Any Thoughts on Chad Ford's "Grantland" piece?

[Jeff Fogle]

Sorry for the confusion. I meant a median for each place where you had used the word average when you were listing the win share data.

As in:
Top guy 12.3
average of top 25th percentile of player population 5.9 (median xxx)
avg of 25-50th percentile 2.9 (median xxx)
avg of 50-75th percentile 1.1 (median xxx)
avg of 75-100th percentile 0 (median xxx)

Did the medians also sink downward to the same degree (like 4-3-1-0)? Or, were the medians consistent across the board (say, 0, 0, 0 for the bottom three quadrants, with outlier guys influencing the averages) and the difference in averages were the result of 1-2 outlier players lifting the averages?

JF

[Crow]

I understand what you are asking now. I'll probably go back and add it but I need to do something else for a bit.

[Jeff Fogle]

Thanks crow. Please don't bother if it's a big hassle.

Just saw this that others of you may have heard about already. Looks like Minnesota didn't do due diligence on their 2nd round pick

http://minnesota.sbnation.com/minnesota ... draft-pick

[Crow]

It would be a bit more spreadsheet manipulation hassle than I am willing to go thru right now to find the 'median WS per 82 games guys' for each quartile but I can find and will list in the above data the 'WS per 82 games' of the 'median guy by total career WinShares' for each quartile and maybe that will roughly help with what you are interested in.

[Crow]

It does appear that the % of top 15 picks (close to the lottery cut Ford used) that were bottom 50% on WS performance increased over time:

early 90s rookies 21.6%
late 90s 19.7%
early 2000s 31.0%
late 2000s 36.0%

This matches up pretty closely with his statement about increased "busts" and it is coming from significantly fewer top 15 picks turning out to have top 25% performance (which matches up pretty closely to the message of his All-Star data).

Second quartile outcomes have only declined slightly across the time periods.

I guess he defined sleepers as late first round and 2nd round (I didn't noticed that earlier). If late first round was about 20th then it appears that the sleepers are achieving top 25-35% percentile performance or thereabouts and this analysis pretty much matches up with his on this part of the distribution too.


It also does appear that the frequency of top 40 picks over 3.0 WS per 82 games is probably falling from period to period, so the % under that level is probably rising and that may fit with Jeff's commentary of leveling of talent with a lot of guys at a modest level at best.


Why the difference between my note about no drop off in the # per decade (roughly) with 30 WS and a WS/48 of .125 over their first 5 seasons (from MikeG's data post) and yet the dropoff that Ford computed and the one I computed in this thread? I am not sure.

Maybe the distribution has a sharp inflection point. There are other differences in that MikeG's numbers were for first 5 years whereas I used career to date. The 90s guys all got their peak years, the first half of the 00s have too but not all of the late 00s guys have or have as fully. Also Mike G use WS/48 whereas I used WS per 82 games. I am not sure whether Ford used a comparable set of early years in player careers or full careers and whether he looked at per minute impact or per season impact on wins or something else. If average minute levels per game changed (especially early in a career) that could be another partial explanatory factor.

If Chad were willing to give a bit more background information here or in a follow-up article that could help. My analysis here could also probably be extended / adjusted. But I'll stop there for the time being.

While I had initial questions and concerns about details of the article and still have some, I do think he is bringing attention to a topic that it would be valuable to understand better and more fully.

[Mike G]

Crow and I both decided to go with 5 year intervals. From this page -
http://bkref.com/tiny/OtCyD
I got per-game career averages, games, and Win Shares for draft picks 1 thru 10, and 11 - 20;
to see if there was any improvement in the selection of the top 10 picks (expected to be good players) and the 2nd 10 (hopefuls).
I generated (approximate) totals and divided by seasons played. This doesn't detect 'busts' with abbreviated careers, but it adjusts for later draftees whose careers are ongoing.

