Page 4 of 13
Re: 2022-23 team win projection contest
Posted: Sat Nov 12, 2022 10:49 pm
by Crow
A lot of analysis (mine and probably others) is really simple net rating based. I didn't look at schedule, simulate a season or look at relative team game to game variation. All appropriate but too much work imo for this contest. Perhaps I shade things some based on background awareness of team game to game behavior & explicitly checking actual vs expected win results in some but not all cases. More in past than I did this time.
Objectively I would be neutral about 3 misses by 3 compared to 2 direct hits and a miss by 9 but it might depend on the "reason" for the miss by 9. If it was injury or unexpected trade, I might not feel bad about the miss by 9. I feel pretty good about direct hits, even if "lucky" to some degree.
Re: 2022-23 team win projection contest
Posted: Sun Nov 13, 2022 12:07 am
by Crow
I have sometimes found significant differences in my performance by conference. Relative domain knowledge, confidence, "expertise".
I have sometimes checked my results by tiers in W-L projections and results (high, middle and low). I have sometimes changed my behavior in specific tiers a bit from year to year because of results but can't say I know the impact of adjustments with precision or the long-term record. I didn't review last season performance.
Re: 2022-23 team win projection contest
Posted: Sun Nov 13, 2022 12:42 am
by Mike G
In the last 14+ seasons (since 2007-08), the Min. T-Wolves have not once won more games than their point-differential would suggest. They've won an average of 3 fewer games per season than they 'should have' in that interval.
Does anyone consciously take into account a franchise's apparent "culture" of winning or losing when making predictions?
( I tried it one year, but found no benefit; and it felt like 'cheating', since I was testing a system of player evaluation.)
Re: 2022-23 team win projection contest
Posted: Sun Nov 13, 2022 3:41 am
by Crow
I'd say few current teams have medium or long-term winning vibes right now, less than at points in past. Couple years, sure. Losing vibes are pretty short too.
Don't know off the cuff the average coaching tenure is with current team but it is probably pretty low relative to other times.
Only 8 coaches in 4th year or more with current team.
Re: 2022-23 team win projection contest
Posted: Mon Nov 14, 2022 12:43 pm
by Mike G
Another crazy night of games and shakeup in the contest:
Code: Select all
. avg err rmse avg err rmse
ncs. 5.77 7.40 TmRk 7.18 9.41
EBPI 6.06 7.62 vegas 7.22 9.53
vzro 6.07 7.40 veg2 7.23 9.51
drko 6.12 7.78 2022 7.36 9.39
dtka 6.50 7.73 nuFi 7.61 9.48
538E 6.64 8.73 LEBR 7.76 9.46
Crow 6.76 8.53 22Py 7.84 10.06
trzu 6.86 8.46 EExp 7.97 9.82
emin 6.98 8.69 MPra 8.10 10.21
538R 7.02 8.03
ncs' lead is less than half what it was 4 days ago.
https://www.basketball-reference.com/fr ... _prob.html
Re: 2022-23 team win projection contest
Posted: Tue Nov 15, 2022 4:57 am
by nbacouchside
Mike G wrote: ↑Mon Nov 14, 2022 12:43 pm
Another crazy night of games and shakeup in the contest:
Code: Select all
. avg err rmse avg err rmse
ncs. 5.77 7.40 TmRk 7.18 9.41
EBPI 6.06 7.62 vegas 7.22 9.53
vzro 6.07 7.40 veg2 7.23 9.51
drko 6.12 7.78 2022 7.36 9.39
dtka 6.50 7.73 nuFi 7.61 9.48
538E 6.64 8.73 LEBR 7.76 9.46
Crow 6.76 8.53 22Py 7.84 10.06
trzu 6.86 8.46 EExp 7.97 9.82
emin 6.98 8.69 MPra 8.10 10.21
538R 7.02 8.03
Lead is non-existent if you go by RMSE (what I built the player projections towards). Tied with v-zero! Also don't expect any lead or tie for first to sustain, but it's been fun being in first for a little while.
Re: 2022-23 team win projection contest
Posted: Tue Nov 15, 2022 7:53 am
by Mike G
RMSE are minimized by cranking up the regression-to-41, which moves every team closer to .500. Then the co-leaders are still tied at 7.17, and all teams are projected between 49 wins (Bos, Phx) and 29 (Det).
It seems pretty safe to avoid predicting very high or low win totals. A couple of entries had no highest/lowest guesses: dtka and TmRk.
ncs was tied for lowest on Mil (48) which looks very good right now.
trzu was only tied for highest on Cle (48) also very good.
vzro was high on Cle, Det, Was; low on Mem -- mostly good.
