2024-25 team win projection contest

[nbacouchside]

An interesting baseline projection to add might be using the Marcel the Monkey projections that Basketball Reference produces for Win Shares per 48 minutes on their player pages (see, e.g., Jayson Tatum's player page and projected WS/48 of .178 under the "2024-25 Projection" section of his page) with the Pelton minutes projections.

[Mike G]

When I "de-regress" the projections at b-r.com, the avgA error becomes worse. Regressing teams toward 41 wins makes us all look better right now.
These 'errors' are relative to the current b-r projections, with all teams heading for 31-52 wins

Code: Select all

.    avg err    rmse    r^2
WShr    5.71    6.76    .32
bpmW    6.16    7.00    .30
eWin    6.36    8.28    .10
perW    6.45    8.27    .09
avgA    7.26    8.22    .24
TmTj    7.50    9.11    .27
dtka    7.68    8.89    .25
DQin    8.04    9.42    .28
emin    8.74   10.30    .19
Crow    8.90   10.20    .22
medi    9.02   10.55    .19

[Crow]

So if you dont expect all teams are heading for 31-52 wins, you probably won't look that good right now...

Last season had 7 teams with 50+ wins and 7 with fewer than 30. This year I projected 8 with 50+ and 7 with <30 and am satisfied with that basic distribution.

One other prediction set has 7 50+, another 6, no one else more than 5. Regressed might be smart but I didn't do it as much as others.

2 others have 7 <30. One has none, 2 have just one.

Either a team is maintaining or fighting for home court advantage or not. At least close to 10th seed or not.

The general spread, broad or tighter, will go a long way to sort the winners and losers.

6 games in should be considerably better gauge than 3 games in. 12 getting pretty good. 24 pretty darn good, I'd think.

Last season's winner took lead on Nov. 16. It could be later this time.

[Mike G]

It took us (me anyway) several years to conclude it was better for your odds to not predict wildly successful or terrible teams. Some of our winners had Zero highest or lowest guesses for teams.
Last year was likely an exception to the rule, but we shall see.

Teemo predicts a total of 1192 wins, which is 38 short of a full season. If I add one win per team, his absolute error is presently unchanged (though it improves at higher exponents, e.g. RSME).

Some dummy entries:
4141 is giving every team 41 wins
24py is last season's Pythagorean wins again.
24pr is Pyth Regressed = (24py + 41)/2
AvgA is the avg of 10 APBR submissions.

Code: Select all

.  avg err   rmse   r^2      .  avg err   rmse   r^2
24pr   5.11   6.2   .27      TmTj   7.50   9.2   .27
4141   5.49   6.5    -       dtka   7.60   8.8   .25
WShr   5.70   6.7   .32      DQin   8.04   9.4   .28
bpmW   6.09   6.9   .30      emin   8.80  10.2   .19
perW   6.38   8.2   .09      24py   8.87  10.4   .27
eWin   6.45   8.2   .10      Crow   8.90  10.2   .22
avgA   7.20   8.2   .24      medi   9.02  10.5   .19

If TmTj doesn't care, he can stick with his original submission. Otherwise I suggest adding 1 win and calling it a rookie error; one-time relaxing of the rules.

EDIT: adding 1 win per team.
This makes 1222 wins, still 8 short of a season. Better than 38 short and a league .485 W%

[Crow]

Maybe go hard one way or the other. If everyone or almost everyone else is heavily regressed, either have to be the best at that or go the other way.

A mix of the two could be best or not. Aggressive on a handful of best cases but deferring to consensus where less sure. Depends on how strong you feel about best takes and what you think of consensus thinking.

[TeemoTeejay]

It took us (me anyway) several years to conclude it was better for your odds to not predict wildly successful or terrible teams. Some of our winners had Zero highest or lowest guesses for teams.
Last year was likely an exception to the rule, but we shall see.

Teemo predicts a total of 1192 wins, which is 38 short of a full season. If I add one win per team, his absolute error is presently unchanged (though it improves at higher exponents, e.g. RSME).

