APBRmetrics

Hello everyone, I'm new to this board. I'm a Mathematician hoping to get a start in this industry. It was suggested to me that I join this board so I can start getting some of my ideas out there.

I have been working on developing formulas for evaluating how well a player will translate to the next level. The basic idea is a weighted mean of specific stats to provide the players overall score, then that score is adjusted by scalar multiples based on some outside factors. These factors include things like competition level or age etc. The formulas are slightly different for each position. The scores usually range anywhere from 0 to 1.6 with 1.6 being the key number. I've plugged the '06, '07, and '08 drafts into my formulas. Of these draft classes the majority of the players scoring legitimately at or above 1.6 have turned out to be All-Stars. For example, in 2008 only Derrick Rose, Kevin Love, and Russell Westbrook scored about 1.6. Of course I have found cases of inflated scores but there is a common trend in almost every single one of these cases making it easy to spot if a score is legit. Unfortunately, I have not yet developed a scale for foreign players so at this point I can only provide my thoughts on the NCAA players entering the draft.

First the point guards:
Kyrie Irving: He scored an astounding 1.9528, however because he played in so few games and the games he played in were primarily during Duke's OCC schedule against weaker opponents we should expect his score to be inflated. Based on such a small sample size it is unfair to conclude anything one way or the other.

Cory Joseph: 1.6319
Kemba Walker: 1.6183
Reggie Jackson: 1.5775
Darius Morris: 1.5771
Brandon Knight: 1.4982

* Joseph and Morris both have one sign of inflation so their legit score is probably slightly lower however with where they are projected to go I look for both to be potential steals. Walker appears to be a legit 1.6 player. Thus I will go out on a limb and say I think Walker, and not Knight, is the second best PG in this draft.

Shooting Guards:

Alec Burks: 1.6037, only the second Sg in the classes I've rated to score 1.6 or above, the other was Brandon Roy.
Travis Leslie: 1.5268
Deandre Liggins: 1.5146
E'Twan Moore: 1.5015
Klay Thompson: 1.4716

* Julyan Stone and Jon Diebler both scored very high as well but they both have all the signs of an inflated score.

Small Forwards:

**Carlton Scott: 1.5763
Derrick Williams: 1.5244
Kawhi Leonard: 1.4913
Tyler Honeycutt: 1.4624

* Guards tend to score higher then forwards and centers so to compare them straight across isn't quite fair. I'm working on scaling them to do away with that problem. Kevin Love was the exception to this trend.

** Obviously the clear question mark here is the remarkably high score by Carlton Scott. He is the one true enigma in this draft based on my scales. I can't really explain away his score but I certainly am not trying to claim he is the best SF in this draft. With where he is projected to go I think he is someone worth watching if he is able to stick in the NBA.

* Finally, I would say my true sleeper of the SF's is Honeycutt. In a draft that is so weak at SF to be able to get him late in the first could be a real nice steal for a team.

Power Forwards:

Derrick Williams: 1.4153
JuJuan Johnson: 1.3644
Tristan Thompson: 1.3612
Malcolm Thomas: 1.3234

* Again it is unfair to compare forwards to guards on this scale. I would say JuJuan Johnson could be a nice find late in the first or early in the second for some team. One other note, Rick Jackson scored a 1.3883. He has some signs of inflation but his legit score should still be fairly high.

Centers

* The majority of the centers in this draft are foreign I could only rate 4 players none of which are highly regarded and their scores would tend to agree.

So I will conclude by saying I still believe Irving and Williams are the top players in this draft but if there is a sleeper All-Star coming from the college ranks I would expect it to be either Alec Burks or Kemba Walker.

If you actually reached this point I appreciate you reading my thoughts and look forward to questions, comments, and constructive criticism.

Hey, there. Welcome. I think you'll find this forum to be very receptive to new ideas, but much more so if those ideas are actually and explicitly shared. With that in mind, could you provide some real details for your method? Otherwise it's just a black box to me. Thanks.

