. 0.5, 0.658, or any other hypothesized value) explains the full distribution of possible series outcomes (0-4, 1-4, 2-4 etc), or estimate the most likely such probability, (the maximum likelihood). That value in this case must lie somewhere between 0.5 and 0.658, but you can’t determinate it without enumerating all 48 actual outcomes and then doing some type of optimization that minimizes the error variance.
Jim, I had a sense that it would not be all that easy, but the calculation was fairly easy to make the comparison as to the Yankees being swept 3 out of 48 times (as if they and their opponent were coin toss equals) versus the Yankees sweeping 9 out of 48 (as if they were the powerhouse that had other major league teams as a de facto farm team or outspent them by 100 million in roster payroll.
That is why in another post, I suggested a look at the Pythagoren Projection to see if the Yankees played as if they were as dominant as .658. Actually, I think a look at the Pythagorean Projection simply obtained by adding up Runs Scored by WS Champs versus those scored by WS Losers (or for any level of postseason play) would give a thumbnail indication of how the odds tip from those of a coin toss.
It doesn’t make sense to combine run differentials from WS games played over the course of 100 years in greatly varied run environments.
Evan, that is most likely true, but Pythagorean Projections work very well within the scoring environments of each season. What I am trying to get a handle on is this – on the average, how much better is the WS Champ than the team they defeat? In Birtelcom’s calculations, the assumption has been that the teams are evenly matched and each game is 50/50. An in lots of situations it is spot on. One way to re-estimate it is to add up all of the Champions runs scored and allowed and do a Pythag Projection. Another way is to calculate a Pythag Projection each year and then average those.
More complex would be to run a Pythag Projection with each outcome: 4-0, 4-1, 4-2 and 4-3. It is not the most statistically rigid analysis, but it would seem to be good for a first estimate.
Yes, the ratio 9:3 gives a rough indication of which of those two possibilities is more likely to be correct, and therefore that 0.658 is likely a better fist approximation than 0.5 for an estimate of a single, constant win probability. What the data really tell you though, is that there is not likely any such constant win prob value (because if 0.658 is the “true” value, you are very unlikely to be swept three times because (1 – .658)^4 = once every 73 series). This puts one in the realm of estimating some unknown *mixture* of win probs over time, which turns into an analytical nightmare real quickly.
Jim, I was looking at it more from this angle: when the Yankees were “destined” to be swept 4-0, how much “better” was the opponent versus when the Yankees were destined to sweep, how much better were they? Sure, there is a lot of small sample size “noise” from 4 games played in one year. But the Yankees have played in a sizable portion of all post season games ever played and the advantages they have had over the rest of the league have held fairly constant over time when they are in contention.
I wanted to return to a variation of question I posed a few weeks back. If you look at the powerhouse teams of the past (let’s say, a WP of at least .667), in season, how many of them had losing streaks of 4 or more games? It’s obviously not a perfect match, because a WS opponent is more likely to be tougher and roster usage is different during the regular season, but maybe that helps you contextualize.
Referring to your first paragraph there btw.
When you said Casey, I thought Sean.
He was after all, the mayor of first base! Walked the dog here in Detroit while waiting for the delay, beautiful day turned into awful night.
Obs @ 11,
The Casey reference was to the fictional ballad “Casey at the bat”, published in 1888. The Mudville nine lose when their star slugger Casey strikes out to end the game (as Ibanez did in game 3).
If the Yanks and Tigers don’t get the game in tonight and the Yankees are able to extend the series, the off day between 5 & 6 would disappear. If the Yankees were to force a game 7 Sabathia would be unavailable to start unless they are willing to bring him back on 2 days of rest. These are a lot of ifs, but it’s supposed to rain for quite awhile in Detroit and I’m waiting for the NL game to restart so idle speculation seems like a good diversion.
And now the Yanks and Tigers are officially postponed.
It cuts both ways, Evan. By not starting the game in spite of playable weather at game time (which lasted at least 1.5 hours), they avoided New York’s worst-case scenario — a repeat of last year’s ALDS opener, when CC got interrupted by rain after 2 IP and then was unavailable until game 3.
I’m pretty sure the Tigers would have been happy to get tonight’s game started and take their chances with the weather.
I don’t disagree. Though in retrospect, knowing the forecast and the possibility of a rainout or lengthy delay, Girardi might have been better off using Sabathia on short rest yesterday to ensure that he’d be able to pitch in 2 of the final 5 games if they take place.
For the record, the Yankee postponement is a good thing. After yesterday’s doubleheader of stress (debate, Yankee game) I think I needed an extra day’s rest. I was a little stiff warming up in the bullpen before the scheduled start of the game…
This is exactly why (well, one reason why) the Yankees should have started Sabathia in game three. It’s not all that hard to look at the 2 to 3 day forecast and factor in the possibility of a rain out and how it might change your plans. Could turn out to be a fairly serious mistake.
I’m starting to get the feeling that Girardi doesn’t heed my suggestions. Just trying to help the Yanks out–God knows they need it.
29 teams have started a series 3-0 to set themselves up for a sweep (not counting 1907/1922 WS which had early ties). 23 of them finished the job in the next game.
Of the six teams that lived to fight another day, half were knocked out in Game Five (All in WS: 1910 Cubs lost to A’s, 1937 Giants lost to Yankees, 1970 Reds lost to Orioles).
The 1998 Braves and 1999 Mets both won Games Four and Five of the NLCS before falling in Game Six.
Of course, the 2004 Red Sox were the first team to even force a seventh game, let alone come all the way back and win it.
If the Yankees are entertaining any thoughts of making an unlikely comeback in this series, then tonight’s rainout is bad news. They were set up with Sabathia, Pettitte, Kuroda and then Sabathia again for games 4-7. All have been great throughout the postseason. Now they’ll be forced to go with Phil Hughes and his bad back again vs. Verlander if there was a game seven. If Phil Hughes is DL’d, then they could bring back Nova or Garcia. Death by fire or water?
The hole they’ve dug just got deeper.
The ’99 Mets led Game Six in the 10th, but some closer couldn’t hold it.
However much I loathed having said closer on my team, I was glad he stuck around long enough that I got the chance to boo him in the flesh.
So why is the High Heat Stats logo that appears as a default in all posts (unless you have your own avatar) gone away on some postings? It no longer appears on mine since yesterday, and I’ve noticed it for others, yet for most it still works.
Wish I could answer that, MikeD, but if I knew anything about avatars, I’d probably have my own. 🙂
Yeah, I also used to have an avatar, that linked to my link.
Used to know how to change it, too.
Thanks. I’m not even trying to advance to my own avatar yet, happily living with whatever default logo HHS, or prior B-R, would give me!
Seeing that “X” next to my name reminds me of an insurance TV commercial from twenty years back that would open with the picture of a family, accompanied by an off-screen, ominous-sounding voice asking the unpleasant question, “what would happen if you were gone?,” as a big X was then being drawn through the man in the photo.
In the Giants post-game show on the radio, Jon Miller said that he spoke to Lon Simmons (the previous “voice of the Giants”) on the phone during the rain delay.
Lon said that a star getting knocked out during the LCS is never good for the Giants.
He recalled how in 1971 somebody got hurt and a nobody named Bob Robertson came into the game and hit Three Homeruns!
And in that same series some other Pirate who was from the Santa Clara area couldn’t go and his replacement pitched a two-hitter!
So, I just looked it up – and there was no two-hitter, and Bob Robertson was the Pirates starting first baseman the entire year.
Even broadcasters get worn out by 210 minute rain delays, folks.
And they end up quoting 90-year old men without checking their facts.
http://www.baseball-reference.com/boxes/SFN/SFN197110030.shtml
______________________
Let’s Go Freak!
Lon Simmons was probably mixing up the 2-hitter Nellie Briles threw against Baltimore in the ’71 World Series. It is true that Briles had not been in the LCS rotation. BTW, Pittsburgh started used 6 different SPs in the ’71 Series.
http://www.baseball-reference.com/boxes/PIT/PIT197110140.shtml
P.S. Briles was born in Dorris, CA and went to Santa Clara University.
Dorris, CA, Siskiyou County, State of Jefferson. Not quite in the middle of nowhere, but as the saying goes, you can see it from there…
Klamath River; Mounts Shasta and McLoughlin; Modoc Plateau; edge of the Great Basin. More antelope and mule deer than people.
God’s country, to the core. Man do I miss it.
Ellis
Johnson
Blass
Walker
Briles
Moose
Blass
Looks like Dock Ellis.
His wiki page mentions elbow pain after Game 1.
His wiki page is also the most tragic and entertaining of any baseball player.
http://en.wikipedia.org/wiki/Dock_Ellis
Voomo,
If you find Dock’s wiki page interesting you owe it to yourself to read “Dock Ellis in the Country of Baseball” by the poet Donald Hall. Well worth your time.
