THE BOOK--Playing The Percentages In Baseball

playoffodds.blogspot.com has the odds as 54.5% for the Rays in the WS.  Now, Home Field Advantage is somewhere between 53.5% and 54% I think (maybe more this year… didn’t it start off around 57% for the first half?).  And it’s probably greater in the world series because of how much more having a DH helps the AL as opposed to the NL.  So really, the entirety of that 4.5% advantage (and maybe more) is covered by extrinsic factors.  Do the stats really have TB and PHI as equals, or even PHI as the better team?  Sportsbooks, according to THT, have it as 58% for the Rays, which makes sense to me.  4 or 5% I’d apply to HFA and DH rule, maybe 2% for the league difference, and 1-2% for TB being a more talented team.  Anyways just curious what others would set the odds at for this series.

Mike/231 - You’re forgetting that the HFA only comes into play if the series reaches game 7. If you have two evenly matched teams, there’s around a 33% chance that they’ll play a game 7. Therefore, the HFA for the series is whatever the HFA for an individual game is divided by 3 (i.e. if the home team should be expected to win 55% of the time, they should be expected to win about 51.7% of the time in a 7 game series).

#232 is correct.  If both teams are even with a 5% HFA (winning 55% at home), then there will be a game 7 31.5% of the time.  And the team with the game 7 HFA (in this case, TB) will win the series 51.6% of the time.

HFA this year I think went down to 55% overall, not a whole lot higher than historically.  Whether that is just a fluctuation or a result of something else (like something to do with fewer PED use) which would persist in the post-season, I don’t know.  Probably just noise.

In any case, I think the playoffodds.com odds is underestimating TB a little.  I am not sure what they are using to calculate the odds.  At this point in the season, anything that does not use projections for the actual players who are going to play (and especially pitch) in the WS, is a little silly wouldn’t you think?  For example, should we be using statistics from the season when guys like Eaton and Howie Kendrick pitched a lot of games for the Phils?  Should we be using stats from the season, when Crawford and Longoris missed a good portion of the season?  And no David Price?  Using season records and stats is nice “toy” for predicting post-season odds, but it should NOT be taken seriously by any stretch of the imagination.

Vegas has TB around 58%.  Supposedly they made it a little higher than they think they really should be, to offset some of the money the books would lose if TB wins the series (since many people bet them to win before the season started a long odds).

In any case, I think most rational people are in agreement that the strengths of the teams are fairly similar, even considering that the AL is much better overall.  Actually that is interesting considering that the Rays won 5 more games than the Phillies in the AL which corresponds to around 9 or 10 more games after adjusting for the relative strength of the leagues.

If TB were truly 10 games better than the Phillies, they would be around a 63% fave to win the series.  I guess people don’t believe that their respective records are indicative of their true talents (and they would be right), and/or there is not that big of a disparity between the leagues (and they would be wrong).

Earlier I said 60/40 Rays, but just pulling numbers out of my A.  The 67% chance you calculated is probably right.

Last year BPro had Colorado as a favorite over Boston, for what reasons I can’t even fathom, but I think they changed it before the series actually started.  Around here somebody (probably MGL) said it was 70/30 Red Sox for the series.

Rays are probably not quite as good as the 07 Sox, true talent-wise, and Philly is probably ahead of the Rockies, but the Rays should be heavy favorites.

The Rays pushed their series to the limit, while the Phillies win in 5 and get a full week off.  Does this change the odds?  I did a study once on teams that swept their LCS opponents vs those that went the full 5 or 7 games in the other LCS.  Meeting in the world series, the two team groups were very close to 50/50, so I don’t think we have a predictive effect here.  If anything, the teams that swept should be considered the better team going in (selective sampling) so if all they could manage was 50/50, perhaps the long layoff hurt more than it helped.

Tango, Vegas is a solid -138 for TB for the “no juice” line (e.g., -143/+133), not -125.

I really don’t know the difference between the NL and AL this year.  I only guess at it and I usually use a conservative estimate (it is always correct to be conservative in the “default” direction).

I agree that BP’s numbers are ridiculous no matter how you look at it, especially once you factor in the relative strengths of the leagues.

Here is a quick assessment of each team’s true talent, not including any league differences (IOW, each player’s numbers are relative to players in their own league - e.g., a zero in the AL is better than a zero in the NL).

