[more synopsis -- Eric M.Van posted some work in which he noticed a
correlation between a teams manager and it's $H (relative to the
league's $H?) he suggested that $H was influenced by
    park effects
    managerial tendencies -- positioning of fielders pitching around
        batters, etc.
    team defense -- how talented are your fielders
    luck -- probably still the biggest component
    pitcher skill -- probably the smallest component
If this sounds really obvious, it's because it is, except for the notion
that pitcher skill has minimal influence on the %age of batted fair
balls that fall for hits.
]

: 1) Much of the reason why $H varies so much from year to year, where
: K% and BB% don't, is *because the sample size is smaller*.  Too small.
: Note that HR%, which has a very similar sample size, also has a large
: variance from year to year.   The additional lack of correlation in $H
: can be attributed to larger team defense and park effects.

Why stop there? How do we know that, say, an entire *team's* $H aren't
subject to noise in the "league-wide" $H? IOW, isn't it possible luck
becomes a large influence on team $H, so large that we can't reliably
detect team defense on batted fair balls (especially considering park
effects)?

:
: It's trivial to simulate multiple seasons of $H in Microsoft Excel for
: a pitcher with any theorized innate level of $H.  The results are
: eye-opening.

I'm not sure what this [what I did] shows, but I used this method to
simulate a *full league* of $H, and there's still a good bit of variance
(I hope I'm using that word right).

For the 2000 season, the observed NL $H was .28750. I then plugged this
"innate" $H into 4500 cells, and ran 16 trials.

On the left are the actual $H numbers from 2000; on the right are
simulated $H numbers with a league wide $H of .28750
CIN .27502  .27333
STL .27773  .28222
LOS .27860  .28222
SDG .28009  .28222
MIL .28261  .28311
NYM .28303  .28667
ATL .28357  .28733
PHI .28516  .28800
CHC .28739  .28933
SFO .28860  .29133
ARI .28958  .29244
FLA .29047  .29333
COL .29514  .29444
HOU .29803  .29511
PIT .30126  .29511
MON .30351  .29822

[apparently, my random number generator blows -- .28222 three times!?!]

I'm not at all sure what this proves, but it seems plausible that a good
bit of the impact on *team* $H (not just individual pitcher $H) is
simply luck.

[there are many interesting things about these numbers, but I'm hesitant
to draw conculsions from the fact that the expos give up 5% more hits on
batted fair balls than league average, and the reds give up 3% less]

I just realized that I haven't looked at one of the best pieces of
information we have on the subject: correlation between team $H in year
N and team $H in year N+1 (!). Of course, I don't have this data on me,
nor as much of a grasp on correlations, regressions and spreadsheets as
others who read r.s.bb or r.s.bb.f :-). Does anyone have any data to
support/refute the idea that "team $H has strong correlation from year
to year"?

=======
The rotisserie implications of "team $H is not a reliable indicator of
team $H in the coming year" (which might not be true!)
====

In the thread "Rotisserie implications of $H", Voros McCracken and Eric
M.Van both suggest that you can help yourself out by looking for
pitchers who have high $Hs relative to their team $H. The idea is that
these pitchers are likely to see a decrease in their $H (and therefore
WHIP & probably ERA) in the coming year. It doesn't mean that they're
getting "better" necessarily, it means they're getting "luckier".
Inversely, you should avoid pitchers who have low $Hs relative to their
team's. You're not going to fool anybody on Greg Maddux here, but it
might help you make the right pick (or, more likely, help you avoid making
the wrong pick) in later rounds of the draft/auction.

This is getting abstract, so lets pick a concrete example:

         HInP   InP     $H
Rusch     178    587   .30324
NYM      1234   4360   .28303

Glendon Rusch quietly put together a pretty good year last year. He was
27th in WHIP in an ML universe, and 52nd overall. Of course, Rusch is
not a "name brand" pitcher (read: Proven Veteran (TM)) yet, so some
rotisserie managers will look at last year and say "he got lucky". To
the contrary, Rusch was *unlucky* and still turned out to be an
effective starter (there may be some bias in here, because for a while,
IIRC, BobbyV was using Rusch as the 5th starter, and he may have more
starts against low-slugging chump teams, thus lowering his ERA).

If he had gotten average luck, then he would have given up 10-12 fewer
hits, enough for about a 0.20 era difference (and a .04 WHIP
difference). Of course, nothing says that Mr. Rusch will have average
luck next year -- he might get shelled in his first few starts and be
sent to the pen -- but odds are that he will have better luck next year
(thought it might not be much better).

Let's put Rusch and the Mets up against the NL as a whole.
         HInP   InP     $H
Rusch     178    587   .30324
NYM      1234   4360   .28303
NL      20624  71736   .28750

Rusch, it would seem, had bad luck even by league wide standards. So
he's probably a good bet to see his luck improve; thus he's likely to be
morevaluable in the coming year than in later ones.

Let's take another example: Andy Benes.

         HInP   InP     $H
AnBenes   144    504   .28571
STL      1207   4346   .27773
NL      20624  71736   .28750

Hmmmm .... Alan Benes's brother got bad luck relative to his team, but
slightly good luck relative to the league. If the more important factor
is team defense, then there there are slight odds that Mr. Benes will
have better luck next year. If the more important factor is league
offense (really the only way to interpret the NL $H in this context),
then you're basically flipping a coin -- he might get better luck, he
might get worse, and a priori there's no way to figure out which is more
likely (if you can, go tell Tony LaRussa).

