## Re: The Problem with Possessions-Based Linear Weights

Expand Messages
• First of all, some good comments have continued while I ve been sleeping... I ll address the below, though. ... function of teammates.) ... Inordinate is a
Message 1 of 42 , Mar 26, 2004
First of all, some good comments have continued while I've been
sleeping... I'll address the below, though.

--- In APBR_analysis@yahoogroups.com, "dan_t_rosenbaum"
<rosenbaum@u...> wrote:
> --- In APBR_analysis@yahoogroups.com, "Dean Oliver" <deano@r...>
> wrote:
> Let's first hear a good argument that the unmeasured contributions
> are more (more what?) than the measured contributions. Everything
> rests on this statement. If unmeasured contributions are relatively
> unimportant (which I believe they are on the offensive side, not the
> defensive side), then you can use an approach like mine.
>
> DeanO
> -------------------------------------------------------------------
>
> What do we measure on the offenive end?
>
> 1. We measure which player touched the ball last on every field goal
> attempt and we measure the outcome of those field goal attempts.
> 2. On successful attempts, we sometimes measure the player that
> touched the ball second to last.
> 3. We measure personal fouls on a particular player when those
> personal fouls lead to free throws and we measure the outcome of
> those free throws.
> 4. On failed field goal attempts, we measure the player who regains
> possession of the ball.
> 5. And finally, when possession turns from one team to the other
> without a field goal or free throw attempt, we measure who is
> responsible for that "turnover" of possession.
>
> That is a lot and that is much better than what we measure on the
> defensive end. But what contributions to scoring or not scoring do
> we not measure?
>
> 1. We do not measure which players successfully navigate the ball to
> the frontcourt. (not hard to do at NBA level)
> 2. We do not measure which players initiate an offense with an
> effective non-assist pass. In fact, we fail to measure all of the
> non-assist passes that contribute to scoring (or non-scoring), such
> as all of the passes that lead to shooting fouls. (Easy)
> 3. We do not measure which players get themselves open in out of
> bounds situations. (Easy)
> 4. We do not measure screens on the ball or off the ball. (Easy)
> 5. We do not measure which players keep the floor spaced leading to
> fewer turnovers and higher percentage field goal attempts. It is
> pretty tough to have a successful field goal attempt when you are
> double teamed because of poor spacing. (Relatively easy, but a
function of teammates.)
> 6. We do not measure which players tend to hold onto the ball for an
> inordinate amount of time leading to forced shots or shot clock
> violations. (Everyone on a team contributes to time off the clock.
Inordinate is a judgment call.)
> 7. We do not measure which players correctly run plays and which
> ones do not.
> 8. We do not measure players failing to get open leading to a
> turnover for the player holding the ball.
> 9. We do not measure players with good hands grabbing an errant pass
> that would have been a turnover for the passer.
> 10. We do not measure the player who keeps a possession alive by
> tipping an offensive rebound to a teammate or by blocking out an
> effective defensive rebounder.
>

Most of these unmeasured things aren't that hard to accomplish (or to
avoid, if they're negative). I can go out and set picks. A lot of
these 10 unmeasured things are taken as givens. Guys know how to do
these things and, if they don't, they aren't as important as the
measured things. That's the conventional wisdom. Perhaps not right,
but I think there is a significant burden in showing that these
unmeasured factors are more important than the measured ones.

For instance, I think one way would be by showing that predictions of
player performance from season to season are much worse using measured
stats than using unmeasured stats. I know Dave Berri showed that 80%
of variance in stats is explained by previous stats. That is,
predicting measured stats is done well with measured stats. Is that
the right test to apply? I think so. It means that the unmeasured
ones aren't significantly impacting measured stats and you'd think
they would if they were so important in winning.

> statistics. All that I am in essence doing is arguing that we
> should focus more heavily on these correlations, because there are
> lots of contributions to successful or failed possessions that are
> not accounted for in our measured statistics.

I have no doubt that correlations do exist and that we should look for
them. They're just not high on my list ON THE OFFENSIVE SIDE.

>
> What is interesting is if you do the same exercise for baseball, the
> list of measured versus unmeasured contributions towards a run
> scored is much more heavily tilted towards the measured
> contributions.

Depends on how you make that list. It's ALWAYS easier to make a
longer list of unmeasured things than measured things. For baseball,
things that affect whether a run is being scored:

1. The signs flashed by the 3rd base coach.
2. Whether the man on first is running on the pitch or not.
3. Whether the man on first saw the signs.
4. How the fielders are positioned (now starting to get measured).
5. Whether the hitter has that black stuff under his eyes or not.
6. Whether the pitcher is in the sun and the hitter is in the shade.
7. How good the hitter is at reading speed of pitches.
8. How fast a hitter gets out of the batter's box.
9. Whether the hitter is swinging for the fences or for a base hit.

etc.

My point is that you can break down the games of baseball or
offenses are broken down pretty well by stats. What's left over are
small variations of strategy or training. Do they matter? Yes, but
do we miss a significant amount of value by not measuring them? I
don't think so.