All numbers are runs per 150 games.  I tweaked some of the numbers.  For example, Upton’s UZR projection in CF is negative. Longoria is zero, Bartlett is still a little hurt, I think.  So is Crawford.  Etc.

PHI

Player offense defense

Rollins 0 1
Werth 17 6
Utley 19 12 (monster player)
Howard 23 -1
Burrell 14 -12 (like Manny, his defense pretty much negates his offense.  Should be a 1B or DH.)
Victorino -2 10 (good defense, not the offensive player he was made out to be during the post-season.)
Dobbs/Feliz -9 -9/-18 9
Stairs/Bruntlett (DH) 0/-19 (I don’t know who is going to DH for Philly versus Kazmir.)
Ruiz -9/x (no defense for the catchers)

Total of around +66, or +42 offense and +24 defense.  That is .28 per game on offense and .16 (runs saved) on defense.

Starting Pitching (4.50 is an average starter)

Hammel 3.50
Myers 4.50
Moyer 4.80
Blanton 4.30
Hammel
Myers
Moyer

So, the average starting pitching would be:

.23 runs above average per 9 innings per game, or .15 runs saved per game (assuming 6 IP per game per starter).  The Phillies do not have a good starting pitching staff for a WS team.

Total for pitching, offense, and defense is .28 for offense and .31 for pitching/defense.  At 4.5 rpg, they will score 4.78 and allow 4.19.  That is a .565 team.  Again, relative to a .500 NL team.

TB

Iwamura -2 -1 (not the defensive whiz he is sometimes made out to be, at least according to UZR.) Not much of a hitter either.
Upton 13 0 (UZR has him at -7 in CF believe it or not.  Everyone is talking like he is the next Willy Mays on defense, so I made him a zero.  I can’t justify much more than that.)
Pena 25 -1
Longoria 15 3 (I had him at zero projected UZR, but he is regarded as much on defense than that.)
Crawford 0 5
Navarro -11 x (That is really bad hitting for a switch hitter.)
Floyd/Gomes (DH) -2/13 (That’s right, Floyd is NOT a good hitter anymore.  That is terrible for a DH!)
Baldelli/Gross 0 0/5 0 (Baldelli does not look healthy to me.)
Bartlett -9 4

That is +35 on offense and +10 on defense, or .23 per game on offense and .07 runs saved on defense.

Starting pitching:

Kazmir 3.80
Shields 4.10
Garza 4.20
Sonnastine 4.70

Kaz
Shields
Garza

That is .52 runs saved per 9 IP, or .35 per game (6 IP per game).

So TB, in the AL, is .23 rpg on offense and .42 rpg saved on pitching and defense.

That is 4.73 scored and 4.08 allowed in a 4.5 rpg environment, or a .573 team in the AL.

Add in another .005 wp edge for TB on base running.

I think the bullpens and benches are about the same, although I don’t think it matters much.  Lidge is a very good closer. TB does not really have one (they might use Price), but they have lots of quality relievers to use late in the game.  Maybe you can give an edge to the Philly pen, I don’t know. OK, let’s cancel out the Philly pen advantage and the TB base running advantage.

So we have a .565 team versus a .573 team, both relative to their respective leagues.

If the AL is .525 and the NL .475 overall, then we have around a .530 team playing a .470 team.

That makes the TB wp versus PHI .560 per game.

In a 7-game 2-3-2 format, that makes them 64% to win the series.

If I sim out each game separately, with all the R/L matchups, I get TB around 58% for the series for some reason.

Whether the long layoff helps or hurts, either:

a) no one has any idea, or;

b) it neither helps nor hurts to any significant degree (other than how it might change the pitching rotations).

Both are essentially the same thing.

Trying to find empirical evidence one way or the other is an exercise in futility, because of the sample sizes.

I always say that if you are trying to find subtle effects (at least you think that if the effect is there, it is probably small), don’t even bother looking at empirical data unless you have an enormous sample size.  That is especially true if the result you are looking at is filled with noise in the first place (e.g., win loss records).

Insightful, thanks #2 and #3!

MGL, fantastic, thanks.

A -138 non-juice implies a .580 series win record, or .537 single game record.

***

MGL’s sim, which may or may not have the league adjustment, gives us a .580 series record.

***

MGL’s player-by-player analysis, presuming a .5267/.4767 league split turns Phillies’ .565 into a .544 record in a .500 league.  The Rays’ .573 becomes .599 in a .500 league.  And .599 against .544 means .556 single-game odds for the Rays, or a .621 series odds for the Rays.