Let's try one more: Jeff C. D'Amico. I'll just put $H here:

DAmico  .25344
MIL     .28261
NL      .28750

This one's a relative no-brainer. Mr. D'Amico is unlikely to have luck
this bad again. If he had gotten average luck for a Brewer, he would
have had 8-11 hits more, which likley bumps his ERA by about .3 (he
pitched fewer innings than Rusch), and gives him a .06 increase in WHIP.
If he had gotten NL average luck, then he would have looked even worse.
D'Amico is more likely to have worse luck than better luck.

The question remains: which metric is more important? The team $H, which
can somehow measure team defense/park effects, or the league $H, which
measure league wide "average on balls in play"? Or the pitchers career
$H (there seems to be little evidence for this onea unfortunately)?

Allright, I'm going to bed now. I'll figure out a better way to look at
this ... later.

Cheers,
Nick

PS Many thanks to Voros for suggestiong Glendon Rusch and Jeff D'Amico
as over/underrated candidates. Saved me lots of hunting!

Thanks also to Voros and Eric for this discussion; it's really neat.

--
ni-q
.
bomb president allah marx encryption revolution Pat Buchanan unabomber occult

In any event the correlation is greater though it's tough to say what the
reason for it is.

As roto implications go, the point I keep trying to make is that we're
really just trying to assess pitcher ability and regardless of what stats
we use, we're going to run into some level of error and most likely a
greater level than we do for hitters.

The point I'm making is that even if we accurately assess that one
pitcher's ability level is .003 higher than another, that information is
pretty useless in anything but the longest of runs. The stat is
constructed as such to where such a difference may not even manifest
itself over a period of eight years much less one.

The key is not to assume you have a level of certainty on hits that in
actuality we don't have. The point is not that Glendon Rusch's $H total is
going to go down, it's that we really don't know what it's going to be and
so the best guess is somewhere around average (or maybe team average or
halfway between or whatever).

In any event, multiple seasons worth of data are important for pitchers
especially in the areas of Home Runs and (if you choose to count them
anyway) hits.

I wouldn't mark Glendon Rusch down as necessarily being better (though
that's certainly a very possible outcome) as by eyeballing things it looks
like his Home Run rate was a bit lower than his previous established
level.

In any event I think he should be at worst a slightly above average
pitcher and at best a very good one.

--
Voros McCracken
vo...@daruma.co.jp
http://www.baseballstuff.com/mccracken/

This sort of statistical analysis has the noise filter built in.  The most
certain result in the whole output is Fenway Park at an average +0.87 standard
deviations per year, which is typically 40 hits.  In general, the results follow
common perceptions of park effects remarkably well.

". . . from that day forward she lived happily ever after.  Except for the dying
at the end.  And the heartbreak in between." - Lucius Shepard.

If nothing else, a larger sample size?

: > The question remains: which metric is more important? The team $H, which
: > can somehow measure team defense/park effects, or the league $H, which
: > measure league wide "average on balls in play"? Or the pitchers career
: > $H (there seems to be little evidence for this onea unfortunately)?
:
: Well the pitcher's career $H presents a problem for a number of
: reasons: One, many pitchers haven't pitched for very long, so it's useless
: for them. Two, those that have pitched a while aren't showing very big
: differences between each other in the stat. Three, the cause of those
: differences is a topic for much debate.

Four: the pitcher is getting older, traded, having the defense aroun
him change, etc.

:
: As roto implications go, the point I keep trying to make is that we're
: really just trying to assess pitcher ability and regardless of what stats
: we use, we're going to run into some level of error and most likely a
: greater level than we do for hitters.
:
: The point I'm making is that even if we accurately assess that one
: pitcher's ability level is .003 higher than another, that information is
: pretty useless in anything but the longest of runs. The stat is
: constructed as such to where such a difference may not even manifest
: itself over a period of eight years much less one.
:
: The key is not to assume you have a level of certainty on hits that in
: actuality we don't have.

I hope I didn't sound like I had any certainty

: The point is not that Glendon Rusch's $H total is
: going to go down, it's that we really don't know what it's going to be and
: so the best guess is somewhere around average (or maybe team average or
: halfway between or whatever).

Yes. That's really all I'm trying to say. It's not that Rusch *will*
come out of next season with better roto stats (I'm talking mostly about
ERA and WHIP of course), but rather that he's *likely* to come out with
better roto stats (if we assume his walk and HR rate will stay roughly
constant).

: In any event, multiple seasons worth of data are important for pitchers
: especially in the areas of Home Runs and (if you choose to count them
: anyway) hits.
:
: I wouldn't mark Glendon Rusch down as necessarily being better (though
: that's certainly a very possible outcome) as by eyeballing things it looks
: like his Home Run rate was a bit lower than his previous established
: level.

Right: if HR rate is subject to this sort of sampling error as well,
then you should take any improvement in it w/ a grain of salt.

Cheers,
Nick

--
ni-q
.
bomb president allah marx encryption revolution Pat Buchanan unabomber occult