As for the position player, I have no idea what he is talking about. The only position that seems to have a player who was the “regular” during the season not play very much is shortstop where Jackie Hernandez was playing for Gene Alley, but Alley did play a little in both the LCS and WS and it seems to me that it was a concious choice to play Hernandez over Alley, not something to do with an injury.
Giants killed themselves last night by stranding what…11 runners in the first 6 innings? IMO, Cain should’ve had a chance to swing away. I’m so tired of St. Louis and their random players with big hits, it’s disgusting but that’s usually how playoff baseball works out. Shut down the big guys and the little one’s beat you. Hunter Pence is a rally killer and they won’t pitch to Posey. Time to make adjustments if your Bochy.
It’s all on Timmy now…I hope he can come through for us.
By the way, I hate the complete bias on fox for the Cards. McCarver(one of the worst color guys EVER! and Joe Buck who’s just so full of himself, seriously?). Bad enough when you play them but to have hometown guys calling it sucks. Also, the pregame with A.J. Pierzynski and Eric Karros is terrible. Karros’ lisp(Former Dodger) and A.J.(who nobody likes…anywhere)is just icing on the cake for me as a Giants fan.
Buck I would acknowledge as a Cardinals homer, though if you read anything from Cardinals’ fans, they think he doesn’t show his true colors enough. But McCarver? Sure he started his playing career with them, but as far as I can tell, never worked for them as an announcer (his local stints as an announcer were with the Phillies, the Mets, the Yankees and, ironically, the Giants). Cardinals’ fans definitely do NOT think of him as favoring them at all, quite the opposite, part of which stems from his work with two of their NL East rivals, the Mets and the Phllies, in the 80s.
McCarver just adds no insight whatsoever, which for a former catcher, is very lame. I never learn anything from him. The ESPN announcers are just light years ahead of FOX. I’m listening to the ALCS on ESPN radio and virtually everything Hershiser says is insightful in some way, and Dan Schulman, unlike Joe Buck, does not appear to made of cardboard.
Schulman is also an outstanding NBA announcer. Also, Ron Darling on TBS is very good.
Hershiser was great to listen to. Insightful and effective with his words
Another guy to avoid is Rick Sutcliffe. Somethimes has useful insights, but his voice is hard to listen to, and he places too much emphasis/significance on the points he makes. Same critique for Boomer Esiason doing the NFL.
I feel your pain, Jeff. As a Cubs fan, I not only get the Cards, Buck and McCarver but A.J. f***ing Pierzynski thrown into the deal as well.
Go Timmy and go Giants!
BTW, the phrase “St. Louis and their random players with big hits” is pure gold.
They’re not pitching to Posey.
Only Pablo can protect.
I hope Bochy flips 3-4.
As far as I can tell, fans of all teams have a dislike of the national broadcast teams from Fox to TBS to ESPN. I join them in their dislike. Fans of all teams are convinced the national broadcast teams have a bias against their home teams, which of course is impossible. They can’t biased against everyone.
John Smoltz is the one who drives me crazy on all the broadcast teams. Part of it is because I think he should be smarter, especially when talking about pitching. He never ceases to amaze me in how unprepared he is, with at least one major head-scratching comment every broadcast. They’re Joe Morganesqe. One example. The other day he was talking about how Phil Hughes has difficulty getting ahead of batters, but once he does he knows how to put the batter away. I was sure I misheard him, but he repeated it several times. It’s the exact opposite. That leaves the viewer wondering if he’s even trying. Yet when he was pitching he would certainly understand the strengths and weaknesses of various players. How can he not now as a broadcaster?
Mike, you just described in good detail a problem that plagues all national broadcasters: even if they study and prepare, they’re not really going to understand the true intricacies of a team well enough to get near the truth of what’s going on with certain players. Hardcore fans know these things. Broadcasters can’t religiously follow every team out there, and it’s quite possible they’re as prepared as can be but can be entirely wrong on many, many things about a particular team or player.
I think that’s part of the problem, and I don’t know that it’s fixable. Obviously, ESPN vastly improved their Sunday night telecasts this year and the overall product is so much the better for it. Still, even though Orel Hershiser and Terry Francona are quite good, I winced every time this year the Braves got a Sunday night game because I would have to listen to broadcasters who don’t really understand the underlying dynamics of a team all that well. Even though the Braves broadcast team is average at best, I’ll take them over ESPN, Fox, or TBS national broadcasters any day because they know the team and players so much better.
Over on MinorLeagueBall they took at look at how the final four playoff teams have acquired their players. The Cardinals have the most drafted players by far with 16. By contrast the Tigers and Giants have only drafted 10 of their players and the Yankees only 8. Summary is here:
http://www.minorleagueball.com/2012/10/14/3504194/origins-of-the-playoff-players
I tried posting links to the more detailed analysis for each of the 4 teams but HHS thought I was posting spam. So if anyone is interested, they can go to:
http://www.minorleagueball.com/
Only at game 3, but this is already only the third LCS round ever with two different pitchers recording a 2-inning save. Others were 1991 (Duane Ward, Bob Walk) and 1976 (Pedro Borbon, Steve Mingori).
Brad Lidge (2004), Mariano Rivera (2003), Byung-Hyun Kim (2001) and Dennis Eckersley (1988) have all recorded two 2-inning saves in the same LCS series.
Off-Topic: RIP Eddie “The Walking Man” Yost. Led the league in walks six times and finished his career with a .396 OBP.
More off-topic -what’s with all of the “Eddie’s” in the 1940s/50s who walked a lot: Eddie Joost, Eddie Stanky, Eddie Lake? That not even mentioning Eddie Mathews, who was in a different class as a hitter.
Apparently _someone_ was paying attention to OBA BBJ (Before Bill James).
Most of the walks in the 40s/50s were from Teddy and the Eddies.
On the serious side eight of the ten highest seasonal walk rates (since 1901) occurred for the years 1947, 1948, 1949, 1950, 1951, 1954, 1955, and 1956.
Richard, I think that you can infer that tells you what managers and GM’s must have valued during that period.
And those GMs and Managers got what they valued.
Looking at rolling 10 year averages across MLB, the 13 highest walk rates per team game are for the decades ending in the years 1948 to 1960, ranging from 3.49 (1939-48) to 3.74 (1947-56).
Good one Richard, and half of them are HOFers.
Ah, for the stat minded in us, Edward was the 9th most popular boys name for babies both between 1920-29. It’s the 131th most popular for boys born between 2000-2009.
Here are the only players having careers (min. 3000 PA) with OBP more than 50% higher than BA.
– Max Bishop
– Eddie Stanky
– Eddie Yost
– Gene Tenace
– Adam Dunn
– Mickey Tettleton
– Eddie Lake
– Eddie Joost
That’s funny…50% of the players having careers (min. 3000 PA) with OBP more than 50% higher than BA were named Eddie. Ah, what’s in a name?
Career OBP >= 0.375 – 182 players. Here’s how they break down by OPS+
175+ – 5 (Ruth, Williams, Gehrig, Bonds, Hornsby)
150-174 – 23
140-149 – 26
130-139 – 42
120-129 – 43
110-119 – 28
100-109 – 13 (incl. Yost)
< 100 – 2 (Billy Goodman, Rick Ferrell)
For everyone’s amusement. I’ve coming in in the train and we have just gone over the Harlem River Bridge. Underneath is an eight man rowing scull. Rowing as fast as they can away from Yankee stadium. The stench must be overpowering
OK, one last try for my construct. The ’27 Yankees were a pretty good team, probably the best team in history. If any team would have been expected to sweep their opponent in the World Series, it would have been the ’27 Yankees. So, what happened? They faced the Pirates and outscored them 23-10 with a 4-0 sweep. The Pythagorean Projection was a .841 winning percentage. So, I’ll use that as their expected odds of winning each game (instead of the default coin toss of 50/50). Thus the Yankees odds of sweeping the Pirates were (.841)^4 which is .500. I think it is very reasonable to assess the ’27 Yankees as having been so much superior to the Pirates that it was 50/50 that they would sweep them.
I don’t know how to populate spreadsheets with B-R data. I think a sum total of all of the Champs runs scored versus their runs allowed and running the Pythag Proj on those would give a good first estimate of how much better a WS Champ compared to the loser.
Larry, I think I be misunderstanding you – don’t the odds we’re interested in need to be prospective, rather than retrospective. It seems to me that what you’re saying is that the odds of a team that has outscored another team 23-10 over the course of four games winning each and all of those games is 50/50; that says something about the manner in which de facto superiority was likely to have been distributed over those games, but nothing about the relative strengths of teams going in, I think. A team that sweeps will always outscore the other team overall. Were the odds of the Miracle Braves sweeping the A’s good in 1914? The Braves outscored the A’s 18-6: it was .900^4, or 60/40 that they’d win each game and they did, yet the histories tell us people were surprised.
I *may* be misunderstanding you. (I be ungrammatical – no doubt there.)
or, simpler, since we are trying to estimate probabilities of teams winning ex ante , by definition , we must use only data that was available before the start of the relevant series.
