I created a ratio to show how efficient a player is on off.

[bgassassin]

This is a ratio I came up with when I made a thread on the Grizzlies messageboard (not on RealGM) looking to see if Rudy Gay should be utilized more in our offense. It utilizes Hollinger's Usage Rate which is my favorite stat from him. I wanted to show how offensively effective and/or efficient (E/E) a player when doing something with the ball in their hands compared to their level of usage on their team. In other words a ratio of Offensive Output to Usage Rate. I made a simple formula to calculate Offensive Output:

[(PPG+APG) – TOPG]/UR

First I want to mention that thanks to some discussion elsewhere I was able to see that there are two outliers this season if you take into account all players. Monta Ellis and Gerald Wallace both played over 40MPG, but Hollinger's UR uses per 40. When I did this for the top ten scorers I couldn't understand at first why Ellis was as high as he was, but a poster indirectly pointed it out to me. Since this ratio doesn't utilize true MPG, it didn't dawn on me at first that was the reason. So when you see him, don't pay him too much attention. Anyway those two got a boost in their E/E. So other than them everyone else was fine. A couple samples I will give you are the top ten scores this regular season, and the top ten rookies in UR. This will allow you to see the contrast in players and the effect of their true stats versus assumed usage. I also included their UR as a reference point.

Player (UR) - OO:UR

Top Ten Scorers
Durant (29.3) – 1.01:1
LeBron (32.2) – 1.08:1
Carmelo (30.5) - .93:1
Kobe (30.5) - .94:1
Wade (33.2) - .90:1
M. Ellis (27) - 1:1
Nowitzki (26.1) - .99:1
Granger (25.9) - .94:1
Bosh (25.9) - .93:1
Stoudemire (24.4) - .88:1

Top Ten Rookies by UR
Jennings (25.6)- .73:1
Evans (25.4) - .90:1
Flynn (23.4) - .64:1
Collison (23) - .67:1
Marcus Thornton (23) - .66:1
Beaubois (23) - .31:1
Hansbrough (22.8) - .39:1
A.J. Price (21.3) - .38:1
Terrence Williams (21) - .46:1
Curry - (20.9) - .97:1

Now for some explanations. Not only does LeBron have the highest ratio, his OO or E/E is greater than his usage level. Simply put Lebron is so efficient when he does something with the ball that his offensive output clearly exceeds his usage. Same goes for Durant on a lesser scale. Dirk almost put out numbers equivalent to his usage. Don't let the numbers for guys like Kobe and Carmelo fool you. Early indications from using this stat show that their ratio is very solid.

Stoudemire and Wade are on the lower side of their group, even though it's still decent. What is important to note is that Wade had the league's highest usage rate for the second straight season. While his UR fell slightly, his stats fell off much more dramatically. To compare, Wade's ratio last season was 1.01:1. I'm sure Heat fans can give reasons as to why the fall off. Stoudemire was also more E/E the previous season at .95:1. While some would assume that they could just look at Wade's stats and make the assumption based on their drop, that's not always the case. Zach Randolph is a very good example of a player who put up nearly identical stats this season and the previous season, but had a big jump in his ratio. He went from .80:1 to .89:1. Stoudemire also saw minimal change and went down in OO.

As you can see I like this stat because it can help show how E/E a player is instead of just taking traditional stats at face value since with some you can't always get an accurate reading.

So what's your take?

I guess while I'm at it I'll give you the numbers for the Grizzlies starting 5 as well.

Gasol (16.1) - .93:1
Gay (21) - .92:1
Mayo (19.8) - .92:1
Randolph (22.9) - .89:1
Conley (18.9) - .80:1

[mysticbb]

Your "ratio" actually makes zero sense. A player needs the ball in his hand to commit a turnover. Turnovers are a part of the efficient equation, subtracting them will give you a missleading impression like the one about Monta Ellis.

