APBRmetrics

I pretty much went off the cuff with this for a few hours last night - and this is what I posted:

http://www.pointguardu.com/f136/statman ... post382787

Followed with some results and a little analysis:

http://www.pointguardu.com/f136/statman ... post382793

I want to do more - actually, much more. My files got corrupted emailing them to myself, so I won't have access until Saturday night at the earliest. So, I'll have to wait until then to do a breakdown of all the players drafted (and a few extras) with their ratings broken down into various skillsets: scoring in various ways (2pt, FT, 3pt), rebounding, handles & passing, and defense - all withing the context of team SoS & pace. That might shed more light about players in differing specifics that my ratings above kinda gloss over in their general single "Score" rating.

The results look really impressive, Statman! They certainly pass the smell test. Basically, it still looks like Norris Cole is a serious sleeper.

I note a ton of seniors to the bottom of the list. Red flags for Brandon Knight?

Here's the funny thing. I'm starting to hear that Kemba Walker is falling, maybe even out of the lottery. I'm beginning to think the Warriors would make a big mistake passing on him even though we have Curry. I know there are mixed opinions on him, but I can see him as a Tim Hardaway type player. I fully believe in the tier strategy, and if Kemba is close to the top tier and is on the board when we pick, we'd have to do it, I think. Right?

Having said that, I've also really come to like Tristan Thompson. If both him and Kemba were still on the board, that would be a tough decision for me.

The Warriors will probably get Marcus Morris, who I don't really like, but your stats suggest will be good.

You know, I think I would just use your "potential" here, rather than also incorporating production. Essentially, what you've done is added an aging curve. The existing rating probably predicts the next year pretty well....

One thing I have found regarding the draft, that may surprise people, is that younger rookies do not perform worse, given their draft slot! In other words, players theoretically drafted for potential have, by the next season, already realized enough of it that a 19 year old #5 pick is projected to perform the same as a 23-year-old #5 pick.

I like the general approach and the final scale.

I wondered though if the ratings of say the top 10 guys of each year's class (the guys most likely to matter in the draft) varied by as much as the entire classes do. You provide the data so I checked. They appear to vary less, I think by at least 25-30% less on "Rating". That might be an argument for reducing the class based Draft adjustments. But I leave it to your consideration and possible fine-tuning.

DSMok1 wrote:One thing I have found regarding the draft, that may surprise people, is that younger rookies do not perform worse, given their draft slot! In other words, players theoretically drafted for potential have, by the next season, already realized enough of it that a 19 year old #5 pick is projected to perform the same as a 23-year-old #5 pick.

This is probably more true since the rule that draftees must be at least one year out of high school. Apparently that one year is sufficient to put kids on the cusp of their potential to be NBA players. Training camp and the rookie season itself then tends to indicate their talents.

DSMok1 wrote:One thing I have found regarding the draft, that may surprise people, is that younger rookies do not perform worse, given their draft slot! In other words, players theoretically drafted for potential have, by the next season, already realized enough of it that a 19 year old #5 pick is projected to perform the same as a 23-year-old #5 pick.

I actually think this may be related to that player's ability to adjust to an immediate step up in competition. That 19 year old drafted usually is a kid that just came off a season in which they had a big step up in competition and were good to great immediately (1 & dones, 2 & dones). They show more of an ability ("talent"?) to handle that immediate competition upgrade. They didn't shy away from it from HS to college - and they don't shy away from it as much from college to the pros.

The older kids often have been playing against the same competition for a few seasons (junior/seniors in college), they've learned to get better (gotten comfortbale with?) through time against the SAME level of competition, and it's been a while since they've made that big step up. So, those first couple years in the pros are something they aren't used to - something they can't relate to as well since it had been 3 to 4 seasons since they made a similar move in competition upgrade.

Also, teams drafting 18/19 year olds might show a much longer leash through the growing pains - they want to get that kid his minutes to hopefully reach his potential quicker. The teams drafting older players might be drafting need - looking at the draftee as a specific role guy who may eventually develope in to something better through time.

I think the "potential" side of your ratings most accurately reflects the draft value. I would use 2 parts Per Minute, and 1 part Rating. However, seniors in the potential rating are still "overrated" from a draft perspective, because most players that rate highly here that are seniors are seniors only because of some perceived deficiencies in translating their game to the next level. Otherwise, they would have come out earlier.

DSMok1 wrote:I think the "potential" side of your ratings most accurately reflects the draft value. I would use 2 parts Per Minute, and 1 part Rating. However, seniors in the potential rating are still "overrated" from a draft perspective, because most players that rate highly here that are seniors are seniors only because of some perceived deficiencies in translating their game to the next level. Otherwise, they would have come out earlier.