And yes I know the game is in Detroit
Mike D,
John Smoltz got to pitch the bulk of his career to a 30″ plate. So he basically knows jack squat about pitching. It would be like having a pitcher from the 45′ era where the batter got to call for a high strike or a low strike commenting on the modern pitcher. He just needs to shut up and give thanks for all the gift strikes he and his staff mates milked out of the system.
Octavio Dotel is something else. Every single post season, there he is, albeit for a different team each year. Quite a unique career. I think he was a starter, then a closer until Brad Lidge turned the lights out on that phase if his career, then a very serviceable talent for set-up.
The Yankees have not had a lead the entire series. In the 1989 WS of San Francisco vs Oakland vs the earthquake, SF did not come to bat a single time after Oakland had an at bat that they weren’t behind.
Tigers win!
Bring on the Cardinals or the Giants!
Delmon Young number 1500 om B-R among all-time batters. Number 1501: Joe Girardi! Gotta love baseball.
Just checked- the only 2 Tiger holdovers from ’06 are Verlander and a retread Omar Infante. The Cards have 4 with Molina, Schumaker, Wainwright & Carpenter. I’d love to see a rematch.
And as much as I appreciate Delmon’s post season heroics I’m really hoping the Tigers are willing to part ways this coming offseason.
Also Ramon Santiago, who like Verlander here every year. Maybe we re-sign Delmon and he fakes an injury for the regular season. He’s like the John Salley of the Tigers, or a poor-man’s Robert Horry.
Or like Charlie Victory Faust.
http://www.baseball-reference.com/bullpen/Charlie_Faust
For some reason I think of the 2003 Ramon and the Ramon of today as two different people. Good catch.
Seems like a lot of turnover in 6 years but the reality is that if there hadn’t been we wouldn’t be in the World Series today.
Wow.
Wow.
Wow.
Ooof.
Congrats to Hartvig and John A.
Back to politics I guess.
This Yankee club looks a lot like the ’64 Yankees who weren’t heard from again for a dozen years
How? CBS bought the Yankees after 1964 season (well, 80% of it),and I think _that_ was the main cause of their decade+ decline. Also, the amateur draft was introduced in 1965, which put the other MLB clubs on a much more even playing field when competing with the Yankees for the top prospects. No such drastic event has happened (yet).
Also, the Yankees still have the advantage of two gigantic revenue streams, in their TV/radio contracts and their new stadium, that no other MLB can match.
The big advantage the Yankees still have is that they can absorb the cost of getting rid of (most of) their free agent mistakes, that might cripple most other clubs with smaller payrolls.
Lawrence, I think you crystallized the reason(s) that baseball fans who hate the Yankees hate them with such a passion. “Oh crap, A-rod is more interested in chasing tail than playing baseball and we still have years to go on the biggest contract in the history of sports. No problem, we’ll just absorb the cost and move on.”
But baseball thrives when the Yankees thrive so the rest of us just have to grin and bear it and take up the dutiful role of being the Washington Generals so Yankee fans can wallow around in their largesse.
@ EPM and BryanM,
I agree that a Pythag Proj is not a “prospective” way to predict things for the future. However, it is a su
(continued) it is a superb way to retrospectively see how a team should have done, or what the odds were that they would have done what they did. That is a pretty good start. For example, the Astros got swept by the ChiSox in the 2005 WS. But it was very close every game with the ChiSox outscoring by only 20 to 14. That is a Pythag Proj of .671 with the odds of a 4 game sweep of only 20%, which seems exactly right.
With 100 years of post season results we can use Pythag Proj to see what the odds were to have what happened to happen – better than just a first guess 50/50. Then, if two similar clubs face each other, we can use those odds to make predictions.
I think WAR is kind of like that. Runs created are turned into Wins by use of a Pythag Projection. Then the WAR is used to compare players and predict who should go into the HoF or who got the better of a trade.
And remember, at a quantum level, causality goes back and forth. The Pythagorean forces are so great in baseball that they seem at times to be prospectively causal, odd as that may sound.
So, what were the odds of Pittsburgh winning the WS in 1960?
Or, if Washington had managed to hang onto their 9th inning lead in game 5 of the NLDS, what would have been the odds of winning that series, despite being outscored 29-16.
For an example of the prospective nature of Pythagorean forces, consider the nature of “regression to the mean”. A team that over or under performs in relation to its Pythagorean Projection has a tendency to get closer to the mean (closer to the Projection) the next season. That is why you see a Tony Pena getting manager of the year one year and quitting a couple of years later:
Larry, Well, we do use the pythy projection retrospectively for the regular season of 162 games, but it’s not about odds of winning, it’s an assessment of performance over talent, and one that works because 162 games is a span long enough for regression to the mean to work its magic. Four games, a Series . . . I don’t know.
As for regression to the mean, I’m not sure that concept is being used properly in the contexts we’ve been using it. Regression towards the mean is a reflection of level-playing-field conditions. Talent differentials, payroll differentials, these sorts of factors tilt the mean within various sectors: for example, the Yankees’ overall W-L record has resisted regression towards the mean for many, many decades, and the Browns never regressed – progressed. Regression towards the mean in career performance for many players isn’t that at all: it’s the effects of aging. (It would be interesting to add an age-factor to WAR formulas.)
As for mixing regression with Pythagoras, I’d need evidence to agree that a team that deviates from its projection by N in Year 1 will deviate by <N in Year 2, though I imagine you could do that if you found an optimum margin of error to apply and only consider N beyond that, but you'd have to find that margin empirically. But I don't think that's actually regression towards the mean either, since the teams in Year 1 and Year 2 are significnatly different teams, with new or aging players, playing different opponents. It seems to me the question would be whether *any* teams – or how many teams – year-to-year exceed the margin of error.
However, I'm with you 100% on how indeterminancy and direction of causation in baseball resembles quantum mechanics. I don't understand either of them.
epm,
I’ve been debating some of these topics with Jim. My latest missive is here.
http://www.highheatstats.com/2012/10/where-have-all-the-good-teams-gone/#comment-41761
In activities that combine skill and luck, and hence necessarily exhibit reversion to the mean, I contend that skill can best be regarded as a hindrance to the reversion process. (Jim disagrees.) But reversion to the mean is a tricky concept. What’s crucial to understand about the process is that change within the system occurs at the same time that the system itself doesn’t change. Change and no change operate side-by-side: the change part obviously is reversion to the mean, but we can’t assume that all results revert to the mean, which would effectively mean that the standard deviation of outcomes narrows over time. This is not true. I think a lot of people make that assumption and it’s what trips them up.
The Browns, I think (I don’t know their long-term record), did “regress” – because “regress” in this sense just means “draws closer to the mean,” which here would mean to improve. It’s why I prefer the term “reversion” over “regression” – it just seems semantically better.
And you’re dead-on about quantum mechanics. Trying to understand that stuff is just a nightmare.
I’ve been following the back-and-forth between you and Jim, tag, and I agree with you in granting the premise that components of randomness are governed by the rule of reversion/regression towards the mean, but holding with Jim that far less in baseball is random than it may appear. On the basic issue, I think I gave this one my best shot at #55 on the same thread you and Jim were debating on, trying to catch birtelcom’s attention (with trepidation: it’s presumptuous to argue with the avatar who commanded baseball’s Law!). But I also think there is a basic category error going on, reflected in Larry’s posts, that conflates prospective and retrospective action: “Luck” is basically an explanatory fiction for the causes of past events that we can’t identify and which are personified collectively as a powerful agent, stripping actors in past events of the particulars of agency that they actually possessed. The way I see it, when applied beyond a narrow range, “Luck” is just “Fate” brandishing a finite math textbook.
As for the Browns, there were times it did appear that their luck was evening out: 1906, ’08, ’16, ’21-’22, ’25, ’28-’29, ’42, ’44-’45 — all of those sunnny years when they exceeded .500 in the 51 seasons after their initial stong debut in ’02. Their 18 sub-.400 seasons are also pleasingly spread out. Like their descendants, the 2012 Orioles, they broke the Law in their persistance, but in their case, there wasn’t much mystery about it. They fielded really bad teams.
““Luck” is basically an explanatory fiction for the causes of past events that we can’t identify … stripping actors in past events of the particulars of agency that they actually possessed. The way I see it, when applied beyond a narrow range, “Luck” is just “Fate” brandishing a finite math textbook.”
BINGO!!!!!!!!!!!!
Beautifully stated epm. This has been (part of) my point all along, and I made a similar (but not nearly so eloquent) statement maybe a month back or so. This is really the crux of the biscuit here as far as I can tell, and is exactly why I have said that people have to be very careful in their use of the term “luck”. What people are doing is confusing random variation and/or sampling error, with luck, and throwing around terms like “regression to the mean” (which, apparently, James and Tango and others, are very fond of), without careful distinctions about *exactly* what that phrase means in terms of signal to noise ratio (i.e. in ability to detect cause and effect relationships), and the *temporal scale* at which the concept should be applied and/or evaluated with actual data. This is extremely important.