Re: 2022-23 team win projection contest
Posted: Tue Nov 15, 2022 3:40 pm
by dtkavana
I use the "predictive" version of Crafted +- to make my projections. I use my own minutes projections which take into account the last two seasons, and then I go and adjust each team's rotations manually.
I then basically do pretty much exactly what 538 does
https://projects.fivethirtyeight.com/20 ... s/celtics/ on their team pages to get the wins totals. I used to simulate the season using the real schedule, but found that this worked just as well.
The only manual adjustments I make are to slightly downgrade teams I think will "embrace the tank" and then re-allocate those wins elsewhere.
https://craftednba.com/projections
Re: 2022-23 team win projection contest
Posted: Tue Nov 15, 2022 3:42 pm
by dtkavana
Oh, and as for the Warriors, Crafted +- is just not high on many of their younger rotation players, nor is it as high on Klay as conventional wisdom.
Re: 2022-23 team win projection contest
Posted: Thu Nov 17, 2022 4:42 pm
by tarrazu
Not sure what we use for projected wins but think TR does a good job for their season long win numbers:
https://www.teamrankings.com/nba/
Re: 2022-23 team win projection contest
Posted: Thu Nov 17, 2022 10:24 pm
by Mike G
I've been using the b-r.com projections here; but today the site has not updated, so for this occasion we are seeing current W-L + mov-suggested W% of remaining games.
To deliberately minimize avg errors, I'm using 70% that ^ number and 30% "41".
Code: Select all
. avg err rmse avg err rmse
ncs. 5.59 7.10 veg2 7.18 9.22
drko 5.63 7.24 TmRk 7.22 9.09
vzro 5.70 7.06 vegas 7.22 9.25
dtka 5.95 7.25 nuFi 7.39 9.24
EBPI 6.01 7.34 LEBR 7.42 9.10
trzu 6.30 7.94 EExp 7.80 9.38
Crow 6.38 7.93 2022 7.84 9.54
538E 6.53 8.41 MPra 7.84 9.75
emin 6.67 8.34 22Py 8.01 10.02
538R 6.75 7.69
There are distinct tiers with this edition. Almost everyone on the right is worse, but the overall avg error is unchanged.
UPDATE Nov.18 (back to b-r.com forecast)
Code: Select all
. avg err rmse avg err rmse
ncs. 5.42 6.94 veg2 7.05 9.05
DRKO 5.60 7.17 vegas 7.11 9.06
vzro 5.81 7.05 TmRk 7.11 8.93
EBPI 5.89 7.24 LEBR 7.20 9.00
dtka 5.98 7.13 nuFi 7.40 9.11
Crow 6.17 7.82 EExp 7.63 9.25
trzu 6.27 7.80 2022 7.89 9.47
538E 6.50 8.48 MPra 7.91 9.63
538R 6.75 7.72 22Py 7.95 10.09
emin 6.79 8.36
Seeing only insubstantial differences between these projection models. And I am gratified to see all APBRMetrist entries in the above-avg section.
UPDATE Nov. 19
Code: Select all
. avg err rmse avg err rmse
ncs. 5.62 7.08 veg2 7.17 9.17
EBPI 5.67 7.12 vegas 7.19 9.21
DRKO 5.68 7.26 TmRk 7.20 9.08
vzro 6.01 7.18 LEBR 7.21 9.03
dtka 6.02 7.20 nuFi 7.51 9.21
trzu 6.25 7.84 EExp 7.77 9.43
Crow 6.34 7.99 AnBa 7.77 9.41
538E 6.50 8.50 2022 7.83 9.44
538R 6.70 7.74 MPra 7.85 9.66
emin 6.93 8.48 22Py 7.86 10.02
It's abruptly a 3-way race, with 2 of them being outsiders.
Overnight, ncs was off by .20, while EBPI improved by .22
vzro leads with exponents >2.25; DARKO at .22<e<.86; EBPI below that.
-- On page one of this thread, DarkStar48 posted a link to predictions by Andy Bailey of Bleacher Report. He's now included, nestled neatly among the also-rans.
UPDATE Nov 20 -- new front runner
Code: Select all
. avg err rmse avg err rmse
EBPI 5.67 7.10 LEBR 7.09 8.97
ncs. 5.71 7.11 veg2 7.11 9.16
DRKO 5.73 7.31 vegas 7.15 9.22
dtka 5.99 7.22 TmRk 7.16 9.07
vzro 6.05 7.23 nuFi 7.57 9.24
trzu 6.17 7.85 AnBa 7.73 9.36
Crow 6.29 7.98 EExp 7.73 9.43
538E 6.58 8.54 MPra 7.82 9.67
538R 6.76 7.74 22Py 7.83 10.00
emin 6.94 8.52 2022 7.86 9.45
The LEBRON entry has risen from last place, to top of the 2nd division, in <2 weeks.
