2024-25 team win projection contest

[Mike G]

Big schedule and big shakeup.

Code: Select all

.  avg err   rmse   r^2      .  avg err   rmse   r^2
KPel   5.89   7.6   .59      24pr   6.59   9.2   .34
bpmW   5.90   8.0   .50      medi   6.74   9.6   .42
WShr   6.02   8.3   .46      Crow   6.88   9.2   .45
TmTj   6.05   8.4   .50      emin   7.20   9.6   .41
avgA   6.10   8.3   .48      eWin   7.29   9.5   .31
eExp   6.16   8.3   .55      24py   7.52  10.8   .34
DQin   6.51   8.5   .50      perW   7.97  10.0   .24
dtka   6.52   8.2   .51      bmgm   8.13  11.0   .37
vegas  6.53   8.9   .46

https://www.basketball-reference.com/fr ... _prob.html
Changes from "yesterday" range from -.13 for KPel to +.37 for medi.

[Crow]

BRef is still projecting Celtics to lose at about a 75% higher rate going forward than to date. Cavs now to lose at a ridiculous 300% higher rate. Raptors, Hornets and Wizards all still to win roughly twice as frequently. Thunder to lose 50% more often. Jazz to win 50% more often. Still highly doubt and don't understand all these tail cases.

We'll see what it looks like after 50, 75, 82 games.

[DQuinn1575]

BRef is still projecting Celtics to lose at about a 75% higher rate going forward than to date. Cavs now to lose at a ridiculous 300% higher rate. Raptors, Hornets and Wizards all still to win roughly twice as frequently. Thunder to lose 50% more often. Jazz to win 50% more often. Still highly doubt and don't understand all these tail cases.

We'll see what it looks like after 50, 75, 82 games.

So doesn't it make the leaderboard partly dependent on how well our projection systems line up with whatever projection system b-ref is using? So if my system is more similar to theirs than yours, I will have less of an error on that part of the equation?

[Crow]

Of course.

I could be wrong but I don't see the teams mentioned above regressing to the mean the rest of the way that much. Does anyone agree with most or any of these 7 cases for major regression? I could see 1 or 2 happening maybe but not 7 this much.

Restated that opinion but maybe not again for awhile.

[Mike G]

BRef is still projecting Celtics to lose at about a 75% higher rate going forward than to date. Cavs now to lose at a ridiculous 300% higher rate...

Cavs have lost 4 of 27 games, a "loss rate" of .148 of games. B-R.com projects they lose 18 of remaining 55 games, or .327. This would be losing 2.2 times as frequently, or 120% higher.
Their pythagorean (based on point differential) record suggests 6 losses in 27 games, or .222 L% -- so that .327 is 47% higher than to date.
Further, their strength of schedule has been the weakest in the league by a full .50 ppg; remaining SOS looks to be 2.73 ppg tougher.

These factors are likely bigger right now than regression to the mean for Cle.
Their top 5 players have missed a total of 5 games so far. Extraordinary good luck to date.

Celtics with identical SRS and future forecast. Also projecting 50% more frequent losses the rest of the way, with almost 1 ppg tougher SOS ahead.
OKC with none of these issues projecting to lose 39% more frequently. They were +15.5 ppg in 9 games with Holmgren and +10.0 since.

[TeemoTeejay]

BRef is still projecting Celtics to lose at about a 75% higher rate going forward than to date. Cavs now to lose at a ridiculous 300% higher rate. Raptors, Hornets and Wizards all still to win roughly twice as frequently. Thunder to lose 50% more often. Jazz to win 50% more often. Still highly doubt and don't understand all these tail cases.

We'll see what it looks like after 50, 75, 82 games.

So doesn't it make the leaderboard partly dependent on how well our projection systems line up with whatever projection system b-ref is using? So if my system is more similar to theirs than yours, I will have less of an error on that part of the equation?