You’re the only one I’ve seen so far who seems to understand this point.
Just logging back on . . . Thanks for the kind words, Jim. I appreciate them. I have to say, though, that if I found that the only person agreeing with me was me, I’d be worried. (It’s more hopeful for me, since the other person on my side of the argument is you!)
The problem with ascribing randomness to events in the manner being suggested here, it seems to me, is that it makes all results random.
E.g., DiMaggio’s 56-game hitting streak is just a random variation; i.e., if you set an infinite number of greater apes up with bats and balls and had them play an infinite number of baseball games, one of them would hit in 56 straight games. Therefore all human endeavor is essentially meaningless, being governed by, you guessed it, regression to the mean, and Joe D didn’t really do anything that was special compared to myself who can explain why his efforts had no bearing on what happened.
In other words, sometimes HHS means Happy Horse Patooties.
Re baseball, randomness in any meaningful sense stops with the first pitch, mathematical and philosophical hindsight notwithstanding.
nsb: Arguing for the presence of randomness does not mean everything in baseball is random, only some portion — baseball is a game of probabilities, where talent and randomness intersect. Joe D. was far more likely to have a 56-game hitting streak than most players, because he was one of the greatest hitters who ever lived. He was far more likely to have a 56-game hitting streak than many other great hitters because he didn’t strike out a lot (far less frequently than, say, Babe Ruth) and he didn’t walk as much as great hitters such as Ted Williams or Barry Bonds. So DiMaggio had a much higher probability of putting together a long hitting streak than most players in history. That he actually reached 56 did, though, take a certain amount of randomness falling his way. Joe had just one hit in a game 34 times during the streak. Surely it took a certain amount of luck that not one of those 34 lone hits didn’t end up as a liner caught or a hard grounder to the infielder instead of up through a hole (in fact the streak ended on two such hard hit balls that were fielded for outs but could easily have been hits). The reason DiMaggio’s streak is so wonderful a landmark and such a unique event is precisely its combination of extraordinary skill and lightning-strike chance. If all it took were skill and effort to get a 56-game hit streak, others would have matched DiMaggio. But it took skill and effort plus some favorable falls of fortune to create such an extraordinary accomplishment.
birtelcom,
Nicely put. It’s what I’ve been arguing for the past three months. I even linked to several studies about these things, including one about DiMaggio’s hit streak that directly addressed the confluence of skill and luck needed for such a streak.
Long streaks of success in sports tend to be held by very skillful players. The NBA record for consecutive field goals made is held by Wilt, at 18. The Stilt made 54% of his FG attempts over his career and set the single-season FG% record of 72.7%.
In the NHL, Gretzky, the league’s leader in career goals, assists, and points, holds the record for most consecutive games with a point.
And in baseball, who is second in the consecutive hit streak? Pete Rose (I think), the all-time leader in hits. And I think third is Molly, isn’t it? Another Hall of Famer singles hitter.
You have to have a lot of skill to accomplish these records. But you also need a good deal of luck, more in the case of baseball than in basketball or hockey. I just don’t see why people think skill and luck are mutually exclusive.
tag, by Molly do you mean Paul Molitor? He’s seventh all-time on the hitting streak chart.
Rose is third(44) with Wee Willie Keeler second(45), and Bill Dahlen(42) is fourth, then Sisler(41), Cobb(40), and Molitor(39).
Well, I always thought Rose TIED Keeler’s 44-game streak (the NL record) but Baseball Almanac, my original source, lists Keeler at 45 games.
Not sure what’s going on there. Other sources say Keeler and Rose are tied at 44.
Yes, bstar, by Molly I meant Paul Molitor. I thought he’d made it a little higher, but seventh is good.
I always thought Rose had reached second place too, but Wee Willie Keeler will work just as well.
Keeler hit safely in the last game in 1896 and then the first 44 of 1897. So whether he has a 44 or 45 game hitting streak depends on whether you include the last game of ’96 in the streak.
“I just don’t see why people think skill and luck are mutually exclusive.”
Just a wild guess but it might be because NOBODY IS IN FACT SAYING THAT, but rather that you keep thinking, for unknown reasons, that they are.
What some people have definitely been saying (or implying), by contrast, is that the Orioles will necessarily do one or more of the following due to the supposed iron law of “regression to the mean”: (1) decline in their overall W-L record to something predictable by a pythagorean expectation, or (2) decline in their ability to win one-run, extra-inning or otherwise close games.
And people are further concluding, that if one of those things does happen, that the explanation for this year’s Oriole success therein is therefore due to “luck”, without ever even defining exactly what does, and does not, constitute “luck” in baseball games. You just use it as some catch-all phrase for post-hoc explanations of things you can’t otherwise explain, as epm and I have both noted.
For birtelcom, briefly:
When you talk about the end of DiMaggio’s streak as a result, more or less, of randomness catching up with him, you make it seem as if Ken Keltner was just a cog in the the great randomness machine, and not a player out there doing his job. I don’t see randomness on the game level in this at all, unless one ignores the human factor. The truth is that human will and effort aren’t random except when viewed from the detached perspective of the social scientist—all us rats running around in our respective mazes, how interesting!
Jim @123: I think there is some risk of confusion in mixing discussions of the role of randomness in long hitting streaks vs. the role of randomness in teams exceeding pythagorean expectation. Long hitting streaks are a combination of a fundamental baseball skill (the ability to get base hits in a comparatively large percentage of plate appearances) and an area with a large swath of randomness to it — the timing of when successful hits happen to occur in a small sample. On the question of teams exceeding PE, in contrast, the fundamental skills involved (runs scoring and run prevention) are already being controlled for, and the only variable is the timing element. A comparable “how much skill/how much randomness” question for hitting streaks would only arise if we put the question as something like this: “If we control for a player’s talent in getting base hits as a percentage of his plate appearances, then how much of the remaining variation in his likelihood of achieving a long hitting streak is randomness and how much is a skill in controlling the timing of his hits?” The answer to that limited question is again, I believe, much more randomness than skill.
I’d like to try using this hit streak theme to clarify what seem to me divergent ways of constrasting luck/skill. As a thought experiment, let’s posit two different worlds on 6/21/41.
In World 1, DiMaggio faces Dizzy Trout in the fifth. Trout throws a curve over the outside part of the plate – DiMaggio slaps the pitch sharply towards right for a single, passing inches beyond a diving Gehringer’s glove. The Clipper’s streak reaches 34.
In World 2, DiMaggio slaps the pitch sharply towards right, just within the reach of the Mechanical Man. Joe is out at first and his streak ends that day at 33.
The question is, in World 1 (meant to be ours, although I made up details), what caused the ball to pass inches to the right of where it did in World 2? Was it luck, or was it the specific motor responses of the DiMaggio who lived in our world, as opposed to the motor responses of the disappointed Joe in World 2?
I think the “luck” response would have to rely on the idea that in baseball, because of the difficulty of hitting Trout’s curve, DiMaggio was no longer able to exercise full control, and so his swing was, say, 57% governed by his agency and 43% governed by things other than his agency (or choose any other numbers). I’m not sure what those other things would be, but let’s say: random motor responses that do not reflect his focused control or his history of training – so, in the end, 57% skill, 43% luck: he gets credit for slapping the ball sharply; he’s just lucky that the motor elements he did not control produced or did not misdirect a hit-worthy vector. The difference in the two worlds lies largely or entirely within the 43%: there were a range of swings that would all be DiMaggio at his 1941 skill level, and so long as he was within that range, everything else is out of his control.
Since I don’t admit that randomness can itself be regarded as a cause, I’d be forced to make an analysis something like: 57% DiMaggio’s skill and 43% Trout’s skill. (We’ll leave Gehjringer out for simplicity.)
If the Detroit Free Press in our world reported that just the moment after DiMaggio swung there was a gust of wind from the west – while in World 2 there was no such report – both approaches would recalibrate by adding, say, a factor of 16% random with respect to our world’s Joe and Diz. In that respect, we’d all agree that Joe got lucky in our world.
I think the main point of disagreement may lie in the way in which the two approaches either isolate outcomes with respect to individual players or view them as products of competition. But there may also be a root difference in the picture of how well MLB players’ training and execution actually permit them to control events on the field in a gross sense. In my view, the major impediment to their control of outcomes is the skill of the opposition. If you trace their histories back to initial Little League games, the greatest impediment to control there would be inadequacy of skill, and randomness seems everywhere.
birtelcom’s comment @126 came in while I was typing @128 (as did some others, I was slow). I just wanted to add that I agree that the PE the 1-run game issues needs a different analysis.
Jim,
You wrote: “What some people have definitely been saying (or implying), by contrast, is that the Orioles will necessarily do one or more of the following due to the supposed iron law of “regression to the mean”: (1) decline in their overall W-L record to something predictable by a pythagorean expectation, or (2) decline in their ability to win one-run, extra-inning or otherwise close games.”
