Re: NBA stats site
- Great work, Roland. Confirms a lot of my thoughts about +/-, both
good and bad.
--- In APBR_analysis@yahoogroups.com, "roland_beech" <roland@t...>
> Well, as a newcomer to this group I've been enjoying readingthrough
> some of the archives. There's definitely an impressive collective
> body of knowledge here.
> I've already talked with Bob some on his work, and Dean and John I
> guess have books coming out which I will track down.
> From my end though, I just launched a site with all kinds of funky
> NBA stats including plus/minus number comparisons for players on
> court versus off court, specific 5-man units and how they perform
> together, clutch play stats, etc.
> I would be interested in getting the feedback of the sharp minds
> here on some of the numbers, recognizing it's still early and the
> site will be evolving rapidly with the onset of a new season
> The site is 82games.com
> or http://www.82games.com if you want the full URL
- --- In APBR_analysis@yahoogroups.com, bchaikin@a... wrote:
> nice work..... can you show your source for the data you have used
> divulging any proprietary work?...system?....
> how is this sytem any different than the winston/sagarin WINVAL
Yes, very nice work. The system looks pretty simple actually, not
trying to do what WINVAL does (and does not quite right). Basically,
he's got what lineups are on the floor all the time and adds things
up. Very good pieces of data. And it is data even though
the "Roland Ratings" implies something more, it is mainly data, which
is GOOD. The fact that Greg Ostertag ends up looking good doesn't
mean that the system is wrong. It means that Utah played a lot
better when he was on the floor. Other guys may have been on the
floor at the same time, causing some confounding problems, but it's
just data, not a ThisGuyIsTheGreatest thing. Ostertag is not the
best or even 2nd best center, as we'll see this year, but he
complemented Stockton and Malone fairly well apparently (and his
backup did not). In fact, these ratings do reflect a lot on who was
backing someone up. Kidd had no decent PG backup in NJ, so he looks
extra special. If his backup was Bobby Jackson, it would lower his
Play-by-play data is going to spawn an entirely new wave of
analysis. This is that wave.
----- Original Message -----
From: "Dean Oliver" <deano@...>
Sent: Thursday, October 09, 2003 5:41 PM
Subject: [APBR_analysis] Re: NBA stats site
> Play-by-play data is going to spawn an entirely new wave of
> analysis. This is that wave.
Indeed. I'm already itching to try a player rating based on changing game
states, like baseball's Player Game Percentage.
Speaking of play by play, I wish I could see the attempts for the clutch
shooters. My quick study on Tim Duncan showed that he only had about 300
attempts in "clutch time" defined by Roland's page, not enough for
>One other thing I like about this is that it seems to support myThe shooting percentages definitely change as the shot clock winds
>long-standing but poorly supported statement about shooting with 3
>or fewer seconds on the clock being much poorer. I can't check it
>for sure with their posted numbers, but they can. Hint, hint.
down, as does the type of shot a team is likely to take. We're
doing some research on this to see if the better teams are primarily
better at executing in the later seconds on the clock, get to that
situation less often, or if everyone struggles with time running out
and it's a matter of the good teams being stronger in the earlier
shot clock segments...likewise the defensive numbers are interesting
>Perhaps the Pythag would be informative here. Using Tim Duncan as anagree completely that the 50-16 vs 5-11 is a nicer look than the
>OffPts DefPts Min Win% W-L
>OnCourt 97.0 88.9 3151 75.6% 50-16
>OffCourt 91.7 97.9 775 29.9% 5-11
>50-16 is a number people can easily understand, while points scored
>and allowed appeals only to us statheads :-)
others we show...of course it's a projected stat versus an actual
one, but we can certainly look to add this for future page updates
>BTW2: Roland, nba.com shows Duncan as playing 3181 minutes. Typo,We calculate playing time to the second and accumulate as opposed to
>or is there a rounding error? Not that a 1% error is a big deal.
the NBA which I believe rounds to the nearest minute on a game
basis. So, that explains part of it. I'm sure we will differ from
official stats in certain places
>The 5-man units provide enough information for now -- I could spendA good point too, and particularly in-season you might have in the
>hours looking at them! I would suggest showing the floor time as
>absolute minutes, as well as the percentage of total team minutes.
first month a unit with something like 2% of total minutes
translating into 5 actual minutes together...not a big sample!
>That data set could lead to a flood of ideas on what to do with it.We have an ever-expanding list of research ideas to pursue,
including such obscure things as: performance of players in
different foul states (eg playing with 5 fouls), does getting a shot
blocked change a player's subsequent shot selection, rookie play pre
and post-wall, where does home court advantage actually come from,
and in particular detailed analysis of five-man unit traits, how
these characteristics match up against opponent five-man units, etc
- Very interesting data on 82games.com. A couple of questions:
1) What is the source of the raw data from which the +/-
summaries were generated?
2) Apparently I am misunderstanding the +/- definition in the
tables. By my way of thinking when you take the +/- for a given
player when ON the court and multiply it by his fraction of playing
time and then add the +/- for the same player when OFF the court
and muliply it by his fraction of non-playing time, what you should
get is the team plus minus, no? In doing a loose check on Paul
Pierce, I find the implication that the '03-'03 Celtics were
outscored (which may be the case) but also that the team
implied figure based on Mr. Battie's +/- differed, which cannot be,
according to my conception of the stat. Please advise.