Some dummy entries:
4141 is giving every team 41 wins
24py is last season's Pythagorean wins again.
24pr is Pyth Regressed = (24py + 41)/2
AvgA is the avg of 10 APBR submissions.

Code: Select all

.  avg err   rmse   r^2      .  avg err   rmse   r^2
24pr   5.11   6.2   .27      TmTj   7.50   9.2   .27
4141   5.49   6.5    -       dtka   7.60   8.8   .25
WShr   5.70   6.7   .32      DQin   8.04   9.4   .28
bpmW   6.09   6.9   .30      emin   8.80  10.2   .19
perW   6.38   8.2   .09      24py   8.87  10.4   .27
eWin   6.45   8.2   .10      Crow   8.90  10.2   .22
avgA   7.20   8.2   .24      medi   9.02  10.5   .19

If TmTj doesn't care, he can stick with his original submission. Otherwise I suggest adding 1 win and calling it a rookie error; one-time relaxing of the rules.

Sure, I don’t mind!

[Mike G]

Well I have a train wreck to report.
Nobody noticed or mentioned that all 4 of my Bucks' projections were several wins short of everyone else's.
The reason is that I got my rosters off of b-r.com, and they did not list Gary Trent, Delon Wright, or Stanley Umude.
While I found them in the K Pelton minutes list and copied their minutes, I failed to generate their various 'wins'; so the team added up suspiciously low.

Code: Select all

Mil      min   eWin  perW  WShr  bpmW 
Trent   1966    2.7   2.8   1.6   2.0
Wright  1584    2.0   2.9   3.5   3.4
Umude    152     .2    .2    .3    .2
total   3702    4.9   5.9   5.4   5.6

Adding these in and re-standardizing the other 29 teams (to avg 41 wins), regressing and de-regressing, these 4 entries for Milwaukee should have been:

Code: Select all

stat:   eWin   perW   WShr   bpmW
wrong   36.6   35.5   39.6   38.3
right   43.1   45.2   43.6   43.7

All other teams/stats are reduced by 0.1 to 0.4 wins.
Bucks currently project to 34 wins, so this isn't helping anyone at the moment.

[Mike G]

If TmTj doesn't care, he can stick with his original submission. Otherwise I suggest adding 1 win and calling it a rookie error; one-time relaxing of the rules.

Sure, I don’t mind!

I meant (of course)...
... adding 1 win per team.
This makes 1222 wins, still 8 short of a season. Better than 38 short and a league .485 W%

Is anyone going to dig up Pelton predictions? anything we can call "Vegas"? anyone else??

Code: Select all

.   avg err  rmse   r^2      .  avg err   rmse   r^2
24pr   4.99   6.3   .25      TmTj   7.44   9.4   .25
4141   5.05   6.5            dtka   7.59   9.1   .22
WShr   5.83   7.0   .26      DQin   8.11   9.8   .22
bpmW   6.28   7.3   .25      emin   8.51  10.3   .18
eWin   7.03   8.5   .07      24py   8.67  10.6   .25
perW   7.04   8.6   .06      Crow   8.71  10.1   .23
avgA   7.15   8.4   .21      medi   9.02  10.6   .18

[Crow]

"Vegas" will vary some but one of these should serve the purpose:

https://www.covers.com/nba/win-totals-odds

https://sports.betmgm.com/en/blog/nba/n ... tals-bm23/

[TeemoTeejay]

If TmTj doesn't care, he can stick with his original submission. Otherwise I suggest adding 1 win and calling it a rookie error; one-time relaxing of the rules.

Sure, I don’t mind!

I meant (of course)...
... adding 1 win per team.
This makes 1222 wins, still 8 short of a season. Better than 38 short and a league .485 W%

Is anyone going to dig up Pelton predictions? anything we can call "Vegas"? anyone else??

