2021-22 team win prediction contest

[Crow]

Congrats to upcoming winner. And any others relatively pleased with performance

[Mike G]

While there are scenarios which would tighten this race, none are likely to change the lead.

Code: Select all

.   avg err   rmse          avg err   rmse
trzu   6.55   8.24      lisp   7.18   8.72
cali   6.79   8.50      emin   7.19   9.29
shad   6.87   8.68      pewb   7.46   9.67
5.38   6.91   8.50      bpmW   7.73   9.95
avgA   6.97   8.71      WShr   7.84   9.61
dtka   6.98   8.79      21re   7.95   9.15
vegas  7.08   9.01      2021   8.18  10.14
Crow   7.12   8.88      eWin   8.51  10.11
.                       perW   8.95  10.42

https://www.basketball-reference.com/fr ... _prob.html
trzu leads at exponent from near zero to over 3

[Mike G]

According to their season point differential, the Celtics "should have" won 8 more games this year.
The Suns' MOV would normally have them winning 5 fewer games.
So without some luck -- or whatever determines who wins/loses the close ones -- they'd not be separated by 14 games in the standings, but by a single win.
https://www.basketball-reference.com/le ... anced_team

Uta, SAS, and Ind should have 7 more wins than they do. Por and Chi should have 6-7 fewer.
Here's a side-by-side with the traditional "wins based" rankings on the left, and Pythagorean-expected wins on the right.
"Luck" is attached to the columns at left, and is simply the difference (pyth-minus-projected)

Code: Select all

err.   proj  "luck"    err.   pyth
trzu   6.59   .65      5.38   7.05
cali   6.80   .47      dtka   7.10
shad   6.91   .26      emin   7.14
5.38   6.92   .13      lisp   7.15
avgA   7.01   .36      Crow   7.15
dtka   7.05   .05      shad   7.17
vegas  7.10   .23      trzu   7.24
emin   7.12   .02      cali   7.27
Crow   7.12   .03      vegas  7.33
lisp   7.19  -.05      avgA   7.37
pewb   7.55   .49      pewb   8.04
bpmW   7.84   .49      21re   8.07
WShr   7.87   .25      WShr   8.11
21re   8.07   .00      bpmW   8.33
2021   8.23   .15      2021   8.38
eWin   8.62   .26      eWin   8.88
perW   9.04   .25      perW   9.30

avg.   7.47   .24             7.71

trzu looks like the 'luckiest'; lisp is least lucky; and without the luck, it's a much closer contest, 1 thru 10.

[Crow]

Interesting.

By Pythagorean, I would be ahead of the 3 leaders based on actual wins. One of the lowest on "luck".

But "luck" isn't necessarily all luck and the contest wasn't for net rating. It could be though.

Or with regular season subject to various forms of luck and somewhat devalued by load management and indifference, maybe there could be a contest for the playoffs. Would probably involve and / or reward different methods, maybe more subjective, maybe more decisively influenced by luck in a shorter framework.

[tarrazu]

Need to consider a team's actual Win Pct % as well, although net rating is still the main driver. I'd imagine Chris Paul's teams have often been "lucky" relative to their net rating.

[Mike G]

I'd imagine Chris Paul's teams have often been "lucky" relative to their net rating.

A quick look indicates this is so -- aside of his Clippers teams, which were a bit short on wins.
A discrepancy between expected (from MOV) wins and actual wins isn't really just about winning the close games, but also how blowout scores happen. A 35-point loss not only puts an L in your record but indicates you "should have lost" one that you won.

The Bulls this year were a very average team that somehow has a 45-36 record. They've won 3 games by 20 points or more, and they've lost 10 such games. In games decided by 3 or less (or OT), they are 5-5.

[Mike G]

Final results:

Code: Select all

.   avg err   rmse          avg err   rmse
trzu   6.49   8.22      lisp   7.19   8.70
shad   6.83   8.62      pewb   7.52   9.64
5.38   6.87   8.52      bpmW   7.79   9.92
cali   6.89   8.50      WShr   7.86   9.60
vegas  6.97   8.92      21re   8.04   9.17
dtka   7.00   8.85      2021   8.17  10.09
avgA   7.01   8.70      eWin   8.63  10.10
Crow   7.13   8.86      perW   9.00  10.43
emin   7.13   9.26              		

The average of all these MAE hit 7.55 yesterday, worst of the entire season.
Congrats to tarrazu, who won decisively at exponents 1 and 2.

[sndesai1]

Ranked on RMSE and including some entries not in the official contest.




The 538 number looks completely different between ours - I have them performing extremely poorly while you have them near the top. Mine are pulled from Oct. 18 - any chance yours are a different date?

