Brooklyn Nets
D5 Analytics - Rating Correlations
Jan 10, 2018 7:56:11 GMT
Ian Noble, Walt Frazier, and 6 more like this
Post by Andrei Kirilenko on Jan 10, 2018 7:56:11 GMT
D5 Analytics
#2 - Rating Correlations
This was originally going to be one massive post, but I have decided to break it up into two separate posts, with the next coming next week. This first section will cover basic correlations between player ratings and player statistics. The second part will dive deeper and will include the element of "player style" in the discussion.
My interest over this topic was originally sparked through a discussion with Chauncey Billups about his team's poor performance to start the season. I headed over to the statistics section of the forum and scrolled down to the Sacramento Kings on the most recent post. Woah. Andre Drummond is shooting only 39% from the field, Thaddeus Young is shooting 29%, Otto Porter is shooting 37%, Avery Bradley is shooting 41%, and Damian Lillard is shooting 44%. These guys are all rated well overall, but they all suck in the sim. I went over to the most recent player ratings thread to investigate, and didn't see anything too unusual about any of his team. In fact, when looking back at the statistics, it appears that a lot of teams seem to be shooting unreasonably low percentages this year. I thought about writing an analytics article on this fact alone, but instead decided to try to find the root cause and figure out just how our rating changes are impacting in-game play.
Thus, I took all of the player ratings (Jan 4) and all of the player statistics (Dec 1 in game) and compiled them into one nice big excel file. Then, I removed any player who has played less than 10 games and any player who plays less than 20 minutes per game. I didn't want the low-minute guys to alter my results. I was left with a database of 153 players who met the criteria of 10+ games and 20+ mpg, which I think is a pretty good population to use for my formulas. I then ran some correlations between our player ratings and their D5 stats. Here are some of the more interesting things I found.
Assists
This is the highest correlation I found in all of my results. The “Pass” rating holds a 93% correlation with assists per game. This means that it is super easy for us to increase someone’s assists simply by increasing the Pass rating. Even a guy like Tyus Jones, who is a 69 overall rating, is averaging 8.1 assists per game. Tyus has the 16th highest Pass rating in D5.
Something else I will point out here is that the Pass Rating actually had a higher correlation with fewer TO per game (-50%) than did the Handle rating (-32%). Our league leader in turnovers per game is Avery Bradley, who has a respectable 79 handle but only a 29 pass rating. In fact, none of the top 15 highest-turnover players have over a 50 pass rating, but almost all of them are guards with high handle ratings. The handle rating doesn’t actually seem to correlate with anything, which leaves me wondering if it is really even utilized in-game. Conclusion? If we want to reduce turnovers, we need to increase Pass. But that's also going to increase assists.
Rebounds
The second highest correlation is for rebounds. I added together OReb and DReb ratings, and ran a correlation versus rebounds per game. The two data sets are correlated at 86%. We actually have 23 players averaging above 10 rebounds per game, as compared to just 11 guys doing it in real-life. Our biggest underachiever is Enes Kanter, who has the 9th highest total rebound rating but averages only 7.9 rebounds (not even top 50). I thought that this surely has to do with physical attributes, as teammate Joel Embiid averages over 10 rebounds per game but has lower rebound ratings, but ran a correlation of Jump rating vs Rebounds per game and actually found a negative correlation of -20%! (meaning a higher Jump rating corresponds to lower rebounds). Perhaps this is because our “jump” ratings are typically allocated to guards instead of bigs. As such, I ran a correlation between our Strong rating and RPG, and found a solid correlation of 74%. This again may be due to the fact that “strong” is typically probably allocated to big men who already have high rebound ratings. I don't really have an explanation as to why Joel Embiid is outrebounding Enes Kanter, except maybe it has something to do with Play Style.
Blocks
Also highly correlated, at 84%. We have 13 guys averaging over 2 blocks, whereas only 5 guys do that in real-life. LeBron averages 2.4 blocks in the sim with only a 58 block rating, but he has 99 D Aware and great athletics. On the other end of the spectrum, Giannis averages only 0.5 blocks yet is a top 15 blocker in the league rating-wise. I think this may be because his rating was recently changed. The physical attribute that seems to impact Blocks per game the most is the Strong rating, which is correlated at 72%. Marc Gasol is an interesting case here, as he only has a 65 Block rating, yet averages 2.6 blocks per game (4th in the league). His Strong rating is an 81, and DAware is an 85. These are respectable ratings, but he still seems to be vastly overperforming.
Steal Rating vs. Steals per Game
63% correlation. It actually got worse when adding in the "Quick" rating. No real comments.
Free Throw Shooting
Free throw rating is a decent indicator of FT%, with a 79% correlation. I also theorized that the “FG” rating somehow impacts free throw shooting as well, so I added both FT + FG together and compared it to free throw percentage, and got a worse correlation at 72%. I don’t really have any conclusions here except that maybe some of the physical attributes impact FT shooting. I tried a few different physical attributes out (namely, Fatigue) but got much lower correlations than just using FT%.
Field Goal Shooting
The Field Goal Rating only correlates 22% to a player’s field goal percentage. There are obviously then a ton of factors that go into FG%. In fact, the Inside Scoring rating is a better correlation with FG% (51%), which makes sense as inside players typically shoot better from the field. Surprisingly, CJ McCollum has our highest FG rating in D5, ahead of Durant, Curry, and others. McCollum also has our third highest 3p rating, and second highest FT rating. However, CJ shoots just 46% from the field and 43% from three in D5. Field Goal percentage will be the main topic of my next post.
Three Point Shooting
To run this test, I did not include any players with below a 20 three point rating, as they do not shoot many threes and would skew the results. Three point percentage only held a 32% correlation with three point rating. What!? Even when doing the same second test as I did for freethrows (adding 3p rating + FG rating together to run the correlation versus 3p%), I got a worse correlation at 28%. Our highest 3p rating in the league is Steph Curry (96 rating), and he shoots 37.7%. Our best three point shooters are Mirza Teletovic (55.6%, only 75 rating), Wilson Chandler (52.2%, 75 rating) and Malcolm Brogdon (50%, 89 rating). I’m not sure what to conclude here except that there must be a lot of other factors impacting 3p%. I think my data may still have not been great, even after removing players with below a 20 rating.
My overall conclusion is that this stuff is confusing and there is obviously a lot going into the sim engine. Hopefully my next post including Play Style leads to some more definitive analysis. Stay tuned!
Boston Celtics
Minnesota Timberwolves
And I've always acknowledged it's not perfect, but I've attempted to be consistent, at least. That said, I do notice ratings out of whack and over time have been decreasing block ratings overall as I noticed we get too many blocks here,for example. 
New Orleans Pelicans
Detroit Pistons