For your Grizzlies you are running into the same trap. You need to calculate a version of points created. A method would be:

PTS_C = (PTS+AST*1.06-0.508*FGM)*36/MP

1.06 is the points per made shot devided by 2 on the Grizzlies. 0.508 is 1.06*0.479. 47.9% of the Grizzlies made baskets are assisted. Thus you will give half of the value of a basket to the passer and subtract the same amount from the scorer. Not perfect, because each player has a different value for points per fgm, but we are correcting a part of this via some sort of efficiency factor:

EF = TS%+(1-To-R)

After that we can get something, which are you interested in:

ER = PTS_C*EF/USG

USG in that case is normalised to 36 minutes, unlike Hollingers 40 minutes. The result would be for the Grizzlies S5:

Code: Select all

Name      ER   ORtg
Gasol    0.535 112.0
Conley   0.466 112.3
Mayo     0.450 110.6
Randolph 0.423 111.3
Gay      0.410 111.3

That ORtg is the teams ORtg while the players are on the floor.

The overall conlusion: Take the ball out of Gay's hands, let Conley run the show and involve Gasol more often to make your offense better.

[bgassassin]

Well the fact that your conclusion says let Conley run the show already makes your version invalid towards me. But I would like to see your results for the top ten scorers and rookies for comparison's sake. Especially the rookies which you'll understand why later in this post.

I'm confused as to why it makes no sense. Maybe it's because it's simpler than something you are used to based on your response. TOs are subtracted to penalize the player for a negative result with the ball in their hands. Adding them would just make their ratio better. Ignoring them doesn't penalize them. That to me is important to include both the negative and positive aspects. Plus Hollinger's UR includes TOs. Like I said Monta's issue deals with the minutes he played. He in essence got a "bonus" due to that. Using TS% defeats the purpose in that I'm looking strictly at a final outcome. UR doesn't take into account any type of % and in turn I'm not either because using any type of percentage to show "how they got there" isn't what this was about. UR utilizes attempts as a part of the formula and I'm comparing that to the final results of those attempts as a part of mine as well.

I also didn't want to include any other per minute type of adjustments outside of UR because that can get misleading for those whose true MPG are under the per minute adjustment. Look at those rookies who played fewer minutes, but had a pretty high UR. That's probably why Conley had a jump in your version since he was the only starter clearly under 36MPG. At least with the one I'm using there are more players under 40MPG than there are players over 36MPG for your version. The purpose was to look at true output against an assumed level of usage (I say assumed due to per 40 adjustment). That's why a lot of those rookies had a much lower ratio than others. And as a Grizzlies fan those numbers in my ratio line up very well with what I watched.

This is the type of input I was looking for though to test the validity of the idea.

[mysticbb]

Well the fact that your conclusion says let Conley run the show already makes your version invalid towards me.

Well, the major factor would be have Gasol involved more. Also his OnCourt ORtg is the best of those S5 players.

I'm confused as to why it makes no sense.

Yeah, I would be confused too, because that was obviously a complete stupid comment by me. I completely messed it up and thought your were actually adding the TO which would increase the players efficiency. Sorry for that.

But still your results contradicts every offensive metric. The Warriors with Ellis on the court a worse offensive team than without him.

I will add a list with the requested numbers for the Top10 scorers and the rookies in a few minutes.

Top10 scorers:

Code: Select all

Durant     0.506
James      0.552
Anthony    0.421
Bryant     0.429
Wade       0.477
Ellis      0.374
Nowitzki   0.477
Granger    0.448
Bosh       0.478
Stoudemire 0.466

Top10 Rookies:

Code: Select all

Jennings   0.364
Evans      0.419
Flynn      0.371
Collison   0.435
Beaubois   0.492
Hansbrough 0.305
Price      0.413
Williams   0.317
Curry      0.464

[bgassassin]

Well, the major factor would be have Gasol involved more. Also his OnCourt ORtg is the best of those S5 players.

True. Both of ours can be used to confirm that.

But still your results contradicts every offensive metric. The Warriors with Ellis on the court a worse offensive team than without him.

I think this might help me see why you are having a problem with my ratio and hopefully at the same time will allow me to address your issue. The ratio was not designed to be concerned about whether the team benefited from having the player on the court or not. This was strictly an individual-based stat. If I understand the part in bold correctly you're having a problem with it because you are looking for something it was never designed to address. The logic behind this stat was that when said player has the ball in his hand, this is how effective he is. Not how much of an effect he has on the team. That's also why even if I could adjust Hollinger's UR to accommodate a higher UR, (You wouldn't by chance know where I can find the formulas for league pace and team pace would you? I've been looking for them off and on for a good while now for other things.) Ellis might still have a decent number because it's a ratio that looks directly at what he does as an individual regardless of how the team might benefit from it.