Let me frame it one other way. From a team standpoint the value of
the four factors are

1. Shooting % (10)
2. Turnovers (6)
3. Offensive rebounding (5)
4. Getting to the line (3)

For individuals, we measure how well they shoot. We have one
measurement of a factor that does contribute to shooting better:
assists.

We have individual measurements of turnovers (thank god). We have
occasional cases where a player's turnover is due to the fault of a
teammate, but individual turnover rates really don't have large
variations from season to season.

We measure offensive rebounds for individuals. Things that might
contribute to an individual's offensive rebound rate are, well,
primarily whether that is his designated responsibility, which we kind
of represent through position. On an individual play, a lot of
factors matter, but in the long haul, it's a role.

We measure getting to the line. Double-teams created by other players
can change this factor somewhat, but I wouldn't see that as big.

On the defensive side, we have no individual measurement of shooting
percentage allowed (the single biggest flaw in basketball stats, as
far as I'm concerned). We have only a partial representation of
turnovers created (through assists). We do the best with rebounding,
tracking defensive rebounds. We measure personal fouls, but we lump
offensive fouls in with shooting fouls and non-shooting fouls.

DeanO

Dean Oliver
"Basketball On Paper aims to lay out new measures for examining team
and player performance and must be considered the finest effort in
this regard yet seen for basketball." 82games.com's Roland Beech
• I think wimpds has a point here. If a particular player with limited offensive skills tends to play at times when his teammates shoot more poorly (because they
Message 42 of 42 , Mar 30, 2004
I think wimpds has a point here.

If a particular player with limited offensive skills tends to play
at times when his teammates shoot more poorly (because they are in
essence playing 4 on 5) than their WINVAL ratings suggest they
should, that player would end up with more offensive rebounding
opportunities than a typical player.

But let's suppose a guy is on the floor for 80 possessions in 40
minutes and he drops his team's shooting percentage down 5
percentage points (a pretty big drop). I would assume that would
result in about three to five extra offensive rebounding
opportunities for this player and if he collects 10 percent of those
extra rebounds, this amounts to 0.3 to 0.5 extra offensive rebounds
per 40 minutes. Not trivial, but also not a huge difference.

--- In APBR_analysis@yahoogroups.com, "wimpds" <wimpds@y...> wrote:
> But if teammates are more likely to miss shots when they play with
you
> than when they play with other players, that will depress your
WinVal
> rating, right? If this attribute is correlated with individual
> offensive rebounding, won't it have a negative effect on the value
of
> the coefficent?
>
> On the defensive side, we only have a few variables to contribute
> significantly to your WinVal rating. [I'd be interested in seeing
> regressions run separately on the offensive and defensive WinVal
> ratings.] Defensive Rebounds coefficient might be picking up
> something more than simply rebounding ability, it could be getting
at
> defense somewhat more generally.
>
>
>
> --- In APBR_analysis@yahoogroups.com, "Michael Tamada"
> wrote:
> > It's a good thought, although the regression should already
> > account for this by already measuring the impact of
> > opponents missing shots and one's team missing shots.
> >
> >
> > --MKT
> >
> >
> > -----Original Message-----
> > From: wimpds [mailto:wimpds@y...]
> > Sent: Monday, March 29, 2004 3:13 PM
> > To: APBR_analysis@yahoogroups.com
> > Subject: [APBR_analysis] Re: The Problem with Possessions-Based
Linear
> > Weights
> >
> >
> > Another thought on this subject ( though I may not be thinking
through
> > the regression carefully enough) is that offensive rebounds are
> > probably correlated with having teammates missing shots while
> > defensive rebounds are correlated with the opponent missing
shots.
> > This effect would seem to be even stronger if your teammates are
more
> > likely to miss because you're not attracting any defensive
attention.
> >
> >
> >
> >
> >
> > --- In APBR_analysis@yahoogroups.com, "carlos12155"
> > <carlosmanuel@b...> wrote:
> > > --- In APBR_analysis@yahoogroups.com, "Mike G" <msg_53@h...>
wrote:
> > > > --- In APBR_analysis@yahoogroups.com, "Kevin Pelton"
> > > > <kpelton08@h...> wrote:
> > > > >
> > > > > Rodman specialized in rebounding -- not offensive
rebounding. He
> > > > was
> > > > > great at both ends. I don't know if I can think of someone
who
> > > > > specializes specifically in offensive rebounding.
> > > >
> > > > The original poster meant, I'm sure, that on Offense, Rodman
did
> > > > little more than rebound missed shots, at least by the time
in his
> > > > career when he was with the Bulls.
> > > >
> > > > Obviously, he went to the other end of the court on defense,
and
> did
> > > > other things there.
> > > >
> > >
> > > That was exactly my intention. It seemed to me that if
offensive
> > > rebounds are at some level a negative, Rodman performance
would be the
> > > perfect example; after all that was the only thing he did on
offense.
> >
> >
> >
> >