2024-25 team win projection contest

[Crow]

Caught ESPN experts, at least for a moment. Will see if it lasts. Pretty close to Vegas.

[Mike G]

Code: Select all

.  avg err   rmse   r^2      .  avg err   rmse   r^2
TmTj   6.76   9.0   .50      emin   7.56   9.8   .44
dtka   6.79   8.9   .51      eExp   7.58   9.8   .51
avgA   6.92   9.2   .47      Crow   7.59   9.7   .46
medi   7.21  10.1   .42      WShr   7.77   9.7   .41
KPel   7.24   8.8   .57      24pr   7.99  10.4   .32
bpmW   7.27   9.2   .47      eWin   8.50  10.8   .27
DQin   7.32   9.4   .47      24py   8.52  11.5   .32
vegas  7.43   9.7   .44      perW   8.74  11.0   .24
.                            bmgm   9.03  11.9   .33

Worst errors yet for most and for the group.
A nice constellation of co-leaders:

Code: Select all

e = .10           e = .20          e = .50
3.89   Crow      4.72   medi      5.52   dtka
4.43   medi      4.72   dtka      5.63   TmTj
4.45   dtka      4.86   Crow      5.63   medi
4.80   TmTj      5.00   TmTj      6.13   bpmW
5.04   bpmW      5.33   bpmW      6.16  vegas
                     
e = 1.12          e = 1.75         e = 5.00
7.04   TmTj      8.39   dtka      11.8   KPel
7.07   dtka      8.46   KPel      12.3   dtka
7.46   KPel      8.46   TmTj      13.3   bpmW
7.52   bpmW      8.75   bpmW      13.4   WShr
7.59   medi      8.94   DQin      13.6   TmTj

[dtkavana]

Let's Go! Getting closer and closer to 1st place. I'd love to win for the 2nd time in 3 years. Relatedly, I am currently building a season projection tool, similar to how 538 used to do it. It seems like there's an opening there since their site was shuttered. The projections are based on Crafted+-, which is what I use for my dtkavana entry in this pool each year, so I think they should be pretty strong. Here's a preview of what I have right now ->

[Crow]

Spot checking 10 teams, you and BRef seem to be projecting within 1-2 team wins of each other. But how much difference would that make on the contest measurement?

[Mike G]

So the Dubs at 55% for the season are expected to win 70% the rest of the way?
Is there a recency factor affecting their season record?

Any reason to not include a decimal place in the forecast?
Is it fair that Den, Uta, Chi, and Det get 83 games?

[Mike G]

At least for now, dtka has seized the rmse lead from KPel.

Code: Select all

.  avg err   rmse   r^2      .  avg err   rmse   r^2
TmTj   6.76   9.0   .50      emin   7.60   9.9   .44
dtka   6.77   8.85  .51      eExp   7.62   9.8   .51
avgA   6.87   9.2   .47      Crow   7.64   9.7   .46
medi   7.21  10.2   .42      WShr   7.71   9.7   .41
bpmW   7.25   9.2   .47      24pr   7.94  10.4   .32
DQin   7.30   9.4   .47      eWin   8.49  10.8   .27
KPel   7.34   8.86  .57      24py   8.51  11.5   .32
vegas  7.54   9.7   .44      perW   8.84  11.0   .24
.                            bmgm   9.04  11.9   .33

[Crow]

Best I can say at the moment is tied for 8th on rmse, 8th on r2 and only .88 behind leader on average error.

Ahead of ESPN and Vegas on one measure each.

Would like to get into a top 5. Need to see some forward movement soon.

[dtkavana]

"So the Dubs at 55% for the season are expected to win 70% the rest of the way?
Is there a recency factor affecting their season record?

Any reason to not include a decimal place in the forecast?
Is it fair that Den, Uta, Chi, and Det get 83 games?"

The projections are built mostly from the player level using CraftedPM (Jimmy Butler is good!), but we also use their season net rating up to this point, as well as their net rating in their last ten games. There's quite a bit more to it, but that's the quick and dirty.

83 games is a bug

Do you think we should show a decimal place? I like the clean look aesthetically, but the decimal does provide a more granular projection.

[Mike G]

Do you think we should show a decimal place? I like the clean look aesthetically, but the decimal does provide a more granular projection.

That's what b-r.com shows, even knowing that no team will end up with a fraction of a win.
I thought maybe 51.5 and 30.5 (Den) both got rounded up to a projection of 83 games.

Right now, you've got the lead at exponent from 1.37 to 2.14. This includes RMSE (e = 2).
What's your feeling about the value of RMSE vs MAE in such a contest? I'm leery of stats that I can't explain to the average Joe. But also skeptical of the sacred status of e = 2. Something between (or beyond) 1 and 2 may be significant?

[dtkavana]

This is completely outside of my level of expertise, but here's what Chat GPT 4o has to say:

This is a great question, and it's worth questioning why RMSE (e = 2) is the default error metric in so many contests.

RMSE vs. MAE
RMSE (Root Mean Squared Error, e = 2): Penalizes larger errors more heavily than smaller ones. This is useful when you want to emphasize avoiding big misses.
MAE (Mean Absolute Error, e = 1): Treats all errors equally, making it easier to interpret and explain, but less sensitive to outliers.
The Case for e ≠ 2
You're right to be skeptical of RMSE's dominance. The choice of exponent (e) determines how harshly large errors are punished. If you set e < 2, you're reducing the penalty for big misses, while e > 2 makes the metric even more extreme.

