Half season rap-up: Home/road splits, one-run
games, normalized records, and how the playoff picture is shaping up
We have now completed half
of the inaugural Brassworld season, and it has been loads of fun so far. With a
reasonable number of games played, we are now in a position to determine who
the good teams are and how the playoff picture shapes up. There are two
interesting questions we may want to ask ourselves at this point: (i) which
teams are most likely to improve or regress in the second half? And (ii) which
teams appear to have the best chances to win it all? This is all speculation at
this point, but that’s the fun of it, right?
One
thing that needs no speculation is that, judging from their current overall
records, there are some very strong teams in this league. To name but one per
division, Syracuse, Toontown, Virginia, and Maryland all look pretty good, both
performance-wise and on paper.
However,
won-lost records don’t always tell the whole story. Baseball is a game of skills,
but also a game of luck, a fact that is conveyed in strat by the random rolls
of the dice. Whether they are right or wrong, one result that sabermetricians
(no, I’m not a stathead- not that there’s anything wrong with that …) attribute
to random luck is the outcome of games that are decided by the slightest of
margins, i.e. one run. Still, doing well in one-run games can sometimes end up
being the difference between winning a pennant and a third place finish.
Another factor that tips the
balance of results is, like in any sport, the home field advantage. Generally
speaking, teams perform better at home than they do on the road. This might
even be more of a factor in Brassworld given that home managers, if not playing
via NetPlay, have the advantage of playing against their opponent’s computer
manager. As carefully as one programs his CM, and as much as HAL has improved
over the years, the fact remains that it’s not the optimal way to manage a
team. Managing it yourself is. (That’s why I encourage everybody to use NetPlay
as much as they can. Plus, it’s a lot more fun squaring off with another human
being than it is with a cyber non-being. And, let’s face it, the win is much
more satisfactory and the loss less frustrating.)
While it is difficult, if
not impossible, to factor the overall effects of the random rolls of the dice
out of a team’s record, it is possible with both one-run game results and
home/road records; their effects are quantifiable. In doing so, and assuming that things remain the same from July to September, we get at least
an idea of each individual team’s likelihood to improve, regress, or stay put in
the second half. More precisely, the idea is that the less a team has benefited
from the home field advantage and one-run game results (negative splits), the
more likely it is to improve in the second half, while a team that has
benefited the most (positive splits) is more likely to regress or stay put. In
the process we may also be able to identify which teams will be the toughest
come playoff time.
The method is not
scientific, but it has some merit as it uses some kind of personalized version
of the standard sabermetric approach random luck measurement (left out of the
equation is the Pythagorean calculation of run differentials; I just don’t know
the formula, and I’m not sure I’m a firm believer either). There are some
caveats, though (more details below): The sample size is a little small, and
the method is not foolproof. For one thing, the teams’ April-June overall record
may have been affected by “bad luck” altogether in spite of neutral home/road
and one-run game splits. Still it’s a fun way to break down the numbers and to
look at each team’s individual performance beyond its overall won-lost record.
Besides it still gives us a fairly accurate idea of the relative strength of
the current .500+ teams.
On with the show.
Home/road splits
When determining how much a
team deviates from the norm in any given performance measure, the first thing
to establish is, of course, the norm itself. In the case of home/road splits,
we want to know how all teams perform at home on average, and compare each individual team to this average in
order to see whether a given team performs better or worse than the norm.
One-run games
This is pretty straightforward.
Assuming that one-run games are determined by shear luck (and, yes, this is
debatable), a team not affected by random results should have a .500 record in
these contests. A better record than .500 is a negative (luck involved), a
worse record a positive (bad luck involved).
Illustration
Let’s illustrate with a
fictitious example.
Say Team A has an overall
record of 60-40, with a 34-16/ 26-24 home/road (H/R) split, and has a 10-6
record in one-run games. Let’s further say that the league’s average home-road
differential is 2, i.e. that, on average, teams are 2 games better at home than
they are on the road (e.g. a team with
a 20-30/18-32 split falls within the norm).
Team A’s home record is 8
games better than its home record, meaning that it performs 6 games better than
the league average (8 minus 2); its 10-6 record in one-run games also says that
Team A performs 4 games better than the norm. (IMPORTANT NOTE: Keep in mind
that the fact that a team is x number
of games over .500 in one-run games means that it has actually won x/2 games wore than it should have; for
example, if 2 of the 16 one-run games had been decided in favor of its
opponent, Team A would have played for .500, i.e. 8-8; the same reasoning
applies to the H/R differential; as we will see shortly, this is taken into
account in the overall calculation since all differentials are ultimately
divided in 2). This is summarized below:
Games H/R game diff (=
8) One-run
game diff.
Over
.500 minus
league diff.