That is true but not true of me. I have not stated nor suggested either of those two things, and in fact wrote that, while reversion to the mean affects baseball, and does so more than basketball, it is a tricky concept and is often misapplied.
I will plead guilty to using luck in a catch-all way. But I defined what I meant by luck and what I meant by skill, using the former neither in the colloquial nor the accepted scientific way. I gave it a specific definition for the circumstances being described. Perhaps I should just call it non-skill because what I am interested in, and have only and always been interested in in this discussion, is isolating how much, a percentage expression if that can indeed be determined, the outcomes of baseball, and specifically in this case the outcomes of the O’s extra inning record, are the result of skill alone. The point is to try to understand better and make a guess at whether / to what extent their record might indeed be repeatable.
Well, economists have won their Nobels by showing that our tendency as humans to seek patterns, a trait which has benefited our species greatly, tends frequently to go overboard: we find them where there are none, and we just don’t like to view random events as random. We draw bad inferences from small samples and persist in harboring an illusion of control over things that we have very little or no control over at all. So luck could indeed be explanatory fiction or cold, hard, uncomfortable fact. 🙂
Yep, apparent pattern does not always equate to actual pattern.
And the flip side of that coin is that apparent randomness does not always equate to actual randomness.
It’s largely an issue of temporal scale and we have statistical methods for evaluating the likelihood that a signal actually exists amongst the noise. Indeed, that’s what statistics *is*.
tag, maybe it works that way for economics, but the vast majority of Nobel Prizes are awarded in the sciences precisely BECAUSE patterns were ingeniously recognized by the perceptive mind that escaped the notice of others.
Dear Steinbrenner Family,
I understand that I am no longer part of the demographic you care about.
I would not spend $1,000 on a ticket to a baseball game if Ty Cobb, Babe Ruth, Dizzy Dean, and Billy Martin were re-aminated and promised to brawl at 2nd base.
I do not own a television – in fact, I no longer spend ANY money on your product, ever.
My toddling daughter just got her first baseball cap – of the San Francisco Giants.
That’s right.
This Bronx-born dork, who wore a Yankee cap every single day of his childhood,
who, despite being 130 pounds at age 15 and in a high school for the freaking gifted, still had one and only one goal in life – which was to be a starting pitcher / pinchrunner for the New York Yankees
-despite the fact that I went to 30 games a year, every year, and smelled my first second hand marijuana smoke in your RightField Bleachers
-despite that fact that I actually had a poster of Don Mattingly in my college dorm room wearing a Godfather-inspired pinstriped suit, and holding a bat like a Tommy Gun with the caption “Hitman”
– despite the fact that, still, 200 days out of the year my emotional arc is partially described by the fortunes/misfortunes of my hometown team
Despite all this, I am actually more worn out by The Freak’s failure tonight than by the inconceivable fact that
(aside from 20 magical Yankee minutes in Game One)
we scored 2 runs in 38 innings, and they were both created by a third-string shortstop who can’t field.
Listen, Hal, Hank, Fredo, et al,
I have followed this team unrelentingly for 32 years.
I’d bet that I know more about the Yankees than Hank.
Please, even though I am typing this from deep in the forest 3000 miles away and I have no intention of ever setting foot in NYC again, please consider my advice.
BLOW THE WHOLE THING UP.
Spend the winter jackhammering half the bleachers so that you can push the fences back to dead-ball-era distances, and do the following:
____________________________
TRADE Curtis Dunnderson, Ivan Nova, Eduardo Nunez, Melky Mesa, Francisco Cervelli, and Dellin Betances for
Aroldis Chapman and Billy Hamilton
__________________________________
TRADE Mark Teixeira, Kevin Whelan, and Cody Eppley for
Jose Altuve
___________
TRADE Alex Rodriguez, Corban Joseph, Manny Banuelos, Kevin Russo, and Pat Venditte for
Jose Reyes
__________
TRADE Robinson Cano for
Gio Gonzales
__________
Move Jeter to 3rd.
Resign Ichiro
Resign Mariano
Resign Russel Martin
Resign Kuroda
Resign Petitte
Sign Kevin Youkilis to play 1st
Sign Mike Napoli
(Napoli backs up Martin and Youk and does some DH)
Sign Grady Sizemore (why not?)
Sign Carl Pavano (just kidding)
Here’s your lineup:
1 Jose Reyes
2 Brett Gardner
3 Ichiro
4 Mike Napoli
5 Kevin Youkilis
6 Derek Jeter
7 Russell Martin
8 Jose Altuve
9 Billy Hamilton
1 CC Sabathia
2 Gio Gonzales
3 Hiroki Kuroda
4 Michael Pineda
5 Pettite/Hughes
Bullpen:
Already great, plus a Cuban freakin Missile.
____________________________________________
And this:
Fire Girardi.
Just give the job to Jeter.
___________________________
And one last thing.
Look, the world is scheduled to end in two months.
Can we just agree to let your best-in-the-world multi-millionaire grown men employees grow their beards and hair if they goddamn feel like it?
Frank Sinatra is dead, he’s not going to judge anybody’s sideburns anymore.
Wow. I can feel your anguish all the way here in North Dakota.
You’ve certainly got some interesting ideas going here. I don’t know enough about some of the minor leaguers involved to know how realistic your proposals are but I think you’d certainly have a competitive team on the field. That said, there are a few things that I would do differently.
First, with an aging & coming off of injury Jeter at 3rd, an injury prone Reyes at short and a very young Altuve at 2nd I would probably try and find a way to keep Nunez around
I also don’t think that I’d resign Ichiro. I’m assuming that the Yankee’s will be eating a fair amount of payroll in the deals that you propose and I don’t know how that counts towards going over the salary cap but assuming that it doesn’t there may be enough money to look at BJ Upton. If not, it’s also possible that Melky Cabrera might be available at quite a discount. Otherwise maybe a couple year deal for Torii Hunter might be a better gamble. Victorino may be another possibility for a bargain.
And anyone named Billy Hamilton immediately needs to have Slidin’ added to the front of it.
Billy Hamilton is so fast he doesn’t even need to slide.
And yes, I mentioned some of those minor leaguers for the amusement of Yankee fans who might be paying attention. You’re right about Nunez, though.
Even an old Ichiro is fast and a great defender and doesn’t strikeout.
I want a home outfield so vast that the opposing team will have to bench that one slow good-hitting old guy they’ve got on one of the corners.
And no, with me at GM, the Yankees don’t eat any payroll.
That’s the great thing about a fantasy life.
Although, I rethought the Jeter-as-Manager concept.
Too complicated with him in the field.
Mariano Rivera as Manager.
The question remains whether Billy can get on base to do this thing. The Cubs have the lightning fast Tony Campana and he steals at will. (He does slide but I think it’s only for show.) The problem is that he doesn’t get on base enough.
I hope Billy is able to slap enough hits and draw enough walks to blister the base paths.
I so curious to find out.
Maybe he’ll be poor offensively like Vince Coleman.
Or maybe he’s as good as Tim Raines.
Coleman at 21 did this in A ball:
.350 .431 .399
145 steals
Hamilton just spent age 21 splitting A/AA.
.311 .410 .420
155 steals in 20 more games
Coleman jumped a level to AAA at 22 and…
.257 .323 .334
But so what?
Coleman is his youth was truly fun to watch change a game. Slash lines really don’t apply to a player with that kind of speed.
Here’s the slashes from Van Go’s age 24 season:
.232 .301 .280 .581
And he scored 94 runs.
With zero homers.
In the 1987 World Series:
.143 .200 .214 .414
And 5 runs.
And 2 RBI.
I don’t think Campana is a strong comparison.
He didn’t get to AA until age 24.
And though he hit well there, his SB numbers were only
48
20 CS.
Just so he’s not the next Joey Gathright
Voomo,
I agree that Campana is not an ideal comparison. But check out his SB rates in the majors. Last year 24 SB and 2 CS. This year 30 and 3 and one of those CS was a pretty egregious blown call.
It’s always very, very difficult to predict whether these guys can hit enough to make it at the MLB level. I sure hope Hamilton is able to.
@ Tag, EPM – what a great discussion! Thanks for clarifying the concepts.
I am trying to recall, but I think I first came across “regression to the mean” in a baseball context in a book Bill James wrote about managers. I think the context was the hypothesis that a manager might possess some skill or attributes that would make his team play better than its PyPj. Hence there might be a way to use that as a way to rank managers. Bill James showed that there was no correlation for a manager to be able to replicate success in that. Also, managers who by all accounts were/are the best didn’t show it with consistent having a Winning Percentage that exceeded the PyPj. Indeed, the largest effect on variance of PyPj was “regression to the mean”. I think he also showed that happens when teams do especially well in one run games (let’s watch the Orioles next year!) regress to the mean and don’t do as well the next year.