I think what also is lacking is more reasoning behind the stat. The Grizzlies showed an offensive philosophy change halfway through the season the emphasized more mid-range iso with Zach Randolph and more pick and roll with Conley while at the same time de-emphasizing our wings. Some of us noticed this shift and began to complain about it. So in my hopes to prove why I felt Rudy should be our main guy or used way more, I came up with the idea for this stat since I as already pretty familiar with UR. I know Gasol came out on top in both of our stats which would mean to me that he's both effective individually and has the greatest effect on the team, but looking at how the Grizzlies are built I wouldn't want him used as the main guy. To me Rudy would be the best choice for that and this ratio helps show that on our team Rudy might be capable of doing that.

I hope that clears some things up.

[mysticbb]

Well, yes, only stats are useful which help determine how much a player helps a team winning. Going away from that makes a stat rather useless, especially when I incorparate advanced stats. If I want to see how much a player produces I can also just look at the raw data.

You called your stats "a ratio how efficient a player is", but it is hardly efficiency you are measuring, but production. Your conclusion is that Gay gets not enough touches, but overall he has a very low efficiency. Giving someone like that more touches will decrease the offensive efficiency of the whole team. Monta Ellis is a perfect example of how flawed your idea is. He is doing great in your metric, but he isn't anywhere close to an efficient player. In a really good offense he is destroying the offensive efficiency for the Warriors. And that game is about how efficient a team can use a possession. Each team will have roughly the same amount of possessions in a game, using them more efficient than the opponent will give your team the win. And that is the main goal of advanced metrics, determine how to win games by using stats. Everything else is imho just playing with numbers.

[bgassassin]

Haha. To put advanced metrics on that kind of pedestal, and say everything else is just playing with numbers is kind of elitist. Assist/TO ratio has become a highly used stat, but according you it's just playing with numbers. This elitism would also apply to saying the only useful stats are the ones that help determine how much a player helps a team win. To me that's absurd and we should just eliminate tracking all other stats then. Of course though you can't get your advanced metrics without those "useless stats" if we did that now could we?

Speaking of which Assist/TO ratio doesn't tell us how said player effects the team. It's about the individual. Same thing with my stat. Now if you want to attack my logic supporting the stat, that's fine. But to attack/putdown the stat itself when I already pointed out that Ellis is an outlier due to his MPG is pointless. I said in the beginning that Ellis and Wallace would not be correct and why, and you trash the whole thing still due to one of them. Show me how it's wrong away from what I already pointed out.

Outliers happen. There wouldn't be a term for them if they didn't happen. I bet there are more players that would be off with your stat than my ratio. To me the fact that Conley has a rating as high or higher than a portion of the top ten scorers in your formula makes yours even more "flawed" because yours actually looks at team effect as opposed to mine. Conley has nowhere the effect those guys have even for the Grizzlies. And with you using a per 36 adjustment it's just going to make it misleading on a lot of these players who don't play a lot (see Beaubois). You're left to ask, "If this player really played 36 minutes, would he have the same effect?" Mine just says, "This is what he did

To say that I'm looking at production instead of efficiency is subjective IMO. Looks at the definition of efficiency and tell me why I'm wrong using that term. I also pointed out in the OP that this ratio shows you can't just look at "raw data" as you put it and gave examples to back that up.

I just want to make sure that you know I'm not offended in anyway and have enjoyed the discussion. I'm just the type that if I'm wrong on something, then I need to be clearly shown that I'm wrong. And as you can see I need a pretty strong reason to change my tune in regards to something like this.

[bgassassin]

Thanks to you and two other posters on two other boards that indirectly confirmed roughly the same thing, I made a change to the formula once I got some free time.

The new formula looks like this.