(= 2)
Team
A 20 +6 +4
The first impression we get
from Team A’s splits is that, although it has a very solid record, it has also
been somewhat “lucky” by performing better than the norm in both home and
one-run games. It’s still a good team, but what we’d like to know is how good
it really would be if we strip away random luck - or “over performance” - from
its won-lost record. One way to achieve this is to deduct the over performance
numbers from Team A’s overall record.
In this particular case the calculation goes like this:
20 – 6 - 4 = 10
That is, 20 games over .500,
minus the 6 games over the league norm in H/R splits, minus the 4 games over
.500 in one-run games. The resulting number is 10. This is what I call the Balanced Performance Indicator (BPI).
This indicator tells us how many games
over .500 a team should be if random luck is taken out of the equation. In
Team A’s case, the BPI suggests that it should be 10 games over .500 instead of
20, i.e. its normalized record
should be worse than its actual record.
To calculate Team A’s normalized
record, we first establish what the .500 record should be given the number of
games played by Team A. For 100 games played, the .500 record is 50-50. In
order to get to the desired 10 games over .500 normalized record, we need to
divide the BPI in 2 and add the result to the 50 wins, while subtracting it
from the 40 losses. Dividing 10 in 2 is 5, and adding 5 to the wins and
subtracting 5 from the losses yields a 55-45 record, 10 games over .500; that is Team A’s normalized record.
Projecting Team A’s
normalized record over 162 games and comparing it to a 162-game projection of
its actual record gives un an idea of what Team A’s record would be if random
luck did not enter the picture over an entire season.
x number of (normalized) wins in y number of games à z
number of wins in 162 games
we get a projected record
(based on its actual 100-game record)
of 97-65, and a normalized record of 89-73. That’s an 8-game difference, in
other words, the potential difference between a pennant and a fourth place
finish.
Let’s now look at the teams
that are playing for .500 or better so far in our league.
Brassworld’s current .500 teams
Before we get into the
teams’ records as such, we need to establish how home teams in Brassworld fare
on average.
At this point, there have
been 972 games played. Home teams have won 500 of those games, losing 472, good
for a .514 average. The overall game difference is +28. Spread among the 24
teams, we get the average home team
differential of 1.17. In other words, with respect to their overall records
teams perform an average of 1.17 games better at home than they do on the road.
Teams with a higher differential “over perform” at home, while teams with a
lower differential “under perform”.
Concerning one-run games, it’s a zero-sum thing. As
many teams win one-run games as they lose them. Consequently, I will just
compute the number of games over .500 a team has in those contests.
Let’s now look at the teams
with .500 or better records so far (I use this cut-off point both for
simplicity (and laziness) as well as an indication of a team’s likelihood to
make the playoffs as things stand right now; it does not mean that teams presently playing below .500 have no chance to
make the playoffs; we still have a long, long way to go; I come back to this
below). The splits for those teams look as follows (teams presented in
alphabetical order):
Table 1. Home/road and
one-run game splits for teams playing .500+
in the first half
One-run
H#R game games
Overall pct. Home
pct. Road pct. diff diff.
Abilene 44-37 .543 24-18 .571 20-19 .513 +2.5 +1
Annadale 41-40 .506 21-21 .500 20-19 .513 -0.5 -1
Aspen 48-33 .593 22-17 .564 26-16 .619 -2.5 -1
Buckeye 42-39 .519 24-18 .571 18-21 .462 +4.5 +8
Greenville 46-35 .568 26-16 .619 20-19 .513 +6.5 +13 y
Lafontaine
Park 45-36 .556 23-18 .548 22-17 .564 -0.5 +3
Maryland 51-30 .630 26-13 .667 25-17 .595 +2.5 -1
Plum
Island 45-36 .556 27-12 .692x 18-24 .429 +10.5 0
Syracuse 52-29 .642 29-13 .690 23-16 .590 +4.5 +6
Toontown 51-30 .630 27-12 .692 x 24-18 .571 +4.5 -7
Virginia 47-34 .580 21-21 .500 26-13 .607 -6.5 -2
Waukesha 48-33 .593 27-15 .643 21-18 .538 +4.5 +1
West
Oakland 45-36 .556 22-17 .564 23-19 .548 +0.5 +4
x- League
high; league low: West Bend, –9.5
y- League
high; league low: Taggart, Silver, -9
To
illustrate how the game difference works, let’s look at my own team’s splits:
Lafontaine Park W L pct. Diff.
Road 22 17 .564 ---
Home 23 19 .556 0.5
This
means that Lafontaine Park performs one half game worse at home than it does on
the road.
From
the data in Table 1 we see that:
Those are among the
performances that deviate from the norm. So the question is, what should everybody’s
record be if “everything else was equal”, that is, if all records were
normalized to the league average? Let’s see.
The
effect of random results on a team’s overall record is reflected in the BPI. As
mentioned above, this indicator shows how
many games over .500 a team should be if random luck was taken out of the
equation. Let’s look at the table below (team presented in order of BPI,
the calculation of which is explained in details right below Table 2)).
