I think more research needs to be done on "Luck" and "expected" wins using run differential. Each game has its own context, it sure seems like the same group of teams are lucky year in, year out. (Angels, Twins, CWS....etc)

F this.

Can we at least wait until we know how they do in the playoffs before calling them the luckiest team of all time? Anything less than a World Series appearance and I can guarantee you nobody who plays for, roots for, or is associated in any way, shape, or form with this team is going to be feeling lucky.

It really doesn't matter if this team is really a 100 win, 85 win, or 110 win team. Either way, what they have done so far is merely what they were supposed to do: win the division. In the end it really doesn't matter if you win the division by 3 games or 20 games.

We'll see how "lucky" they are next week.

F this.

Can we at least wait until we know how they do in the playoffs before calling them the luckiest team of all time? Anything less than a World Series appearance and I can guarantee you nobody who plays for, roots for, or is associated in any way, shape, or form with this team is going to be feeling lucky.

If 162 games won't persuade people about the relative strengths of the Angels as opposed to their run differential, I fail to see how an extra 11 to 19 games will do the trick.

I think it was Gary Huckabay who wrote a BP piece a while back showing that after 150 games or so, a team's actual record becomes more predictive of their future record than their Pythagorean record.

The implication is that if a team significantly beats the record one predicts from RS/RA (or components, or whatever) over the course of a whole season, then there's something structural about the team that has them outperform their Pythag, instead of dumb luck.

Couldn't the team's recent performance be more predictive than either their older performance OR the combination of the two?

Particularly for a team that may have, say, added Mark Teixeira at a point along the way.

(this should be true whether you use record or pythagrecord.)

Conversely, are the Jays the most unlucky? They're nine game over .500 but their Pythag is 90-69.

Just to use one example, the Jays are Rays each scored 64 runs against each other this year but the Rays won the season series 11-7. They played 14 games decided by two runs or less and the Rays were 10-4 in those games. Before the Jays swept the final series of the year, the Rays had won 10 of 12 of such games.


The thing is, the Jays have been "unlucky" for a few years now too.

Their Pythag last year was 87-75 (Jays went 83-79) and 88-74 in 2005 (Jays finished 80-82). Since 2001, they've only beaten their Pythag twice and each time it was just by a single game.

It seems the Jays are the anti-Angels/Twins etc.

Best Regards

John

I think more research needs to be done on "Luck" and "expected" wins using run differential. Each game has its own context, it sure seems like the same group of teams are lucky year in, year out. (Angels, Twins, CWS....etc)

Angels wins - pythag 2000-2008, from B-R.com

1
-2
-2
-3
1
2
5
4
12

the WS winning team was one of the -2's; Angels have been "lucky" the last 3 years, not previously.

White Sox, same years

3
2
-5
-2
-1
8
2
5
-1

the WS winning team was the +8 ... other than that one year, they've been even

Twins, same years

0
4
8
5
5
-1
3
-1
-1

Twins had a run of luck 2001-04, not recently.

In short, the data do not support your statement. The White Sox were "lucky" one year and won the WS, but no net "luck" outside of that year (the +5 and -5 cancel). The Twins had a 4-year run of "luck" several years ago, but not recently. And the Angels have only been "lucky" the last 3 years. In fact, their "luck" periods are mutually exclusive:

2001-04 Twins
2005 White Sox
2006-08 Angels

A million rally monkeys behind a million typewriters couldn't come up with this.

I think it was Gary Huckabay who wrote a BP piece a while back showing that after 150 games or so, a team's actual record becomes more predictive of their future record than their Pythagorean record.

I've been wondering about that piece; while insightful, I'm not sure you can say much about future predictons after 150 games when that means there are only 12 games left in the season. Hello, samll sample size!

However, if it says (going from memory) that by 80-100 games in that future success is a mix of RS/RA and W-L records to date, that at least says luck is not 100% of the diff, that there is a real effect somewhere going on.

Certainly it's always seemed to me that the most straightforward way for a team to beat its Pythags is simply to call in the mopup man in blowouts and leave him out there. When I play sim baseball, I always prefer to lose by 20-3 than 9-3; I'd rather save even my decent middle relievers for games I have a chance to win. Similarly, I rest my hitters in games where I have an overwhelming lead, so I don't get my own share of 20-3 wins against opposing mopup men to cancel out the effect of the blowouts I lose. My Pythagorean record is thus often around .550 even as my team wins 105-110 games.