Code: Select all

.   avg err  rmse   r^2      .  avg err   rmse   r^2
24pr   4.99   6.3   .25      TmTj   7.44   9.4   .25
4141   5.05   6.5            dtka   7.59   9.1   .22
WShr   5.83   7.0   .26      DQin   8.11   9.8   .22
bpmW   6.28   7.3   .25      emin   8.51  10.3   .18
eWin   7.03   8.5   .07      24py   8.67  10.6   .25
perW   7.04   8.6   .06      Crow   8.71  10.1   .23
avgA   7.15   8.4   .21      medi   9.02  10.6   .18

Yeah sure
Sounds like we’re getting destroyed rn

[mediocre]

Mike, would you mind posting the updated full projections with the Bucks + TmTJ update?

[dtkavana]

Mike,

In year's past you have used projections from various metrics in the contest, such as Kevin Pelton, LEBRON, RAPTOR, NumberFIRE, etc. I used these to compare my projections which use CraftedNBA to other metrics. Any chance you will be adding any of these to the contest this year? Also, just for reference, I use my own minutes projections which start with last years minutes played per game, and then I go team by team and adjust as how I imagine the coach might set the lineups.

Thanks

[DQuinn1575]

Well I have a train wreck to report.
Nobody noticed or mentioned that all 4 of my Bucks' projections were several wins short of everyone else's.
The reason is that I got my rosters off of b-r.com, and they did not list Gary Trent, Delon Wright, or Stanley Umude.
While I found them in the K Pelton minutes list and copied their minutes, I failed to generate their various 'wins'; so the team added up suspiciously low.

Code: Select all

Mil      min   eWin  perW  WShr  bpmW 
Trent   1966    2.7   2.8   1.6   2.0
Wright  1584    2.0   2.9   3.5   3.4
Umude    152     .2    .2    .3    .2
total   3702    4.9   5.9   5.4   5.6

Adding these in and re-standardizing the other 29 teams (to avg 41 wins), regressing and de-regressing, these 4 entries for Milwaukee should have been:

Code: Select all

stat:   eWin   perW   WShr   bpmW
wrong   36.6   35.5   39.6   38.3
right   43.1   45.2   43.6   43.7

All other teams/stats are reduced by 0.1 to 0.4 wins.
Bucks currently project to 34 wins, so this isn't helping anyone at the moment.

lol, maybe this error will wind up with by far the most accurate win projection for them

[Mike G]

Mike, would you mind posting the updated full projections with the Bucks + TmTJ update?

Yes, if nobody objects.

In year's past you have used projections from various metrics in the contest, such as Kevin Pelton, LEBRON, RAPTOR, NumberFIRE, etc. I used these to compare my projections which use CraftedNBA to other metrics. Any chance you will be adding any of these to the contest this year? Also, just for reference, I use my own minutes projections which start with last years minutes played per game, and then I go team by team and adjust as how I imagine the coach might set the lineups.

Yeah in years past it's mostly other people with access thru paywalls or who just have the time to find these. I asked if anyone would.
I did my own minutes as well, and my original predictions are looking a lot worse than with the Pelton minutes.

Once again, I'm sitting uncomfortably outside a public library in my neck of the woods, without phone service/internet at home. Laptop is also falling apart

[Mike G]

Seeing no objections to the late updates, here we are:

Code: Select all

tm    bpmW  emin  perW  DQin  eWin  TmTj  medi  Crow   WS  dtka
Atl   40.6   34   44.2   39   38.6   37   37.7   35   41.1   41
Bos   55.7   62   52.5   61   52.3   57   60.8   59   56.7   60
Brk   29.2   23   30.1   29   31.2   24   25.8   18   33.8   26
Cha   28.4   33   34.1   30   30.9   26   23.5   31   29.5   29
Chi   34.7   27   34.1   28   34.2   31   25.1   29   35.1   30