I definitely see that argument for sure - but I don't think I'd want to take it further in adjustments (yet?).

HOPEFULLY, by next season - I'll have a good 12 seasons of college data - and I'll have a much better idea how to look at this. I agree - the "potential" rating seems to reflect what GMs are looking for. I don't want to almost completely discount the very last season - because all these guys are well below their prime age - and the last season was their latest performance (not projected).

I have another way to look at it, I will delve into it Saturday night - it MAY shine an even brighter light on the younger kids. We'll see.

Part of the reason you see so many seniors high even in potential is that there just weren't that many good early declarers this season. Sullinger of Ohio State and Jones of UK would have been at the top (Sullinger 1 or 2, Jones 4 or 5). Barnes & Lamb of UConn would have been higher than Knight. Also - I'm just plain using my college rating - MAYBE skillsets should be weighted differently for NBA potential - I'll have a much better idea about that when I get the past college seasons and crunch the numbers.

Does anybody besides Draft Express and other sites like that have access to Synergy stats for college players? I think those stats would tell you just as much about what players will be successful in the NBA, because certain skillsets are more valuable than others.

My latest post:

http://www.pointguardu.com/f136/statman ... post383378

I'll copy here - since it doesn't have charts (which I never seem to be able to format right):

Well, I had a eurika moment last night in my looking at a different approach on ranking players in terms of "potential" based off my college ratings. I kinda went all over the place the day of the draft trying to best work it ( http://www.pointguardu.com/f136/statman ... post382793 ) - it took the day AFTER the draft when a simple (relatively speaking), yet I think quite sound, approach hit me. I crunched the numbers - and they completely passed the laugh test as far as I'm concerned.

The basic issue is how to best figure out "potential". Well, I have all my college players from last season above 300 total minutes ranked. I have ratings for all of them, that I can seperate into individual statistical skillsets. I know the college class of each player. So, I decided to figure out where each player ranked (by my ratings) by their class in each of 7 different statistical skillsets - 2pt scoring/efficiency, FT scoring/efficiency, 3pt scoring/efficiency, rebounding, ball control/passing, defense, and intangibles (pretty much playing time adjustment relative to team quality). I gave each player a % score for each skillset relative to their ranking in their class - 100.0 would be the #1 ranked player in their class, 0.0 the worst player, everyone else in between (#2 like 99.8, #3 99.7, and on down). I now have establish rating totals from last season in each skillset for each class at different % plot points. From there, I could reasonably project a theoretical "potential" rating (the rating you'd expect to see them with for a senior season) for every underclassman. The seniors obviously just keep the rating they achieved last season.

I know that sounds confusing - so I'll use one example to illustrate. Here's Derrick Williams % rank last season in each of his skillsets among sophomores:

2pt: 99.4, FT: 100.0, 3pt: 84.7, Reb: 96.5, Hand/passing: 5.3, Def: 75.0, Int: 0.2

So, he's exceptional from the field, the best in getting to and converting from the line, a very good rebounder & three point shooter (he needed to shoot MORE threes to get in the 90s), solid in the defensive stats, quite poor in terms of his assists & turnover rates, and had one of the worst "intangible" scores.

Again - intangible is just a playing time adjustment relative to a players per minute production an team quality. It's merely the difference between a player's rating per minute and the final rating. Derrick was a 225 per minute (over 2.25 times more productive than an average player), a 192 final rating (only 30 mpg) for a -32 intangible (after round off). -32 is one of the lowest in the nation, a product of being an insane per minute producer while playing lesser minutes than one would like.

Anyway - we have Derrick's % ranks - so we can take those and see what rating totals we get for seniors at each skillset at those points. Actually, it's not always seniors - I actually took the HIGHEST rating at each plot point (at or above the respective player's class) - since sometimes the very best weren't seniors, although usually they were.

OK, back on point - Derrick's actual rating for each skillset, and then his "potential" one:

Actual : 84 2pt, 66 FT, 36 3pt, 52 Reb, -23 BH, 9 Def, -32 Int, 192 Rating (sum of skillsets)
Potential: 94 2pt, 66 FT, 40 3pt, 52 Reb, -19 BH, 11 Def, -32 Int, 212 Potential Rating (sum of skillsets)

Notice his FT stayed the same - he had the best FT score in the nation - it was higher than the upperclass plot points, so he keeps his rate (having it go down would make no sense). Same with rebounding - rebound rates actually change very little on average through each class - kids usually rebound about the same per minute (generally speaking) each year as they go through college - so rebounding doesn't tend to go up much or at all in "potential". Expected improvements would have been 2pt, handles, 3pt, and defense.

As you can see, certain skillsets have a small potential for increase (rebounding, 3pt shooting if you don't take threes, and defense among already elite shot blockers/ball theft guys). Others - 2pt, FT (except for Derrick & Kyrie Irving), maybe hands/passing, maybe intangibles can be expected to improve pretty solidly.