Somewhat but at the end of the day if they’re just using SRS I don’t think that doing it on, let’s say projecting SRS or net rating into wins, does that much differently than doing wins directly

That it’s holding in constant 30 games in as if those things will continue to hold for sure, is much different from projecting what it will look like at the end of the season, imo. Hard to see the suns and lakers both being 10 and 11 by the end of the year assuming reasonable health to the top of their lineup given how they often have late season surges for example

[Mike G]

Daily diary of who (or what) is leading in the contest. Days at #1 in a streak, and biggest lead over #2:

Code: Select all

.  November           December
#1   days  big     #1  days  big
24pr   2   .38     bpm   6   .10
4141   1   .26     WS    2   .01
24pr   4   .42     bpm   2   .04
bpmW   4   .17    *KPel  3   .04
WShr   4   .08     bpm   3   .00
24pr   1   .02     WS    2   .05
bpmW   3   .12     TmTj  2   .05
WShr   1   .00     KPel  1   .01
bpmW   8   .13     bpm   1   .05
WShr   1   .02     TmTj  3   .11
bpm    2..         KPel  1.. .02

* -- Pelton came in at #1 when we included him.

A couple of years ago, I de-regressed the b-r.com projections early in the season; which is to say all teams were moved further from 41 wins. Gradually I reduced the amount that I did this. It was kind of an annoying and arbitrary process.

The spreadsheet still has this capability, but I've just used the b-r.com number for these reports. Every time I've applied it this season, it makes everyone's error worse.
So here are current b-r.com projections and successively de-regressed results. The standings change, and we all get worse with this adjustment.

Code: Select all

dr = 1            dr = 1.06        dr = 1.12        dr = 1.18        dr = 1.24
bpmW   5.90      TmTj   6.22      TmTj   6.46      TmTj   6.75      TmTj   7.17
KPel   5.96      bpmW   6.38      bpmW   6.85      DQin   7.14      DQin   7.45
WShr   6.02      WShr   6.42      WShr   6.88      bpmW   7.32      medi   7.68
TmTj   6.07      KPel   6.43      DQin   6.91      medi   7.34      bpmW   7.80
avgA   6.16      avgA   6.54      KPel   6.91      WShr   7.34      WShr   7.81
eExp   6.22      eExp   6.69      avgA   6.95      avgA   7.39      avgA   7.83
dtka   6.57      DQin   6.70      medi   7.07      KPel   7.39      vegas  7.87
DQin   6.58      medi   6.83      eExp   7.15      vegas  7.53      KPel   7.87
vegas  6.59      vegas  6.87      vegas  7.19      dtka   7.59      dtka   7.97
24pr   6.61      dtka   6.89      dtka   7.21      eExp   7.61      Crow   8.03
medi   6.70      Crow   7.09      Crow   7.37      Crow   7.68      eExp   8.07
Crow   6.93      24pr   7.10      24pr   7.60      24py   7.97      24py   8.15
emin   7.27      emin   7.49      emin   7.73      emin   7.98      emin   8.24
eWin   7.30      24py   7.64      24py   7.78      24pr   8.10      bmgm   8.56
24py   7.55      eWin   7.70      eWin   8.10      bmgm   8.39      24pr   8.61
perW   8.03      bmgm   8.23      bmgm   8.30      eWin   8.52      eWin   8.99
bmgm   8.16      perW   8.42      perW   8.81      perW   9.20      perW   9.63

At the end of the season we may check and see if the order or the errors are better predicted by these various levels of adjustment. The final column is with OKC winning 68 games and NOP 13.
dr is the multiple applied to (proj - 41).

[Crow]

My above comments on BRef projections were quick rough estimates. 24-4 current goes to 36-6 if things stay the same but they projected 17 losses and 17 is almost 3 times 6.

"300% higher" should have been 300% as high. I (and others) sometimes don't catch this phrasing issue. We know what we meant but "as high" is the more accurate, though perhaps somewhat "odd / not normal" sounding way to say it. The point is often the real final comparison not the incremental change.