***

So, in summary, we have the single game odds for each one as:
.579 Baseball Prospectus implied by Pct3
.556 MGL player-by-player analysis
.537 Vegas
.537 MGL simulator
.521 MLB Playoff Odds
.516 Cool Standings
.493 Baseball Prospectus official Odds

***

.550: If a team in the AL faces a team in the NL with identical W/L or RS/RA records.

I think the most reasonable of all the numbers above has to be MGL’s player-by-player analysis, or BPro’s Pct3; the minimum has got to be .550.  The Rays, relative to the AL, must likely be better than the Phillies, relative to the NL.

Agreed on the layoff.

Given the sample size of teams we have to study, we could still say there was an effect if the results were overwhelming.  If in 20 chances, the team that was better rested won 18 series then we should consider there is something to it.  If the advantage was smaller, like 12-8 or 13-7, then I don’t think it’s significant, and quite possibly due to chance.  When we get results like 10-10, well, nothing to see here.

If somebody tells you Philly should win because they are better rested, the proper response is “Oh yeah?  Well TB will win because they are sharp and the Phils have gotten rusty.” Or vice versa.

I run MLB Playoff Odds.  I’m a professional baseball handicapper, and these are the estimates I worked out.  I wrote up a summary of how I calculate these odds for THT; you can read it by clicking my name.  Of the methods summed up in #8, it’s most similar to MGL’s player-by-player analysis; this is why my .545 estimate is similar to his .556.

As for why the Rays are a smaller favorite than some think they should be, it has a lot to do with MGL’s point in post #3.  Edwin Jackson isn’t a good pitcher, but a 4.42 ERA from your number 5 starter is much better than most teams get.  Meanwhile, Eaton and Kendrick were terrible, even for number 5s.  Sonnanstine is also a better number 4 than Blanton, but they only start once each.  When you take those factors out, the gap between the teams narrows considerably.

James, you have .521 on a game-by-game basis, and .545 on a series basis.  That does not compare at all to MGL’s, unless I’m misreporting your numbers.

If I recall right from last year, BP was using a team’s 2008 won-loss record versus LHP and RHP to drive their postseason projections. That of course is not at all predictive for so many reasons and I suspect that the same thing is happening this year. Clay wrote this over at their postseason odds page:

“Philadelphia’s WS odds are down precisely because of those L/R splits; with LH Moyer and Hamels, they should match up better with TB.”

My only objection to James’ article above is Dane Cook did his “October” commercials last year.

James, are you stuck in a time warp?

Tango you are right.  James’ numbers are nowhere near mine. At least not by my player-by-player ones.  Mine is .560 per game (I don’t know where the .556 came from), which is .640 for the series.  His is .545 for the series.  Quite a difference.

Having TB .545 for the series implies that they are .515 per game, which implies that they are quite a bit worse than PHI relative to their own leagues (assuming that you think that there is a large difference between the leagues).

I think the notion that the Phils are quite a bit better relative to their league than the Rays are relative to their league is quite a stretch by any way of thinking.  Certainly my numbers do not bear that out.

As I said and showed, the offenses and defenses, relative to each team’s respective leagues, are quite similar, and the pitching seems to be a little better for the Rays.  I am not sure that too many people would argue with that.

Nothing against James at all, and he may be one of the .1% legitimate handicappers, but telling someone you are a “professional handicapper” as evidence of your competence to come up with credible odds for a game or series, is like telling someone that because you are a professional politician, surely your integrity and honesty must be second to none.

99.9% of all “professional handicappers” are charlatans or incompetent but honest analysts.  If James is at all in or near that .1% category, I am sure he would agree with that.

Nothing against James at all, and he may be one of the .1% legitimate handicappers, but telling someone you are a “professional handicapper” as evidence of your competence to come up with credible odds for a game or series, is like telling someone that because you are a professional politician, surely your integrity and honesty must be second to none.

Well, Wayne Root is running for vice president!

Here’s a blog that posts their odds, and offers an EXE file as well.  I know nothing about it, so if someone wants to try it out and report back, that’d be wonderful:

http://dodgersims.blogspot.com/

The game odds look OK for game 1, but the total runs is way, way, way too high. The Vegas total line is 7.5 under, which is around 7.3 runs, which implies an average rpg of around 7.95.  I don’t know how they can be off so much, but I have no idea of their methodology.  Or maybe they are right and Vegas is wrong (I doubt it).

