2020-21 team win projection contest

[liminal_space]

Measuring stick is now 6 part, including last season actuals (from BRef)? 6 part is fine by me. Using last season actuals may have its place but was surprised by it returning. Eventually new B-Ref projections will replace this old data, I assume.

the bballref components are not last season actuals -- they are based on SRS and pythag as you can see here. https://docs.google.com/spreadsheets/d/ ... 1416875271

the arbitrary measuring stick right now is:
0.15 * bballref SRS + 0.15 * bballref pythag +
0.15 * ctg net rtg pythag + 0.15 * ctg exp wins +
0.3 * teamrankings +
0.1 * 538 RAPTOR (since we think this one is bad?)

[Crow]

Guess I misunderstood the old BRef's page continued presence in the workbook, partly because BRef was being used twice but apparently in 2 new ways rather than one new / one old as I assumed.

Keep us up to date on any further measuring stick changes.

[Mike G]

For whatever reason, errors (defined by liminal) have gotten much worse in the last 2 days.

Code: Select all

MAE:    2-08    2-06
shad    4.52    3.96
bbst    4.63    4.00
avg.    4.66    4.20
TeRa    4.82    4.19
dtka    4.83    4.39
EExp    4.86    4.20
vegas   4.90    4.45
trzu    4.90    4.33
Crow    4.94    4.27
538R    5.14    4.73
cali    5.21    4.84
KPel    5.35    5.04
lisp    5.37    4.84
DSM1    5.37    4.93
emin    5.59    5.17
538E    6.14    5.65
EBPI    7.54    7.41

Are current wins and losses part of the projection? They are set in stone, while the SRS and pythag elements are fluctuating.
I wish they'd schedule games 37-72, so b-r would pick up.
Are 'play in games' just not counting in stats or standings? Will we count them for this contest?
[Ja Morant scored 35 in that game, but his high for the year was 30, and his PPG remained at 17.8]

[liminal_space]

For whatever reason, errors (defined by liminal) have gotten much worse in the last 2 days.

Code: Select all

MAE:    2-08    2-06
shad    4.52    3.96
bbst    4.63    4.00
avg.    4.66    4.20
TeRa    4.82    4.19
dtka    4.83    4.39
EExp    4.86    4.20
vegas   4.90    4.45
trzu    4.90    4.33
Crow    4.94    4.27
538R    5.14    4.73
cali    5.21    4.84
KPel    5.35    5.04
lisp    5.37    4.84
DSM1    5.37    4.93
emin    5.59    5.17
538E    6.14    5.65
EBPI    7.54    7.41

Are current wins and losses part of the projection? They are set in stone, while the SRS and pythag elements are fluctuating.
I wish they'd schedule games 37-72, so b-r would pick up.
Are 'play in games' just not counting in stats or standings? Will we count them for this contest?
[Ja Morant scored 35 in that game, but his high for the year was 30, and his PPG remained at 17.8]

Yes current W/L are part of the projection if I built the spreadsheet right...

They got worse because I removed the 0.05 weight to our preseason avg projections

No play in games for the contest

[Mike G]

Aha. Good to know; I wasn't comfortable with 'our projections' as part of the projection.
OK then, the contest ends after 72 games -- we hope.

[Mike G]

Code: Select all

.    avg err    rmse
shad    4.52    5.91
bbst    4.58    6.02

avg.    4.71    6.16
EExp    4.74    6.09
TeRa    4.74    6.15
vegas   4.83    6.21
dtka    4.85    6.29
trzu    4.94    6.49

Crow    5.05    6.22
cali    5.15    6.17
538R    5.23    6.67

KPel    5.39    6.81
DSM1    5.39    6.65
lisp    5.48    7.01
emin    5.56    7.35

538E    6.17    7.55

EBPI    7.44    8.94

That top 2 has been unshuffled for 5 days now. Maybe they are very similar?


update Feb. 12 -- a stiff challenge from a pair of outsiders

Code: Select all

.   avg err   rmse       .  avg err   rmse
shad   4.43   5.79      Crow   4.96   6.08
EExp   4.46   5.86      538R   5.09   6.69
TeRa   4.48   5.95      lisp   5.32   6.90
bbst   4.51   5.89      KPel   5.38   6.86
vegas  4.59   5.98      DSM1   5.39   6.66
avg.   4.67   6.06      emin   5.50   7.09
dtka   4.84   6.23      538E   6.25   7.62
trzu   4.84   6.39      EBPI   7.55   8.96
cali   4.87   6.06            

[Mike G]