{[(PPG+APG) – TOPG]*(lg_pace/team_pace)}/adj_UR

After finding what I needed I made a small adjustment to Hollinger's UR by using the player's actual MPG instead of the per 40 he uses. I also added a pace adjustment to the player's stats so that players in a more uptempo system do not benefit and players in a slower system are not punished. I think you'll like this version better.

Durant – 1.02:1
LeBron – 1.13:1
Carmelo - .95:1
Kobe - .97:1
Wade - 1.02:1
M. Ellis - .89:1
Nowitzki - 1.06:1
Granger - .98:1
Bosh - 1.02:1
Stoudemire - .99:1

I also have no problem admitting that with this version I was wrong in my premise about using Rudy Gay more. I prefer to get the ratio right beyond anything else.

[mysticbb]

Sorry for not replying to your previous post. I read it, but didn't have the time for a detailed answer and later forgot about it. If you want an answer to specific things, let me know.

And yes, the results are looking more in line with my observation and with other metrics. Especially Ellis with his rather low efficiency is represented rather good. Suprisingly Nowitzki is scoring better in your rating than Wade or Bosh, while he is usually right on the same level with those two. If you don't have really large outliers (did you look up the top point guards, for example), the results are looking solid. The pace adjustment is always really helpful. How much of an impact has the minutes adjustment for the usage rate? I would like to see the differences here.

I also tip my hat to you, making an analysis, listen to critics and make proper changes. Nice work.

[bgassassin]

That's no problem at all as I'll forget to respond to a post as well. And thanks for the compliment. I just hope I didn't put you off too much with my debating. I'll debate if I don't feel something is valid enough, but since I saw a enough similarities between you and the other two I accepted there were a couple of flaws and used those discussions to pinpoint the areas of concern and fix them.

Yeah Nowitzki jumped out at me as well, but after looking closer he had a very low amount of TOs. In fact if you look at his total TOs compared to the six guys above him in scoring, his total is 149 and the next closest to him was Carmelo at 209. Just to see how it looked since you mentioned Dirk is normally in line with Wade, I gave Dirk the same amount of TOs as Wade and he went from 1.06:1 to .95:1.

Minutes played seems to have an interesting affect on UR due to total minutes being in the dividend and MPG being in the divisor. If all other components remain constant to the player and he sees either an increase in total minutes played or drop in MPG (all this while the opposing minutes stat remains constant), his UR would drop which would increase his ratio. And vice versa. To help look at players with different minutes played, here are the new ratios for the rookies from before.

Code: Select all

Player          MPG      OO:UR

Jennings       32.6     .91:1
Evans          37.2     .96:1
Flynn          28.9     .85:1
Collison       27.8     .97:1
M. Thornton    25.6    1.03:1
Beaubois       12.5    1.04:1
Hansborough    17.6     .84:1
A.J. Price     15.4     .95:1
T. Williams    22.6     .83:1
Curry          36.2    1.00:1

As for the top PGs, here is what they looked like:

Code: Select all

Nash       1.16:1
Paul       1.22:1
Williams   1.13:1
Billups    1.11:1
Rondo      1.08:1
Rose        .99:1
Kidd       1.18:1
Davis       .99:1
Westbrook   .93:1
Harris      .96:1

I'm assuming based on your question that you were most concerned about guys like Nash and Paul. I was surprised to see Kidd slightly higher than Nash since you wouldn't assume that looking at the surface. To be honest I think Paul was helped slightly due to playing fewer games.

[longfellow44]

It also seems to measure how much time a player spends with the ball in their hands. Bosh is the type of player that dribbles a lot more than other bigmen that's why his number is higher than other bigs. That's also why the pg's have the highest numbers as well. I'm sure if the numbers were done for guys like Ray Allen who spends most of his offense on the catch and shoot the number would drop dramatically. I don't think that this metric measures effectiveness at at all. I think what it really measures is the amount any given player dominates their respective offensive system.

That's why Chris pauls numbers are sky high. He's involved in almost every possesion weather creating for himself or for others.

[bgassassin]

Thanks for the post as different views help me test it.

Ray Allen was 1.08:1. Right there with Rondo and slightly better than Dirk.

Don't forget that Ellis has the ball a lot and was very low. And also Kidd was much higher than Dirk.