For one thing, post-season games are lower scoring even after you factor in the colder weather and the starting pitchers and batters who play.  The reason is that teams play more small ball and managers use their bullpens much, much more efficiently than in the regular season.

#11: Whoops, I can’t read.  I suppose this isn’t a legitimate claim in favor of my analysis.

#13: I completely agree with the 99.9% statement.  When I call myself a professional handicapper, I’m explaining why and how I developed the numbers.  I am NOT saying “Hey, I do this for a living, so I’m right and you’re wrong.” Note that I’m not attempting to sell anything.

#15: I don’t mean to insult the author of Dodgersims or his methods, but he includes a link to AccuScore on his blog.  I personally think AccuScore is awful, as I wrote here a year ago:

http://gambl0g.blogspot.com/2007/06/accuscore.html

(The article referenced in the post has since been removed from their site, but the quotes are genuine.)

In 2008, the W/L record was .591 (!) for the AL over NL.
2006-2008 was .582.
2004-2008 was .558.
2002-2008 was .533.
2000-2008 was .533.

Here’s the 5yr average, with the year listed being the last year in the time period:
2001 0.499
2002 0.505
2003 0.494
2004 0.502
2005 0.502
2006 0.519
2007 0.531
2008 0.558

Starting in 2005 is when the split starts to happen:
http://mlb.mlb.com/mlb/history/interleague/records.jsp

Preferably, I’d like to see the RS per 27 outs and RA per 27 outs.

Btw, over a 5-yr time period, 1259 games, 1 SD = .014.

The 3 year is 1 SD = .018.

So, I don’t see how we can expect to say anything lower than .550/.500 as the true split.

Unfortunately the RetroSQL schema doesn’t have an “interleague” flag - maybe something to think about adding?

I think I have a table set up to map Retro team IDs to BDB team IDs. I’ll see if I can wire something together.

I really don’t see how you can knock someone like AccuScore without having done some kind of analysis as to how accurate they are to actual results compared to Vegas and how one would’ve done had they used AccuScore as a tool to help wager.  Just because they have grammar errors and typos in their FAQ doesn’t mean anything.  I am not saying they are great, I am just saying the analysis of their site in #17 is poor.
vr, Xei

I don’t mean to insult the author of Dodgersims or his methods

Hey, if there is any place that you can responsibly “insult” someone by criticizing their work, it is here.  As Tango likes to say, if he (or anyone) writes or says something that I think is crap, it is irresponsible of me to say otherwise or not say anything at all.  Ditto for me writing crap.  People criticizing your criticizing because they don’t like your “tone” is B.S., but that is another story (your - James - tone was fine of course).  It is the equivalent of one person arguing a point about something, a second person offering a cogent rebuttal, and then the original person saying, “Oh yeah, well, your mother is ugly!”

Anyway, I’ll check out that link and see what I think.  As I said, the “total” for the game is not even close to what it should be, the Vegas line notwithstanding.

And, BTW, if you profess yourself to be a sports handicapper and you are not selling anything, then you go from 99.9% BS to 98% BS (numbers are a WAG).  I guess that is something!

Clay wrote back, and confirmed it’s because of his LH/RH adjustment.  He said that if he removes the adjustment, he has the Rays as .580 to win the series.

The single-game win% becomes .537.  Here is the updated list:

.579 Baseball Prospectus implied by Pct3
.556 MGL player-by-player analysis
.537 Vegas
.537 MGL simulator
.537 Baseball Prospectus, no LH/RH adjustment
.521 MLB Playoff Odds
.516 Cool Standings
.493 Baseball Prospectus official Odds (with hand adjustment)

I really don’t see how Clay can be handling the splits to come up with those results. Here’s the platoon splits I came up with:

http://statspeak.net/2008/10/the-first-casualty-of-batting-left-handed-is-innocence.html

Brian and I had a long chat on how to improve those, so I wouldn’t take them as gospel, but… well, I suppose I should just run the numbers player by player if I really want to comment on this.

Sigh.

Clay is using the team record against LH/RH.  I already mentioned my displeasure with that approach.

This is one of those things where I think “nothing is better than something”.  Same with (unregressed) park factors.

