+/- numbers - how accurate/relevant? (Luis Scola in game vs. Hornets)

So, he wrote that Marion was better than the player on his team who won the MVP? hmmm....

That's beside he doesn't tell me any out of sample information.

 

There were actually a lot of people that year who said Marion was the real MVP of the Suns. Hollinger wasn't alone on that.

And I don't understand what your second sentence means.

 

Mutombo has one of the best rebound-rates in the entire league over the last number of years with the Rockets. So stats do tell me that Mutombo is a very good rebounder, much better than Yao for example, and that he'd be more likely to get a rebound like that than most other players.

 

sure i'll agree with that but what im saying is stats will just record 1 rebound no matter how hard or easy the rebound was.

like a game winning buzzer beating 3, stats will just say its a 3 pointer and its 3 points, but that 3 is definitely more than that. stats is just overrated in general if u know what i mean.

 

So what stats are of worth to you? Boxscore stats I suppose? What +/- basically tells you is the ideal rotation to use in that game.

So when I see that Scola was a -6 on court and a +17 off court, that tells me he probably wasn't part of the best rotation the Rockets used that game. Adelman made the right move in having Landry out there instead of Scola at the end. Landry's athleticism basically gave us 2 extra possessions (thus 2 fewer for New Orleans) with those late offensive rebounds. I'm not sure if Scola would've been able to get those.

 

If you defend the validity of using such stats with too much "just that season/game", it defeats the purpose of why people are interested in them in the first place.

 

Here's what I typically hear when people try to dismiss various statistical measures as useless:

"Oh no, numbers! My head hurts. Ooh a slam dunk, pretty."

 

I agree that, ultimately, being able to make predictions is what's important. But a lot of sports talk is centered on who's had the best season or who's the most valuable player. Those type of questions aren't answered with forecasting.

Hollinger does make predictions, a lot, using his own methods. He's got his playoff odds, his PER projections at the beginning of the year, he tries to forecast the numbers of players from the European leagues, he does trade analysis at the trade deadline, and so on. For years, he put out an annual book called Pro Basketball Forecast/Prospectus. That was all about predicting what will happen in the following season.

 

I do know what you mean, if you're talking strictly about box score stats. There's very little context in there. But team tracks many more specific events than simply "points scored" or "three pointers made". I'm sure the Rockets track, in great detail, buzzer beating shots by their players. They probably statistically track pretty much everything you would observe with the naked eye. Then, they put their team of analysts to work in mining all that data and trying to draw some useful information from it that may not have been known otherwise.

 

To be honest I think even the adjusted +/- is lacking or incomplete as well. The thing with regression is, you "average" out a players performance as if its constant when its anything but that year to year. On newbs like AB adjusted +/- is useless because of the small sample data and the fact that the learning curve will be sporadic due to his adjustment on the NBA game. On the other hand, if you have "years and years" of data then that dude will probably be on the tail end of his career already.

A good example would be Jason Kidd. Say he had like 2 years as a good player, 8 years of dominance as an elite level pg and now 1 1/2 years as an slightly above average pg. The 8 years of dominance and the 2 years of good playing would cancel or overpower the current level of performance, making Kidd seem much much better than he really is (i.e. someone who can push the Mavs from contenders to champions). I suppose you can add weight to the present years as some sort of correcting factor but how do you know the exact weight to use? I'm not even sure if the weighting itself will skew the "unknown" variable of the +/- computation that we're trying to remove in the first place lol. Anyway, I think adjusted +/- is a good starting point, but I think you need to tweak or add something to it because if you don't you end up with Cuban's wonky system that somehow shows that Kidd is the second most productive player in the league right after LBJ.

 

In one game, i'm not sure i care for any stat. You can watch the game and see all kinds of things that might skew a stat. Now, if there is a trend that can be tracked over the course of multiple games, i'd find that somewhat useful.

 

I think you're misreading +/-. All its saying is that at that the Rox got outscored by 6 pts when Scola was playing and outscored NO when Scola is off the court. That's it. You can't draw anything else except that single sentence, much less making "blanket" statements like Scola is not part of our best rotation. For all you know Scola's own production was +25, but Artest's is -10, then West is +10 (-10 for us) and so on and so forth....

 

Well, I think it's probably just one measure to look at, and it might reflect some things that are easy to miss that don't show up in a boxscore. But obviously it shouldn't be something you exclusively rely on, because the results are noisy and, you're right, there are simplifications the model makes. To me, its an interesting idea and it might be used to strengthen or weaken one's views on a player, but I don't know how useful it really is in practice.

 

You'd probably also have to factor in which Hornets players were on the court when Scola was on the bench, as well as their +/- numbers.

But my feeling is that +/- numbers may make more sense for evaluating groups of players rather than individuals. For example, which lineup had the most success over a period of time.

 

It seems you don't have a problem with stats, you have a problem with the box score. I'm not sure if you're familar with our GM Daryl Morey, a stats guru who has a team of analysts, but he completely agrees with you.

Box scores are vague. Two points were scored, but how? Was there some measure of luck? Did the player receive help from teammates from set picks and screens? All of that is lost.

While boxscore stats aren't useless, they are only a part of the greater picture at hand. The stats they have derived are more focused on seeing how well the team plays when any given player is on the court. Ball movement, percentage of rebounds grabbed, shots contested. This way they can gather what exactly the player is doing on the court to receive his boxscore stats while accounting for who's on the court with any said player.

 

Can u believe Finley's +/- number today? +11. He scored 0 pts, 4 rebs, 0 ass, 1 to. Got abused by Artest all game long and the guy had the best +/- on his team?! It just does not seem right.

http://www.nba.com/games/20090322/HOUSAS/boxscore.html

 

Yeah, Artest was abusing him, so they brought in Bowen. Bowen didnt give them any more offense, but was really good against Artest making him work. Go figure. Finley is a classic case of playing with the starting unit. Though, Scola had a really great +/- today.

 

This. And the same logic applies to all statistics. When people complain about missed freethrows, +/- stats, PER, etc. for ONE GAME, it really grinds my gears. Even 5-10 games is a pretty small sample. No stat over that timespan is going to be very meaningful. I could flip coins throughout a game of basketball and get a disproportionate number of tails.

 

Your link is for the game Rockets vs Hornets.

Scola is +14 in today's game.

http://scores.espn.go.com/nba/boxscore?gameId=290322024

 

It started out as a discussion of Scola's +/- number but it has turned into a general talk about the +/- number itself.

If I read it right. most agree it's a very misleading stat if you only use 1 game.

I 'll take it further and say it is probably not a good stat even if you consider the whole season. My argument is if it's not true for one game, what makes it true for many? When u average it out, would it not be simply an avg of bad calculations in either directions?