APBRmetrics

Just plugged Sindarious Thornwell into my model and he rated top 5, despite my model fading the importance of scoring at an old age. Such good stats for a major conference player (16 BPM!)

Dr Positivity wrote:Just plugged Sindarious Thornwell into my model and he rated top 5, despite my model fading the importance of scoring at an old age. Such good stats for a major conference player (16 BPM!)

Does your model take into account his first 3 years as well?

I adjust for previous years in scoring but not the other categories. So Thornwell being mediocre earlier in his career as a scorer hurts him, but overall it's possible my model isn't scared enough of his major recent leap across the board. His shooting rates as really good because of the 83% FT (on a high volume of FTAs as well notably) in addition to the 39% 3pt but it's possible that he jumped in an outlier way that's not accounted for as well. The reason I do things that way is it led to better results in previous drafts for me so it could end up being the right call with Thornwell, but we'll see. I follow a guy on twitter @big_wafe who's draft statistical work I like and he told me he's a lot lower on Thornwell than I am because he adjusts for earlier seasons more

Thanks man!

For me the one guy projected lotto in this class I find overrated is Frank Ntilikina.

- Weak production: 12 PER in French Pro A, 10th highest on his own team. By minutes per game he is the 9th man on his team. Only 18 but that's a lot lower than I like seeing for a high level Intl prospect, and makes me think at best he's going to be a multi year project. Those numbers would suggest to me 2nd round flier

- It seems like half the reason people love him is they love the idea of a huge PG (6'5 with a long wingspan) and the defensive potential that brings. Except he's playing SG and averages barely any assists, he's 10th on his own team in both Ast/40 and Ast %. Just based on that alone, Frank being a huge PG instead of a SG sized SG seems like it's built on a house of cards. NBA teams seem to care more about him putting up assists at U18 level than what he's done as a pro, I'm not sure that's how it works, it's like caring more about an NCAA prospect's high school career than college, and if they only played a few games (Or only caring about Nike Hoop Summit/McDonald's game, etc.)

Dr Positivity wrote:Just plugged Sindarious Thornwell into my model and he rated top 5, despite my model fading the importance of scoring at an old age. Such good stats for a major conference player (16 BPM!)

If anyone here follows me on twitter - they know I've been calling Sindarius Thornwell the best player in the nation this season for the last few months. My ratings, which do account for pace, SoS, team defense, playing time relative to teammates, etc. have the guy as a college superstar. This was before the tourney - when I warned everyone that South Carolina had a chance to cause serious damage because they will always have the best player on the court - & he'll be playing increased minutes.

Like all models, my draft model obviously won't rate a 22 year old college senior as the #1 NBA prospect unless there just aren't any good/very good <20 year olds. That being said - I'm thinking he very well could be top 5 in my model as well (or AT LEAST lottery), despite the depth of talented freshmen & his lesser weighted Junior year included. I will know after the NBA season ends (updating my NBA age progressions) & commence to project ALL the careers of all the seniors & guys that declare.

No matter - in my historical ratings a guy who is THAT good in one year in college (being north of 200 in HnI is extemely rare) no matter the age ALWAYS gets PT in the NBA (even if 2nd round pick - no player rated that high in a season has gone undrafted), & more often than not ends up a very good/great NBA player. The fact that Draft Express & others don't even have him in the mocks is a joke. Herd mentality - soon someone at the mocks (say the ESPN guy) will get word that some team is looking at him late 1st - & all the sudden all the scouting sites will mock him late 1st & make their write ups better reflect that projection. They often aren't really scouting anything - it's just bracketology except for the NBA draft - caring about trying to predict draft position (& making their writings reflect that position) instead of actually best trying to truly scout, rate, & rank prospects.

Dr Positivity wrote:I follow a guy on twitter @big_wafe who's draft statistical work I like and he told me he's a lot lower on Thornwell than I am because he adjusts for earlier seasons more

I like ALL the draft modeler guys because they are willing to step out & try to produce results that may fly in the face of convention & open them to ridicule.

That being said - @big_wafe follow each other. His results seem REALLY HEAVY <20 year old leaning. They also appear to have mocks melded into the projections - to allow the results to come much closer to mimicking the mocks. I worry when I see results that get too close to the mocks, it seems like one would have to ignore a TON of quality info (like actual college dominance across an entire statistical spectrum for an entire season) & too heavily weight info that may have little value (but help correlations to mocks) to make that happen.

I notice many draft modelers seem to be going this route - I guess they figure if their results come close to conventional wisdom, their model will look strong. It makes me worry that they're veering away from why someone should be trying to create a quality draft model to begin with - to find those crazy outliers. Outliers are the key imo.

Now I could be wrong, maybe his model just somehow comes up with results that mimic mocks, but then that means his results will come fairly close to mimicking actual drafts - and those results traditionally suck. Teams have sucked at drafting for a long time, I'd hate to have my model resemble it too closely - at least not until many, many teams start drafting much better.

If someone is running a draft model & does not have Thornwell as a 1st round rating right now - I have serious doubts about the model's value.