Actually, in many cases, something is worse than nothing.  For example, un-regressed one-year park factors may be worse than nothing.  Team records against LH and RH pitchers - ugh, that is ugly!  I think nothing is better than that.

BTW, those odds that Tango is giving is for each single game (he does say that). I don’t know why you are listing it that way, although it doesn’t really matter. It is confusing though.  After all, each game has its own separate wp and we are talking about the odds for the series. Why not list everyone’s wp for the series?

I also show the Series record in an earlier blog post.  The reason I show the single game record is that I want to see it against the .550 line, which is what the average AL team does against the average NL team.  I don’t know that if I say that a best 4 out of 7 between an average AL against average NL implies a .608 record that people would “believe” that, but that they would believe .550.

***

http://sports.espn.go.com/mlb/playoffs2008/news/story?id=3656676

Diamond-Mind gives the Rays a 70.7% chance of winning the series.  The implied odds of that for each game is .598.

Interestingly, they list the actual split, and it does not come out to .598 at all.  It comes out to .576, which is obviously impossible after 2000 simulations.  If you flip their chart for the Phillies around (Phi 7, Phi 6, Phi 5, Phi 4 being the correct order), you get… .598 as the single-game odds.

Here is the updated list:

.598 Diamond-Mind Baseball
.579 Baseball Prospectus implied by Pct3
.556 MGL player-by-player analysis
.537 Vegas
.537 MGL simulator
.537 Baseball Prospectus, no LH/RH adjustment
.521 MLB Playoff Odds
.516 Cool Standings
.493 Baseball Prospectus official Odds (with hand adjustment)

Rays: 774/671 (pythag .565)
Phillies: 799/680 (pythag .575)

That’s each team’s runs scored and allowed. 

AL: 10844 / 10609 (pythag .510)
NL: 11741 / 11976 (pythag .490)

As noted, if we put them all in the same league, the AL team should be around .5267 and the NL would be around .4767.

The Rays are +.055 compared to the average AL team, so adding that to .5267 gives us:
Rays: .582

Phillies are +.085 compared to the average NL team, so adding taht to .4767 gives us:
Phillies: .562

This gives us a single game win% of .520 in favor of the Rays.  This doesn’t consider the harder schedule the Rays faced, or the change in rosters for the playoffs.

Under this analysis, it seems that Vegas’ .537 for the single-game win% is just about right.

How about that…

Definitely agree on the LH/RH issue - I think you’d be far better off ignoring that. Or at least regressing it substantially based upon lineup composition.

Tango you are right.  James’ numbers are nowhere near mine. At least not by my player-by-player ones.  Mine is .560 per game (I don’t know where the .556 came from), which is .640 for the series.  His is .545 for the series.  Quite a difference.

Tom, shouldn’t MGL’s single game number be .560?

If you use his numbers and apply the Odds Ratio Method, you get .556.  His .560 was a quick addition.  No biggie either way.

Or at least regressing it substantially based upon lineup composition.

Well, once you know lineup composition, you’d be silly to use the actual records at all.  I mean if I told you that team A had 4 lefties and 5 righties (that’s a lot of lefties for a typical lineup) and their record against lefties was .540, but against righties it was .480, would you really want to use those numbers and then regress (even heavily)?  I wouldn’t think so.  I think you would simply take their overall record and assume that they “should” be better against righties than lefties (assuming the same quality righty and lefty).

Dolphin has TB at .601 at home and .521 away vs Phi on a single-game basis.

That’s roughly .560, depending on number of games.  Does Andy give the final Series %?  (His site is barred from work.)

.598 Diamond-Mind Baseball
.579 Baseball Prospectus implied by Pct3
.560 Andy Dolphin
.556 MGL player-by-player analysis
.537 Vegas
.537 MGL simulator
.537 Baseball Prospectus, no LH/RH adjustment
.530ish Runs Scored/Allowed, league adjustment
.521 MLB Playoff Odds
.516 Cool Standings
.493 Baseball Prospectus official Odds (with hand adjustment)

AccuScore fwiw, had the Rays with a 68% chance of winning the series before game 1.
vr, Xei

.598 Diamond-Mind Baseball

.585 Accu-Score
.579 Baseball Prospectus 1 of 3 (implied by Pct3)

.560 Andy Dolphin
.556 MGL player-by-player analysis

.537 Vegas
.537 MGL simulator
.537 Baseball Prospectus 2 of 3 (no LH/RH adjustment)
.530ish Runs Scored/Allowed, league adjustment

.521 MLB Playoff Odds
.516 Cool Standings

.493 Baseball Prospectus 3 of 3 (official Odds, with hand adjustment)

That’s just from the “predict score” feature on his page (http://www.dolphinsim.com/ratings/mlb/index_pred.html), which includes playoff games already played.