Code: Select all

Team	RAPTOR	Elo
ATL	51.0	48.0
BOS	49.0	40.0
BRK	50.0	52.0
CHA	33.0	34.0
CHI	38.0	40.0
CLE	25.0	26.0
DAL	49.0	45.0
DEN	50.0	48.0
DET	24.0	29.0
GSW	36.0	43.0
HOU	20.0	25.0
IND	44.0	40.0
LAC	48.0	52.0
LAL	42.0	45.0
MEM	43.0	44.0
MIA	44.0	42.0
MIL	56.0	57.0
MIN	40.0	34.0
NOP	47.0	38.0
NYK	43.0	45.0
OKC	19.0	22.0
ORL	24.0	24.0
PHI	53.0	52.0
PHO	53.0	55.0
POR	48.0	48.0
SAC	32.0	35.0
SAS	36.0	38.0
TOR	41.0	35.0
UTA	54.0	52.0
WAS	38.0	41.0

[Mike G]

Yeah from page 1 of this thread, it looks like I got a 538 prediction on Oct. 14, and I don't know which of those it was (Raptor, Elo)

[DarkStar48]

Yeah from page 1 of this thread, it looks like I got a 538 prediction on Oct. 14, and I don't know which of those it was (Raptor, Elo)

Yeah. It’s the RAPTOR version. Link: https://fivethirtyeight.com/features/wi ... ba-season/

Teams were still cutting/signing guys onto rosters before the start of the regular season on Oct. 19 (e.g., Avery Bradley — GSW to LAL). The most prominent change between the 538 RAPTOR projection from Oct. 18 compared to Oct. 14 was the news of the Zion Williamson injury.

[Mike G]

So I recorded the 538Raptor prediction for the Pels at 44 wins.
After " the news of the Zion Williamson injury" they figured 47.
And Elo guessed 38.

[DarkStar48]

I am not sure where you got the 44 predicted wins.

However, the 538 RAPTOR projection on Oct. 18 with all the changes (including roster modifications as well as the Zion absence) shifted about 1-2 wins across many teams, and then the Pelicans went down from 48 to 47.

[Mike G]

I got the 538 predictions from the same link you posted above. Of course, it's changed since Oct. 14 2021

Code: Select all

538  10.14  10.18 Elo
Atl    50    51    48
Bos    46    49    40
Brk    49    50    52
Cha    37    33    34
Chi    41    38    40
Cle    27    25    26
Dal    49    49    45
Den    50    50    48
Det    24    24    29
GSW    40    36    43
Hou    22    20    25
Ind    43    44    40
LAC    47    48    52
LAL    40    42    45
Mem    45    43    44
Mia    44    44    42
Mil    56    56    57
Min    39    40    34
NOP    44    47    38
NYK    44    43    45
OKC    19    19    22
Orl    23    24    24
Phl    52    53    52
Phx    51    53    55
Por    47    48    48
Sac    33    32    35
SAS    34    36    38
Tor    38    41    35
Uta    55    54    52
Was    39    38    41

Cha and GSW were downgraded by 4 wins in the interim, Chi by 3.
Bos, NOP, and Tor were kicked up 3. That is all I know about it.

[DarkStar48]

Ah, I see what happened here. The first 538 RAPTOR model output released on Oct. 14 (technically run on Oct. 13) had the Pelicans with 48 predicted wins. Sometime on Oct. 14 in the morning, they must have included the “Zion update,” which altered the team totals league-wide by 3-5 wins and reduced NOP’s record to 44-38. Finally between then and Oct. 19, other roster and playtime changes (speculated Zion return?) simulated the Pelicans back up to 47.

[Crow]

So I am going to use DARKO even more going forward.

Recently did a twitter thread checking trends on about 40 players on DARKO. No one commented on any of course.

RAPTOR and LEBRON were nothing special here and EPM was quite weak. I may still check them but I am not giving them the same respect as DARKO at this time.

BPM may still be a quick draw casual reference for NBA but I probably should pull from elsewhere. Maybe the craftedNBA blend for descriptive use. Will probably still use for college players but may look more into some of the other college metric options I am less familiar with.

Porpagatu and Adjusted Net Rating may be decent college metrics. But probably mostly stick with BPM for now.


May be too late to retrofit DRIP. Probably should track next season, if still available.

BPI did well but I am probably not going to look for it.

Can / will anyone make a comparable to above "luck" assessment on DARKO? That would be very important imo.

addendum:
Fwiw, By sndesai1's list, I tied for 2nd least 10+ win misses.