Question for tremaine

Where is the link to the original location of the formula you use? You called it the "ESPN rating of a player:" and listed it as

Points + Rebounds + 1.4*Assists + Steals + 1.4*Blocks - .7*Turnovers + # of Field Goals Made +1/2*# of 3-pointers Made - .8*# of Missed Field Goals - .8*# of Missed Free Throws + .25 *# of Free Throws Made

 

NBA.com's efficiency rating

<div class='quotetop'>QUOTE </div><div class='quotemain'>NBA.com evaluates all players based on the efficiency formula: ((Points + Rebounds + Assists + Steals + Blocks) - ((Field Goals Att. - Field Goals Made) + (Free Throws Att. - Free Throws Made) + Turnovers)).</div>

 

I found it at the bottom of http://sports.espn.go.com/nba/seasonleaders

 

You found it. I knew you could find it with no problem.

Funny you should bring it up, because on a day like this when I am spending some real time on basketball, I am plotting how I will improve on this formula a little, without going overly complicated as Hollinger does. For at least the team I am covering, which looks like it will be the Raptors, I am thinking seriously about modifying the ESPN formula a little, assuming I can get Excel to do everything it is capable of doing.

Don't get me wrong though, even with no improvements at all, this basic ESPN formula is pretty damn good, and is better than every other combination formula out there. I have seen about a dozen other combination formulas that use everything, or almost everything, that is tracked by scorekeepers, on various other sites, and this one is as good or better than any of them.

Even though I can't find the time to figure out the hard to understand Hollinger formulas, I am already aware that his standardization around the number 15 is not a good thing, because there is no true, good reason for making all the players average out to 15. What if all League players are better one year than the other year? You would never know it from the Hollinger rating system. A player could be better one year than the other, but show up as worse in the year he was better, according to the Hollinger per, due to the standardization of all ratings around 15.0.

 

<div class='quotetop'>QUOTE (tremaine @ Jul 29 2008, 09:59 PM) <{POST_SNAPBACK}></div><div class='quotemain'>Don't get me wrong though, even with no improvements at all, this basic ESPN formula is pretty damn good, and is better than every other combination formula out there. I have seen about a dozen other combination formulas that use everything that is tracked by scorekeepers, on various other sites, and this one is as good or better than any of them.</div>

Prove it.

<div class='quotetop'>QUOTE </div><div class='quotemain'>Even though I can't find the time to figure out the hard to understand Hollinger formulas, I am already aware that his standardization around the number 15 is not a good thing, because there is no true, good reason for making all the players average out to 15. What if players are better one year than the other year? You would never know it from the Hollinger rating system. A player could be better one year than the other, but show up as worse in the year he was better, according to the Hollinger per.</div>

There is a very good reason to set the average number at 15. The formula for PER is dependent upon factors for each season. For example, the amount of total rebounds and points changes each season.

The "problem" you bring up is only a problem when people that don't understand the statistic attempt to use it. That happens the majority of times with stats anyways.

 

Looking at this formula

ESPN rating = PTS + REB + 1.4*AST + STL + 1.4*BLK -.7*TO + FGM + .5*TGM -.8*(FGA-FGM) + .25*FTM - .8*(FTA-FTM)

1) There is no rational reason for a block being worth the same amount as an assist

2) There is no rational reason for a turnover being discounted

3) A Free Throw miss should not be the same value as a Field Goal miss

4) A steal, a block and a turnover all end a scoring opportunity, yet they have different multipliers

 

That's the point: if the number of points or whatever goes up League-wide, why should every individual player have the measurement of his points scaled back? No, Hollinger is wrong about this. A player who does more in season B than in season A should automatically get a higher rating, regardless of whether the League as a whole had more overall production. Year to year comparisons are more valid if the rating is not standardized around a particular number, not less valid as Hollinger wrongly claims. The standardization is a needless academic type of thing.

 

<div class='quotetop'>QUOTE (tremaine @ Jul 29 2008, 10:22 PM) <{POST_SNAPBACK}></div><div class='quotemain'>That's the point: if the number of points or whatever goes up League-wide, why should every individual player have the measurement of his points scaled back? No, Hollinger is wrong about this. A player who does more in season B than in season A should automatically get a higher rating, regardless of whether the League as a whole had more overall production. Year to year comparisons are more valid if the rating is not standardized around a particular number, not less valid as Hollinger wrongly claims. The standardization is a needless academic type of thing.</div>

Because PER is an evaluation of how far away from the "average" player an individual is. That is why it is both per-minute and pace adjusted.

You are taking PER out of context and then complaining that it doesn't fit your context.

