Method to order rankings

[josecarlos]

Hi!!!

For a personal project, I'm trying to make a ranking for players for each stat. But I have a problem with players who have played few games or minutes and have good stats.

So, I would like to know if there is any standard to do this rankings. I have thought about it and in my opinion I have to discriminate for games played, a percentage of games played for example. So, from the total of player we will get two segments of players. One above that % and another underneath that %. Then we need to know if player belongs to one of those segments of players and finally to get the ranking inside this segment.

That's is what I thought, but I would like if there is any standard to do this type of rankings.

Thanks in advance

[Crow]

Check the first 2 links on this page https://www.google.com/search?client=ms ... s-wiz-serp

[Mike G]

You could certainly do a ranking of "rebounds above average" for a player, which would be along the lines of (Reb%-10)*Min/48
This would give more credit to someone who gets more minutes, if they keep rebounding at that rate. And this isn't all bad. A guy who ONLY rebounds (late Rodman) is likely to have higher Reb% than he would in a bigger role (early Rodman). It doesn't mean he became a better rebounder, but it probably means he gets fewer minutes.

In Denver, centers Jokic and Plumlee have identical TReb%; but Jokic does a lot more scoring and passing, and he gets 2.4 times the minutes. If Jokic reduced his scoring and passing roles, he would likely do more rebounding.
Meanwhile, Plumlee has sort of a "plum" job of filling in after Jokic has worn down the opposition, drawn fouls on their big men, etc. He seems not to get as many rebounds when he starts games (per min.)
https://www.basketball-reference.com/pl ... plits/2019

If Jokic and Plumlee play 1-on-1, I would guess the Joker easily gets more rebounds. Not 2.4 times as many but enough to notice.

[josecarlos]

Thank you so much to @Crow and @Mike G for your appreciated help!!!

[vzografos]

Along the lines of what MikeG suggested.

The basic problem you have is that the stats, being usually a statistical average of a quantity, depend a lot on the "amount" of that quantity.

So in other words, the more data you have the more accurate your stats are (i.e. the less unstable your averages).


Take two players and let's compare their average FG% per season. Player A has FG%=0.5 with 40 games played and Player B has FG%=0.48 with 80 games played (out of a max 82 games per NBA season). Or take the extreme case that Player A has FG%=0.9 with 1 game played.

How do you rank the two? Is Player A better than Player B? Well maybe.

I think the best way to deal with this problem (of insufficient data) is to weigh the two stats by the number of games played (or minutes like MikeG suggested, if you are looking at per-game stats rather than per-season stats). This "continuous" weighting avoids the manual, discrete threshold like your segmentation that you suggested.

So in the above example your weighted stats would be

Player A= 0.5*40/82 = 0.2439
and
Player B= 0.48*80/82= 0.4682

which would indicated that, given your existing observations, PlayerB should rank above PlayerA due to the uncertanty in PlayerA's stats.

In my view, it is better to be more conservative with insufficient data rather than making decisions on their face value.

[josecarlos]

Along the lines of what MikeG suggested.

The basic problem you have is that the stats, being usually a statistical average of a quantity, depend a lot on the "amount" of that quantity.

So in other words, the more data you have the more accurate your stats are (i.e. the less unstable your averages).


Take two players and let's compare their average FG% per season. Player A has FG%=0.5 with 40 games played and Player B has FG%=0.48 with 80 games played (out of a max 82 games per NBA season). Or take the extreme case that Player A has FG%=0.9 with 1 game played.

How do you rank the two? Is Player A better than Player B? Well maybe.

I think the best way to deal with this problem (of insufficient data) is to weigh the two stats by the number of games played (or minutes like MikeG suggested, if you are looking at per-game stats rather than per-season stats). This "continuous" weighting avoids the manual, discrete threshold like your segmentation that you suggested.

So in the above example your weighted stats would be

Player A= 0.5*40/82 = 0.2439
and
Player B= 0.48*80/82= 0.4682

which would indicated that, given your existing observations, PlayerB should rank above PlayerA due to the uncertanty in PlayerA's stats.

In my view, it is better to be more conservative with insufficient data rather than making decisions on their face value.

Very well explained and useful!!! Thanks vzgrafos!!!

[Crow]

https://docs.google.com/document/d/1h3z ... p=drivesdk

I found this somewhere. For what it might be worth, assuming you can see it.

[vzografos]

All the maths on this paper have been messed up when it was converted to Doc. Do you have a link to the original?

[Crow]

No, I dont have the original link and a brief search provided no answer.

[josecarlos]

https://docs.google.com/document/d/1h3z ... p=drivesdk

I found this somewhere. For what it might be worth, assuming you can see it.

Thank you!!! I can see it. I have downloaded to read it with attention. Thanks for your appreciated help!!!

[vzografos]

https://docs.google.com/document/d/1h3z ... p=drivesdk

I found this somewhere. For what it might be worth, assuming you can see it.

Thank you!!! I can see it. I have downloaded to read it with attention. Thanks for your appreciated help!!!

I would be very interested into hearing about your summary of that paper and conclusions.

[josecarlos]

https://docs.google.com/document/d/1h3z ... p=drivesdk

I found this somewhere. For what it might be worth, assuming you can see it.

Thank you!!! I can see it. I have downloaded to read it with attention. Thanks for your appreciated help!!!

I would be very interested into hearing about your summary of that paper and conclusions.

Of course!!! I just only a little bit of time because I'm making a new version of my website of statistics for women basketball in Europe.

[josecarlos]

https://docs.google.com/document/d/1h3z ... p=drivesdk

I found this somewhere. For what it might be worth, assuming you can see it.

Hi!!!

I have started to read the document and maybe there is a codification problem. In the first page, third paragrah which starts with "As an index of effect size " I can read ...

Thus (8):
1 2 j j μ μ - = Δ (8)

Do you read the same or am I only who the text appears like that?

[Crow]

It reads that way from the version I have too.

I looked and looked for original without success. I can't mess with this anymore.

If the text doesn't give general ideas, I'd suggest to just forget it on this one.

[Crow]

Published academic journal basketball analytics isn't always good or useful but it seems rarely searched for or used. Conference presentations sometimes highlights articles published or to be published but not always.

Very rare for professional media to ever mention / review / use. Sometimes they do) go gaga over broad, pop social science. Good and / or overhyped. That isn't enough.

I have occasionally looked, offered links, commented on. Might look around again sometime in future.