Thanks for the SOS & SRS check. We'll see what actually happens. Actual wins & losses are the real final measure. Strength of schedule and margin influence them but the actual W-L behavior to date may be the best indicator. Last season, the Cavs were .571 win on December 21 and ended up real close at .585, slightly higher. The net margin changed a lot more %wise first part of season to remainder than the actual win%.

SOS & SRS are not perfectly reliable for projecting win%. There is noise / error in such calculations. Behavior in blowouts or potential blowouts varies and doesn't always fully impact win% as suggested. There probably would be a helpful way to adjust net margin data for this side behavior to produce more accurate actual win% projections. Yes margin matters and you want to prevent a lead from potentially bring erased late but a lot of it comes down to how much a team takes its foot off the gas, feeling safe enough. Coaches vary. Players hunt stats and wins. In the end, it is flat out about wins.

Early streaks can have luck and may not be fully sustainable, but in general teams try for playoff seeds or tank and might lead to wider distributions as the season progresses. I haven't seen such a study, just guessing it might go that way. I assume the signal is more powerful than the noise.

We'll eventually see. Til then it is guessing. Desired, appreciated, improving guessing but guessing.

[Crow]

By the various degrees of de-regression, Pelton, for example, would be either 2nd, 4th, 5th, 7th or 8th. Somewhere in there. The less de-regression the better and vice versa.

[DQuinn1575]

By the various degrees of de-regression, Pelton, for example, would be either 2nd, 4th, 5th, 7th or 8th. Somewhere in there. The less de-regression the better and vice versa.

lol, And I am the Anti-Pelton, going 8,7,4,2,2

[Mike G]

Code: Select all

.  avg err   rmse   r^2      .  avg err   rmse   r^2
TmTj   6.17   8.5   .51      medi   6.85   9.5   .44
KPel   6.23   7.8   .56      Crow   6.92   9.2   .46
avgA   6.27   8.4   .48      24pr   7.25   9.5   .34
bpmW   6.29   8.3   .49      emin   7.34   9.7   .41
WShr   6.33   8.6   .46      eWin   7.48   9.7   .31
DQin   6.56   8.7   .50      24py   7.64  10.9   .34
eExp   6.60   8.6   .55      perW   8.12  10.2   .25
vegas  6.61   8.9   .48      bmgm   8.36  11.2   .36
dtka   6.63   8.2   .52               

Without the 'correction' we imposed on Teemo, he'd be in 6th; and very close to 9th.

UPDATE Dec. 27

Code: Select all

.   avg err   rmse   r^2       .  avg err   rmse   r^2
KPel   6.12   7.79   .56      medi   6.91   9.48   .44
bpmW   6.14   8.28   .48      24pr   7.00   9.45   .33
TmTj   6.15   8.41   .51      Crow   7.01   9.29   .45
WShr   6.17   8.53   .45      eWin   7.48   9.63   .31
avgA   6.30   8.39   .48      emin   7.49   9.74   .40
eExp   6.54   8.57   .54      24py   7.70  10.95   .33
vegas  6.67   8.90   .47      perW   8.19  10.16   .24
dtka   6.68   8.24   .51      bmgm   8.44  11.28   .35
DQin   6.75   8.76   .49               	

UPDATE Dec. 28

Code: Select all

.  avg err   rmse   r^2      .  avg err   rmse   r^2
TmTj   6.22   8.5   .51      medi   7.00   9.6   .43
bpmW   6.24   8.3   .49      24pr   7.13   9.5   .33
WShr   6.25   8.6   .45      Crow   7.14   9.4   .45
KPel   6.35   8.0   .56      emin   7.46   9.7   .41
avgA   6.38   8.4   .48      eWin   7.52   9.7   .31
eExp   6.59   8.6   .53      24py   7.71  11.0   .33
vegas  6.62   8.9   .47      perW   8.22  10.2   .24
dtka   6.70   8.3   .52      bmgm   8.46  11.3   .35
DQin   6.76   8.8   .49               

[TeemoTeejay]

Lakers 55 wins was probably pushing it but barring injury I think it’s safe to say they’ll outperform their srs this year

[Mike G]

If we used some exponent of each game's MOV, like .85 or .90, this would reduce the weight of blowout contests. Lakers this year have lost several of those and not won many.
Another way to adjust the rating would be to weigh recent games more heavily. And then we'd have exceptions like a star player about to return from injury.