Games H/R
game diff. minus One-run
BPI over.
500 league diff. games
/ 2
Virginia 13 -7.67 -1 21.67
Toontown 21 + 3.33 -3.5 21.17
Maryland 21 +1.33 -0.5 19.72
Aspen 15 -3.67 -0.5 19.17
Syracuse 23 + 3.33 +3 16.67
Waukesha 15 +3.33 +0.5 11.17
Lafontaine
Park
9 -1.67 +1.5
9.17
West
Oakland 9 +0.67 +2 6.33
Abilene 7 +1.33
+0.5 5.17
Annadale
1 -1.67 -0.5 3.17
Plum
Island 9 +9.33 0 -0.33
Greenville 11 +5.33 +6.5 -0.83
Buckeye 3 +3.33 +4 -4.33
Let’s
illustrate in details how we arrive at the BPI number using Aspen as an
example.
So
far, the Rainmen are 15 games over .500, while performing below the norm at
home (-2.5) and in one-run games (-1). What we would like to know is, what
would Aspen’s record be if the team had performed within the league’s norm? The
idea here is that the two aforementioned factors (H\R splits, one-run game
differential) should be credited to the team’s overall record in order to reflect
the record it should have everything else being equal. So Aspen gets a 3.67
games credit for H/R splits (which is Aspen –2.5 H/R differentia minus the
league H/R differential 1.17, i.e. (-2.5) – (1.17). In other words, Aspen has
performed 3.67 games below the norm at home. Crediting this number to the team puts
them at the league average in H/R splits; they also get a 0.5 games credit for
the one-run game differential, that is, 1 divided in 2; this makes Aspen a .500
team in one-run games. The calculation comes to an overall credit of 4.17.
Added to the 15 games over .500, the total comes to 19.17. This represents how
many games over .500 Aspen should be if random luck was factored out of the
first half results. With this, the Rainmen’s record is now normalized.
Actual record projected normalized
record
W L Adj. W Adj. L
Syracuse 52 29 49 32
Toontown 51 30 51 30
Maryland 51 30 50 31
Aspen 48 33 50 31
Virginia 47 34 51 30
West
Oakland 45 36 44 37
Greenville 46 35 40 41
Lafontaine
Park 45 36 45 36
Abilene 44 37 43 38
Plum
Island 45 36 40 41
Buckeye 42 39 38 43
Annadale 41 40 42 39
Now,
it would be interesting to see what the full effect of random luck would be
over a full season. In other words, he next question is, what would each team’s
record at the end of the season if be if the second if the second half played
out like the first half? To see this, we need to project the normalized records
over 162 and compare the results to 162-game projections of the teams’ actual
records. Here’s what we find (rounded to the next decimal):
Actual normalized normalized
increase
Virginia 94 102 8%
Aspen 96 100 4%
Annadale 82 84 2%
Toontown 102 102 0%
Lafontaine
Park 90 90 0%
Maryland 102 100 -2%
West
Oakland 90 88 -2%
Abilene 88 86 -2%
Syracuse 104 98 -6%
Buckeye 84 76 -10%
Plum
Island 90 80 -11%
Greenville 92 80 -13%
THIS IS IMPORTANT: The number of projected
normalized wins is not a prediction of the actual number of wins
a team should have at the end of the year. Rather, it’s a way to measure the
likelihood of a team to improve or not in the second half, and to identify
those teams for which it could be more difficult to maintain their current
level of performance. To be clear once again, I’m not saying, for instance,
that Buckeye should win only 76 games overall. Luck or not, what was gained was
gained. I’m only saying that stripping random luck off Buckeye’s first half
won-lost record, and projecting the results over 162 games, we come to 76 wins.
Comparing this figure with Buckeye’s projected actual record gives us an
indication on whether we should expect the team to improve or not during the
second half.
The normalized increase column is the important one
in Table 4. Assuming that things eventually even out, the larger the difference
between actual and normalized record, the more likely a team is to experience a
change in performance. According to this calculation, there is one team that
sticks out as the one most likely to improve in the second half: Virginia,
followed by Aspen and, marginally, Annadale. All three teams have been unlucky
on both fronts. They have a worse record on the road than at home, and have
lost more one-run games than they have won. All three teams also play for over
.500 on the road, which is a good sign for future performance. What sets
Virginia apart as the team most likely to improve is the highest negative
(positive for the Patriots) H\R splits of the current .500 teams. If this
somewhat corrects itself in the second half, we’re in trouble (by “we” I mean
my fellow National Leaguers and I).