Cle   45.7   49   43.5   48   44.7   48   50.1   50   46.0   52
Det   31.7   28   32.5   26   33.0   32   25.6   27   31.5   28
Ind   43.0   44   47.8   42   41.0   44   47.0   47   42.5   46
Mia   43.3   47   40.8   42   44.8   42   42.3   40   42.8   43
Mil   43.7   48   45.2   49   43.1   45   43.7   48   43.6   50

tm    bpmW  emin  perW  DQin  eWin  TmTj  medi  Crow   WS  dtka
NYK   42.8   58   38.4   43   40.7   50   48.4   53   43.7   50
Orl   40.9   42   33.9   43   38.9   43   47.5   49   43.7   46
Phl   45.6   43   52.4   47   52.6   51   55.1   46   46.1   47
Tor   31.7   32   32.2   29   30.1   29   29.4   25   32.1   30
Was   28.1   19   33.1   14   33.9   15   17.5   19   28.1   21

Dal   49.7   43   53.0   56   52.2   52   46.5   51   47.5   47
Den   46.8   53   49.1   49   50.6   46   51.3   49   47.3   51
GSW   45.8   47   40.1   49   40.8   49   48.4   52   44.2   45
Hou   44.8   33   43.2   36   46.9   44   46.1   44   43.6   38
LAC   40.5   32   34.8   45   31.4   36   35.4   33   43.9   40

tm    bpmW  emin  perW  DQin  eWin  TmTj  medi  Crow   WS  dtka
LAL   39.1   42   36.9   43   36.4   45   39.4   44   38.2   44
Mem   41.1   48   41.5   45   42.9   41   42.5   49   34.2   43
Min   50.0   57   46.0   55   53.3   48   52.0   51   50.4   50
NOP   44.4   49   44.1   45   43.8   44   49.4   48   43.8   42
OKC   54.4   58   52.1   61   49.2   59   62.0   54   55.1   54

Phx   47.0   46   47.9   43   45.4   51   49.9   48   47.0   49
Por   27.2   23   27.8   26   31.0   24   23.9   26   26.0   22
Sac   49.0   44   48.0   47   47.4   47   42.0   50   46.5   47
SAS   36.1   38   37.2   31   36.2   34   30.6   29   34.0   31
Uta   28.9   28   33.3   29   32.5   27   31.1   26   32.0   29

The most outlying projections (guesses), relative to the APBR average (avgA)
"Also" are the biggest outliers for entries with none >6.0

Code: Select all

over  high   tm   guess  avgA      over   low    tm   guess  avgA
11.2  emin   NYK   58    46.8      -9.0   Crow   Brk   18    27.0
11.0   eW    Was   33.9  22.9      -9.0   emin   Hou   33    42.0
10.3  PER    Was   33.1  22.9      -8.9   PER    Orl   33.9  42.8
7.8   DQin   LAC   45    37.2      -8.9   DQin   Was   14    22.9
6.8    WS    Brk   33.8  27.0      -8.6    WS    Mem   34.2  43.0
6.7    WS    LAC   43.9  37.2      -8.4   PER    NYK   38.4  46.8
6.5   medi   Phl   55.1  48.6      -7.9   TmTj   Was   15    22.9
6.2   Crow   Orl   49    42.8      -6.8   emin   Dal   43    49.8
6.2   DQin   Dal   56    49.8      -6.7    eW    OKC   49.2  55.9
6.2   Crow   NYK   53    46.8      -6.1    eW    NYK   40.7  46.8
6.2   Crow   Mem   49    42.8      -6.0   medi   Cha   23.5  29.5
6.1   medi   OKC   62    55.9      -6.0   DQin   Hou   36    42.0
.                                  -6.0   PER    GSW   40.1  46.1
over  also   tm   guess  avgA                  
5.2    BPM   Was   28.1  22.9      over   also   tm   guess  avgA
4.3   dtka   Cle   52    47.7      -4.0   BPM    NYK   42.8  46.8
4.2   TmTj   LAL   45    40.8      -4.0   dtka   Hou   38    42.0