[longfellow44]

Remember though that he hasto share the ball with curry and he was very turnover prone which helped to lower that number. And despite everything goldenstates offense is about spreading around the touches and Ellis doesn't shoot many 3pt shots which lowers his true % because a 2pt shot is simply less efficient than a 3.

[bgassassin]

I'm not understanding the premise of your debate since Dirk, Kidd, and Beaubois all clearly exceeded Ellis and they were on the same team. I'm assuming you're still focused on the having the "ball in their hands" argument, but that's not really relevant. Again look at Beaubois and how many MPG he averaged. He's not getting a lot of touches yet he's outproducing his usage. Curry was at 1:1 so why wasn't he hurt since he's in the same system? Also you see Rondo and Allen had high numbers and are on the same team. Dominating the ball only helps if you are doing something with it. Kobe dominates the ball and yet his output does not equal his usage.

Being turnover prone is a good part of why his ratio was so low. That was part of the idea. To punish players for something negative on offense. True shooting percentage includes FTs which Ellis take more than enough to make up for his lack of three point shooting, yet not really that relevant to this ratio.

Here's a look at Boston's starting five since you brought up Allen as a concern.

1.08:1 - Rondo
1.08:1 - Allen
1.05:1 - Pierce
1.04:1 - Garnett
0.84:1 - Perkins

Four of their five starter have an output that exceeds their usage so I think it can again be assumed that sharing the ball or dominating the ball has little to do with having a good number.

[old rem]

Remember though that he hasto share the ball with curry and he was very turnover prone which helped to lower that number. And despite everything goldenstates offense is about spreading around the touches and Ellis doesn't shoot many 3pt shots which lowers his true % because a 2pt shot is simply less efficient than a 3.

Ellis' game is based on speed which is most effective if he's playing 32-36 min...not 45-48,and then he gets double dipped on "pace adjustment",which is a bit odd since GSW had 6-8 players much of the year and the pace...for most, faded when there wasn't rest or guys played hurt.

Perhaps there should be an "adjustment" for 3 rookies in the lineup...and 2 just arrived a couple weeks ago? When you got the 3rd string in the game...for 20-30 min,and they sort of just recently showed up,you have to guess they MIGHT be in the right place. At one point, with Mikki Moore and Vlad Rad getting major minutes...a major part of the game plan was Monta and Maggette each score 30 because otherwise the numbers won't add up. Efficiency is great but if half your rotation is missing, then you try to get LUCKY.

[mysticbb]

I'm assuming based on your question that you were most concerned about guys like Nash and Paul. I was surprised to see Kidd slightly higher than Nash since you wouldn't assume that looking at the surface. To be honest I think Paul was helped slightly due to playing fewer games.

Nah, not that surprising at all. Nash plays more risky passes, which gives the teammates better looks, but also can end up in passing turnovers. Kidd gets a couple of assists per game just by dumping the ball into the midpost area to Nowitzki. Nash also creates via dribble penetration which can lead to turnovers, Kidd isn't doing it. And the last point would be Nash creating his own shot more often, while Kidd is standing at the perimeter for an open set shot created by a double team against Nowitzki.

Paul is usually really efficient, his numbers aren't surprising at all.

@old rem

Fact is Monta Ellis is an inefficient player, he is scoring way below league average and he is turning the ball over like hell. That is not only showing up in his boxscore stats, but also in his On/Off Court numbers. Ellis has the worst Net+/- of all players on the Warriors (-11.5) and the 2nd worst adjusted +/- (-6.5). The Warriors actually played their worst basketball with Ellis on the court. He made them weaker on offense and was a defensive liability. The APM numbers are adjusted for the lineups and showing exactly what this by bgassassin created rating is showing, Ellis shouldn't get many touches as long as he isn't back to his pre-accident efficiency.

[Mikistan]

PTS_C = (PTS+AST*1.06-0.508*FGM)*36/MP

1.06 is the points per made shot devided by 2 on the Grizzlies. 0.508 is 1.06*0.479. 47.9% of the Grizzlies made baskets are assisted. Thus you will give half of the value of a basket to the passer and subtract the same amount from the scorer. Not perfect, because each player has a different value for points per fgm, but we are correcting a part of this via some sort of efficiency factor:

EF = TS%+(1-To-R)

Alright sorry to jump in with a different subject. Mystic, you're doing a great job to help me understand advanced statistics but I am having trouble getting my head wrapped around the assist subtraction system.

