kevin's weighting system
- --- In APBR_analysis@yahoogroups.com, "Kevin Pelton" <kpelton08@h...>
> > Using linear weights generated from team statistics, I suspectthe article describes a different approach (though i'd actually wish
> > that a player such as Iverson would get a relatively low rating,
> > even though they may be quite valuable to their team.
> I've tried to account for problems like this by putting individual
> players' statistics in a team context -- an imaginary team of them
> and four average teammates. For this specific issue, I make an
> artificial adjustment to teammates' efficiency based on percentage
> of possessions used. That seems to work fairly well, though there
> are still major problems with the system overall.
> That link explains it in painful detail, if anyone has interest.
there was more detail to explain why you feel (if you do) that the
results of your formulas more closely fit reality than linear weight
systems. but certainly the new outcome measures are interesting and
useful conceptual twists.
i read the rebounding stuff in particular and understand what you are
saying about roles and other teammates. but perhaps instead of
adding 4 average teammates wouldn't be cleaner to just compare
individual player performance to league average player at that
position? isnt that the real intent?
one of my reactions was that your approach seems to blur and
somewhat minimize rebounding, i.e. the other 4 average players will
do 70-90% of it. and the results of adding in the 4 average players
will be the stat number yielded will be far more heavily composed by
what they do rather than the player in question. it can also end up
rewarding those that rebound beyond the norm of the position (say a
guard who pulls done 5 or 6) more than those that might rebound
higher but are in positions that are supposed to rebound higher (say
a power forward that "only" pulls down 7 or 8) . but isnt a rebound a
your position comparative approach appears to do what you do with
rebounding for all stats which is consistent and appropriate and
produces a useful if somewhat different analytical product. maybe it
is a compliiment to, not a replacement of indivdual player data only
ranking systems. with about 80% (give or take) of the data weight
in that product not being from that player so it would seem to mute
the range between top and bottom or at least require looking at top
and bottom level stat outcomes differently. small differences
(around 1/5th the size of differences in individual player data only
rating systems?) in performance ratings would mean big differences in
performance and comparing across positions would end up mixing the
stat value of those two players with the comparative stat value of
the two positions involved (pg vs sf, etc). i dont see how this
particularly helps a lot. it does allow comparisons of player
performance compared to the norm for their position but it is not
necessary to introduce the 4 average players to do that.
leaving out the theoretical 4 other players out and just using
individual data seems easier and closer to reality to me. you can do
that with your formulas just fine and your weights are your weights
and can be compared and debated just the same as any linear weight
system. rewarding the individual rebounder fully still seems fair
me even if it produces a big man heavy in top rankings end result.