Larry, I think this can be an interesting counter-argument when it comes to managers (because they presumably retain much the same tactical-decision skill set from year to year), but maybe not when it comes to one-run games (because the personnel and social dymanics of teams change from year to year). My recollection is that James was the one who developed the idea that PyPj deviations were manager-caused and also the one who debunked the idea (I could easily be wrong in that memory).
Actually, I’m in way over my head, but the way I’d try responding to these sorts of regression-to-the-mean counter-arguments is to suggest that a team far exceeding its PyPj reflects a rare team dynamic of resiliant self-confidence that is season-bound, while a team that substantially underperforms its run-producing talent generally undergoes significant changes between seasons. That’s why results rarely are replicated from season to season. While to test the first claim would require anecdotally based qualitative research, I think both those claims are, at least, falsifiable – they are hypostheses that you could test. Explaining (as opposed to describing) the phenomena as a function of regression to the mean doesn’t seem to me to produce a testable claim; I think what’s often involved is a description of de facto data reinterpreted as their cause.
I guess I’m not sure why a “team dynamic of resiliant self-confidence” that causes over-performance compared to the Pythagorean Expectation should not show some evidence of survival across seasons, if it exists. If that evidence doesn’t show up, it seems to me Occam’s razor, or the null hypothesis, or whatever one wants to call it, would point us to a default assumption that the regularity of non-recurrence of PE over-performance is the result of regression to the mean, rather than a hypothetical team dynamic that for some reason regularly self-destructs after a season.
I’m open, as always, to a hypothesis that in the particular case of this year’s Orioles, for example, there was some team talent-based ability to win close games. But let’s say next season the Orioles, with similar basic personnel and management, don’t show a recurrence of this year’s unusual success in close games. Which hypothesis do I then default to — an evanescent team-based ability or regression to the mean? I’m afraid I’ve have to choose the regression to the mean alternative.
This question gets us to the crux on this issue. I was just about to ask everyone what interpretation they would have if the O’s go say, 19-19 in one-run games next year, but you asked the perfect leading question.
Let’s return to the original comment that started this entire discussion, which was made by Andy (It’s all Andy’s fault!) some 3 months ago maybe, in which he noted that the Orioles were, at that time, 19-6 in 1-r games, very likely due to luck and thus not likely to be maintained over the remainder of the season. John jumped in to say he’d bet the O’s would not likely win more than 1/2 of their remaining 1-r games, which led to an actual bet being made with a player to be named later, and before you can say Batting Average on Balls in Play is the mother of all statistics, poor John was blasting his TV with a shotgun, which at the very least must have been hard on the vacuum cleaner and the nearby drywall, and cause for concern amongst the neighbors.
So why did I take this bet and why is it relevant?
I took it because I knew–even without doing the binomial prob calculation–that such a record was such a large departure from expectation based on a random process, as to be…wait for it…
…unlikely to be due to a random process (and therefore unlikely to “regress to the mean” over the remaining games of this season). The risk of losing the bet, in my mind, was not so much that the Orioles had really done this 19-6 thing by chance, as it was that the number of remaining 1-r games was unknown and might thus turn out to be very small and thus subject to the randomness that accompanies small sample sizes. But such randomness could equally well have worked in my favor too.
So the issue is this. If the Orioles accomplished their 29-9 record by chance this year, then we might expect them to do no better than .500 in such games next year, maybe quite worse (but also possibly quite better). OK, so that represents a hypothesis: the Orioles are winning 1-r games by chance this year and it’s therefore likely to come to a screeching halt next year.
The problem with this assumption is that we can apply the exact same logic to different portions of the season *this* year to test the very same hypothesis, which has the decided advantage that the team’s makeup (personnel, attitude, gestalt, whatever) is more likely to be constant over the course of a single season than it is across multiple seasons.
Using for example, three distinct data points in this regard, the O’s were: (1) 11-6, (2) 19-6, and (3) 29-9. So, after they were 11-6, they “should” have gone .500 over the next 8 games, but instead went 8-0. And from there, should have gone 6-6 but instead went 9-3.
We can in fact carry this logic down to a micro level where we can compute the likelihood of any possible departure from .500, for any small sample of consecutive games (>= 2), assuming a random process. This is computable, which is to say we can compute the likelihood of steaks where the O’s win (or lose) a number of 1-r games in a row, because “streakiness” is *constrained* when the generating process is truly random. But that aside, if we just use the three time points mentioned, the combined probaility of going 11-6, then 8-0, then 9-3, is p < .0002.
So I ask, in the face of this evidence, how is one warranted in concluding, at the end of 2013, assuming the O's have gone 19-19 in 1-r games, that their performance in 2012 was due to chance, and that 2013 represents a regression to the mean?
Jim,
The answer, as I’ve always maintained and wrote at length about, is that baseball in general, and winning one-run games in particular, is never based wholly on luck. Baseball, like many things, is a mixture of both skill and luck. The only question is what is the blend of the two. That’s what the whole two urn method emphasizes. You can mirror any point on the skill/luck continuum with it. You surely aren’t suggesting that the outcomes of baseball games result only from skill, are you?
tag, when you make comments like that, it indicates to me that you’re not understanding the analytical methods involved in resolving this type of issue, and also that you have the fundamental nature of the debate that’s going on here perfectly backwards.
To wit: as I wrote above, certain individuals began this debate by repeating, innocently enough (no problem with that), the oft-promulgated storyline that success in a certain category of games is the result of chance/randomness/luck/voodoo whatever word you want to label it with (in your case “luck”). I have been trying ever since to basically say “Whoa, not necessarily, the data may or may not indicate what you claim, and you have to look at it more carefully and thoroughly”. So you’ve got the nature of this debate entirely backwards. I’ve never once said that baseball is entirely skill based; my point is that quite a lot of people think that randomness plays a larger role than is warranted by the full suite of evidence.
Of course, baseball involves both player skill and elements of randomness. That’s not up for debate. The point of any statistical analysis is to try to extract signal from noise, if indeed there is such a signal. The degree to which it can be extracted is what power analysis is all about.
The only way you can evaluate a hypothesis is formalize in quantitatively (mathematically) and then compare observed results with those predicted from the hypothesis. That’s why you set up a hypothetical model based on random processes–because you can then evaluate the likelihood that such processes are in fact responsible for (i.e. sufficient to explain) the observed results. That’s what’s going on here.
Jim,
I understand the methodology you are describing completely. What I’m saying is you have to go beyond it to get at what we’re (or at least I’m) interested in trying to ascertain, which is: How skillful are the O’s in one-run games?
You do binomial distributions and tell me that what the O’s did can’t be explained solely by luck. Agree 100%, always have. To me the only thing notable about one-run games is that they (probably) involve even more luck than the outcomes of non-one-run baseball games. But certainly, definitely, their outcomes are not the result solely of luck.
So: what the O’s did can’t be explained solely by luck. Given. But what they accomplished also can’t be explained solely as a result of superior skill, either, as can a chess champion’s or Roger Federer’s. The trick is in trying to isolate how skillful the O’s are in one-run, or let’s say extra inning, games. So let’s say we simulate 100 extra inning games 100,000x using the O’s current roster against a full range of MLB opponents, with the O’s as both the home team and the visitors. What would their record be across these simulations? Could they play .700, let alone .800, ball, as they did in extra frames this season, in any meaningful number of them? Now these simulations would not map the real world one-to-one, clearly. And maybe you believe there would be all kind of hidden factors (confidence et al.) not captured by them. I don’t think so but fair enough: a debatable point. I still think they would give us a decent idea of what the O’s true skill level is over a fair sample size. And my hypothesis is that there would be a meaningful amount of reversion to the mean in their record. And that’s what I have been contending all along.
Jim,
As I said I’m not trained in this stuff, and don’t even like it, but I’ve been forced in my profession to read a lot about it.
http://gradworks.umi.com/33/64/3364332.html
Here’s a work I had to read a few years ago in researching certain points and, if I remember correctly, it bears out a lot of what I’ve been saying about reversion to the mean.
OK, then I have some questions for you.
1. If the Orioles perform much worse in close games (by whatever definition) next year, what is your cause and effect explanation of why this occurred.
2. If the Orioles’ record in close games (however defined) had a significant element of chance to it, then why did they not “regress to the mean” in such contests this year after certain points in the season, such as the two I mentioned above, or any others you’d like to use?
3. What exactly, in terms of specific events, do you–and do you not–have in mind when you say that baseball has a significant component of “luck” to it?
4. Is random variation synonymous with “luck” in your mind, and if not, what is the distinction?
…and one note:
As for simulations, I am 100% for them. The problem is that such simulations are very difficult to do, for the reason you mention, i.e. you have to know that you have included, and quantified correctly, all the relevant variables, which is frequently impossible. That’s one reason for leaning heavily on hypothesis testing using null models and expected outcomes, as I’ve been describing.
