OU-UBC Statistical Plus/Minus

[DSMok1]

Well, I thought I'd roll out a new evaluation method.

Remember the comprehensive statistical plus/minus I calculated for last year? The same can be done for a single game to estimate a player's play and contributions.

Basically, it is a proxy for doing actual +/- calculations for the game... which, I might add, could be doable from the soonersports.com play-by-play data. Actually, SPM is a proxy for adjusted +/- calc's, which can't be done with a single game.

Anyway, SPM is calculated in a per-40-minute form. If a player is a +40, then the team should have been on pace to win by 40 in the minutes he played. As you can imagine, the numbers from yesterday were rather gaudy. Here are OU's SPM's and total margin contributions, from a SPM perspective:

Player			SPM	Contribution
Warren, Willie		52.0	36.4
Mason-Griffin, Tommy	43.1	24.8
Gallon, Tiny		38.4	22.1
Crocker, Tony		40.1	22.1
Davis, Cade		55.2	20.7
Fitzgerald, Andrew	31.2	19.5
Willis, Ray		47.4	19.0
Pledger, Steven		37.2	16.7
Wright, Ryan		23.9	13.8
Franklin, T.J.		28.9	2.2
Gerber, Beau		28.0	2.1
Randolph, Ryan		27.6	0.7

Note: I applied the team adjustment (to cause the team total SPM to equal the actual final margin) equally to all players (every SPM was bumped up equally).

Basically, WW and Cade had the best per-minute calcs. None of the posts played exceptionally well; Gallon was solid and the others a little less so.

Josh Whyte was the best player for UBC.

 

[DenverSooner]

that is certainly a different way to view basketball. The plus/minus is one of the primary statistics in hockey. In hockey you are going to be +/- one or two typically and you would never see these huge double digit numbers.

 

[DSMok1]

that is certainly a different way to view basketball. The plus/minus is one of the primary statistics in hockey. In hockey you are going to be +/- one or two typically and you would never see these huge double digit numbers.

Yeah... that's because the scores in hockey are so low! If your team wins by, say, 2 goals, the total of all the +/- ratings for the team should be 10 (unless there was a powerplay goal, which isn't counted on +/-)--since there are 5 on the ice at a time.

In basketball, similarly, if your team won by 40, the total of the team's +/- should equal 200.

However, both of those stats don't account very well for the opponent and who else was on the court/ice at the same time. In basketball, it is possible to disentangle each player's contributions from the others, at least approximately. That does take, however, a lot of games with a lot of lineups playing versus lots of opponents.

The result is "Adjusted +/-." Raw +/- could show wierd things (say, for instance, Anthony Parker is mostly on the court with Carmelo--and thus gets credit for his feats.) But adjusted +/- tries to solve that. It mostly works, and is the only stat that can really measure defense in basketball with any degree of accuracy. Here is a site that calculates NBA adjusted +/- daily: http://basketballvalue.com/topplayers.php?&year=2008-2009 (I'm linking to last year's final numbers).

The only problem is that college is a massive project to do the same with--it is hard to collect enough play-by-play data to actually run the adjustments and sort things out. An alternative is to attempt to correlate the boxscore data to the +/-. That is the Statistical Plus/Minus, which is a formula that gives an approximate +/- based on the boxscore numbers. The formula takes in per-minute stats and gives out a per-minute estimated +/-. So I gave the per 40 minute numbers here, along with the actual estimated adjusted +/- for the game itself. No, it's not completely accurate, but at least it gives us an idea...

 

[OUHoops]

Interesting stuff, Mok. Thanks for sharing.

 

[tomkar98]

Yeah... that's because the scores in hockey are so low! If your team wins by, say, 2 goals, the total of all the +/- ratings for the team should be 10 (unless there was a powerplay goal, which isn't counted on +/-)--since there are 5 on the ice at a time.

In basketball, similarly, if your team won by 40, the total of the team's +/- should equal 200.

However, both of those stats don't account very well for the opponent and who else was on the court/ice at the same time. In basketball, it is possible to disentangle each player's contributions from the others, at least approximately. That does take, however, a lot of games with a lot of lineups playing versus lots of opponents.

The result is "Adjusted +/-." Raw +/- could show wierd things (say, for instance, Anthony Parker is mostly on the court with Carmelo--and thus gets credit for his feats.) But adjusted +/- tries to solve that. It mostly works, and is the only stat that can really measure defense in basketball with any degree of accuracy. Here is a site that calculates NBA adjusted +/- daily: http://basketballvalue.com/topplayers.php?&year=2008-2009 (I'm linking to last year's final numbers).

The only problem is that college is a massive project to do the same with--it is hard to collect enough play-by-play data to actually run the adjustments and sort things out. An alternative is to attempt to correlate the boxscore data to the +/-. That is the Statistical Plus/Minus, which is a formula that gives an approximate +/- based on the boxscore numbers. The formula takes in per-minute stats and gives out a per-minute estimated +/-. So I gave the per 40 minute numbers here, along with the actual estimated adjusted +/- for the game itself. No, it's not completely accurate, but at least it gives us an idea...