Evan, thanks and certainly that makes sense. Basically what I am doing is taking specific stats that I feel say something about a given player. The theory behind which stats I choose has come kind of been a combination of what I feel, what I have read, and insight I have gained from personal contacts. I started these formulas a few years ago in college and got together with an assistant coach from my college's team about preparing a report for him. Then I altered my formulas slightly based on some of his insight. Basically, what I am looking for is quality over quantity. For example, a guy who scores a lot but needs to shoot a lot to do so wouldn't score well.

The two things I look for in a stat are how well would it translate to the next level and does it provide some additional insight about a player that can't be directly derived from another stat. For example, let's say you want to know how well a player will perform when the game's on the line. Well to my knowledge they don't keep track of things like points in the last 3 minutes of college games. So something you could look at to possibly gain an insight about a players clutch ability would be free throw percentage. I believe this because free throws are one of the few times where the game slows down and the player can beat himself by over thinking.

What I'm striving for is complete objectivity. I hope this paints a more clear picture for you, if there's anything else you want me to elaborate on I would be more than happy to do so.

I believe what EvanZ meant is that we typically expect that you share the formulas, or at least enough of your methodology so that we can meaningfully comment on your work.

While your goals are admirable, and it's great that you have consulted with an assistant coach, neither of those things help us understand how you arrived at the numbers in your original post.

A number of draft translation systems that have been published have only given general methodological descriptions and not all the exact weights. That limits discussion but the situation is not unique. What to specify exactly and what to describe generally is ultimately up to the author. Depends on how much help or understanding you are seeking.

It would be possible, if you were interested, to compare your projections to Hollinger's projections and perhaps other metrics in a summary table. That would draw further attention to your ratings, give a sense of how it is similar or different and be handy for looking back and evaluating the degree of success each had.

I briefly compared your position rank order projections to Statman's college performance rankings (two different measures but I wanted to compare them anyways just to see how they matched up) http://www.pointguardu.com/f136/statman ... post355594
and they seemed to coincide more on the PGs than other positions.

for discussion & 2010 rankings see
http://sonicscentral.com/apbrmetrics/vi ... hp?f=2&t=8



Predicting All-Stars is one measure. Predicting starters or significant rotation players or busts would be addition tests of helpfulness.

If you want to determine and advertise predictive power you willing probably want to do a correlation of predictions for the large group vs actual (at same point in their careers).

A regression to confirm or further refine the weights used in the predictions might be a useful next step if you haven't already applied that rigor.

A review of the false positive predictions might be helpful, for past seasons and later with this one. In public or in private.

The 0 to 1.6 was a scale that happened, that you could work with, but if you want to promote the metric to the media or teams you might take it to a different scale. Multiply by 60 and it would close to a 0 to 100 scale. Or set the median at 100 like Staman does or whatever.

Taking the decimal out to many places won't throw or disturb everyone here but it isn't ideal to me. It not as easy to read and suggests greater precision that is probably warranted.

I noticed the first and fifth guys by position seem to vary in your rating by only 5-10%. That seems small. Maybe too small. Somehow that difference is probably going to getting magnified and perhaps a lot when things move from prediction to actual.

Crow wrote:I briefly compared your position rank orders to Statman's rankings http://www.pointguardu.com/f136/statman ... post355594
and it seemed like you agreed more of the PGs than other positions.
.

BTW - I should be posting my final 2010-11 college basketball player ratings in pretty extensive detail later tonight - and using those ratings, trying to look at a number of factors (class, rating breakdown, etc) for the NBA draft probably wednesday night.

Staying at college performance rankings explained in detail or taking them to NBA performance predictions?

I revised what I said above to try to be more careful about that distinction.

Thanks everyone for the tips and analysis.

Philosopher, I have some reservations about putting the actual formulas out there being a public forum and all. I will say this, the stats that are most heavily weighted are Points Per Shot and Assist to Turnover Ratio. There are 8 total stats that I input and 2 adjustment factors, those being the players age and the competition level.