Jim @125: Let me take a shot at defining “luck”, or at least “good luck” and “bad luck”. Let’s say that “good luck” is defined as randomly variable results occurring such that one is rewarded in some way for those results. And the reverse: “bad luck” is the case of randomly variable results occurring such that one is penalized for those results. So then we back into a definition of “luck” in general: randomly variable results that result in either rewards or penalties. In life in general, lots of random things happen every moment that have no particular effect in terms of rewards and penalties, so those are just random events and don’t involve ‘luck’. But baseball is a very artificial environment in which most measured events do have a distinct reward or penalty element (fair/foul, ball/strike, safe/out, run/ no run, win/loss, etc.) Therefore in baseball using “randomness” and “luck” interchangeably is, to me, a natural linguistic tendency and nothing to get overwrought about.
Jim, answers to your questions @125:
1. I don’t really have one. I simply don’t trust small sample sizes. I think we could arrive at some correlations but I don’t think I could give you cause and effect. Roster changes, key guys a year older and potentially less skilled (though younger guys could also be potentially better), Buck damaging too many brain cells, potential reversion to the mean, though I am not expert enough to quantify this.
2. The O’s are clearly an outlier, and if they are indeed a .615ish team in one-run/extra-inning games due to bullpen, home run ability, so many factors I can’t name – I have no idea whether they can truly play at his level, thus my call for simulations (and I realize they’re difficult to do) – then I think their actual record would fall within the standard deviation, would it not?
3. Non-skill. Everything above and beyond it. Yes, I plead guilty to using it as a catch-all phrase but I did define it before, neither in colloquial nor scientific terms. I defined it for the circumstances I was using it in. It’s characterized by a distribution that has an average of zero and tends to be transitory.
4. Luck, as indicated above, includes random variation and many other things.
Jim,
Let me backtrack a little on answer 4 here. Random variation means “outside the realm of noise,” correct? Which (it’s been a while) is, what, 2 standard deviations off the mean? Which means that an outcome has, um, less than a 5% chance of happening randomly? With 30 MLB teams, probability would say that 1, maybe 2 teams has the potential for a low-probability outcome like playing 2 SD above the one-run mean to occur? Which I guess I wouldn’t consider luck by my definition then.
But today we had unseasonably beautiful weather and I played three hours of tennis, and while trying to answer your questions in good faith I’m not sure I’m up to this last one.
Jim,
As a sort of meta-answer to your questions, I think that a lot of your objections to certain concepts raised here should not be directed at birtelcom, me, John A, et al. but at your fellow (I’m assuming you work for a living with stats in some capacity) professional practitioners of statistical science. We don’t posit in our posts any radical new concepts that aren’t used by lots of stats pros (now certainly our application of them can be called into question, yet even here I would characterize what we’ve been saying as considered, reasonable, “responsible,” given the limitations of our training and analytical ability.)
You mocked (and I use the word advisedly) PE in your initial posts. Birtelcom defended the idea, and I posted a study by a Williams College math/stats prof that I take to confirm Bill James’ conception of Pythagorean wins and expectation. To my mind, if you don’t see much sense in PE, your beef really is with him, not birtelcom, whose understanding of PE is, I would hazard to guess, the mainstream stats community understanding of it. I highly value your nuanced explanations of stats matters, but I don’t think you’ve given us any real basis for not seeing PE as James and the Williams Ph.D do.
Now no one expects you to write 20-page rebuttals challenging complex algorithms – you’ve got enough real work to do and you already are plenty generous in pointing out errors in our posts, and we probably couldn’t understand such arguments anyway (I’m taking the liberty of speaking for you, b.) – but I think you can see that we’re not just pulling these things out of our asses. I, for one, read up on the ideas I put forward, consulting credible sources (and, yes, as with many other technical matters, I’m at the mercy of experts and have to take certain conclusions on faith). I have maintained that “reversion to the mean” is complicated and easy to misconstrue and -use. I don’t think I’ve done so (I tend to limit my application of it to individual players) and pointed to a dissertation by a Stanford Ph.D candidate which I believe bears out my understanding and use of it. I can link to others.
The thing is, I feel pretty darn confident in predicting, based on how they’ve already compared other stats and baseball outcomes, that a good number of your fellow stats pros, not just website amateurs, will say or at least agree that, if the O’s go .500 next year in one-run/extra inning/close games, this outcome does indeed demonstrate (at least to some degree) reversion to the mean. I just wonder, given what you’re saying, which shouldn’t be hard for fellow stats pros to follow: why?
tag, I know I’m kind of butting in, and perhaps making a point different from Jim’s, but I don’t think he’s disputing in any way that if the O’s go .500 in 1-run games next year they will have reverted towards the mean. I don’t think that’s disputable – it’s built into the definitions of the terms we use. I think he’s pointing out the problem of treating the tendency to revert to the mean as if it were a cause of the reversion, just like explaining the Birds’ record this year as luck would be like saying that a large deviation from the mean occurred because large deviations from the mean do occur (which is true).
When a deviation as large as the O’s’ this year occurs, it should be ok to ask what caused it. If someone answers “luck,” it’s equivalent to saying that nothing caused it – “these things sometimes just happen” – as is pointing to next year’s likely reversion as a prospective demonstration of luck. That treats the O’s record as (loud) statistical noise.
Maybe it was. But what’s the evidence? The only evidence in arguments so far has been the claim that statistical noise occurs in baseball (true enough, but not proof of a specific instance) and skepticism that the record could be anything else (fair enough, but not an argument).
epm,
You’re not butting in at all. And I appreciate your perspective. The way I’ve had it explained to me, and this was from a math prof, is that reversion to the mean is “built into” any activity whose outcomes have at least some element of luck attached to them. The more luck, the more likely you’re to see reversion and the more rapid it will be. (For instance, the performances of chess players don’t revert to the mean. The players earn a rating, which is a solid proxy for their skill and based on their game results. A player rated, say, 200 points higher than his or her opponent is expected to win 75% of the time, and does.)
Now Jim disputes this – the point above the parentheses, not the chess example. Fine, but I can’t really debate it with him because I don’t know enough math or statistics. I’m going by the math prof discussion I had five or six years ago, and what the guy said certainly seemed logical to me.
But let me just reiterate, so there’s no misunderstanding on what the prof told me, that “built in” does not mean that all outcomes revert to the mean. There are still outliers: the standard deviation of outcomes does not narrow over time.
None of that seems problematic, tag. I take Jim to be noting that through the season, we were all expecting the lucky O’s to revert to the mean, and pretty fast — but then they didn’t, which seemed a likely indicator that the role of luck was not that large after all. Now if they revert to the mean after the winter, and we say, “It must have been luck after all,” treating the seasonal divide as if it were identical to a mid-season reversion to the mean, then we’re proposing a theory that luck comes in season-long quanta, as opposed to a theory that says that long in-season 1-run/xtra-inning victory streaks may be largely skill/agency based.
For my part, I see it as more likely that what would evaporate over the winter for a grossly overperforming team is an unusual performance skill (effort/optimism/focus levels) rather than its luck. We could understand the first as an analogue to ordinary experiences (as I wrote @114 below, picking up on Larry’s point), but so far as I know, when luck overperforms during the season, it doesn’t spend its winters getting older, exchanging personnel, and sliding into the new frames of mind and body that a four-month layoff can induce. Players and teams do those things.
If that on-the-fly hypothesis fails an empirical test, birtelcom, then I suppose the null hypothesis is the default. But the hypothesis actually predicates discontinuity across seasons, so if none occurs, the hypothesis hasn’t failed; the problem is that I didn’t convince you it was interesting. (No argument!)
The “evanescence of team-based abilities” may or may not cause “the regularity of non-recurrence of PE over-performance,” but my understanding is that regression to the mean is itself the effect, and so cannot be its cause. So it seems to me that the alternative you’re offering may be: the phenomenon of regression to the mean in cases of PE over-performance is “the result of regression to the mean.”
My favorite baseball movie is 61*. There is a scene which captures the very essence of my mental image of “regression to the mean”. Roger Maris is in the locker room after hitting #61 standing next to Sal and the twit reporter who says, “Roger, Roger, what about next year? Do you think you can do it again next year?” To which Roger tosses his head back with a “c’mon man” grin and there is a collective groan in the room. That moment captured the essence of what “regression to the mean” is in a sports psychology sense. Which is different from its meaning in a purely statistical sense, I believe. It is in the same domain as “being in the zone”, being the pin that pops the bubble.
I think if you asked Buck and his Orioles if they could duplicate their one run record next year, you might get the same reaction. Or maybe not. But the Orioles are not going to repeat that one-run record next year.
I think that’s it, Larry, and I’d expect everyone here has had some analogue experience, where we’ve made a rare sustained effort to keep our momentum in some significant skill-based task going far beyond our norm for success. The next time out, the chemistry is rarely there for an immediate repeat – to try for it again, rather than for ordinary success, may even feel like a path to failure and disappointment. It seems to me a normal aspect of human effort and focus.