With the Phillies up three games to one, I am going to go with the following World Series win probability for the Rays.

Game 5: .38
Game 6: .63
Game 7: .58

Win Prob: 13.9%

I reserve the right to modify these game win probabilities, especially the Game seven number.

vr, Xei

If the AL is much better than the NL (and all the data seem to point in that direction), why such a low percentage for TB in game 5?

With the large difference in pitching between the leagues, Kazmir cannot be much worse than Hammels.  And with the overall difference between the leagues, including on offense, the rest of the Rays team almost certainly has to be better than PHI.

Tampa Bay has a rather healthy HFA, while Philadelphia has a moderate one.  The difference between Game #1 and Game #5 odds (Kazmir vs Hamels in both games) is almost all based on the HFA or lack there of.  As to why the game win probabilities are the way they are, my best answer would be that that is just how the talent plays out.  The current Phillies team may be better than their final regular season record indicates.
vr, Xei

Tampa Bay has a rather healthy HFA, while Philadelphia has a moderate one. 

That would make PHI even less of a fave in game 5, no?

Why do you have different HFA for both teams.  Each team’s HFA (over and above the usual 54 or 55% of course) this year has exactly:

ZERO

predictability.

And if it had any, you would regress one year’s HFA a lot anyway (like almost 100%).  We’re talkin’ about two samples of 81 games here (record at home versus record away) folks.  Whether there were any unique HFA or none, 1/3 of all teams will miss their true mark by almost 8%!  How would you ever find differences of 1 or 2% amid all that noise?  You can’t!

So are you saying that all teams in baseball have the same HFA?  And a team being an extreme outlier for three years in a row is just noise?  How about other sports like basketball that have even less games (sample size), are they all suppose to be measured with the same HFA?  Not trying to argue, just curious as you’ve seemed to have done some extensive research on this matter.
vr, Xei

No. All teams in baseball have roughly the same HFA, other than COL, BOS, and MIN (presently).  In the past, SEA (in the Kingdome) and HOU (in the Astrodome) had a large HFA.  The only thing that makes for a larger HFA in baseball is a quirky park.  Actually there were some teams in the 70’s and 80’s that had a lower than average HFA.  Those in cookie cutter parks, like CIN, PHI, STL, and PIT (off the top of my head).

If a team does not have something unique about its stadium to suggest a larger or smaller HFA, you can regress any one-year or even 10-year sample HFA almost 100% toward the mean.

True in basketball to an even larger degree.  Since all courts are basically uniform, only Utah and Denver have different “true” HFA due to the altitude.

I don’t know about football, but it is common among betters to assume that all teams have the same HFA in the NFL, at least among the smart betters (the “public” betters think all kinds of goofy things).

MGL -
I realize the increase would be small, but if the Metrodome, Kindgome, and Astrodome all had above average HFAs, isn’t it reasonable to speculate that Tropicana Field has one? I’m not sure what makes certain domes more quirky than others.

Also, a year ago, Tango listed the home/road records for every team since 1994. The Rays had the 4th highest split. (click my name for the link)

My guess is that if you were betting the Rays this year, using a 4.2% HFA you would be betting the road team to beat them quite often, which given the Rays amazing home record wouldn’t have worked out too well this year.  Could still be noise, but they have one of the best HFA track records the past couple of seasons and do play in a dome fwiw.  vr, Xei

Here’s a study from a few years ago of the persistency of home field advantage. I split the average home field advantage over a three-year period into five groups, and then checked the HFA for the teams in each group in years 4-5 (I don’t have the date range, but there are just under 400 teams in each group—I believe it is 1900-2005):

Avg years 1-3    Avg years 4-5
    +.154            +.091
    +.106            +.080
    +.080            +.079
    +.054            +.076
    +.015            +.080

The regression equation is .094*(years 1-3) + .073, with an r2 of .006.

Colorado had a HFA over 2003-2005 of +.185, so the 2006 estimate was +.090. The Cubs were at -.013, for a 2006 estimated HFA of +.072. So at the end of 2005, the observed difference in HFA for the Cubs and Rox was almost .200, but the TRUE difference was about .020.