Code: Select all

Draft Picks    Career     Per Season Averages

#1-10      yrs     G     Pts    Reb    Ast   Stl   Blk      T      WS
1990-94   11.9    725    789    341    168    60    46    1404    3.97
1995-99   12.0    780    963    376    206    65    48    1659    5.30
2000-04    7.9    489    813    347    147    53    49    1409    4.28
2005-09    3.8    248    851    324    187    56    41    1459    4.10
                                    
#11-20     yrs     G     Pts    Reb    Ast   Stl   Blk      T      WS
1990-94    7.8    425    482    206    105    43    21     857    2.41
1995-99    8.8    491    564    233    114    42    30     983    3.06
2000-04    7.0    390    505    243     93    37    29     908    2.39
2005-09    3.5    198    457    186     73    34    24     774    1.81
                                      
ratio      yrs    G      Pts    Reb    Ast    Stl    Blk     T     WS
1990-94   1.52   1.71   1.64   1.66   1.61   1.37   2.15   1.64   1.65
1995-99   1.36   1.59   1.71   1.61   1.81   1.55   1.62   1.69   1.73
2000-04   1.13   1.26   1.61   1.43   1.58   1.43   1.69   1.55   1.79
2005-09   1.10   1.25   1.86   1.74   2.56   1.66   1.69   1.89   2.27

In the first 2 tables, T = Pts+Reb+Ast+Stl+Blk
At both levels, the 95-99 era looks like much better drafts than the previous era. And weaker since.
The 3rd table, ratio of top/2nd tier draftees, would seem to indicate the last 5 year interval has been the best at discriminating draftee talents.
Picks 1-10 tend to go to the worst teams, and WS have been continuously higher relative to picks 11-20.
On a weaker teams, a given player will get more minutes, but Win Shares per minute are harder to come by.

[Jeff Fogle]

Mike G., do you think there's enough evidence in your chart to suggest that the quality of the talent pool is diminishing in this decade?

Win Shares Sums 1-20
1995-99: 8.37
2000-04: 7.67
2005-09: 5.91

Or, is that sum decline for players 1-20 just a result of having fewer years and games in the sample?

[Mike G]

Jeff, those are per year averages. Indeed, it seems the late '90s put a lot of great players into the league, some of whom are still going strong.
Your 'flattening of talent' seems to have been a decent intuition.

Still, that youngest group may not have had many years in their prime.

[Mike G]

[A less hurried response]

Win Shares Sums 1-20
1995-99: 8.37
2000-04: 7.67
2005-09: 5.91

This does indeed look like dramatic decline. But Win Shares are funny. If in fact teams with 1-10 picks are getting better with their scouting, then the best players are going -- with fewer exceptions -- to weaker teams. And such an environment will be quite the drag on a player's WS.

In his first 2-6 years, it's likely that the majority of a player's career has been with the team that drafted him. In 7-17 years, increasing likelihood that his teams have been above average -- if he's an above-avg player.

Per-season average (Pts+Reb+etc) productions show a much less precipitous trend:

Code: Select all

drafts    #1-10   10-20   Tot.   ratio
1990-94    1404    857    2262    1.64
1995-99    1659    983    2642    1.69
2000-04    1409    908    2317    1.55
2005-09    1459    774    2233    1.89

Again with the ratio of (1-10)/(11-20) productions for each interval.
That late '90s bunch: Garnett, Duncan, Dirk, Allen, Pierce, Billups, Marion, Sheed, Carter, Iverson -- and in the 2nd 10, Kobe, Nash, Peja -- is proving very hard to beat.

The 1.89 ratio at the end would seem to be quite a spike in improved drafting. It's a sample of 50 1-10's and 50 11-20's, and it's not really 'lottery picks' vs the rest. Rather, it's a consistent number. The number of lottery picks has risen, from 11 non-playoff teams in 1990, to 15 currently. This would seem to dilute the sample in recent years.

[Jeff Fogle]

Would this be a fair assessment?

*When comparing this decade to the last, the talent pool has declined, perhaps because the '95-99 hunk was an outlier featuring an abnormally large volume of high impact star talent.

*Within that declining talent pool, scouting has done a good job of separating the top 10 picks from the second ten picks. As you said, "The 1.89 ratio at the end would seem to be quite a spike in improved drafting."

*And, scouting has done a good job of finding more sleepers beyond that range. As Ford said in his piece, "In the '90s a typical draft produced 5.6 sleepers in the late-first or second round....This decade? That number jumped to 8.3 sleepers per draft."

So, rather than "information overload" making things worse, there's compelling evidence that the increased emphasis on analytics has improved drafting. The top 10 have been better than the next 10 in this decade. There are more sleepers this decade than in the past.

Plus, I guess, reduced access may not be a real penalty. Drafting is getting better once you adjust for context. Any set of reasons given for "drafting getting worse" have to be questioned if "adjusted for context" analysis shows that drafting is actually getting better.

