Basketball Tracking the + / - for each hoops player this season
- Thread starter Richie O
- Start date
Well yeah, that’s my point. People are using it to argue which players are better.
well I'm a HUGE stat guy but I personally consider this a Bull Shit stat. It has WAY too many variables that effect the numbers from who your playing, who you're playing against, who your playing with, what the game situation was, blah, blah, blah.
Come up with the one showing optimal on court productivity mix based on game time, opponent defensive ranking, score margin when inserted, etc. and I'll be your huckleberry.
Tell me who should be playing together last 5 mins +/- 5 points
...then again if AI is the answer we don't need a $4 million coaching staff to use their brains.
Come up with the one showing optimal on court productivity mix based on game time, opponent defensive ranking, score margin when inserted, etc. and I'll be your huckleberry.
Tell me who should be playing together last 5 mins +/- 5 points
...then again if AI is the answer we don't need a $4 million coaching staff to use their brains.
This stat isn’t normalized for PT relative to overall MOV for the game - correct? Fluox - is there any way to incorporate this?
If I’m understanding the math on this metric - if player A comes into the game for 30 seconds, commits 2 turnovers that are converted for 2 back to back unanswered baskets before getting pulled for the duration - that player’s +/- for the game would be -4. Yes? If player B plays the entire game and we go on to lose by 12, player B +/- is -12 - worse that player A. Do I have this right?
If I’m understanding the math on this metric - if player A comes into the game for 30 seconds, commits 2 turnovers that are converted for 2 back to back unanswered baskets before getting pulled for the duration - that player’s +/- for the game would be -4. Yes? If player B plays the entire game and we go on to lose by 12, player B +/- is -12 - worse that player A. Do I have this right?
Better, which it is, barely, and much better are different things. J WIll is really just a bigger, stronger, more mature Simpson who has grown out of the mid range nonsense and gets to the rim. Both want to dribble, have pretty good asst %'s and rebound well for guards.
Doing it by lineup rather than by player is also a better approach. You can then look at the lineups that make up 80%+ of the minutes
If each player has a equal weight on a game the players + / - would include 80% of data that had nothing to do with the player.
Per minute +/- relative to the final game MOV seems necessary to ascertain meaning from the metric.
I missed this game, but +/- seems meaningful enough in this regard only to tell
me Gavin must’ve had a pretty good relative game. He played 22 minutes and while he was in the team was +6. The final MOV was -17 so that means the team -23 in the other 18 minutes of the game. But +6 wouldn’t impress me at all in 22 minutes in the LIU game (for example) where won by 22. In contrast, that would mean that the team only built a 6 point lead in the 22 minutes he was in but managed to pile on a 14 point extra lead in the other 18 minutes (did better with line ups that excluded him - this is an example only). Without this context, isn’t the stat useless? What am I missing?
I’m not sure how it ought to be done to get better meaning, but as is, the data just doesn’t seem very useful to me.
No player is on the court alone. The supporting cast around a player has a lot to do with their level of performance/productivity.
For example, Omoruyi had better production when on the court with Spencer/Mulcahy than with Fernandes/Davis.
The plus/minus should be for a unit rather than a player because this is a team sport, not an individual one.
Yes absolutely this would be more reliable though there would still be limitations of it’s not adjusted based on minutes played. Comparing what one unit does in 3 minutes of PT vs another unit in 26 minutes is also going to be flawed.
I’m just not getting the running total concept of +/- at all without factoring in time. If Cliff plays 40 minutes in a game we lose by 6, and Oskar plays for 40 seconds and turns the ball over twice in a row for an easy 4 points and then Pike pulls him right out - that’s an example where the -4 MOV really was hypothetically about the player and not the unit. Meanwhile Cliff, who (in that example) played with every unit finishes with an even worse +/- of -6 but that’s over an entire game stretch not 40 seconds. How do you compare those stats straight up? You can’t.
This was my point when people freaked out about Gavin's +/- against Princeton.
If Clif gives up back to back 3s to his man that is a -6 for Gavin (and others on the court).
In general, individual +/- is worthless and I'm not sure anyone actually uses it as a point of serious analysis.
Not completely worthless, but not worthful.
Better application is to see 5 player units
I think there are ways it could be useful over the course of an entire season (meaningful sample) but not in the way it’s currently calculated.
I’ve given this a bit more thought. To be useful, I think the data for each game should be normalized over 40 minutes and then compared to the game MOV. That would be a true +/- metric for an individual player relative to the overall team performance in the game.