Hey Dr. Positivity - do we follow each other on twitter?

Don't know hardly anything about draft yet but here are a couple of questions: Are more international draft picks going to NCAA first? I assume the answer is yes but is the change in last 5 years modest, moderate or big? Among the pure internationals, is this a very weak draft class? 3 pure international guys in the DX mock top 20 have really weak pro stats. If they are the best, yuck. How do the two groups compare in NBA performance?


My first instinct on the Thornwell debate is to split the difference between the high and low draft projections, or a bit less. Who are his most and best similars?

I like ALL the draft modeler guys because they are willing to step out & try to produce results that may fly in the face of convention & open them to ridicule.

exactly...

and those that openly criticize/ridicule are rarely those that also do their own modeling - because anyone who has poured through the college player stats for 340+ college teams (and for multiple years) knows what an exhaustive chore it is...

the online experts/prognosticators that bemoan the rankings of modelers will not rank players themselves, only criticize those that do. they are after-the-fact experts...

I have not done any draft modeling outside of analyzing college stats but for those who have, is any weight put on high school data? For the most part, I'm talking about the 1-and-done players.

I imagine you run into the issues of adjusting for competition, actual data (outside of the top Nike/Adidas tournaments?), and eventually enough meaningful college minutes to wash out most high school data. Might something like FT% and 3PT% benefit from some additional high school data when the sample size is lacking? Is there play by play data with any level of detail at the high school level?

There are different approaches to draft evaluations. Either you want to do the full deal and you want others to consider it (including hopefully insiders) or maybe you're a casual by choice who doesn't want to invest that much time, perhaps because of the high likelihood insiders aren't going to use it.

Of 70 top 10 picks in last 6 drafts, only 17% are at or above .100 on WS/48 (league average) so far in career. http://www.basketball-reference.com/pla ... r_by_asc=Y
Only 6% at or above .150 (what I consider very good).
With so few average and very good performers at the top of draft, it is somewhat hard for me to get that excited about the main focus of the draft. The teams of fulltime scouts and post scout evaluators sure don't have a lot of great results to wow with on average for their collective tens of thousands of hours of effort. Casuals can equal or beat that pretty easy.

About 35% of these top picks are at or below .050 which is bad, somewhere near replacement level. Look at the number of players super hyped by the insiders and big name media folks in the bottom part of this list or at the very bottom. I laughed when Kris Dunn was taken so high. Though my basis for doing so was only a half hour of reading and clip watching, I was pretty sure he was overrated.

Picking out better players later in the draft is somewhat fun, in a minor way. Brogdon was my best find last time. Got him in the 50s in a couple of sim drafts. A few other good guys with late picks I think, fwiw (D Jackson, Brice Johnson). Waiting on coaches to play them.

DX now provides stat splits by level of competition. I have advocated for that in past so it is nice to have it conveniently. Will any draft analytic models use the breakdown and weight them differently? The past answer I got was usually no but I lean toward at least trying weighting higher on best competition and seeing how it affects the projections vs. not.

tarrazu wrote:I have not done any draft modeling outside of analyzing college stats but for those who have, is any weight put on high school data? For the most part, I'm talking about the 1-and-done players.

I imagine you run into the issues of adjusting for competition, actual data (outside of the top Nike/Adidas tournaments?), and eventually enough meaningful college minutes to wash out most high school data. Might something like FT% and 3PT% benefit from some additional high school data when the sample size is lacking? Is there play by play data with any level of detail at the high school level?

I have yet to see anyone include actual high school stats in their draft models, but certainly would be an interesting exploration. Obviously, reliability of the data is not at the same level of college stats, and even if it has improved, retrieving it would be difficult for previous years, which you would need if you really wanted to thoroughly test its performance. Then, as you mentioned, you run into the adjusting for competition issues that derive from the millions of leagues, tournaments and international competitions. While more data is always great, college FT% is already one of the best predictors of shooting ability in the NBA. If there was a big sample size issue in 1-and-done players, college FT% would not be as strong of a predictor of NBA shooting ability since 1-and-done players represent a good portion of players in the NBA (of players that played college basketball). The most popular way to try and account for high school performance in a draft model is through RSCI scouting ranking but obviously that does not help with the sample size.

Crow wrote:There are different approaches to draft evaluations. Either you want to do the full deal and you want others to consider it (including hopefully insiders) or maybe you're a casual by choice who doesn't want to invest that much time, perhaps because of the high likelihood insiders aren't going to use it.

Of 70 top 10 picks in last 6 drafts, only 17% are at or above .100 on WS/48 (league average) so far in career. http://www.basketball-reference.com/pla ... r_by_asc=Y
Only 6% at or above .150 (what I consider very good).
With so few average and very good performers at the top of draft, it is somewhat hard for me to get that excited about the main focus of the draft. The teams of fulltime scouts and post scout evaluators sure don't have a lot of great results to wow with on average for their collective tens of thousands of hours of effort. Casuals can equal or beat that pretty easy.