“Which hypothesis do I then default to — an evanescent team-based ability or regression to the mean?”
birtlecom, I really think this is the source of the conceptual problem here, i.e. those two things are not *necessarily* mutually exclusive. Now many, many times, something out away from the center of the distribution one year (or month, or decade or whatever time period is of interest) that returns toward that center the next year does *indeed* represent the result of a random process. But that does not mean that it always does, and in the absence of finer scale analyses, it does not even mean that it *usually* means that.
I suppose I sum this up by saying that the types of analyses that have been cited by some as “proving” that certain phenomena are stochastic in nature and thus “regress to the mean”, are insufficient to warrant that conclusion. They are. They really are.
We’re probably going around in circles here, but I look back at studies like Bill James’s study of the 100 largest positive and 100 largest negative variations from Pythagorean Expectation over a full season (discussed, and with a link here: http://statspeakmvn.wordpress.com/2007/10/15/still-more-pythagorean-musings/) and see a year over year average decline in variation from eight wins to nearly zero. You would think with that sample if there really was some sort of skill (or lack of skill) related to overperforming (or underperforming) PE, something more than a very minimal level of year-to-year persistence would show up on average. I absolutely agree with you, Jim, that baseball is a marvelously foggy amalgam of skill and randomness — that is part of what makes it endlessly fascinating. But the actual evidence suggests to me that in the specific arena of pythagorean expectation, the randomness/skill balance happens to be heavily on the side of randomness.
And one last example.
Imagine a run of the mill regression line through a set of say 100 data points, with a positive slope of the line and some reasonable amount of noise, say an R squared value of 0.50, representing a relationship between two variables over some defined time period. Definite signal plus some definite noise.
Now you measure that same relationship in some next time period and you get a nearly identical result–both in terms of the slope of the line and in the departure of each individual point from that line.
Does the nearly identical departures of the points from the line in the two different time periods represent the result of a random process, or of some cause and effect process?
EPM et al: you know, this is where this site really shines. Another facet is when a contributor finds a “this has never happened before” and the parameters are not at all that esoteric to avid baseball fans. Indeed, certain talents and attributes would otherwise go completely unnoticed.
But the statistical expertise here is phenomenal. I am completely in over my head with the rigors of statistical mathematics, but at least I can follow the logic.
Anyway, one thing I have noticed empirically is how well the PyPj works EVEN in small sample size environments. For example, I tracked it through the Little League careers of two sons from T-Ball to Senior League with 16 to 20 game seasons and it was rarely off more than one game. Actually, there seems to be more deviation at the MLB level where skill and strategy can trump PyPj forces. It seems to have a “self-correcting” mechanism (runs scored and runs allowed) that make it relatively valid even in a small sample size environment – say, a quarter if a season or even the length of all the post-season. Do I have any statistical basis for my intuitive sense? Thanks!
Larry, several studies have been done on this. You’d have to look them up, but I think at least one or two of them have shown that, at the end of May, PE correlates better with a team’s final regular season record than its actual record at that time does. Or something like that. It’s not all that much better of a correlation if I remember correctly, but it works slightly better as a predictor.
PE’s also being used in pro football, whose seasons of course are 16 games. Pro Football Outsiders, a bunch of stats geeks who graduated from Brown and created a website with their own proprietary stats, work with it a lot I think.
birtelcom (127):
The distinction between luck and random variation is in fact very important in this discussion, crucial in fact. The latter is a statistical concept and the former is a colloquial concept with subjective meaning. Random variation = noise = stochasticity = unpredictable variation = chance outcomes, in any variable of interest, as distinguished from a cause and effect, deterministic signal. As I said some while back, the term noise is simply a place holder, a descriptive term to describe variation for which we do not yet understand the cause of. This is true in all statistical analyses. The fact that we do not yet know the cause does not of course mean that there is NO cause. This point is crucial. Noise is simply a label of convenience we give to variation for which we lack an explanation, relative to the system under study.
The easiest illustration of this is just your standard linear regression line–the line fitted to the data represents the signal, and the variation of the individual points from that line represents the noise. How much each such point varies is unpredictable–if we analyzed the same system again we would get the same line, but the individual points would not have the same relationship to that line that they did the first time. This is the essence of random variation. The phrase “regression to the mean” simply means that no single point will be expected to consistently fall a given distance from the line–i.e, the points are unpredictable with respect to the regression line, the signal.
The analogy with baseball is that these points represent things like WL records of teams, or records in close games, or any other variable of interest. The fact that they vary in what appears to be a random fashion over defined time periods, does not of course mean that they have no cause. It may instead be that the nature and/or strength of those causes varies with time, the collective appearance thereof being random variation of the points around the signal. THIS DISTINCTION IS ENORMOUSLY IMPORTANT TO THIS DISCUSSION.
When people use the term “luck” to describe this variation, they are implicitly (and often unconsciously) assuming that the types of causes they are interested in (e.g. player skill, team skill especially), are not operating to create that unpredictable variation. But those causes may in fact **very well** be operating, but just with a different temporal mode and dynamic, and it is therefore unwarranted to assume that they are not by ascribing this variation to “luck”. This latter point is why I have emphasized so strongly the within-season performance of the Orioles this year and what that evidence might mean, relative to one run wins and pythag expectation (the two of which are related, not going to get into that right now though), and relative to what might be, and might not be, legitimate explanations of poorer performance in 2013. It is also the reason that epm has said, more than once, that finer scale (and/or different types) of analyses are required when before such premature conclusions can be made, and he is absolutely correct on that. This point is further highlighted by the fact that not a single person here, you or anyone else, has yet described exactly what types of baseball game events they are referring to when they invoke luck as an explanation for “regression to the mean”. It’s just vague hand waving in the end, panglossian.
If the above explanation falls on deaf ears, then I am done, because I can’t explain it any better than that, short of writing many pages, and even then maybe not. It gets to fundamental conceptual issues in signal to noise detection and really, in what statistics and most of science, are all about. It’s epistemic in nature. There’s only so much I can say on a baseball blog, and I’ve said a lot already.
tag @ 131:
1. My question #1 was poorly worded, not what I actually meant, so never mind.
2. That doesn’t answer the question I asked. Simulations will not help if they do not include the necessary variables and target the right time scale, otherwise they just give a restatement of what one already knows. And the O’s were nearly .900 in extra innings and around .750 in one run games.
3. That is not an answer either. I don’t want a catch all phrase but a specific definition (and estimate of the relative importance/commonness), of specifically what type of events you have in mind when you use the term “luck”.
4. Same as above.
@135:
See response to birtelcom at 147. Noise and random variation are synonyms. They contrast with the deterministic signal in a system, ie the part of the variation that can be explained. You are referring in the rest of that answer to the expected deviation from the mean in a normal distribution, but that does not get at the issue at hand, as I explained to him there. It only gets to how rare or common an event is.
@ 142:
I’m sorry but that’s just largely a cop out. I address a specific topic that was raised and then you misunderstand it and drag the conversation off course based on what you thought the issue was, or what you felt like discussing, and then you criticize me for not taking it up with professional statisticians??
And at any rate, the issues I explained in 147 and elsewhere are indeed well known among statisticians and scientists. It’s what we do. We don’t ascribe things that we just haven’t figured out yet to “luck” and “regression to the mean” and go merrily on.
A last short word from my end. I failed to answer your questions satisfactorily. Let me at least give No.2 another shot.
Consider the 1984 Tigers. Famously, they started the season 35-5. In that record were data points of 9-0, 15-4 and 9-1. So: through one-quarter of the season they were playing .875 ball. Over the next three-quarters of it they won 56.6% of their games to finish at 104-58 (.642), five games better than their PE.
Judging by epm’s words, which you’ve endorsed, you accept that, within the season, the ’84 Tigers reverted to(ward) the mean. It didn’t happen in that initial small sample but it did happen over the course of the larger one. The ’84 Tigers, if you accept PE, which I know you don’t, were a .615ish team (.611 to be precise) despite compiling that opening 35-5 record with those stated data points. And in a year when the O’s played .888 ball in extras and the Astros .083 and the Red Sox .167, sorry, I promised to keep it short.
tag, Although I think we’re all tired out on the luck issue in general, at some point the significance of hot/cold streaks for players and teams would be worth a new discussion. The Tigers’ streak (of fond memory) and its end would be a good one to explore for teams. In hot hitting streaks, we could talk about the difference between instances where players might say, “Everything seems to be falling in,” and where they say, “I’m really seeing the ball right now.”
By the way, if I’ve added right, the 35-5 Tigers had a PE of .824 (34-6), so their record wasn’t really that far out of line. Their remaining PE for the .566 portion of their season was .563. So the issue isn’t really PE related, but hotstreak related.
epm at 128, 143, 146 etc:
Right on the money.
I go to see everyday some blogs and blogs to
read posts, but this web site offers feature based writing.