New at the top is the ESPN Experts' prediction

Code: Select all

.   avg err   rmse       .  avg err   rmse
EExp   4.41   6.19      Crow   5.19   6.46
shad   4.53   6.18      538R   5.37   6.99
bbst   4.58   6.25      lisp   5.54   7.23
TeRa   4.65   6.36      DSM1   5.54   7.01
vegas  4.79   6.37      KPel   5.59   7.16
cali   4.82   6.19      emin   5.87   7.60
avg.   4.83   6.42      538E   6.51   7.70
dtka   4.99   6.53      EBPI   7.73   9.26
trzu   5.01   6.71            

Deep in last place is ESPN Bi-Polar Index

[Mike G]

The espn Experts' average leads our APBR "avg." entry by 0.36 MAE currently (and by 0.28 in RMSE)
Team by team breakdown:

Code: Select all

ESPN Experts looking better      APBR members (avg) look better
over   tm   proj  espn  apbr      over    tm   proj  espn  apbr
6.6   Hou   30.4   32   38.6      -3.2   OKC   25.4   21   24.2
3.9   Min   19.9   28   31.9      -2.5   Bos   39.0   44   41.5
3.1   Sac   27.5   28   31.1      -2.4   Was   24.3   35   32.6
2.7   Phx   44.2   40   37.3      -1.7   Mil   48.2   50   48.1
2.5   Cha   32.4   27   24.5      -1.5   Den   42.9   45   42.3
2.2   NOP   33.7   36   38.2      -1.0   Mia   29.8   44   43.0
1.6   Por   41.0   41   39.4      -1.0   Det   25.3   22   23.0
1.3   GSW   38.7   36   34.7      -0.8   Ind   38.2   39   38.2
1.2   LAL   52.3   49   47.8      -0.6   Orl   24.3   31   30.4
0.9   Phl   43.5   44   42.2      -0.5   NYK   33.7   22   22.5
0.4   Chi   32.1   28   27.6      -0.5   Uta   57.9   43   43.5
0.2   Dal   34.3   43   43.2      -0.3   SAS   38.5   31   31.3
0.2   LAC   50.3   46   45.8      -0.2   Tor   38.6   42   41.8
0.1   Cle   20.6   22   22.1     -0.02   Mem   34.0   33   35.0
.05   Atl   34.5   35   35.1      0.00   Brk   44.5   43   43.0

I realize Kevin Pelton affects both sides; if we disown him, we look better.

Feb. 19 update: Major shifting near the top, and a close race.

Code: Select all

.   avg err   rmse
EExp   4.54   5.99
cali   4.59   5.99
bbst   4.70   6.15
shad   4.71   6.06
vegas  4.71   6.14
TeRa   4.72   6.19
avg.   4.88   6.26
dtka   5.00   6.41
trzu   5.02   6.58
Crow   5.08   6.40
538R   5.38   7.07
DSM1   5.39   6.87
lisp   5.56   7.09
KPel   5.57   6.95
emin   5.78   7.38
538E   6.49   7.90
EBPI   8.02   9.44

EXPN was up by .25 a few days ago, now .05

Feb 23 update; several are now within a good day's reach of the lead.

Code: Select all

.   avg err   rmse       .  avg err   rmse
EExp   4.44   5.76      Crow   4.97   6.38
shad   4.49   5.92      DSM1   5.20   6.82
bbst   4.49   6.03      lisp   5.34   7.10
cali   4.52   5.93      KPel   5.39   6.82
vegas  4.55   5.95      538R   5.44   7.00
TeRa   4.61   6.10      emin   5.85   7.40
avg.   4.70   6.21      538E   6.33   7.80
dtka   4.80   6.33      EBPI   7.66   9.27
trzu   4.87   6.55            

I subtracted 0.3 W from each of the Vegas predictions, so they sum to 36 W/tm. This immediately improved their errors, and they along with bbst and cali have their best numbers of the season.
EXPN leads at exponent from 0.86 to 3.5; above that range it's caliban, and below is bbstats

A week or so ago, East teams were underperforming relative to 'our' predictions by almost one win per team (projected to 72 games). Then they started winning their games vs the West.

Code: Select all

West      East
OKC  94 108 Mia
Uta 132 110 Cha
LAL 124 127 Was
Hou 100 120 Chi
Mon: W 1-3 E

NOP 120 115 Bos
OKC 117 101 Cle
Min  99 103 NYK
Den 115 123 Atl
LAC 108 112 Brk
Sac 115 128 Mil
Sun: W 2-4 E

GSW 100 102 Cha
LAL  94  96 Mia
Sac 114 122 Chi
Por 111 118 Was
Sat: W 0-4 E

Mem 109  95 Det
OKC  85  98 Mil
Min  81  86 Tor
Den 120 103 Cle
GSW 120 124 Orl
Fri: W 2-3 E

LAL  98 109 Brk

Hou 113 118 Phl
Min 128 134 Ind
Den 128 130 Was
GSW 120 112 Mia
Wed: W 1-3 E

Den  99 112 Bos
Phx 104 128 Brk
Tue: W 0-2 E

That's a full week of losing efforts by the mighty West, and East teams are now looking an avg 0.2 wins better than expected.