A semi-related question I’ve been meaning to ask.

Is it EVER right to predict/bet on a sweep in a 7-game series? I’ve seen a lot of so-called experts predict a sweep, but that just seems wholly illogical to me. Sure, they happen. But what I’m thinking is that the true talent level between the teams would have to be so drastic that betting on (I guess the odds factor in here, so maybe a prediction where the “reward” for being right is the same regardless of what that right result is) one to sweep is always (99.9999%?) wrong.

Take the 1927 Yankees and the 2003 Tigers. How many times out of a million do the Yanks sweep? 10%? 20%? I have no idea. But no World Series is ever going to be anywhere near that. I can’t imagine the sweep likelihood based on true talent is ever more than 5%, so why predict this if your goal is to be right (and appear you actually know WTF you’re talking about)?

Great job Dackle!  It seems that the HFA’s persistence only exists with the high-range teams.

So, we can present it as this equation:
true = sample^2.5 + .08

Here’s how that looks:
1-3..  4-5..  Estimate
0.154 0.091 0.089
0.106 0.080 0.084
0.080 0.079 0.082
0.054 0.076 0.081
0.015 0.080 0.080

In order to place an even-money bet on a team to sweep in a 7-game series, the single-game win% must be .841.

In order for the sweep to be the most likely of the 4 outcomes, you need to have the single-game win% of .750.  Under that scenario, the chance of a sweep (or ending in 5) is each 31.6%.

So, forecasting for a sweep can only occur under an extreme imbalance of talent between 2 teams.

Might as well do this.  To forecast a win in X games, then you expect the single game win% to be Y.

Here are the X and Y numbers:
4: at least .750
5: .600 to .750
6: .500 to .600
7: exactly .500

So, what you should ask your Nostradamus is to tell you how often you expect the better team to win each game.  If he says .550 and he also forecasts a 5-game series, then he’s full of it. 

With the HFA, the single-game win% at the fringes can put it up or down one slot.

You should never forecast a 7-game series unless you truly think that you have two teams that are equal, or you can give as many reasons that one team is better as the other team is better.

Tango, that is not right because of 2-3-2 format and the HFA.

For example, if a .600 team plays at home first (.645 at home and .555 on road), the series will go 5 games .261 and 6 games .308 - not even close.

If the .600 team plays on the road first, it will go 5 games .278 and 6 games .291.

Either way, one team being a .600 team is not enough.

As well, 7 games is more likely if one team is a .550 team.

Even without the changing HFA format, I don’t know where you are getting those numbers.

If a team is a .600 team each game (not .645 at home and .555 on road), it goes 5 games only .269, 6 games .300, and 7 games .276.

I think you need to redo your calculations. 

And using a regression equation for HFA is fine if you know nothing about a team.  If you know something, then, it is silly.

All of the first term (sample^2.5) comes from MIN, BOS, and COL.  It makes no sense to use the equation for any particular team, unless of course, you don’t know what team it is, I guess.

Xeifrank, if you look at the number of fly ball doubles per ball in play for the road team and home teams in the Metrodome, you will see a difference.  Not so in the Tampa Dome, the last time I looked at the data at least (several years ago).  When I was at the Tampa Dome last week, the first thing I said to my son was that it is easy to see a ball in the roof, which is kind of gray.  In the Metrodome, it is while and difficult to see a ball.  In modern domes, that is what gives the home team an extra advantage (or not).  In the old Astrodome, it was the very poor lighting. I don’t think you have poor lighting in modern domes.  In any case, that is evident from the home and road K rates.  In the Astrodome it was “night and day” the difference between the home and road team’s K rates.

If you made a mistake in your calcs, Tango, rather than me, I suspect it is by including, for example, a WWWWL in your chances of series going 5 games (and the same thing for all the other chances).

I’m using the following number of combinations for each series:
4: 1
5: 4
6: 10
7: 20

So, the 5-gamer one would be:
LWWWW
WLWWW
WWLWW
WWWLW

The equation I am using is:
(games-1)! / (games-4)!3!

So, for 5 games, that’s:
4! / 1!3!
= 4

For 7 games, that’s
6! / 3!3!
= 20

And my final calculation for each number of games, from 4 to 7 is:

p^win * (1-p)^win * Combination

If someone wants to correct this, please feel free.