Crow, I think you made some great points. The idea of multiplying the scores to make them more meaningful to the reader is something I considered doing before and I think will certainly do going forward. You mentioned it might be helpful to compare my rankings to Hollinger's. His recent rankings actually gave me the confidence to post this because of some similarities. For example, he is the only expert I have seen that likes Kemba Walker over Brandon Knight. Also, he ranks Tyler Honeycutt as his 13th best prospect and mentions him as a sleeper to watch. He has Alec Burks as #9 overall and the top SG even though a lot of analysts have him slipping over the past few days. He doesn't seem to be as high on Cory Joseph, Travis Leslie, or JuJuan Johnson as I am.

Here's the link: http://insider.espn.go.com/nba/draft201 ... ter-110620

You mentioned some past failures. One that comes to mind immediately is Mike Conley. He scored as the highest player taken in the 2007 draft at 1.8875. Breaking apart his score I found that one third of it came directly from his AST:TO of 2.77 which is remarkably high, especially for a freshman. I attributed this to the fact that he played with Greg Oden just about his entire life. Another example from the same draft would be Sean Williams, the big man from Boston College. With a score of 1.5864 he was the best big in the draft which is clearly not true. The problem here is he was slightly below all the other players in just about every category except he averaged 5 blocks per game. So clearly I need to implement some sort of cap on certain stats to keep one stat from overshadowing mediocrity in the others. A crud solution would be something like anything over 3 bpg is just entered as 3 bpg. I'm not thrilled about doing this but it might end up being my best option.

Thanks for the mention of a few false positives. There will always be some. If you can refine the formula from it, great.

Perhaps working with measurements in terms of standard deviations from average would be useful in the translation of college performance data to NBA expectations. For the outlier cases like Sean Williams and more broadly.

IMO, the answer from any young formula is another perspective and probably should not seek to be the dominant guide.

Though perhaps a formula tested over 7-10-15 years of data with detailed review at individual player level of past college performance, prediction and later NBA performance for a 100/200/400+ players for a few hundred or even thousand hours with many rounds of formula refinement from that review and analysis could eventually become a strong perspective to consider in a draft pick operation.

You indicate "the formulas are slightly different for each position". i think it is good that they are different but I'd probably make them quite different in contrast to those who prefer to look at "basketball Players" and not be heavily influenced by position or position expectations. I'd look by position and probably by the main types of roles played at that position.

There are different ways to succeed in the NBA and maybe it is a fairly broad spread of ways to deliver value for players at every position but I think what works well enough to win consistent big minutes from NBA coaches probably tends to cluster around certain stat sets for positions or position / role types.

A draft projection system could either try to recognize and respond to position or position / role clusters for the league as a whole or to the specific patterns or preferences of a specific team and coach. It would be challenging but I think it ultimately could be more rewarding to try than to stop short of doing this NBA focused work and also carefully considering how college position or position / role cluster stats differ from NBA stats because of the different expectations of those position or position / role clusters in college with and against college level talent and under college rules (on 3 pointers and the zone).

Sean Williams was actually very good for the first half of his first season.
http://www.basketball-reference.com/pla ... lits/2008/
This basically proves he was quite good when he wanted to be. At least to the age of 21.3

Somebody did some sort of look at NBA outcomes for players who faced college discipline. I forget the reference at the moment. It makes sense to pay attention to past red flags though the significance of the red flag could vary. Williams getting booted off his team made teams wary and apparently with good reason.

Crow, that's a very good point. In Hollinger's article he mentions that big men in particular tend to fizzle for off-court or motivation issues. He attributes this to the fact that size is rare and those who have it can get to advanced levels with much less work than a guard or a wing. It's kinda tough to say if there is a way to read an effort or motivation problem from the stats. One idea I toyed around with is looking at the players game-by-game rebound numbers. Theoretically someone that lacks motivation may tend to have more games with poor rebounding numbers. Possibly removing outliers and re-averaging someone's RPG might paint a better picture.

The links to my 2011 NBA draft breakdown - which I will definitely go more in depth in a few days - is in this thread:

http://sonicscentral.com/apbrmetrics/vi ... ?f=2&t=257

I wouldn't really call Mike Conley a false positive, especially after this year.

I don't mean that he's not a very good player. What I meant is he scored much higher than I feel is accurate. He was by far the highest ranked player in that draft class and off the top of my head I think the 2nd highest in any of the draft classes I plugged in. Maybe a better way to put it is false over-positive.