Anyway - all that being said - I really think this is the best way for me to approach this. We aren't increasing everyone's overall rating the same dependent on class - we are PROJECTING where players may be relative to their skillsets. Certain types of players (based on how their skillsets break down) will have greater potential for a ratings increase as they progress through college.

Now, I CAN and will improve this through time. When I actually get all the past college seasons, I'll have tens of thousands of plot points spanning the classes - which will make the projections even more accurate. Derrick Williams' FT rating potential would increase if I had him plotted with all sophomores since 1998, against all seniors & juniors since 1998. Also, this would give me the possibility of projecting player performance from one season to the next (rating wise) - which would be fun to play with.

The fact that I can do NBA ratings, and seperate them among all the same skillsets, will allow me to start being able to project how player's skillset ratings change (on average) from college to the NBA - allowing me to better tab why TYPES of players have more NBA upside.

Also, I can do the same approach for NBA players from one age group to the next, allowing me to project a player's entire career (rating wise) as he ages - or project the entire next season of NBA player ratings (after another year of aging), for preseason team predictions.

Like I said - a "eurika" moment. I'll post my revised 2011 NBA draft player rankings here shortly.

Here were the results - ranking the players by my new potential rating outlined above:

http://www.pointguardu.com/f136/statman ... post383387

After thinking about this methodology you have just outlined, I don't think it works right. Why? Selective sampling. The best freshmen leave after one year. The best sophomores leave after 2 years. The best juniors leave after 3 years. And the seniors graduate.

Therefore, comparing the top freshmen to the freshman class is not the same as comparing the top seniors to the senior class. Most of the players who WOULD HAVE been top seniors have long since jumped to the pros. For instance, Blake Griffin would have been a senior this year. Therefore, a player that is a top senior, the top senior, is not as good as a top freshman.

I liked your previous methodology better; a coherent aging curve would work even better.

DSMok1 wrote:After thinking about this methodology you have just outlined, I don't think it works right. Why? Selective sampling. The best freshmen leave after one year. The best sophomores leave after 2 years. The best juniors leave after 3 years. And the seniors graduate.

Therefore, comparing the top freshmen to the freshman class is not the same as comparing the top seniors to the senior class. Most of the players who WOULD HAVE been top seniors have long since jumped to the pros. For instance, Blake Griffin would have been a senior this year. Therefore, a player that is a top senior, the top senior, is not as good as a top freshman.

I liked your previous methodology better; a coherent aging curve would work even better.

Completely agree - except I do believe with a much larger sample including previous seasons - this could work VERY well. I think as it is - it still probably underprojects underclassmen because of the very reasons you stated. Now, there is something to be said about imediate impact - the freshman's ratings is a 'potential" rating, unreached - the seniors is their current rating. So, I'm ok with the results for now - since it's all I have to work with.

Here - I'll quote what I said to another poster on that board about my plans with this - tell me if you think this extension of the current simplified methodology would work - at least a bit better:

Agreed. This is what I have to do with just one season's worth of data - one could suspect the underclassmen here to even outperform these projections - since the very elite 22 year olds that would have been seniors are pretty much in the pros and not in this data set.

When I get all the past seasons - I'll be able to set up my groups for my data points in which you have the exact same players from frosh to soph, then the same sophs to juniors, etc. Thus you won't have a possible more talented freshman group that sends some guys to the pros progessing to a slightly lesser talented soph group, and so forth to the seniors. I could then take a new freshman, plot him - and most likely having so many plot points from one class to the next, using the same data groups from one class to the next, I'll get a much more true projection of a player's potential. This type of projection I would suspect would give younger guys like Kyrie, DWill, Thomspon, Knight, etc. an even larger potential rating. I don't think it'll be alot larger, maybe 5-10% - who knows really until I do it - but it'll also be a better projection across the board (all the skillsets) in general. It would definitely project true 1 & done type talent better - or should I say with less margin of error.

An example would be DWill. If I had, say, 14 seasons worth of data - I'd have, I dunno, probably over 7000 qualifying freshmen that went on to be qualifying sophomores. I could plot DWill's ratings (he won't be hurt and projected a little lower by having to be plotted with quality one & dones - they weren't qualifying sophs), and then get the corresponding sophomore projections. THEN I could take that result - plot them in a seperate, say, 8000 player group that were qualifying sophs and then juniors, and get the junior projections. So forth to senior - and most certainly DWill could see more of an increase across the board than he did here - especially with his FT rating (which is maxed out here due to his being the best of a one season sample size).

I definitely look forward to seeing your retrodictions of players from previous drafts, to see if this system makes sense after, say, three years.