Trying to figure out NBA Comparisons for our Pick Up Crew

[fejaroorz]

I'm looking for advice on handling this problem I'm trying to solve. If this isn't the place for it, let me know. I only heard there's a bunch of smart people here.

Background:
We have a website where we store all our stats from our (terrible) pick-up games. Due to the way we record the stats, we are able to calculate per 100 for all of the major statistical categories. We also track Reb%, Potential assists, and Unassisted vs. Assisted FGA.

I've also have the methods of how players scored. Think of the Synergy Categories: Put-Back, Post-Up, Isolation, Spot-Up, Ball Handler, Cut, & Transition. I rolled Hand-Off up into Spot-Up since it didn't happen very often and I put Roll Man into Cut since that didn't happen enough too.

Problem:
We don't want to use our own photos for Player Profiles but an NBA equivalent. I have a calculation that is used to compare ourselves against NBA players, but it seems to give expected results about 40% of the time, 20% is someone that doesn't make sense and the rest are consistently pointing to 1 or 2 players (James Johnson and Rudy Gay in this instance)

Ideally, the comparison should work like the examples below.
Example: If you're a player who only makes a good amount of 3s, but nothing else and you primarily score off spot-up opportunities. Your NBA comparison would be a 3pt specialist like Anthony Morrow.
OR
You get a good amount of blocks and rebounds, but score almost solely off rolls & putbacks... Your NBA comparison would be DeAndreJordan.

Current Approach:
Since we can't compare ourselves to NBA players directly since our percentages are low and rebounding numbers are high. I thought it was best to compare groups within themselves. Pick-up Players against Pick-up players and NBA players against NBA Players. Everyone gets ranked in percentages of 10. If you're in the top 10%, you're a 10. If you're in the bottom 10%, you're a 1.
Those ranks get compared across all the categories and we calculate the variance. Lowest total variance is SOMETIMES your NBA Comparison.

Scoring styles are included in that rank method as well. But, we rank the frequency of how you like to score.
If 100% of your scores comes of post-ups. Then you'll get a 10 in that category and 1s in everything else.

Questions:
What stat categories should I include or exclude?
Example: Should I include ORB, DRB and TRB? Or just TRB? Or just ORB & DRB?

How can I best use the scoring styles to calculate player comparison?

Thanks for reading all of this!

[Crow]

It depends on how much work you want to give it (apparently a lot) and how accurate you want to be.

Conceivably you could run multi-variable regressions comparing every pickup player to every pro for the whole set of categories you decide to use and see which offers the highest correlations.

Or pick 3-5 variables, weight them and just compare the weighted products for overall similarity scores.

[fejaroorz]

It depends on how much work you want to give it (apparently a lot) and how accurate you want to be.

Conceivably you could run multi-variable regressions comparing every pickup player to every pro for the whole set of categories you decide to use and see which offers the highest correlations.

Or pick 3-5 variables, weight them and just compare the weighted products for overall similarity scores.

Thanks for replying to my question!

My first challenge is trying to figure out which categories would be applicable.

If I do go down the multi-variable route, what would be the dependent variable?

[Crow]

Your dependent variable could be your custom metric or a standard one.

I was wondering if it could be a pro name but I am not sure it can be without the metric score.
Maybe it could be with the pro name being the label for his matrix of non-aggregated data that you are matching the pick up player to?

Maybe someone else can help further.

[Mike G]

If you know per minute (or per 36) rates of Sco, Reb, Ast, PF, Stl, TO, Blk, I can generate NBA/ABA statistical similars, historical or current.
If you know player TS%, that would help with the Scoring component. I don't break it down by type of shots.
Also, team and opponent Pts and Reb per game and player mpg (or estimate).

[fejaroorz]

If you know per minute (or per 36) rates of Sco, Reb, Ast, PF, Stl, TO, Blk, I can generate NBA/ABA statistical similars, historical or current.
If you know player TS%, that would help with the Scoring component. I don't break it down by type of shots.
Also, team and opponent Pts and Reb per game and player mpg (or estimate).

It's pickup. So we can't do minutes.

We do have per 100 possessions. I've been using the data on BBR for comparisons.

We have TS% labelled, but looking at it more closely looks like we're using a modified eFG% instead. (Modified because we plays 1s & 2s)

We have team and opponent points somewhere in the code. We did calculate Ortg & Drtg. It took a while to realize that it was pretty much pointless since teams are always changing and we always strive to have fair teams to ensure good games. Therefore, the more games you play the more likely that Ortg and Drtg will equal each other.

[Crow]

https://theringer.com/the-nba-personali ... bb371810d4