APBRmetrics

[Jon Nichols]

At Basketball-Statistics.com, I've published plus-minus data for the NCAA Division I season thus far. It is net plus-minus. The official play-by-play data has a lot of errors, though, so some of the numbers may be off. I've tried to account for these errors as much as possible, but some still slipped through. To see the data, go to:

http://basketball-statistics.com/blog1/2010/01/01/ncaa-plus-minus-01-01-10/

[Joe]

It would be helpful if you made minutes and total +/- columns (as opposed to just per game).

Awesome stuff, though. Please update this throughout the season, even if intermittently.
_________________
http://www.hoopdata.com

[Crow]

Jon, you say

"By net plus-minus, I mean that the team’s performance with that player on the floor is compared to the team’s performance overall."


So a player who plays 32 minutes gets compared on a 40 minute basis to 32 minutes of the team with him and 8 minutes without?

That ends up being 8 minutes with compared to 8 without though right?

Why not just say on/off then?

And if it is just the 8 minutes on vs. off do we want that "per game" impact or the per 40 minutes version?

I could use some clarification of what you've presented.

[Jon Nichols]

[Jon Nichols]

[Jon Nichols]

[Crow]

Team’s performance with that player on the floor compared to the team’s performance overall would be
(3* 100/50) - (100/50+100/100)=
3 pts per 150 total team possessions
or =+2 per 100 team possessions at the 1:2 ratio of on vs off playing time. Change that ratio and you change the stated impact. Ok, as long you know what you mean.

On/off is simpler so I'd label & call it that and use that description.


Any interest in presenting a list of values for just those on mock drafts? I didn't recognize any name in the top 250. Luke Sikma was the first name I recognized at #264 on overall +/-. Perhaps the quality of subs is more often real bad at small schools? I scanned the top 1000 and found it hard to find guys I recognized.

Raw +/- against just the top 65 (or some such cut) would be even valuable to me.

A crude form of adjusted +/- might use a power ranking to adjust the data on a team and boxscore basis. If anyone need true, play by play based Adjusted +/- for the NCAA I'd suggest it be against just the top 65 and probably 2 year to get more data and including the Tournament.

[Jon Nichols]

[Joe]

[Crow]

In trying to explain, I guess I might have made it murkier, not doing the equations properly. Sorry.

I wanted to try to display what the statement meant, but I could or should have said

"Team’s performance with that player on the floor compared to the team’s performance overall",

if you wanted explicitly that, would be

100/50 (rate with) * 150 possessions with = +300 +/- (if he played 150 possessions)

minus

100/50 (rate with) * 50 possessions with +100/100 (rate without) *100 possessions without = +200 +/- (for 150 possessions)

or 300-200= 100

divided by 150 team possessions= 0.67 per team possession

or +67 difference in team’s performance with that player on the floor compared to the team’s performance overall per 100 team possessions (at that 1:2 with / without time ratio)

If you had a 150 possession game (implied in the original scenario) then this player has a +100 team’s performance with compared to the team’s performance overall +/- impact per 150 possession game.

Of course this example is extreme.



The simple on/off would be (skipping the detail) +200- +100= +100 per 100 possessions. You originally stated the +/- per possession of +1, without a unit designation, but that is comparable to the simple on/off when you recognize the unit of your original +/- statement.

So your calculation is simple on / off and the original description, for which I provided the calculation to match the description, is something very different.


Most places like 82 games just use simple on /off.

[Crow]

This is shifting the focus

and what Adjusted +/- is for

but

you could look at a player's top 5 or 10 lineups (or more if you have the data and want to go further) and then construct a sample of lineups otherwise the same but for him being out and weight the minutes the same way and compare the average +/- with the player and without, everything else comparable.

Another estimate of his distinct impact.

[jmethven]

I agree that adding minutes played would be a good step once you are able to get around to it. With NCAA Plus/Minus, the raw data will have even more noise than in the NBA, given the divergent strengths of schedule between teams, and even between players on a team! For Illinois for example, the bench players have the best plus/minus numbers on the team, but I suspect that is just a combination of both not having played many minutes and having played the majority of their minutes against the weaker teams on Illinois' schedule.

[Jon Nichols]

[DSMok1]

[DLew]

Jon,

What source did you use for these numbers? Are you parsing play-by-play, and if what play-by-play source did you find that has all the substitution data?