[Crow]

BRef should have a second half of schedule informed playoff projection soon. Other projections should become similarly informed eventually.


3 of the listings on the database have the advantage of being part of the measuring stick. That's ok and won't affect final outcome but should be kept in mind.

Are they and other listings not actually entered directly here candidates to "win"? Opinions may vary. But we'll see the comparative results anyways. Maybe multiple winners in "class" and "at large".

Winning is the goal here but close is respectable. A couple tenths more average error isn't much.

[Mike G]

B-R.com now has full season schedule but as yet no Playoff Prob. Report.
Suddenly a dead heat at the top:

Code: Select all

.   avg err   rmse       .  avg err   rmse
EExp   4.49   5.69      trzu   4.99   6.53
vegas  4.49   5.84      DSM1   5.26   6.78
shad   4.51   5.88      lisp   5.39   7.04
bbst   4.51   6.01      KPel   5.44   6.85
cali   4.60   5.88      538R   5.45   7.14
TeRa   4.63   6.00      emin   5.68   7.22
avg.   4.78   6.15      538E   6.43   7.96
dtka   4.95   6.31      EBPI   7.87   9.41
Crow   4.98   6.30            

The co-leaders' MAE are both 4.48666...
At exponent down to 0.92, Vegas leads the field; up to 4.3 it's E-Experts
Only at 1.00 are they tied.

[Mike G]

This morning, the liminal spreadsheet had everyone's errors worsening by .30 on avg, from yesterday.
This afternoon, b-r.com has their forecast up. The avg error is about the same for 17 entries:

Code: Select all

limnsp  MAE   rmse     bk-ref   MAE   rmse
vegas  4.82   6.20      cali   4.86   6.08
EExp   4.86   6.07      vegas  4.93   6.14
bbst   4.89   6.40      bbst   4.98   6.30
shad   4.91   6.29      EExp   5.03   6.07
cali   4.95   6.25      shad   5.11   6.26
avg.   5.01   6.52      avg.   5.13   6.37
TeRa   5.04   6.40      TeRa   5.19   6.37
trzu   5.19   6.88      trzu   5.22   6.72
Crow   5.22   6.53      KPel   5.24   6.66
dtka   5.22   6.75      dtka   5.33   6.70
DSM1   5.46   7.07      Crow   5.44   6.53
lisp   5.59   7.39      lisp   5.51   7.20
KPel   5.66   7.15      DSM1   5.61   6.95
538R   5.67   7.52      538R   5.74   7.46
emin   6.06   7.61      emin   5.85   7.47
538E   6.64   8.29      538E   6.28   7.93
EBPI   8.24   9.80      EBPI   8.50  10.00

In the list at right, mean absolute error is season worst for EExp, shad, TeRa, dtka, Crow, 538R, and EBPI; season best for KPel.
As I hand-entered the team win projections from b-r.com, it was clear almost every team was regressed toward .500 W%. That has historically been the case with them. Still, I may use only their projections in these updates.


update Feb 27
bbstats is #2 on both the liminal ranking and the b-r.com ranking; #1 on the average of them

Code: Select all

.   avg err   rmse       .  avg err   rmse
bbst   4.84   6.19      Crow   5.26   6.41
cali   4.85   6.09      lisp   5.47   7.15
EExp   4.87   5.93      KPel   5.48   6.84
vegas  4.88   6.05      DSM1   5.49   6.94
shad   4.93   6.13      538R   5.61   7.35
TeRa   5.05   6.24      emin   5.87   7.40
avg.   5.05   6.32      538E   6.33   7.89
trzu   5.14   6.66      EBPI   8.12   9.66
dtka   5.25   6.62            

Note the MAE leader is worst in RMSE among the top 5, which are all essentially tied for first place.
b-r.com projections are thus far buried in the Frivolities section:
https://www.basketball-reference.com/fr ... _prob.html


March 1 Update: We now have 3 entries looking better than Vegas, 2 of which are above the espn insiders.

Code: Select all

.   avg err   rmse       .  avg err   rmse
cali   4.89   6.25      dtka   5.29   6.86
bbst   4.91   6.40      lisp   5.39   7.36
EExp   4.92   6.03      KPel   5.52   7.05
shad   4.96   6.35      DSM1   5.56   7.15
vegas  4.98   6.27      538R   5.71   7.56
trzu   5.05   6.87      emin   6.07   7.71
avg.   5.05   6.54      538E   6.34   7.95
TeRa   5.06   6.47      EBPI   8.16   9.83
Crow   5.19   6.50            

Our 'avg' is now only .13 away from the EExperts. The top half is quite a crowded field.

[Crow]

I'm only 0.27 off the MAE lead today on the liminal space table, less than I have been recently. That is not much.