Is this a valid assessment to this point in the conversation? Anyone want to add in additional evidence/thoughts?

[Crow]

One team finding sleepers (is this more the best teams picking 21-30 or whoever picking picking 31-40, maybe the worst or best depending on pick trades) is another team missing them earlier (at 11-20 or perhaps even higher, the average or worse teams).


If one wanted to try to assess the impact or possible impact of "information overload" and one assumed that teams with designated, known and recognized "analytic staff" would be the most likely "victims" of this, there are folks with the rough memory of when that staff was hired and with that information a draft performance comparison could be done for the last few years of with designated analytic staff and without (and maybe also split to good performing team and not and regarded above average GM and coach and not) or probably better it could be done a few years from now when there is more data about the performance of the picks from the last few years.


There probably are ups and downs in draft talent levels, at the top and the degree of depth and even the degree of predictability that makes it tough to evaluate the evaluators and their methods and working conditions.

It is probably unwise to be too convinced of a pattern or any explanation(s) with 2-4 time periods. Though that is probably often what you have to work with and data from longer ago might start to lose its comparability or usefulness.

[Jeff Fogle]

Not sure if I agree with that crow...if a sleeper later in the draft should count as a "miss" for the teams higher in the draft. Depends on how you define "sleeper" I guess. I mean, if a Dywane Wade caliber talent went 33rd or something, that's obviously a big miss. But, if we're talking about role players who fit a system well, then blossom within that role...that could be attributed to good scouting amidst a very dense sample size of relatively even-looking choices.

Let's assume 5-10 potential draft picks seem to have an edge on the field...but then there are dozens of guys (100 or so) who are hard to differentiate given where they're from, their age, their experience, their mix of strengths and weaknesses within the skill set. There are always undrafted guys who are hard to separate from the last 15-20 choices.

I could see where one might interpret the number of sleepers going up as a strike against the lottery teams. But, I think it's also reasonable to assume that the number of sleepers going up is a sign of improved decision making when choosing amongst the dozens of choices once the "obvious" options are gone. The nature of the beast is that many of the "obvious" choices are still gambles. They're just gambles you're supposed to make given the information at hand (a super-talented kid with maturity issues for example).

To this point in the discussion, if you accept the premise that the talent pool doesn't match what we saw a decade ago, we have some evidence that teams are choosing well within that first hunk of 20 choices (first 10 are outperforming next 10), and that teams choosing later are finding more sleepers amidst the morass of relatively equal options.

Or, this probably sums up the positon best:
"Best Available" has been better evaluated in the first 20 picks
"Fills a Need" has been better evaluated with the remaining picks

Not sure if a "fills a need" success story should be counted as a miss against a lottery team emphasizing "best available." Obviously this simplifies things. The real world is more complicated because lottery teams often aim for a need...and sometimes a talent drops later on into somebody's (the Spurs) lap. Think it's a reasonable enough big picture representation though.

And, I'm very concerned about drawing negative conclusions without adjusting for context. Imagine that in Spring of 2010 major league baseball hitting coaches came up with something that helped batters improve 2%. It was a REAL improvement that caught on across the league and all hitters not only improved, but held onto that improvement in 2011. Comparing the years 2009-2011 would still look pretty crappy offensively compared to the steroid era. And, a writer saying "baseball hitting coaches keep tinkering, but offense is much worse than it used to be" would be doing a disservice to the hitting coaches who created real improvement. The outlier explosion of talent in 1995-99 in the NBA as measured by Mike G. may be influencing perceptions in a similar manner here (not saying it was steroids or anything...just a solar burst of production that happened to come in that time period). It's possible for scouting to get better as the number of superstar draft picks goes down if the talent pool isn't producing superstar options in the first place.

Agree with you that it would be very helpful to have a list of analytic teams that we could use to compare to specific picks...

[Crow]

Yes there are more than one way to look at sleepers.

Many possible descriptions, many partial explanations of details and the whole game.



There can be years where what teams need and what is available when they pick will match up better than others. Easier picks and tougher ones (take BPA or best remaining player who fills a specific need).

I would guess there are some GMs and analytic staffs who have thoughts about future drafts (beyond just the next one), the general talent level and talent level in specific draft segments, based on detailed work and use those thoughts to guide or adjust actions they take long before those drafts occur to try to improve the match-up of need and opportunity and others who don't make as much effort to develop foresight and take advanced action.