Seeing last year's Win (and PythWin) totals have secured the cellar -- and other entries seem suspiciously close in their errors -- I wonder if it's a benefit to guess closer to last year's record.
Correlations between an entry's guesses and
2022 range from around .70 to .90, and the avg is about .80
Code: Select all
corr. w 2022 mae rmse
.898 538E 6.58 8.54
.867 538R 6.76 7.74
.857 vzro 6.05 7.23
.857 EBPI 5.67 7.10
.847 ncs. 5.71 7.11
.845 EExp 7.73 9.43
.820 nuFi 7.57 9.24
.803 dtka 5.99 7.22
.797 AnBa 7.73 9.36
.790 trzu 6.17 7.85
.779 Crow 6.29 7.98
.769 TmRk 7.16 9.07
.760 veg2 7.11 9.16
.759 DRKO 5.73 7.31
.752 vegas 7.15 9.22
.743 emin 6.94 8.52
.719 LEBR 7.09 8.97
.704 MPra 7.82 9.67
corr w 22corr -.35 -.44
...
.950 22Py 7.83 10.00
1.00 2022 7.86 9.45
At this point in the season, it seems beneficial to gravitate toward previous year's W% (as well as toward .500)
Higher continuity goes with lower error, hence negative correlation.
Re: 2022-23 team win projection contest
Posted: Mon Nov 21, 2022 5:46 am
by knarsu3
Can you include the BBIPM projections as well?
Here they are from preseason:
https://imgur.com/a/iCR5aIw
Re: 2022-23 team win projection contest
Posted: Mon Nov 21, 2022 11:44 pm
by Mike G
knarsu --
At that link a table appears with a column headed "Team BBIPM"; and the Win projections are those I've been calling LEBRON all this time.
I got those predictions from this link, posted on page 2 of this thread:
That site also shows the same table, called Team BBIPM, that earlier were called LEBRON ; because bbipm was 'yet unreleased' then.
So I am confused and would welcome some clarification.
Re: 2022-23 team win projection contest
Posted: Fri Nov 25, 2022 5:21 am
by knarsu3
So the issue was that the column header was not named correctly (both BBIPM and LEBRON were named LEBRON in the column header) but if you look at the select metric button, you have 2 options: LEBRON and BBIPM. The 2 different metrics have been there since the start of the season but again, the column header was wrong and was just fixed. So I think that was causing the confusion. The default this whole time has been BBIPM. You can still view the preseason projections in the Preseason predictions tab here:
https://www.bball-index.com/season-team-projections/
To make it easier, I'd just use what I've posted here which is the same thing from the app and also the same thing I posted above with the link to the images.
The
LEBRON projections should be:
Code: Select all
Team Wins
BOS 60.61
MIL 54.00
PHX 50.03
NOP 49.61
MIA 49.55
PHI 49.22
TOR 48.86
MIN 48.62
DEN 47.87
CLE 46.17
ATL 44.98
DAL 44.66
MEM 44.55
GSW 44.20
LAL 42.17
NYK 41.56
CHI 41.22
WAS 40.93
BKN 39.90
LAC 39.83
SAC 36.34
CHA 34.85
POR 32.64
UTA 32.27
IND 31.92
ORL 29.78
SAS 29.14
OKC 26.80
DET 25.02
HOU 22.68
The
BBIPM projections should be:
Code: Select all
Team Wins
MIL 62.78
BOS 62.11
BKN 57.61
NOP 54.56
DEN 52.61
PHX 51.88
LAC 50.49
ATL 50.23
PHI 49.39
MEM 47.23
DAL 46.34
TOR 45.45
MIA 45.07
MIN 44.26
CLE 43.47
GSW 43.20
LAL 40.22
CHI 40.11
WAS 39.17
SAC 39.04
NYK 37.56
POR 34.59
IND 31.96
CHA 30.06
UTA 29.76
ORL 23.99
SAS 23.57
OKC 21.60
DET 18.86
HOU 12.82
Re: 2022-23 team win projection contest
Posted: Fri Nov 25, 2022 6:16 am
by Mike G
Great, that makes all kind of sense. I'll try to work it in for the next summary.
Meantime I suppose I can edit all the previous ones, replacing LEBR with BBIP -- or BIPM, if you prefer -- I just like 4-character column headers.