DCA: Thanks for proving my point on the Angels and Twins.

Angels considered lucky. Turns out they do this quite often. +24 in past 5 years. Their "Worst" luck only measuring a -3, -2, -2. Where is the -12 or -5?

The Twins, the worst their luck ever got in 9 years was -1. But posted +8, +5, +5, +4...

The CWS are perhaps neither lucky or unlucky, so I can pull them from the consistely more lucky than unlucky.
---------------------
There is a lack of even distribution on -/+ for the Angels/Twins. Both teams get bigger pops on the + side of the pythag. You'd expect the - and + to be more evenly distributed if the pythag was this perfect predictor/meausre of luck or "real" talent.

Twins + 23
Angels +18
White Sox +11
-------These teams are frequently considered the "lucky" teams. These teams also play the game a little differently than if BTF ran a team.

In the years you selected DCA. Seems to support what I said. If you have a bone to pick it is "year in year out" but I wouldn't get caught up on that, I'll simply amend that to say "demonstrated a consistent pattern over time where unlukcy numbers are pretty close to 0 and the "lucky" numbers register a greater magnitude over zero."

I take it back, the data you provided shows the CWS are a tad "lucky" over a period of 9 years, just not as "lucky" as the Twins/Angels.

My favorite example is the Nationals two years back who were about 10 games over their pythag at the all-star break and then proceeded to play much worse than it for the remainder of the season. Dozens and dozens of articles about how they knew how to win and Chad Cordero this and that.

Below I've put in a table with my Monte Carlo pythag. This takes the actual scoring distributions (i.e. it treats a 12 and 0 runs scored different than two six runs scored (or allowed)) and then replays the season 1000 times based on those scoring distributing randomly jumbled. Percentile is how "lucky" the team has been. Due to run distributions (probably a couple of blowout losses. The Angels are 3 games better by this measure.

To my eye, the Cubs are the class of the NL by a wide margin and the Red Sox are as well in the AL.

Beano,

You haven't proved causation. It is perhaps more likely that they are just the team that has come up the luckiest (someone will be). That said, we'll never get to causation because we can't run large scale experiments. Correlation is the best we can expect.

I think it was Gary Huckabay who wrote a BP piece a while back showing that after 150 games or so, a team's actual record becomes more predictive of their future record than their Pythagorean record.

I've been wondering about that piece; while insightful, I'm not sure you can say much about future predictons after 150 games when that means there are only 12 games left in the season. Hello, samll sample size!

Plus the teams that have been luckiest have the most to play for at the end of the season generally. Unlucky teams are playing out the string while lucky teams are still in the playoff hunt.

"Conversely, are the Jays the most unlucky? They're nine game over .500 but their Pythag is 90-69."

69? Sounds like they got pretty lucky to me.

You haven't proved causation. It is perhaps more likely that they are just the team that has come up the luckiest (someone will be). That said, we'll never get to causation because we can't run large scale experiments. Correlation is the best we can expect.

I realize this. I never took a positon that this data proved anything. In fact, that is my postion. Run differential is assumed to show "luck". I say we don't know for sure that is true. This subject needs more study. Yet we have 80% of the baseball world using these numbers to "prove" luck.

There happens to be a style of baseball played by the (Twins/Angels/CWS) that seems to result in clubs that create data sets that defy conventional wisdom/explanation.

Here is the Simple Rating System that SR uses on some sites. This takes into account strength of schedule and margins of victory. Here is a description for FB. It starts with your run margin and then iterates to include the quality of your opposition.

http://www.pro-football-reference.com/blog/?p=37

SRS is the number of runs they would beat an average team by and then sos is the number of runs better than average their opponents were.




The AL East is just ridiculous this year. They are .570 out of division (254-192) (by my quick and possibly erroneous calculations). The Orioles are 3 games over .500 outside the division.

The NL was horrendous in inter-league play this year. Which is why the Cubs are down around the Orioles. I don't know if that makes any sense or not, but it how the numbers work out based on all of the games played during the season.

Personally, I believe that either the Rays or Red Sox will win the WS in 4 or 5 games and it won't be close.

Also, I think this points out why you are having the bigger offensive performances in the NL. The league is so much weaker that true stars can dominate to a greater extent.

27 pitchers have pitched in both leagues this year.




Note that this doesn't filter out pitchers batting either, so a larger proportion of the NL AB's are coming from pitchers than from the AL AB's. The AL is just far, far, far stronger as a league.

