- I found this in an old newsgroup post:

"First, minutes/game for each player is fit to a linear

sum of that player's per-minute statistics. The resulting

weights indicate how valuable NBA coaches think various

statistical categories are, as expressed by their

willingness to give more minutes to players who excel

at them.

The Raw Minkoff Player Rating is determined by summing

the weighted per-minute statistics, thus determining how

many minutes/game each player "earned". In practice,

some bit-part players will seem to deserve an amazing

number of minutes (eg, a player who played 1 minute and

shot 1-1 from the field would "deserve" 105 minutes/game).

To combat this, one uses the Minkoff Player Rating

defined as (MPR) = (RMPR)**(2/3) * (min/gm)**(1/3).

That is, a player's final rating is two parts statistical

and one part empirical, based on how many minutes he

actually did play. MPR's below about 13 denote players

not of NBA quality, barring special circumstances

(e.g. Terry Cummings)."

--- In APBR_analysis@yahoogroups.com, "cwfrizzell" <cwfrizzell@y...>

wrote:> I'm sure some of you recall this. Other than some broken links and

a

> few minor references in the group archives, I'm at a loss. Can

> anybody expand on this or provide some actual analysis as to how it

> was constructed? - --- In APBR_analysis@yahoogroups.com, "dlirag" <dlirag@h...> wrote:
> I found this in an old newsgroup post:

I hadn't heard this last part. Actually, I remember Tony suggesting

>

> "First, minutes/game for each player is fit to a linear

> sum of that player's per-minute statistics. The resulting

> weights indicate how valuable NBA coaches think various

> statistical categories are, as expressed by their

> willingness to give more minutes to players who excel

> at them.

>

> The Raw Minkoff Player Rating is determined by summing

> the weighted per-minute statistics, thus determining how

> many minutes/game each player "earned". In practice,

> some bit-part players will seem to deserve an amazing

> number of minutes (eg, a player who played 1 minute and

> shot 1-1 from the field would "deserve" 105 minutes/game).

> To combat this, one uses the Minkoff Player Rating

> defined as (MPR) = (RMPR)**(2/3) * (min/gm)**(1/3).

> That is, a player's final rating is two parts statistical

> and one part empirical, based on how many minutes he

> actually did play. MPR's below about 13 denote players

> not of NBA quality, barring special circumstances

> (e.g. Terry Cummings)."

a straight linear average of RMPR and min/gm, which I never liked.

This weighting is interesting. The cutoff of 13 min/g hints at the

old question of replacement level. The reference to Cummings reminds

us that it is a pain to actually determine.

DeanO - -----Original Message-----

From: Dean Oliver [mailto:deano@...]

Sent: Tuesday, September 02, 2003 9:10 AM

--- In APBR_analysis@yahoogroups.com, "dlirag" <dlirag@h...> wrote:

[...]

>> To combat this, one uses the Minkoff Player Rating

>> defined as (MPR) = (RMPR)**(2/3) * (min/gm)**(1/3).

>

>I hadn't heard this last part. Actually, I remember Tony suggesting

>a straight linear average of RMPR and min/gm, which I never liked.

>This weighting is interesting. The cutoff of 13 min/g hints at the

Yeah, I'd forgotten about that too. It's a weighted geometric mean,

probably a very good idea, for dealing with those 1 minute played,

1-1 players Minkoff mentioned in his example. An arithmetic mean

would still show those players with high "value"; a geometric mean

cuts down the ratings of players with low actual minutes/game (or

low RMPR scores).

--MKT