 

<div class='quotetop'>QUOTE (cpawfan @ Jul 29 2008, 09:15 PM) <{POST_SNAPBACK}></div><div class='quotemain'>Looking at this formula

ESPN rating = PTS + REB + 1.4*AST + STL + 1.4*BLK -.7*TO + FGM + .5*TGM -.8*(FGA-FGM) + .25*FTM - .8*(FTA-FTM)

1) There is no rational reason for a block being worth the same amount as an assist

2) There is no rational reason for a turnover being discounted

3) A Free Throw miss should not be the same value as a Field Goal miss

4) A steal, a block and a turnover all end a scoring opportunity, yet they have different multipliers</div>

I think a block is worth about what an assist is worth. Most blocks prevent scores, while assists by definition lead to scores. Even blocks that go out of bounds mostly prevent scores, because the other team so often can not a good shot off in the remaining time.

When I make small adjustments in this, which will not change ratings and ranking all that much, one of the things I want to do is make the turnover factor 1.0. Another thing is to increase the field goal miss to about 1.2. I also want to equalize the 2-point and the 3-point shot "bonus beyond the points" factors at about .7, not have one at .5 and the other at 1.0 as it is. But again, all of these changes added together are not going to change the relative ratings and rankings much. Of the top 330 players in the NBA, changes I might do might change someone's rank by up to about 10 positions, but no more.

You can debate either way whether it is worth it to make the small changes I am thinking about. I could decide to do nothing and still stand by the Real Player Ratings as valid and reliable, especially when adjusted for estimated forced misses. As others have noticed, the RPR rankings are roughly similar to the Hollinger rankings from his rating system, which reminds you that any per time measurement that includes everything that is tracked, and is half way or more reasonable, is going to have players in very roughly the same order. You are never going to see, for example, a player ranked #30 in the Hollinger but only #130 in the RPR, or vice versa. Unless the Hollinger is even more messed up than I think it is, the maximum difference you might see for a player, between the Hollinger and the RPR, would be about 25 positions.

 

<div class='quotetop'>QUOTE </div><div class='quotemain'>Because PER is an evaluation of how far away from the "average" player an individual is. That is why it is both per-minute and pace adjusted.</div>

Per time is necessary. Pace adjustments are great and recommended, but Hollinger, I suspect, goes overboard with his pace adjustments.

Without a doubt the standardization around the number 15 is not needed, unless you are just an academic trying to show off your mathematical skills. There is no basketball reason for the standardization. There is no such thing as an average player in the Hollinger sense, because the League and the players in it do change from year to year and from decade to decade. Why should you pretend that the players of 1975 or whatever were, overall, exactly as good as the players of 2008? That's what Hollinger is claiming, but it's not true.

Also, you can not fairly compare a player from one year to the next using the Hollinger, but you can do it with the RPR.

 

<div class='quotetop'>QUOTE (tremaine @ Jul 29 2008, 10:39 PM) <{POST_SNAPBACK}></div><div class='quotemain'><div class='quotetop'>QUOTE (cpawfan @ Jul 29 2008, 09:15 PM) <{POST_SNAPBACK}></div><div class='quotemain'>Looking at this formula

ESPN rating = PTS + REB + 1.4*AST + STL + 1.4*BLK -.7*TO + FGM + .5*TGM -.8*(FGA-FGM) + .25*FTM - .8*(FTA-FTM)

1) There is no rational reason for a block being worth the same amount as an assist

2) There is no rational reason for a turnover being discounted

3) A Free Throw miss should not be the same value as a Field Goal miss

4) A steal, a block and a turnover all end a scoring opportunity, yet they have different multipliers</div>

I think a block is worth about what an assist is worth. Most blocks prevent scores, while assists by definition lead to scores. Even blocks that go out of bounds mostly prevent scores, because the other team so often can not a good shot off in the remaining time.

When I make small adjustments in this, which will not change ratings and ranking all that much, one of the things I want to do is make the turnover factor 1.0. Another thing is to increase the field goal miss to about 1.2. I also want to equalize the 2-point and the 3-point shot "bonus beyond the points" factors at about .7, not have one at .5 and the other at 1.0 as it is. But again, all of these changes added together are not going to change the relative ratings and rankings much. Of the top 330 players in the NBA, changes I might do might change someone's rank by up to about 10 positions, but no more.

You can debate either way whether it is worth it to make the small changes I am thinking about. I could decide to do nothing and still stand by the Real Player Ratings as valid and reliable, especially when adjusted for estimated forced misses. As others have noticed, the RPR rankings are roughly similar to the Hollinger rankings from his rating system, which reminds you that any per time measurement that includes everything that is tracked, and is half way or more reasonable, is going to have players in very roughly the same order. You are never going to see, for example, a player ranked #30 in the Hollinger but only #130 in the RPR, or vice versa. Unless the Hollinger is even more messed up than I think it is, the maximum difference you might see for a player, between the Hollinger and the RPR, would be about 25 positions.
</div>

You can stand by it and I will continue to point out the many flaws in it and the lack of real world correlation. As I've stated, it has not proven to be either valid or reliable.