There happens to be a style of baseball played by the (Twins/Angels/CWS) that seems to result in clubs that create data sets that defy conventional wisdom/explanation.

Just curious, what is the consistent style between these teams?

Run differential is assumed to show "luck". I say we don't know for sure that is true

I can't see any reason to assume that it's true. The Angels win baseball games; baseball games are contests of skill. If you go through game by game, they have clearly come through with skillful performances (unless you find bunches of one-run wins where the sun gets in an opposing fielder's eyes, or Bartman materializes to grab the ball, or some pattern like that).

Sean, just to make sure I'm reading the table in #14 correctly... For Boston when you say "Percentile 29.0%" that means their actual record is in the 29th percentile of their simulated results?

If so... whoa.

Tough to believe that the Mariners are better than all but two NL teams.

Sean, just to make sure I'm reading the table in #14 correctly... For Boston when you say "Percentile 29.0%" that means their actual record is in the 29th percentile of their simulated results?

Yes, that is what I'm saying.

71% of the time, the Red Sox distribution of runs scored and allowed leads to more wins than they currently have.

There happens to be a style of baseball played by the (Twins/Angels/CWS)

Don't the White Sox win by hitting lots of home runs?

Tough to believe that the Mariners are better than all but two NL teams.

And yet they were .500 in inter-league play.
ATL:1-2,FLA:1-2,NYM:2-1,SDP:5-1,WSN:0-3

Only two AL teams had losing records in inter-league play. Cleveland 6-12 and Toronto 8-10. KCR went 13-5.

The three NL division leaders were 15-30 in interleague play.

I think it was Gary Huckabay who wrote a BP piece a while back showing that after 150 games or so, a team's actual record becomes more predictive of their future record than their Pythagorean record.

Tell that to the Dbacks

And yet they were .500 in inter-league play.

Not that you can read that much into interleague records, but if the Mariners went .500 against a group of NL teams with a combined winning percentage of .456, that would suggest they'd be closer to the bottom of the NL than the top.

The three NL division leaders were 15-30 in interleague play.

Is that a large enough sample to draw from? Also, interleague play occurs during two short periods during the season, which may not really be representative of how those teams played over the course of the season.

Is that a large enough sample to draw from? Also, interleague play occurs during two short periods during the season, which may not really be representative of how those teams played over the course of the season.

Well, divisions are usually won by around 2-6 games or so, and I think we can usually say that the team that finishes 95-67 is better than the team that finishes 90-72. That 5 is only a small sample of games, too. 15-30 seems big enough and lopsided enough to draw a general conclusion.

71% of the time, the Red Sox distribution of runs scored and allowed leads to more wins than they currently have.

Of course, we don't know how much of this is luck (i.e. non-repeatable) and how much will persist in the playoffs. But I like it nonetheless.

Also, interleague play occurs during two short periods during the season, which may not really be representative of how those teams played over the course of the season.

The Dodgers' interleague schedule was over by the time they got Manny Ramirez.

Also, Sean's sample includes the Phillies, at 4-11 in IL play, but not the Mets, who went 9-6. The Dodgers and Phillies have two of the three worst IL records in the NL (the Padres finished at 3-12). The NL's three current second-place teams finished 22-23 in IL play, but that's not as dramatic, is it?

Only two AL teams had losing records in inter-league play. Cleveland 6-12 and Toronto 8-10. KCR went 13-5.

The Royals have a winning inter-league record over the last 5 years (48-42) while playing .376 (270-447) ball over that time against the AL. Most amusing single-season was 2005 when Royals lost 97 of 144 games against the AL ant went 9-9 against the NL.

Actually, back in post 29, I didn't account for the fact that the Mariners had played the sad-sack Padres twice as much as any other NL team. Considering that fact, the average winning percentage of the National League teams they faced is .445. Given that they went .500 against this group, one plausible guess would be that this indicates that the Mariners would be a .445 team in the National League, which is 72-90.

I find that far more believable than saying they'd be a playoff team in the NL.

The NL's three current second-place teams finished 22-23 in IL play, but that's not as dramatic, is it?

No. But put the two together: the top two teams in each NL division combined for a 37-53 record in interleague play. It's still dramatic.

