Re: Consistency Score
- --- In APBR_analysis@yahoogroups.com, <igorkupfer@r...> wrote:
> But the more I think about it, the whole area lies uncharted. Whatdoes it mean for a
> player do be "consistent"? And if we agree on a definition, wouldthat mean that
> consistency is a valuable trait in a player? Is this evenmeasurable?
>I think "consistent" has been used from a style rather than
performance perspective historically. Stockton, Malone, Jordan have
styles that don't change and are still hard to beat. They are
consistent. Other guys may be consistent in that you know what to
expect from them every night -- they don't try to be MJ one night and
Reggie Miller the next. I actually think we can define it in terms
of performance however we see fit and feel our way to a definition
that describes roughly what we see. It won't ever be perfect and may
not be useful (though I'm thinking that it might be), but it could be
a convenient descriptor.
Is consistency valuable to measure? Maybe. It is useful on a team
scale. Consistently good teams are champs. Inconsistently bad teams
are teams that can upset good teams. You want teams that can perform
with different styles if you're not a great team -- in order to upset
consistently good teams. But I'm not sure that means that you want
inconsistent players. I think you still want consistent players,
just emphasizing roles differently to be able to play as a team with
a different style. Consistency allows a coach to strategize better.
> In order to not get bogged down with these legitimate questions,I'll just ignore them
> for now, and focus on a better measure (see below). Perhaps after Iget something good,
> I can test it to see if consistency is indeed an important asset.I am not sure if there is an easy "test" of its importance. I'll be
>love the simplicity.
> Heh. Does anyone else use the credits? I often use them because I
> But most of the measures I devise are completely neutral withregards to the rating
> being used -- credits works fine, but TENDEX and Individual Winsworks as well.
>If I use linear weights, I use credits because they are easy. Tendex
is the same but has a pace factor in it.
> So the problem then becomes: how do you "isolate" consistency,without rewarding
> players who contribute too little to vary much from their averageperformances? This is
> my second pass at a solution:placed on the mean, and
> CS = 1 / (variance / mean) * 100
> By using the square of the standard deviation, less weight is
This formula seemed to produce better results, "better" from my
perspective at least. It's interesting, too, because I've had a
couple occasions to use variance when standard deviation at first
glance seems better. Your CS now has units (is not dimensionless),
which can be troublesome, but I don't see why right now.
> the CS is concentrated more on the actual variation. To show howConsistency Scores
> attempts to capture the variation, I'll use the hypotheticalexample of 3 players, each
> playing 5 games:consistent -- and
> Games #
> Plyr 1 2 3 4 5 mean sdv var CS
> 1 | 30 12 6 36 6 18 14.1 198.0 9.1
> 2 | 14 15 16 20 25 18 4.5 20.5 87.8
> 3 | 9 10 11 15 20 13 4.5 20.5 63.4
> 4 | 4 5 6 10 15 8 4.5 20.5 39.0
> Players 1 and 2 have identical means, but the first is clearly less
> the CS reflects that. Players 2 and 3 have identical variances, butdifferent means.
> Under my formula, player 2 is rewarded for having the higher mean --but the difference
> is not too severe: 87.8 vs 63.4. Player 4 also has the samevariance as player 2, but
> has a much lower mean. He is penalized more severely.consistency is --
> I think this version of my formula captures more of what I think
> certainly better than my first try.I think so, too.