Last night was busy and resulted in changes from -.10 (bpm) to +.15 (TmTj)

Code: Select all

.  avg err   rmse   r^2      .  avg err   rmse   r^2
bpmW   6.14   8.4   .48      24pr   7.01   9.5   .33
WShr   6.22   8.6   .45      medi   7.03   9.6   .43
KPel   6.34   8.0   .55      Crow   7.18   9.4   .44
avgA   6.37   8.5   .47      eWin   7.57   9.7   .30
TmTj   6.38   8.5   .50      emin   7.59   9.7   .40
eExp   6.57   8.6   .52      24py   7.72  11.0   .33
dtka   6.66   8.3   .51      perW   8.26  10.3   .23
vegas  6.70   9.0   .46      bmgm   8.50  11.3   .35
DQin   6.81   8.8   .49               

[Crow]

BPM and WinShare sets use Pelton minute estimates and they along with his are 1,2,3, then those minute estimates are probably pretty good. His player ratings appear just slightly worse on average than BPM and Winshares at the moment but plenty of time for jostling.

[Mike G]

Update on the over- and under-achieving teams relative to the averages of all our guesses -- known as avgA errors.
From the "improved" portion of my worksheet, I've summed the player eWins ups and downs from last season to this one for each team; multiplied them by 82/G (to make them full season projections) and again by *2 (because that's how eWins works).
Ranked by how much better they are doing than what 'we' expected.
https://www.apbr.org/metrics/viewtopic.php?t=10158

Code: Select all

west  eW++   over     east   eW++   over
Mem    7.8   11.4      Cle    9.2   15.7
Hou   -0.1   10.0      Det    4.5    7.5
LAC   12.0    9.2      NYK    7.8    7.2
OKC    9.4    8.4      Chi    2.3    5.9
SAS   10.8    4.9      Brk    5.2    5.4
Dal   -1.2   -0.2      Orl   10.0    4.8
LAL   -1.6   -0.4      Atl    8.3    2.7
Por   -7.2   -0.6      Mia   -5.6    1.7
GSW    1.8   -1.4      Bos   -3.7   -0.5
Den   -9.9   -2.1      Mil   -3.0   -2.2
Min  -14.7   -6.4      Cha    7.7   -3.3
Uta   -2.8   -7.0      Tor    9.5   -3.5
Phx   -4.1   -9.0      Was    0.7   -3.7
Sac   -9.5   -9.6      Ind  -10.1   -5.7
NOP  -16.8  -26.8      Phl  -16.7  -12.3
correlation   .81     correlation    .67

Not considered in eWins++ are the effect of players missing time or rookie contributions.
The Sixers, with almost everyone playing worse than last season and their best players missing much time, somehow are doing better than all that would predict. Pacers too. Raps and Hornets not so much, but they've also had their stars out of action.
Out west, better correlation but why are the Rockets so overachieving? And the Blazers, Nuggets?

and the daily:

Code: Select all

.  avg err   rmse   r^2      .  avg err   rmse   r^2
WShr   6.09   8.5   .47      DQin   6.93   8.8   .49
bpmW   6.15   8.3   .49      24pr   6.99   9.5   .35
KPel   6.39   8.0   .57      medi   7.04   9.5   .44
avgA   6.41   8.4   .49      Crow   7.17   9.4   .45
TmTj   6.41   8.4   .52      emin   7.54   9.7   .41
eExp   6.57   8.7   .53      24py   7.60  10.8   .35
dtka   6.62   8.2   .53      eWin   7.63   9.7   .31
vegas  6.67   8.9   .48      perW   8.24  10.2   .25
DQin   6.93   8.8   .49      bmgm   8.35  11.2   .36