If you'd rather, the top five NL teams in terms of W-L record to date went 33-45 in interleague play; the top 7 teams went 45-63; the top 8 teams (the top half of the NL) went 50-73. The best teams in the NL played worse against the AL than the Orioles played against the AL. And unlike the NL, the Orioles didn't have the benefit of getting to play against the Orioles.

And National League also-rans Cincinnati, Atlanta and Colorado finished a combined 24-21 against the AL. Imagine if those teams got to fatten up on the American Leaguers all year!

I don't understand the point of segregating out the good NL teams and thinking their performance is more indicative than the league as a whole. IL records are extremely small sample sizes to begin with.

Tom, point taken, but over 200+ games the AL outscored the NL 4.93 to 4.02 per game. That is almost as big as the average margin for the Red Sox this year. The AL flat out dominated the NL top to bottom and I really don't believe sample size caveats apply to this many games. I suspect a statistician would assign a lot of significance to a difference that big.

I agree, Sean. The AL is a lot better than the NL this year. But there's no reason to go overboard and make the problem look worse than it actually is.

I'd like to see tables like Sean's #20 for some of the 1960s NL vs AL leagues, since we didn't have interleague play then.

re: 38, the sdev of RS per game is about 3. Over 200 games, avg rs/g sdev is about 3/sqrt(200)= .21, and when comparing two clubs vs each other, it is .21 * sqrt(2) = .3.

So we'd expect two-thirds (+-1 sdev) of the time for the absolute value of [AL minus NL runs per game difference] to be less than .3.

It is .91; 3 std devs out, which happens randomly 1 time in 300.

I'd say the AL is stronger, based solely on that stat.

Past year stats and Sean's ##s in post 20 substantiate it, as well as guys who have switched leagues in the past few years and have done, in general, better in the NL.

I'd put the AL as about a .560 league vs the NL, which means two 95-win teams meeting in the WS means there is about a 62% possibility of the AL winning a best-of-7, all other things being equal.

There happens to be a style of baseball played by the (Twins/Angels/CWS) that seems to result in clubs that create data sets that defy conventional wisdom/explanation.

Just curious, what is the consistent style between these teams?

I was not necessarily saying each of the 3 shared a particular style, just that these teams had a track record recently of outperforming their Pythag. But maybe the teams that do our perform Pythag do share something? That is worth investigating. I know some reasearch found a great bulpen may be a factor in out performing Pythag.

Incidentally, the Twins and Angels, two teams that had some of the better success in out performing Pythag, both have a organizational philosophy of recent years, of playing sound D, throwing strikes, making contact, etc..

Below I've put in a table with my Monte Carlo pythag. This takes the actual scoring distributions (i.e. it treats a 12 and 0 runs scored different than two six runs scored (or allowed)) and then replays the season 1000 times based on those scoring distributing randomly jumbled. Percentile is how "lucky" the team has been. Due to run distributions (probably a couple of blowout losses. The Angels are 3 games better by this measure.

Do you have that listed anywhere on baseball-reference? You should, it's far better than regular pythag.

So, I don't know if this means anything ...

... if you go to Fangraphs, you can see every pitcher's WPA, and also their run-expectancy-derived "pitching runs" and how those convert to wins. Francisco Rodriguez has prevented 13.77 runs above average this year, by run-expectancy, which comes out to 1.34 wins (REW in the Fangraphs system). His actual WPA, however, is 3.66. So he's provided 2.32 wins above his performance, in a sense, due to (1) his leveraging and (2) his performance while leveraged. I don't know who those wins go to, maybe it's dumb luck, maybe it's Mike Scioscia, but right there are two wins that a runs scored/allowed analysis don't see.

Sean's Monte Carlo analysis is interesting ... but it's based on runs scored and allowed in a game being randomly distributed. They aren't randomly distributed, of course, because managers have an effect on the distribution.

Francisco Rodriguez has faced 285 batters this year. 40 of those are in low-leverage situations, 192 in high-leverage. That's not random; his varied performance across different leverages may be random (I don't actually know; has anyone ever looked at whether closers tend to struggle in low-leverage situations?), but his usage is not. 67% of Rodriguez's opposing PA are in high-leverage situations, and 14% in low-leverage. Darren O'Day, by contrast, faced only 5% of his batters in high-leverage situations and 81% in low. The difference between those two pitchers this year has been .050 runs per PA; that distribution is purposeful.