With teams emphasizing the use of rookie contracts for greatest value, there might be a developing pattern of restocking a position / role with rookie talent / rookie contracts 2-3 years after you last did it even if it was successful. Maybe it was always been that way or maybe the timeframe for restocking has gotten shorter and maybe rookies are being used more frequently?



A really strong and firm description of the world of drafting and discovery of what works better and worse would be desirable but would probably take a lot of experience and a lot of careful reflection and analysis. Probably would be good to have on paper and review and refine it if one was doing it professionally.

Conceptually more information and more analysis would seem helpful but digestion of that information and prioritization and reconciliation of it are phases where mistakes can be made. especially perhaps if the process of decision-making is not clearly and rather precisely defined or understood.


If one wanted to continue to expand the research agenda one could look not just at the average importance of various attributes in the translation from college and international play to the NBA but also try to identify attributes or sets of attributes that appear in higher concentration with the highest performers, biggest sleepers and worst busts.

Ever more than could be considered and perhaps for gain with good analysis and judgment about how to synthesize and use it.

[Crow]

Sort the data Roland Beech presented here by draft year
http://www.82games.com/bestnbadraftyear.htm

for the average actual performance of the specific player pick (using a simple, crude career rating with the issues that raises) vs the expected performance or average performance of all draft pick of the same number and the trend is straight down / declining average actual performance vs expected from 94-99 to 00-04 to 05-08 for the top table for all picks and the second table for just the top ten. The decline in performance from 94-99 to 00-04 is maybe about 4% and about 10% from 00-04 to 05-08.

The data is there for 7 bands of performance. For all picks, the number of superstars and stars picked is straight down though far more in the transition from 00-04 to 05-08 than the first period transition from 94-99 to 00-04. The number of solid picks is straight up though much more modestly than the downward movement in the first two categories. Up more in the first transition. The number in the 4th best categories up modestly then down more significantly. Straight down in the 5th. The number in the 6th best category up modestly, the number in the lowest category up more strongly. The trends are basically the same for just the top 10 picks though there was a higher % of decline in average rating in the first period transition here than in the second one.

For all picks and just the top 10, average actual points vs expected went straight down. Rebounds up. Assists up then down but ending up still above where they started in the first 5 year period.

Did the teams get more rebounds and assists because they prized those activities and skills more or did it just happen? Did they not prioritize scoring enough? Are these draft results in contradiction to the claim that decision-makers generally value scoring too much or are there other things involved like old guys taking more of the shots in later time periods and getting more of the points than the newest guys or something else?


Are there few GMs who were NBA players or NBA players for a significant time and at a significant level than earlier? Do coaches have as much input on draft decisions on average as in the earlier time periods, more or less? If one computed these values, these additional factors could be added to the longer and more complicated narrative.

What to make of the last 4 title winners probably being low of key guys directly added thru their own drafts? Probably not much given what San Antonio and Chicago did earlier with 3 or more key draft picks leading the way but I mention it because there is additional research that could be done on that sub-topic.

Based on this other article http://www.82games.com/nbadraftpicks.htm
the average per game points, rebound and assist performance for the top 10 are higher then the next 10 and then the next 20 and the next 20 and dramatically so ... but it is perhaps more interesting to look per minute. The picks from 11-20 averaged about 12% less points per minute in their career than the top 10 picks and 9% less assists but 4% more rebounds. Rebounders who get to play may get picked more from 11-20 than in the top 10.

Scoring per minute falls off from the first 10 to the second, then more modestly to the middle 20 and the lowest 20 draft picks. Rebounding per minute continues to go up for the 21-40 subgroup and then again for 41-60. Assists per minute go up for the 21-40 subgroup but then drop-off to a sub-group low from 41-60.

There is more than can be done with regard to the biggest overachievers and busts and I found some interesting stuff but I'll hold off on posting that at the moment. And of course a more detailed study beyond pts, rebs and assists could be assembled with further effort.

[gfarkas]

It would be a bit more spreadsheet manipulation hassle than I am willing to go thru right now to find the 'median WS per 82 games guys' for each quartile but I can find and will list in the above data the 'WS per 82 games' of the 'median guy by total career WinShares' for each quartile and maybe that will roughly help with what you are interested in.

In Excel, the MEDIAN function returns the median of the given numbers inputted into it.