From there, rather than just summing up the results to yield a meaningless number - the count function should be used to tally up the games where this output is > 0 as a percentage of total games played. That stat would be a reasonable metric to observe the percentage of games where the team performed better with an individual in the line up vs without.
Last year i believe cliff had a high +/- and Woolfolk had a low one. Now maybe cliff played more with the 1st team and woolf with tje 2nd team but i think it had application
Or maybe it was a coincidence. It’s an arbitrary tally across different (uncorrelated games). If one must look at a number, better to normalize the way I described and then look at the median, lower and upper quartile values rather than summing across games. Would be much more meaningful to see that output. As an example, lower quartile value of -5 would mean that in only 25% of the games the team did 5 points or better (40 min normalized) without player X while a lower quartile value of +2 would indicate that the team was 2 point or more per game better off in all but 25% of the games with player x on the court.
Last edited:
Gee I make this point late last 🌃 night then all you nerds weigh in to say the same thing without a single like or comment showing any agreement.
Must not be in the club.. one day you'll realize common sense is the best computer ever made.
Opponent defensive ranking shouldn’t really matter once you normalize the data over 40 minutes relative to team MOV because the metric is measuring a player’s performance relative to alternative options in the line up. Those alternative players have the same advantages and disadvantages that come with game over game differences in opponent strength.
Oh, couldn't do that over the course of a single game, it'd have to be a season.
I tried to pull it together using box scores a couple years back, but there were gaps from several games. I'd pulled all the points scored for and against Rutgers for each lineup, and tallied the minutes over the course of a year - and then divided by total minutes to get a sense of which lineups had higher/lower offensive/defensive performance.
It only makes sense for the 5-6 most-used lineups, though, that have enough minutes to really see anything.
I'll see if I can find that old thread.
Edit: Found it. It was for the 2020-21 season. https://rutgers.forums.rivals.com/threads/2020-21-lineups.215098/
But over a whole season the +/- metric as a running total is even more problematic because it’s blending data across uncorrelated games.
Ok.
I’m not sure what this is supposed to mean but it’s not a difficult concept. You don’t need an SOS adjustment for this metric. It’s supposed to be a relative metric to compare one player to others on the team (who are up against the same schedule so it’s already apples to apples).
There's no perfect approach - the sample sizes on everything are just too small, but that's true of most things in college basketball.
Looking within one team, and across the lineups most used, they'll all have played against the same opponents (unless you have a situation like Mag, where a player missed a significant chunk of the schedule). It's similar to "per game" metrics, which have some value comparing players on the same team but fall apart a bit when comparing players on different teams (as it doesn't account for things like pace, schedule strength, etc).
I see your point but those per game stats are different because they are not meant as comparative stats relative to the team’s performance on a given day. 10 points on 40% shooting means the same thing in terms of individual performance regardless of the game outcome.
+/- isn’t this way. It has to be comparative because -2 on starter minutes not only doesn’t mean the same thing in a 20 point loss vs. 20 point win - it arguably has polar opposite meaning - in the case of the former it reflects pretty good relative performance while the player was on the court while the latter indicates the reverse. That’s why I say adding laterally the stats from these two types of games as a blend is arbitrary.
It was just a gif to show how confused I was by your response... by SOD I was again saying every player doesn't play against the same line up and who was in the game from the other team when YOU were in matters.
as an example (and not a perfect one) Cliff would play when Edey was in the game and backup center when not. There is way too much individual factors to just throw out these metrics as grades. Like saying student A with a 2.9 GPA was not as smart as student B with a 3.3 GPA when they both took 120 credits but student A was taking hard courses while B was taking lots of guts.
I understand what your saying, but your over analyzing what the metric is aspiring to do there. To be meaningful, it needs a fairly large data set. That’s first of all. Then secondarily, you have to take at face value that substitution patterns are a chess match on both sides in any BB game. Coaches are rotating players at all times with intent to maximize their team’s performance. Yes - you are right in that certain players will end up logging more time against easier rotations than others but for the purpose of what this metric is trying to assess, that’s besides the point. +/- is a horrible metric for comparing player A directly to player B - it should never be used that way. What it can do over the course of a season, if analyzed correctly, is provide prospective on the extent to whether the player (in the context of how they were used) has been a relative asset or a liability to the team’s goal of winning (or let’s instead call the goal - accomplishing the most favorable end of game team MOV as possible). To serve as a reasonable metric for this though - Must be a) normalized over 40 minutes and b) each game must be accounted for as a separate entity not blended.