Basically, I am wondering why only "half of the value of a basket" is awarded to the passer. Shouldn't that value depend on the team's overall assist percentage (47.9% in this case for the Grizzlies)?

I know technically this isn't a perfect tool as well, because depending on who the assist was for, the assist should count for more or less...
Thinking about it literally: If Player A has 100% of all his baskets assisted from someone else, and I get an assist for Player A, it shouldnt be worth as much as getting an assist for Player B who has 5% of his baskets off assists.
AKA, an assist for a player that only scores off assists is (and should not be) worth as much as an assist given for setting up an ISO player for an easy score. Clearly the iso player will benefit more from your pass than Player A and you should be rewarded (assuming the true shooting percentage of the ISO player is higher than Player A).

Ok, I am not sure if you are still following me... Because the more I think about advanced stats, the more I think about how to make them perfect. In a perfect world, you would have a giant database with everything you need to look at assists - an assist is recorded, the player that scores is recorded, the player who got the assist for the certain player is recorded as a separate statistic, the percentage of assisted baskets for the individual scorer is compared to how many times the passer assists them compared to other players on the team, etc. etc.

Maybe I am just thinking about it in a far too complex way now. But back to my original question... If an ISO type player hardly ever gets assisted on his baskets, then should the passer be MORE or LESS rewarded for an assist? Does it even matter?

If basketball is a game of percentages, its not so much that you score at a higher rate than someone else, its more that you decided to CHOOSE to pass to another player that might be in a higher percentage situation than you are making the right play. But say you look at Nash for example. He is a very efficient shooter with high percentages, but he is also a great play-maker, albeit with a knack for turning it over due to the degree of difficulty in his plays. The argument can be made that he NEEDS to pass because it makes situations when he scores that much more efficient because of spacing/defense playing the pass to open up his own offence/ etc etc. How do you quantify for the "triple threat" option or "taking what the defense gives you" factor statistically when raw stats tell you - Nash shoots a 3 at a 35% clip, thats clearly better than passing to Dudley who shoots a 2 at 50%.
Nash clearly can't play all game shooting 3's every time he touches the ball because the game is DYNAMIC.

Don't advanced stats fall apart because of this dynamic variable effect?

[mysticbb]

Basically, I am wondering why only "half of the value of a basket" is awarded to the passer. Shouldn't that value depend on the team's overall assist percentage (47.9% in this case for the Grizzlies)?

Actually that is exactly what happened. The passer gets the points per fgm devided by 2 for each assists. And I substract half of the points for 47.9% of the fgm.

Don't advanced stats fall apart because of this dynamic variable effect?

It gets to complicated in the end and you are not getting more informations out of it. Occam's razor, make it useful and easily to apply. You can try to make it more accurate, but in the end you will run into other problems like human error (wrong assigned stuff by the statskeeper for example). At the end you want to get specific informations out of your formula.

Regarding your ISO question: We know that the ISO player is creating his shot by himself, sometimes he is just using head, shoulder or pump fakes to create enough space for a shot. Should the guy who just dumped the ball into the midpost area get an assist for that?

And overall, as I pointed out before, I don't think that this really matters. Stats are always flawed, making them more complicated will not erase the error. But you can always gain some informations out of it.

[bgassassin]

I decided to see how this would look for an individual game. This is for Game 3. Kinda figured some would skew higher since it was one game as opposed to a full season.

Code: Select all

Boston

Rondo         1.32
R. Allen      0.12
Pierce        1.22
Garnett       1.15
Perkins       0.49
Wallace       0.83
T. Allen      1.54
Robinson      1.25
Davis         0.88

Lakers

Fisher        0.93
Bryant        0.92
Artest        0.38
Gasol         1.00
Bynum         0.71
Odom          1.77
Brown         1.33
Walton        2.26
Farmar        0.60
Vujacic       2.27