Something like e = 1.5 or 1.75 could be an interesting compromise—more punishing than MAE but without RMSE’s harsh treatment of big errors. This might better reflect the importance of getting all teams reasonably close rather than just avoiding a few disasters.

What Matters for the Contest?
For a win projection contest, should a single bad miss (like predicting 30 wins for a team that gets 50) outweigh a handful of small misses? If the goal is consistency across all teams, a lower exponent (closer to 1) makes sense. If punishing big misses is important, e = 2 or higher is justifiable.

[Mike G]

Dead heat.

Code: Select all

.  avg err   rmse   r^2      .  avg err   rmse   r^2
TmTj   6.95   9.2   .49      WShr   7.70   9.7   .41
dtka   6.95   8.9   .50     vegas   7.70   9.8   .44
avgA   7.05   9.3   .46      Crow   7.74   9.8   .45
bpmW   7.25   9.3   .46      24pr   7.96  10.4   .32
medi   7.34  10.2   .42      24py   8.59  11.6   .32
DQin   7.46   9.6   .46      eWin   8.61  10.9   .26
KPel   7.54   9.0   .57      perW   8.97  11.1   .23
eExp   7.69   9.9   .49      bmgm   9.20  12.0   .32
emin   7.69   9.9   .43               

At exponents giving optimal separation between #1 and 2:

Code: Select all

e = .20          e = .50         e = .92         e = 1.80        e = 6.00
4.89   medi     5.74   dtka     6.760  TmTj     8.59   dtka     12.6   KPel
4.98   dtka     5.78   TmTj     6.765  dtka     8.75   TmTj     13.1   dtka
5.09   TmTj     5.79   medi     7.07   bpmW     8.76   KPel     14.4   bpmW
5.20   bpmW     6.03   bpmW     7.09   medi     8.90   bpmW     14.5   WShr
5.57   emin     6.33   DQin     7.28   DQin     9.17   DQin     14.7   DQin
5.58   eExp     6.40   emin     7.40   KPel     9.35   WShr     14.7   TmTj
5.65   DQin     6.40   eExp     7.49   eExp     9.40   Crow     14.9   eExp

[TeemoTeejay]

Me and dtka rn

[Crow]

Not that I am close to top, but I think I am about the closest I've been this season. Maybe once similar or slightly better.

If top teams maintain and bottom teams fall, I might tend to improve further against actual rather than BRef projection.

Projection likely to undergo considerable change. Currently projecting most good teams to win less, some about same. Currently projecting most lower half teams to win more often, some about same. Don't think it will go that way but we'll see.

Final 4+ weeks.

[Mike G]

Code: Select all

.  avg err   rmse   r^2         avg err   rmse   r^2
dtka   7.02   9.0   .50      WShr   7.78   9.8   .40
TmTj   7.05   9.3   .48      Crow   7.79   9.9   .45
avgA   7.13   9.4   .46     vegas   7.81   9.9   .43
medi   7.36  10.3   .41      24pr   7.97  10.5   .32
bpmW   7.38   9.4   .46      24py   8.61  11.6   .32
DQin   7.48   9.7   .45      eWin   8.74  11.0   .25
KPel   7.56   9.1   .56      perW   9.07  11.2   .23
eExp   7.74  10.0   .48      bmgm   9.21  12.0   .32
emin   7.76  10.0   .43               

Code: Select all

e = .10          e = .25         e = .54         e = 1.80        e = 6.00
4.05   DQin     5.08   medi     5.88   dtka     8.67   dtka     12.8   KPel
4.15   KPel     5.09   dtka     5.94   TmTj     8.86   KPel     13.2   dtka
4.66   dtka     5.19   TmTj     5.94   medi     8.86   TmTj     14.6   bpmW
4.66   medi     5.27   DQin     6.28   bpmW     9.00   bpmW     14.6   WShr
4.79   TmTj     5.44   KPel     6.30   DQin     9.26   DQin     14.8   DQin
5.02   bpmW     5.48   bpmW     6.50   KPel     9.42   WShr     14.9   TmTj
5.29   emin     5.73   emin     6.55   emin     9.48   Crow     15.1   eExp

[knarsu3]

Idk if it's too late or not but here they all are as well as documentation that they were done before the season started:

The LEBRON projections were where they always were: https://www.bball-index.com/season-team-projections/ but I guess they never got posted.



and if you need a date to see that they were posted at the beginning of the season: https://x.com/knarsu3/status/1849290393269022918

The DARKO ones were here as well: https://x.com/kmedved/status/1849277426930528489



And finally I did my own here: https://docs.google.com/spreadsheets/d/ ... id=0#gid=0



That I posted about here: https://x.com/knarsu3/status/1849207698547503615 and here: https://x.com/knarsu3/status/1849291390875791821

A primary reason LEBRON and DARKO projections weren't included is because they were not entered here by authors, checked or found.

If projections were made and published before season, please provide link.

Last season, Darko finished 13th of 20 listings, LeBron 18th.

viewtopic.php?t=10034&start=165

4th and 5th in 2022-23:
viewtopic.php?t=9946&start=165#p39853

Darko 1st in expanded check for 2021-22, LeBron 19th of 30:
viewtopic.php?t=9903&start=105


3+ year averages might be interesting. There is some "randomness" involved I'd say.

'20-21: viewtopic.php?t=9842&start=105

'19-20 (my most recent win):
viewtopic.php?t=9715&start=150