2015-16 Team win projections

[Statman]

Why am I doing so much crappier in kmedved's rankings than in Mike G.'s or caliban's ?

I'm just the opposite - near the top in kmedved's, near the bottom in the others.

[Dr Positivity]

After BasketDork and Statman's posts things didn't look right so I plugged in the Average column in kmedved's post into my own spreadsheet and got this (Average Error)

Code: Select all

AJBaskets - 4.7
Nylon Calculus - 4.8
tarrazu - 4.87
ampersand5 - 5.0
kmedved - 5.2
nbacouchside - 5.2
DSMok1 - 5.27
fivethirtyeight - 5.27
Kevin Pelton (Wins) - 5.32
CBSSportsline - 5.53
Crow - 5.6
fpliii - 5.6
rmth - 5.6
caliban - 5.67
AcrossTheCourt - 5.7
LV Westgate - 5.77
Jim Jividen - 5.78
MikeG - 5.8
bbstats - 5.9
ESPN Fall Forecast - 6.01
Sports Illustrated - 6.03
Yooper - 6.13
italia13lacio - 6.17
Accuscore - 6.17
numberfire - 6.2
BasketDork - 6.2
ESPN Summer Forecast - 6.47
nrestifo - 6.8
EvanZ - 6.97
Dr. Positivity - 7.07
Bleacher Report - 7.2
Statman - 7.27
Sports Formulator - 7.3
tacoman - 7.47

I'm not sure if the discrepancy is because of an error in the table, or if the new column for strength of schedule was used to impact the projections

[caliban]

Added and ranked it by RMSE for this week. Outside of statman it looks about identical to what's been displayed.

Once again, if you spot an error on my part in your line or whatever, just let me know and I'll do my best to correct it.

[Statman]

Added and ranked it by RMSE for this week. Outside of statman it looks about identical to what's been displayed.

That's because my genius is unquantifiable....

[Mike G]

Relative to b-r.com's projections, the rankings are quite different if you have an exponent on the errors.

Code: Select all

average errors
AJ   4.92      bbs   5.52      MG   6.19
tzu  5.02      Crow  5.64      yoop 6.28
Cal  5.16      fpli  5.75      EZ   6.75
KF   5.22      rsm   5.97      nr   6.76
DF   5.26      BD    6.12      DrP  7.21
km   5.32      itca  6.16      taco 7.38
DSM  5.35      snd   6.18      Dan  7.44


square root of average of squared errors
km   6.01      DSM   6.90      yoop 7.61
Cal  6.06      fpli  7.02      BD   7.95
AJ   6.39      Crow  7.18      nr   8.34
DF   6.45      rsm   7.18      EZ   8.58
KF   6.52      MG    7.27      taco 8.80
tzu  6.69      itca  7.52      Dan  8.82
bbs  6.74      snd   7.58      DrP  9.31

With an exponent (and root) of 1.5, Caliban is the leader.
When it's 4th power (and root), I'm in the top tier.

[kmedved]

First update in a while:



I'm not quite sure I followed the discrepancy issue. It's possible I'm fat-fingering something somewhere - does it look like there's an error in these numbers? The "Average" column is what's driving all the rankings.

[Mike G]

I typed in all the Average projection values in the table above, to see if that helps on the discrepancies. Most of mine (from b-r.com exclusively) are within .05, all but a couple are outside .10
[mine vs yours]
Statman/Dan 7.30 vs 5.10
BasketDork - 6.37 vs 7.22
rsmth - - 5.68 vs 5.56

You may have assigned Statman's predictions to BD ?
You also have some entries I never saw submitted to this thread. And a 3-way trade may be in order?

[permaximum]

@Crow

I remember you shared the results of the previous prediction contests before. Do you have the RMSE (actual wins) of the winners?

[Crow]

RMSE from actual wins available for last 5 winners:

14-15 7.82 viewtopic.php?f=2&t=8633&hilit=predicti ... &start=255
13-14 8.14 viewtopic.php?f=2&t=8362&hilit=predicti ... &start=165
12-13 5.95 viewtopic.php?f=2&t=8068&hilit=predicti ... s&start=60
11-12 4.683 viewtopic.php?f=2&t=7830&hilit=predicti ... &start=195 (my best mark. RMSE computed at weaksideawareness.com)
10-11 5.5 viewtopic.php?f=2&t=7830&hilit=predicti ... &start=195 (labeling not clear but pretty sure this was RMSE)
Not seeing RMSE for earlier contests immediately. 2010 missing. 2009 has average error data viewtopic.php?f=2&t=121&hilit=predictions+projections

[permaximum]

RMSE from actual wins available for last 5 winners:

14-15 7.82 viewtopic.php?f=2&t=8633&hilit=predicti ... &start=255
13-14 8.14 viewtopic.php?f=2&t=8362&hilit=predicti ... &start=165
12-13 5.95 viewtopic.php?f=2&t=8068&hilit=predicti ... s&start=60
11-12 4.683 viewtopic.php?f=2&t=7830&hilit=predicti ... &start=195 (my best mark. RMSE computed at weaksideawareness.com)
10-11 5.5 viewtopic.php?f=2&t=7830&hilit=predicti ... &start=195 (labeling not clear but pretty sure this was RMSE)
Not seeing RMSE for earlier contests immediately. 2010 missing. 2009 has average error data viewtopic.php?f=2&t=121&hilit=predictions+projections

Thank you. That's what I suspected. We can not say our ability to predict next season at the team level has been improved at all throughout these years. Obviously the sample is too small but I do think it's enough to confirm there has been no significant changes. On the contrary, there's a hint it's getting worse but there's not enough data to confirm that.