About 35% of these top picks are at or below .050 which is bad, somewhere near replacement level. Look at the number of players super hyped by the insiders and big name media folks in the bottom part of this list or at the very bottom. I laughed when Kris Dunn was taken so high. Though my basis for doing so was only a half hour of reading and clip watching, I was pretty sure he was overrated.

Picking out better players later in the draft is somewhat fun, in a minor way. Brogdon was my best find last time. Got him in the 50s in a couple of sim drafts. A few other good guys with late picks I think, fwiw (D Jackson, Brice Johnson). Waiting on coaches to play them.

+1

Let's not forget, "ALL models are wrong, only a few are actually useful." Meanwhile some scouts have proven themselves for years.

I'd also have to question if someone adhere's so strongly to their model when the eye test is wildly awry from your model -- I would really question the cognition of the author of that model.

Getting picks right in the order of the draft is nice if you are trying to get an idea of pick value and who's going to be available. But draft model models, as indicated with this trivial query of the data, should realistically be something like:
-- By the end of their rookie contract (2-4 years, 3rd/4th are options):
-- Tell me which 8-10 players will be part of a rotation.
-- Give me the 3-5 who will be starting or still starting.
-- Give me the 2 best players in the draft.

Aiming at those marks might be a better use of time.

I think Thornwell's defense could make him one of those 8-10. He's in my top 15 consideration (not draft slot per say, but for the top 8-10 players that matter -- not to be drafted that high -- but one of those 8-10 guys who's a rotation player in 3-5 years).

2016 was historically one of the worst drafts ever. I saw a few places say it was the worst. I didn't verify so can't say worst for sure.

Some playful editorial via this post:

I think Kris Dunn will be fine. For him, it's all mental. And he's not a PG, he's a 2-guard. I peg'd him as Tony Allen-esque. If he plays 30 mpg he'll gets 150+ steals and 60+ blocks annually. Just have to get him to calm down on offense. He doesn't finish on drives -- not sure why. He isn't someone who wants to be more than he is. He will play "in his lane", just has to find that. Take his poor offense away and he's the best player in his class by WS/48, BPM, etc. Finney-Smith would be 2nd. Brodgon is is 0.8 ODPM, but is -1.3 DBPM. (http://bkref.com/tiny/pmNYC)

Dunn is the 71st best rookie since 1975 (42 years) who played at least 1200 mins in terms of DPBM at 1.7 ---- all while having no confidence. http://bkref.com/tiny/pChXD I also think he fits the system well. Shrugs. Lots of opinions and fun to pick a horse. Methods should definitely be allowed by all.

Some extra editorial per the original discussion:

I think[my OPINION from watching 13+ games, looking at stats, etc] Lonzo Ball is not only NOT near Fultz as a prospect [Fultz his his own flaws: mostly his effort, casualness, desire, and defense] but I also believe Ball is going to "bust" as well. He could be alright as a low usage SG. But he's never played defense in his life -- unless you count that Free Safety style he's played this past year at UCLA. On offense, 99% of the possessions before UCLA were up tempo, no defense. 33% of UCLA's possessions were fast-break. Ball had more spacing [room to drive] than any of the top prospects in this year's draft, by a mile. He often got walled off by anyone of decent athletic ability, e.g. Texas A&M, Oregon, USC, Arizona, Kentucky. If you watch games vs Dennis Smith in high school and other short, but more athletic guards, he struggles to get shots over them whenever inside 28 feet. He's got no plan with his fadeaways. Make him go right. He has to get get to rim or must pass. Ball can spin off the dribble when going left. He can 'pull up' (see that one mid range jumper vs Arz) when going left (right wing, 15 ft, Comanche was 5+ feet away). I'm honestly very surprised this isn't seen -- or maybe it is, and teams are keeping quiet hoping some team goes and drafts him higher than their pick. I'd be curious if Synergy [or anyone else] keeps track of shot % off the dribble, going right.

Regardless of that...in those 10 games vs the competition mentioned above. His stats:
https://ibb.co/e7sKGF [if image allowed]

Again, this is with the most space of any of the top 100 prospects in the draft.

Here's some more in depth scrutiny "models" based off box score data couldn't see on Ball and his shooting by Kevin O'Connor. The type of ball used in games.
https://theringer.com/lonzo-ball-ucla-s ... eda2ef3e41

Also with Lonzo's FT% he projects a 31-35% 3pt shooter, assuming this year's 3 PT% totals would hold true for a larger sample size. If you go by binning his 3PT%, FT% in college to pro for usage of players with at least 100 3PA in college. I'm working with more tiers (e.g. 1 year 100+ 3PA, 2 years 150+ FGA, etc) and other bins, regressions, etc.

I scraped some shot data from CBS. Was about 2/3 of the games for UCLA. Once I just need to figure out how to plot the dimensions to an image, I'll post that later if I can see anything else from it.

This stuff is fun. I hope everyone can enjoy it. And if they can't model -- no one should put them down -- if they are willing to do some homework -- teach them your model and/or show them how to do their own model.