new statistical methodology to rank players

[INKtastic]

I thought this was a very interesting way to rate players contribution to their teams.

http://hardwoodparoxysm.blogspot.com/20 ... -game.html

I'm curious what some of the other stat guys on here think of this.

[theTHIEF]

weird...

lots of reading to do...

[ropjhk]

Looks interesting, but I'm gonna have to find some other time to read the whole thing.

And for those who probably wondered the same thing after seeing who created the thread:

Lebron James is not ranked #1.

[Volcano]

Can you form a line to beat the top five players?

Dwight
Amare
Lebron
Kobe
CP3

[Bgil]

So you'd rank Amare over Duncan, Dirk, KG, and Yao? And Paul over Wade?

[kno]

[quote="Bgil"]So you'd rank Amare over Duncan, Dirk, KG, and Yao? And Paul over Wade?[/quote

Statistically (considering all of Points, Threes Made, Rebounds, Assists, Steals, Blocks, Turnovers, FT%, and FG%), Amare and Paul were the best players last season. Yes, better than Wade, Duncan, and KG.

[UDRIH14]

theres a difference between a regular season performer and a playoff performer

[NO-KG-AI]

This is RealGM. We don't like any stat that doesn't rank our favorite player where we would like.

But since Chris Paul ranks pretty high, I like this stat a lot, it's far more valuable than.... other stats that don't rank him high....

[Duiz]

This is RealGM. We don't like any stat that doesn't rank our favorite player where we would like.

But since Chris Paul ranks pretty high, I like this stat a lot, it's far more valuable than.... other stats that don't rank him high....

Ew... I didn't expect you to be a CP3 homer more than KG homer.

Well, I don't like the stat because Ronnie Price doesn't rank higher than LeBron James. I feel it is innacurate.

[/thread]
[/RealGM]
[/World]

[maggi mee]

This method has the same flaw as any box-score type ranking - it doesn't do a good job of doling out credit between team-mates, for a given game.
As with PER, win score, etc, it starts off with a box-score stat (MEV) for each player and aggregates into a team value (Total MEV). The innovation over pure box-type stats is that (to a first approximation) MPV takes into account winning/losing margins when weighting game-by-game MEV, so players get differing amounts of credit for each game, depending on margin - and get credit even when the team loses.
Assuming that the econometrics is sound and dealing only with the concept behind the methodology, this is a marginal improvement over straight-up box stats. The problem is that it still doesnt deal with the elephant in the room, which is that MEV, as a box-type stat, isn't a good way to approportion credit between players within a team. As a result, if a player gets an outsize amount of credit for the wins (because MEV is biased), his MPV is overly high when comparing with players (both in his own team, and from other teams.)
If there was a better way to approportion credit between teammates (eg some adjusted form of +/- could be used here) in the first stage, the idea would be somewhat more useful. However, even then, I regard this as no more than a marginal innovation. The first stage of the methodology is a standard regression of box stats. The second stage is essentially aimed at improving the correlation between a team aggregate of box stats and "win contribution". However, we already have very good correlation between team wins and avg winning margin - so we wouldn't lose much by using average winning margin as a proxy for "win contribution", as is already done by existing measures.

[Flash3]

How do we know this doesn't have the same "flaws" that people mention about PER, TS% etc?

I mean, every statistic can be used to one's advantage or disadvantage.

[cb4_89]

so andre miller, jose calderon, iggy and jamison are better than chris bosh/carmeloyao/tmac?

Yeah fail