In these years, people have released new and improved metrics such as new RAPM versions, RPM and BPM. We have been given access to player tracking data in the last 2,5 years. Thanks to the use of better regression methods our minute projection and draft models are better, at least on the paper. But somehow they didn't translate well into real world scenerios.

[Mike G]

Every year, the biggest factor is the inherent unpredictability of team performance. A few (or several) teams always go off the chart, for better or worse, in ways nobody could know beforehand.
It doesn't make sense to say we were all very good in some years and all very bad in other years.

One could study annual roster turnover -- the average among all teams in the league -- and its correlation to this "predictability index".
One could rank teams by their roster turnover ratio, see how that correlates with predictability.

[permaximum]

I agree. That's why I said although there's a hint, probably we didn't get worse in reality.

But I believe this confirms our ability to predict has not been improved meaningfully at all, if it's improved.

[caliban]

I don't doubt it Statman

About halfway there and no matter how you slice the numbers it's incredibly tight at the top. Ranked it by combining the two measurements this week. As a nub at this I have no idea how you usually prefer to determent this stuff in the end and it also might be of some interest, idk.

kmedved, bbstats & nrestifo on the bright side of life this week

General swing teams going forward; ATL, ORL, POR + outliers in play; plummeting BRO (kmedved) & MIN (myself & Jim Jividen)

Until next week

Personal note: Closing in on a line free from double digit errors but slowly getting choked out by the Heat's current road-trip in combination with the Dragic injury. Expecting a downward spiral until the break.

[Mike G]

The Pistons are 23-20, their SRS suggests they should be 24-19, and B-R.com projects them to win 44-45 games.
So how did we all guess low? Our average was 36.0, ranging from 27 to 42.

A retrodiction involving this year's minutes applied to last year's rates:

Code: Select all

2016 Pistons      Min   e484   WS/48   PER    BPM     eW    WS   perW  bpmW
Kentavious C-P   1613    .56   .052   11.2   -1.0     1.9   1.7   2.0   2.7
Marcus Morris    1534    .89   .093   13.7   -0.4     2.8   3.0   2.7   2.9
Andre Drummond   1468   1.80   .147   21.4   -0.2     5.5   4.5   5.2   2.9
Reggie Jackson   1332   1.39   .110   19.8    2.6     3.8   3.1   4.2   4.2
Ersan Ilyasova   1202   1.35   .146   16.8   -0.2     3.4   3.7   3.0   2.4

Stanley Johnson   959    .56   .035    9.2   -2.6     1.1    .7    .7   1.0
Anthony Tolliver  804    .49   .104   11.8    0.1      .8   1.7   1.1   1.7
Aron Baynes       579   1.12   .166   15.9   -1.0     1.3   2.0   1.3   1.0
Steve Blake       450    .33   .066    9.5   -2.1      .3    .6    .4    .6
Brandon Jennings  209   1.54   .136   19.7    1.8      .7    .6    .7    .6

Spencer Dinwiddie 111    .64   .009   10.2   -4.2      .1    .0    .1    .0
Reggie Bullock     86   -.02   .019    6.2   -4.0      .0    .0    .0    .0
Darrun Hilliard    82    .26  -.003    7.2   -8.8      .0    .0    .0   -.1
Joel Anthony       50    .46   .142   14.7    1.4      .0    .1    .1    .1
Jodie Meeks        17    .70   .103   14.1   -0.9      .0    .0    .0    .0

.         totals                                     22.1  21.8  21.6  20.1
.                    in 82 games                     42.1  41.6  41.2  38.3

Rookies Johnson and Hilliard are given this season's eW, WS, PER, and BPM.
All 4 stats, converted to wins, suggest more than 36 wins. Some players have improved, and others have not. I get 1.9 extra wins via player eW improvements.
How are they 8 wins over, even without much of Jennings or Meeks?

[Crow]

Kevin Pelton with his by far best rank ever to my memory. Adjustments to method and / or randomness?
May take several more years to guess on that.

Average of those who were Top 5 in 2014-15 producing pretty solid results. If someone wanted to add yet another form of analysis, it might be interesting to see how much money these sets of predictions would have won or lost vs Vegas line if season ended now using equal bets of $1, 10 or $100.

The average of this season's top 5 leads but only produced slight improvement over the very best solo predictions.

My middling rank is due in large part to bad underestimates of a handful of teams. 7 misses by 8 or more, mostly underestimates. The best entrants and the top 5 blend only have 2-4 off by this much. I probably adjusted too much for coaching judgments. Interestingly though, I am still dead on about the Celtics win rate.