Thanks for feeding my statistical hunger, DSM! haha

 

[ctt8410]

Good stuff, DSMok. Quick question: How do you feel statistical plus/minus compares to a traditional plus/minus system where a player's contribution is pulled straight from the play-by-play data?

 

[DSMok1]

Good stuff, DSMok. Quick question: How do you feel statistical plus/minus compares to a traditional plus/minus system where a player's contribution is pulled straight from the play-by-play data?

What are we measuring? The play-by-play data system is better, IF you can adjust for teamates and opponents. That is not possible in college. The statistical plus-minus is a proxy for actual APM; it is was calculated via a linear regression of player stats onto APM. Thus for the NCAA it is far more useful.

Now if someone would actually compile the play-by-play data for all, what, 5000 games and do the actual adjusted-plus-minus, that would be better.... That's way, way, way beyond me.

 

[DSMok1]

Corrected Single-Game Statistical Plus/Minus

I'm sorry, folks... I realized I had a brain cramp. Actually, for statistical plus-minus, it is like adjusted plus/minus: the sum of the team's +/- equals exactly the point differential--not 5 times the point differential (which is raw +/-).

Which means the numbers actually look like this:

Oklahoma
Player			SPM	Min	Contribution
Warren, Willie		16.1	28	11.3
Davis, Cade		19.1	15	7.2
Willis, Ray		11.3	16	4.5
Mason-Griffin, Tommy	7.1	23	4.1
Crocker, Tony		4.1	22	2.3
Gallon, Tiny		2.5	23	1.4
Pledger, Steven		1.1	18	0.5
Randolph, Ryan		-8.8	1	-0.2
Franklin, T		-7.5	3	-0.6
Gerber, Beau		-8.3	3	-0.6
Fitzgerald, Andrew	-4.7	25	-2.9
Wright, Ryan		-12.1	23	-6.9



British Columbia
Player			SPM	Min	Contribution
Dumas, Denny		15.5	17	6.6
Bains, Balraj		9.6	12	2.9
Yu, Nathan		3.8	12	1.2
Whyte, Josh		-0.8	34	-0.7
Murphy, Alex		-3.2	27	-2.1
Watson, Kyle		-8.0	14	-2.8
Posthumus, Chad		-7.7	19	-3.6
Bath, Graham		-10.8	15	-4.1
Malish, Brent		-9.1	18	-4.1
Mayott, Melvyn		-25.4	7	-4.5
Burke, Kamar		-14.0	25	-8.7

And here is the interpretation: SPM is the number of points per 40 minutes the player was better than the average player on the floor that game. So WW was worth 20 points per 40 minutes more than the average player playing; he contributed 14 points of the final winning margin during the time he was in the game.

I will add that the SPM's for players that only played a few minutes can be very high or very low (if they get a steal and a 3 in 1 minute, their per 40 minute stats look ridiculous). That doesn't mean they could have kept it up!

Does that look better? It does to me!

 

[DenverSooner]

Yeah... that's because the scores in hockey are so low! If your team wins by, say, 2 goals, the total of all the +/- ratings for the team should be 10 (unless there was a powerplay goal, which isn't counted on +/-)--since there are 5 on the ice at a time.

In basketball, similarly, if your team won by 40, the total of the team's +/- should equal 200.

However, both of those stats don't account very well for the opponent and who else was on the court/ice at the same time. In basketball, it is possible to disentangle each player's contributions from the others, at least approximately. That does take, however, a lot of games with a lot of lineups playing versus lots of opponents.

The result is "Adjusted +/-." Raw +/- could show wierd things (say, for instance, Anthony Parker is mostly on the court with Carmelo--and thus gets credit for his feats.) But adjusted +/- tries to solve that. It mostly works, and is the only stat that can really measure defense in basketball with any degree of accuracy. Here is a site that calculates NBA adjusted +/- daily: http://basketballvalue.com/topplayers.php?&year=2008-2009 (I'm linking to last year's final numbers).

The only problem is that college is a massive project to do the same with--it is hard to collect enough play-by-play data to actually run the adjustments and sort things out. An alternative is to attempt to correlate the boxscore data to the +/-. That is the Statistical Plus/Minus, which is a formula that gives an approximate +/- based on the boxscore numbers. The formula takes in per-minute stats and gives out a per-minute estimated +/-. So I gave the per 40 minute numbers here, along with the actual estimated adjusted +/- for the game itself. No, it's not completely accurate, but at least it gives us an idea...

Are you an actuary? If not, what do you do for a living? You clearly have a good understanding of statistics.

 

[DSMok1]

Are you an actuary? If not, what do you do for a living? You clearly have a good understanding of statistics.

I had to look up what an actuary was.

No, I'm not. But in structural engineering (I am a structural engineer), the whole concept of design is actually based on the statistics of failure. (Check out the September 2009 NHI Real Solutions Seminar, available for free, to see how.) I'm sort of a mathematical mind, anyway--I don't really do statistics for a living. I do any number of things besides sports statistics... check out The Top 100 Generals, a compendium I have been working on for several years of the greatest generals in history. And my website. I do a lot of things--I was homeschooled.