Out of curiosity, I added up the difference between WPA and REW for each reliever by team. Due to the way the site is organized, and time constraints, I excluded any reliever who pitched for multiple teams, but I doubt the contributions of those pitchers to their teams' totals would be too large. Note that this isn't necessarily a ranking of bullpen quality, but by how much a bullpen's contribution to wins exceeded its "performance". How well has the bullpen been leveraged (excluding multi-team pitchers)?

Team, WPA-REW, PythW-W
LAA, +4.63, +13
TBR, +4.18, +05
TEX, +3.05, +04
KCR, +2.94, +03
CHA, +2.76, -02
NYA, +2.66, +03
PHI, +2.56, -02
HOU, +2.43, +08
SFG, +1.74, +02
COL, +1.53, +02
OAK, +1.40, +00
PIT, +1.25, +00
MIN, +1.21, -01
CHN, +0.84, -01
LAN, +0.32, -03
SDP, +0.31, -06
BOS, +0.11, -01
ARI, +0.01, -01
DET, +0.00, -04
CIN, -0.28, +04
WAS, -0.41, -03
NYN, -0.42, -01
MIL, -0.52, +04
FLO, -0.69, +03
TOR, -1.05, -06
ATL, -1.54, -05
CLE, -2.26, -04
STL, -2.30, +00
BAL, -2.35, -05
SEA, -5.70, -07

There's a correlation of .68 between bullpen leverage and wins above pythag. The top 10 teams by the BL are 36 games above pythag; the bottom 10 teams are 24 wins below. Breaking it down by 10s, rankings:

01-10: +36
02-11: +23
03-12: +18
04-13: +13
05-14: +09
06-15: +08
07-16: -01
08-17: +00
09-18: -09
10-19: -15
11-20: -13
12-21: -16
13-22: -17
14-23: -12
15-24: -08
16-25: -11
17-26: -10
18-27: -13
19-28: -12
20-29: -13
21-30: -24

It's not perfect, but there is a pattern. Teams that leverage their bullpens well tend to get a boost over their pythag, at least thus far in 2008.

Now, there are a lot of reasons a team can over- or under-perform their pythag, of course, and I certainly wouldn't want to make any broad conclusions based on the results from one season (esp. given that 36 of 491 pitchers have been excluded). So, as I started off by saying, I don't know if any of that means anything. But there it is.

The Angels were lucky to get coverage in the one issue of the Wall Street Journal that just about everyone on earth with money will be perusing...

So, I don't know if this means anything ...

While very interesting, I don't think it's significant. Teams that outplay their pythag are likely to have an inordinate amount of 1, 2, and 3 run win opportunities, so you'd expect that they'd have a better reliever WPA simply because of that.

That may be so, DTM, but there may be something of a chicken-and-egg thing going on, too.

Again, the flaw in this type of analysis is that it assumes the biggest -- and sometimes *only* -- measure of "luck" is exceeding Pythag calculations. There are many, many other manifestations of luck, such as having your entire starting lineup go injury free (as happened with the Mariners last year). Also, it assumes that averages are more signficant than medians, and I'm just not convinced that's always so.

I'd like to see tables like Sean's #20 for some of the 1960s NL vs AL leagues, since we didn't have interleague play then.

I've raised this point before, and been shot down every time, but IIRC from roughly the mid-1950's through at least well into the 1960's, the NL dominated the AL in the Grapefruit and Cactus Leagues at about the same rate as the AL is dominating the NL today.

Yes, yes, these were exhibition games. We all know that. But unless you subscribe to a belief that the NL teams were for some unexplained reason always "trying harder" than the AL in pre-season games, that's an awfully large sample size to brush off.

One particular series always stood out to me, and that was the 1954 Cactus League one between the Giants and the Indians, which took place nearly every day that March, since these were about the only two teams training in Arizona back then.

After the Giants had "upset" the 111-43 AL champs by sweeping them in that year's World Series, Leo Durocher said that he wasn't the least bit surprised. After all, he pointed out, his team had played the Indians nearly every day for a month in Arizona, wih essentially the same rosters, and had beaten them over 60% of the time. And even without counting those Spring games, he also noted that those 111-43 Indians had a losing record (22-26) against the only three teams they'd played that year who had winning records.

Which is one reason I take Sean's statistical gatherings seriously, and why I'm not inclined to brush them off with "sample size" or "random variation." And for the past 4 years, the overall interleague results (576-432) are starting to match the All-Star (11-0-1)and World Series (37-20) outcomes of the past 12 years. This is rapidly starting to look like the NFL of the 80's and early 90's.