... is to know the past. Hi guys. First time poster. Long time reader. Over the past few months, I've spent way too much time creating a model to rate the success of college player in the NBA. In short, I've taken over 500 college players that were drafted into the NBA, and rated their NBA success between 0 (bust) and 100 (All Star). I then compared the stats of the college individual (everything from height, class, school, etc.) to the database of player and found the closest statistical matches. I then take a weighted average to predict the success of the college individual. I used years 2001 thru 2008 as a template to determine the individuals that would be used in the database, and the weighted average values. Using this model, I compared my results to those of the GMs, and found that this model outperformed the GMs by over 7% between 2001 and 2008. Don't get me wrong. This program is not perfect by any means. It picked Adam Morrison as a sure thing at number 1. But it also grabbed Paul Millsap at 6, and Rajon Rondo at 7. Not bad. I then plugged in this years college draft individuals. I've attached the full results. Here are the top 10: 1. Anthony Davis 2. Jae Crowder 3. John Henson 4. Bradley Beal 5. Michael Kidd-Gilchrist 6. Jeremy Lamb 7. Will Barton 8. Damian Lillard 9. Jared Sullinger 10. Draymond Green Some others of note: 30. Austin Rivers 36. Andre Drummond 40. Harrison Barnes These results have a pretty good correlation with many mock drafts. But there are quite a few question marks in there that surprised me. My favorites for the Blazers - Damian Lillard at 6, and either Sullinger or Zeller at 11. Let me know what you think. View attachment 2867 View attachment 2866 View attachment 2865 View attachment 2861 View attachment 2855 View attachment 2862 View attachment 2856 View attachment 2864 View attachment 2857 View attachment 2863 View attachment 2860 View attachment 2859
I thought the same thing Btw, great stuff. Lillard is looking more intriguing by the day. I agree in Sullinger at #11. His advanced stats speak loud and clear (30 PER; 10WS)
Nope. Definitely not Mixum. I'm from West Linn, OR. Ben a season ticket holder with my family since 1974. When I started out working on this program, I didn't have much expectations. I was a bit surprised that I was able to create a model that matches up pretty close to a lot of mock drafts. This program certainly can't see "potential". Just the raw stats of what a player has done. It's all fun.
I'm intrigued, but have some geek questions. Is "bdb score" from your database? And what metric are you using to grade against GMs? I don't have the time, but I've thought of using a winscore comparison...does your model use hard metrics, either stats-based or things like All-Star games, championships, all-NBA teams, etc? Also, I wonder what some granularity would do. You have 0 as bust and 100 as all-star, but there HAS to be some difference between, say, All-Star Jamaal Magloire and wannabe All-Stars like James Harden. And then a tier between those guys and MVPs like Rose, and then a tier for all-time greats like K*be and LeBron.
BdB is my initials. It also stands for Basketball data Base. You're right. Grading the success of an NBA player is quite subjective. Doing it for over 500 players gets grueling. The basics - 100 = 5 + time All Star 90 = 2+ time all star 80 = all star 1 time 70 = high end starter 15 + points and or long successful career. 60 = average starter 10-15 points a game. 50 = low end starter 10- points a game. 40 = high end backup (started less than half their games played) 30 = average backup 20 = low end backup 10 = barely played in the NBA 0 = played less than a full season in the NBA The challenge was determining how to weight the importance of each stat category. I primarily looked at the basic stats such as points, rebounds, assists, blocks, steals, fouls, turnovers, fg%, ft%, 3pt%, 3pt shot selection. I also added a factor for the player's height, so it is primarily looking for an individual similar in size. Equalized everything by per 40, added a class factor, and school factor. Whew. I then added filter criteria to eliminate deviations outside the norm (if it finds 5 similar player rated about 40-50, and one rated at a 10, it will throw out the 10). I also use a per game contribution factor similar to Hollinger's formula to compare per game contributions per 40. If it finds a statistically similar player, but their Hollinger per game number is greater than a certain factor, it throws that one out. Hope that helps clarify a bit. Sorry to bore with the details. I'm an engineer, so I enjoy the math side of this. When I compared the program results to the GMs, I took the top 5/10/15 actual college players selected in the draft, and added up the players scores. I did the same for my program results. For example, If the GMs got a player rated 80, 70, 30, 90 and 40 with the top 5 picks (310 total), and this program grabbed player with scores 15, 40, 80, 90, 60 (285 total), then the GMs beat me by 25. 25/310 ~ 8%. I did this for 8 different draft years, and looked at 3 sections of each draft. I attached results summary pages if you are interested to see the players this program selected each year.
I've been reading this site for years. You guys are entertaining. Never had the urge to contribute until now. The fam and friends encouraged me to share these results. I figured you may enjoy it. So much of what we do is based off of emotion. Rooting for particular schools, etc. It's interesting when you take the emotion (and the "potential" word) out of the equation and see what's left. Personally I don't think Jae Crowder is the second best player in the draft. I also think TRob is better than the 15th rated player. Is Drummond really the 36th best player? I guess we'll see...
Pretty good. I'm an engineer myself, so I get it. Here's another wrench to toss in...can you (or do you need to?) put in a per-draft normalization factor? For instance, a GM should be penalized for taking Darko or Thabeet #2, when there were great players available. But if I'm reading this right (no guarantee at almost midnight) then you're counting this across the years, which means that you could be penalizing a GM for taking Bargnani #1 when in other years Melo was available at #3, which doesn't seem right.
I can tell you some of the matches the program found for Lillard - Jason Terry, Luther Head, Gilbert Arenas, and Juan Dixon are a few.
