APBRmetrics

Great stuff everyone!

here's the link for anyone who didn't see it:

http://www.draftexpress.com/article/Ana ... art-2-5068

A few places I referred to the DX mock, though I should have said DX100 since that is what is being presented here. The mock is worth talking about too but sorry for the mis-reference.

Crow wrote:Compared to the DX mock, the analytic model consensus makes a case for move up attempts for one or more of Jones, Portis, Looney and Wright. It doesn't happen often. Might be most likely for Portis (with Jones's size, Wright's age and Looney's asthma potentially tempering enthusiasm). Who from 8-13 is most likely to be willing to trade down? I dunno. Who is most likely to try to trade up? I'll guess the list includes prominently the Celtics, 76ers, Bucks, Rockets and perhaps Spurs. Some others that would like to probably don't have current assets to pay the bill.

If I were a GM I would probably try to trade down often the moment the few guys I really loved were drafted. Try to stock future picks & still get guys I still like (more than other teams) a little later.

Saw this http://www.nba.com/pacers/made-mockery

Pacers outperforming the Indy Star doesn't surprise. How did they do compared to DX mock? I was going to do both first and second round but second was not worth much for either beyond Stephenson. Looking at first round 2008 to 2014, Pacers took better pick 3 times by career ws/48, DX the other 3. Pacers took the 2 best players and fifth; DX 3rd, 4th and 6th. Advantage Pacers. But it is fairly close and not a rout. How did they do against DX100? I couldn't quickly find that on internet. It might be found with more search or Jonathan's assistance. My impression going in and now is that the Pacers have been above average at drafting. If you did team level analysis for all I suspect from the previous analysis of actual draft order vs. DX mock that DX mock would probably be better well more than half the time, or at least half. I am more interested in analytic model or analytic-mock blend performance of course.

Statman, as one of the longest operating modelers, how does your model's picks for when the Pacers picked perform for this time period or longer by my measure or your own?

According to tothemean.com, Kevin Pelton's model, for better or worse?, has the lowest correlation with the model consensus, beating out Stephen Shea's CPR slightly. Highest correlation? DX100.

If anyone is looking for international data, I created the following spreadsheet with raw boxscore totals including everyone in the DX Top-100: https://docs.google.com/spreadsheets/d/ ... sp=sharing

Let me know if there are other stats you'd like included. I have some more advanced stuff, but it's mostly derived from their team's boxscores.

Thanks, that is handy.

no problem. I went ahead and added whatever we have going back to 2011 in the different tabs now as well if that helps at all

jgivony wrote:no problem. I went ahead and added whatever we have going back to 2011 in the different tabs now as well if that helps at all

Any idea where to best get complete league data (all individual player stats) for these overseas leagues? I'd need that to normalize rates relative to league - all leagues & stat keeping are created different..

Also - there are no foul totals in the stats - I don't know if that's an oversight or those aren't kept as regularly in overseas box scores?

Upon further reflection I now see that the extreme tothemean.com correlations might be reasonably expected given that the consensus has a 25% weight for DX (higher than the 9 plus % for each of the analytic models) and no weight for KP's. But they are still extremes worth noting, as would have anything that could have (but didn't) sufficiently bucked the pre-disposition to turn out this way.

I guess there could be smart trade downs for the guys with analytic model projections above the mock projection or just patient waiting. A lot depends on your projection of other teams' projections and actions. A model of what others will do would be helpful. A mock is a loose model of that, which might be made better with rigorous analysis of past drafting tendencies of teams alongside consideration of current needs. Maybe the best mocks already use historical tendencies to some degree?

Statman wrote:

jgivony wrote:no problem. I went ahead and added whatever we have going back to 2011 in the different tabs now as well if that helps at all

Any idea where to best get complete league data (all individual player stats) for these overseas leagues? I'd need that to normalize rates relative to league - all leagues & stat keeping are created different..

Also - there are no foul totals in the stats - I don't know if that's an oversight or those aren't kept as regularly in overseas box scores?

I have a database that a number of NBA teams subscribe to. It has stats from basically every league in the world. You can see bits and pieces of it utilized around the website (for example the stats on the mock draft) but its mostly private. When you say all individual player stats, what are you missing besides fouls (I'll get that for you). I can maybe provide that for guys in the top-100.

BR has data for 6 leagues / competitions. http://www.basketball-reference.com/euro/ It appears that they are complete but I haven't used a lot to notice potential gaps.

How does offensive rebounding translate from college to NBA for guys 6-8 to 6-9 vs 6-10 or above? Which measurable correlate best for NBA OR%, height, weight, standing reach, no step vert, agility, bench press, etc.? Or what is the best fit algorithm of these?

jgivony wrote: I have a database that a number of NBA teams subscribe to. It has stats from basically every league in the world. You can see bits and pieces of it utilized around the website (for example the stats on the mock draft) but its mostly private. When you say all individual player stats, what are you missing besides fouls (I'll get that for you). I can maybe provide that for guys in the top-100.

Well, my work uses all player stats (every player) in a given league - so I need all players in every league I may be working with. It's how I roll. Actually - I use all that data to normalize stats for the league & each team, as well as then rate/rank players relative to their peers while I'm at it. Totally fine if you don't have all that - I'm not one to look a gift horse in the mouth - the fact you are offering the stats you have already presented is very cool & hopefully will help many.

I'm just a stickler for as complete a context as I can have - it's like taking some random statline from some minor league player in baseball (w/ his age) & being asked how good a MLB prospect he is. Without knowing/calculating the level of minor league, league averages, ball park effects, etc. - I just can't put together a good projection.

If you could include fouls in the spreadsheet you have - it would be great - I would think most of the models use foul rate. If not, they probably should - it can be a telling stat in my projection, particularly if high.

I can pull together league data for the Euros from basketball-reference, I believe it is complete enough for me to have context & team ratings. I really want all the China pro league stats - I think there are enough ex D League, NBA, & college players there that I may be able to create decent conversions (from the projections for '15 for all those guys the work backwards from the China ratings).

All that being said - I very well won't be happy enough with the information I'm working with in this short amount of time to include internationals anyway in my model. For now I am doing the retrodiction of the last 17 NBA drafts of every college player who was drafted or played 1 minute in the NBA. That's 1035 players in which I use 2 weighted seasons of college ratings to project their NBA career potential. It is a beating fixing all those college stat errors - but I'm rolling along - hoping the projections of those 1035 guys to be done Monday or VERY soon after. Lord knows how long the write up will take me - so many interesting ways to present all that information.

All those past projections will give me great context for all the current prospect projections - relative ranking of current players to the past, similarity scores, maybe finding player types that might over/under project, etc.

Best case scenerio - I get the internationals into my projections just before the draft (within two days of the draft). Not nearly early enough for a good write up for you all, however much I wish I could.