At the other end of the spectrum, the
calculation shows that it will very difficult for Greenville, Plum Island, and
Buckeye to improve on their first half records of 46, 42, and 45 wins
respectively. While both Greenville and Buckeye have over performed both at
home and in one-run games, Plum Island comes even in the latter. However, the
Greenheads have the largest H/R split differential in the league. More importantly,
in spite of the fact that they have the highest winning percentage at home in
the league, they are playing under .500 on the road by a healthy margin (for a
contending team - they’re the only one of the 6 teams to have a .600+ record at
home not to play for .500 on the road). Unless there’s dramatic improvement on
the road, it looks like the ‘Heads would have to play at a .700 clip at home
for the rest of the year in order to improve on their overall record in the
second half. As for Greenville, they perform a little better than the norm at
home, but they at least play for over .500 on the road, which bodes well.
However, if those 13 games over .500 in one-run games revert to a more normal
differential in the second half, improvement could be difficult. Buckeye is a
particular case. They have better H/R splits than the norm and have won 8 more
one-run games than they have lost, so there’s a possibility that these two
factors revert back to the norm. In addition, Trader Ray went into a trading
frenzy last month (which was a lot of fun to watch, by the way), therefore
altering his June roster significantly, the effect of which I have absolutely
no idea about.
Generally speaking, then, we may say that
improvement in the second half is more plausible for those teams whose
normalized increase is at or above 0%. The further from this neutral cut-off
point, the more likely a change in performance is, in one direction or the other.
One
thing to keep in mind is that random luck here is only defined in terms of H/R
splits and one-run games. This does not include the overall random rolls of the
dice, although some of it is built into the H/R and one-run game splits. More
precisely it does not include the possibility that some team may have been
lucky or unlucky the whole time both at home and on the road (I hope
it’s clear at this point that good or bad luck only one way, e.g. bad luck only
the road or good luck only at home, is included in the normalization). I have
no idea how to do this, nor do I know if it’s in fact possible; or if it happens
at all. But if you ran simulations of a league’s schedule, say, five times, chances
are that you’d get significantly different results from time to time for one
given team. What I do know is, the more data we have, the more likely that this
factor is to be neutralized. With only three months’ worth of statistics, I’m
not sure we can dismiss it.
Finally,
the projections do not take into consideration the possibility that some teams
improve the rest of the way due to trades, adjusted strategies, well-planned
player usage, and a more favorable schedule.
All of these are factors that may alter the overall picture in Table 4.
Now,
having said all that, although the calculation has some base in reality, this
is all speculation, numbers, probabilities, and possibilities. It doesn’t mean
the rest of the season will unravel as described in Tale 4. That’s because the methodology
used to calculate the likelihood for teams to improve or regress supposes that
random luck eventually evens out during the course of a season. Well, it
doesn’t work that way. Things don’t always even out. And that’s a good thing. After
all, that’s why we play the games.
So,
who’s going to defy the odds?
What’s in the cards for the Diamonds?
OK.
Time to be self-centered now. Let’s look at the one team that I know best, the
Lafontaine Park Diamonds, to get an idea of how things could develop in the second
half, and how they could be different from the first half.
The
team had a solid April at 16-10 (8-5 both at home and on the road) but a
disastrous May at 13-16 (7-7 at home, 6-9 on the road), getting beaten in the
process by scores of 16-9, 10-3, 13-0, and 10-1 (the latter two at home, to
boot)!. Needless to say, the team’s ERA took a beating. The highlight of the
month was the four losses in relief by usually reliable Jamie Walker. But the
Diamonds bounced back in June with a 16-10 record (7-6 at home, 9-4 on the
road).
Overall,
the picture looks reasonably bright for a number of reasons.
First,
the Diamonds have completely neutral splits, i.e. their projected record is the
same as their projected normalized record. This means that luck, at least as defined
in this article, did not play a major part in the team’s record. The team plays
for over .500 on the road; it is fifth in runs scored, third in OBP, sixth in
SLG., and in the middle of the pack in ERA (14th), and aside from
the corner outfield positions the team’s defense is generally solid. The
Diamonds have stolen only 22 bases, but they’ve also been caught only 3 times,
the best success rate in the league, whatever that means; the roster boasts a
collection of regulars who can either play most every day or take their regular
turns in the rotation the rest of the way, and the bench and the platoons are
solid. Finally, the team recently made two much needed acquisitions to bolster
a middling pitching staff: Rodrigo Lopez and Braden Looper (although Lopez was
a disaster in May, posting a 7.43 ERA; he bounced back in June with a 2.84 ERA),
and two solid utility players- and much needed left-handed bats- in Geoff Blum
and Damon Minor.
This
all sounds very nice. But there’s a catch: The second half schedule. The next
two months will be brutal for LPD. The team’s opponents? Among others:
July:
Virginia (8 times), Aspen (3), Waukesha (3), West Oakland (3).
August:
Abilene (8 times), West Oakland (6), Maryland (6), Virginia (6), Aspen (3).
Ouch.