Compared to leader there, I see 7 teams causing me notable slippage. 5 where I have some comparative gain. Probably going to come up short but it might stay fairly close or get closer.

In the end, it will be whose projections are best. Right now that is tempered or possibly clouded by how good the measuring stick is / will be later.

[Mike G]

Here the entrants are ranked by how 'bold' their predictions were. Just counted the number of highest and lowest predictions for the 30 teams; and double counted for guesses that were at least 2 wins more or less than anyone else: outliers.

Code: Select all

hilo   x     out+   Hi   Low   out-
22    538E     5     7     7     3
13    Crow     3     6     3     1
12    emin     1     5     5     1
9     538R     1     3     3     2
9     KPel     0     3     4     2
5     cali     1     3     1     0
4     dtka     0     1     3     0
4     DSM1     0     1     3     0
3     vegas    0     2     1     0
2     EExp     0     0     2     0
1     lisp     0     1     0     0
1     TeRa     0     1     0     0
0     shad     0     0     0     0
0     trzu     0     0     0     0
0     bbst     0     0     0     0

28    EBPI     7     8     8     5

EBPI seems to be a dummy entry created to make EExp look good. High and Low counts for everyone else disregard the EBPI guesses; but IF we include them, these are their #s.
These include "tied for highest/lowest" entries, so the Hi and Low sum to >30.
A few of us have Zero highest or lowest guesses; and only 6 of 15 ventured into any outlier territory.
Our 'avg.' entry obviously would be all zeroes so isn't listed.

Meanwhile the average of 2 systems produces a logjam at the top:

Code: Select all

.   avg err   rmse       .  avg err   rmse
EExp   4.79   5.90      dtka   5.21   6.81
bbst   4.79   6.28      lisp   5.34   7.30
vegas  4.79   6.13      KPel   5.41   7.01
cali   4.81   6.17      DSM1   5.48   7.08
shad   4.85   6.25      538R   5.53   7.47
avg.   4.95   6.45      emin   5.94   7.56
TeRa   4.96   6.35      538E   6.35   7.94
trzu   4.99   6.79      EBPI   8.06   9.75
Crow   5.16   6.40            

In the b-r.com 'system', caliban has a healthy .13 lead over Vegas, then EExp
I made an entry that predicts 36 wins for every team; EBPI is worse than that.
East teams are now looking half a win better than predicted, on avg.


Update March 3
Everyone got worse overnight, by as much as .26 -- except for Crow, who improved by .01

Code: Select all

.   avg err   rmse       .  avg err   rmse
bbst   4.88   6.33      lisp   5.41   7.32
vegas  4.90   6.20      dtka   5.44   6.85
EExp   4.91   5.98      KPel   5.45   7.03
cali   4.94   6.21      538R   5.53   7.43
shad   5.03   6.33      DSM1   5.66   7.18
avg.   5.10   6.51      emin   6.14   7.68
trzu   5.13   6.84      538E   6.37   7.99
Crow   5.15   6.46      EBPI   8.33   9.89
TeRa   5.15   6.45            

EExp with their worst errors of the season.


Update Mar. 4
Another bad night: avg .16 worse than yesterday

Code: Select all

.   avg err   rmse       .  avg err   rmse
vegas  5.01   6.29      KPel   5.53   7.15
cali   5.02   6.30      lisp   5.58   7.51
bbst   5.03   6.48      dtka   5.60   7.00
EExp   5.08   6.10      538R   5.65   7.62
shad   5.19   6.47      DSM1   5.80   7.34
avg.   5.26   6.67      emin   6.27   7.83
trzu   5.33   7.01      538E   6.59   8.14
TeRa   5.33   6.60      EBPI   8.55  10.13
Crow   5.33   6.65            

Gainers among the losers: cali and KPel, both of whom were only .08 off and moved up 2 spots.
Worst numbers of the season for EExp, shad, trzu, TeRa, dtka, DSM1, EBPI, and avg.

[Crow]

Caliban with the lead in the direct entry to contest class.



"Vegas" doing comparatively better than in last few years (as I can recall).

[Mike G]

The Utah Jazz are defying expectations, thanks to their top 8 players all doing as well or better than their previous career numbers.

Code: Select all

.             2020-21 Season              thru 2020
. Utah    Min   PER   ws/48   BPM     PER   ws/48   BPM
O'Neale  1181   10.0   .120   0.4      9.6   .098   0.2
Mitchell 1154   19.0   .134   1.7     17.5   .099   1.5
Gobert   1105   22.5   .220   3.4     21.5   .218   3.8
Bojan    1102   13.3   .114  -0.7     13.7   .082  -1.1
Clarkson  937   18.8   .137   2.7     15.6   .055  -0.7
Conley    853   19.2   .194   3.9     17.5   .131   2.3
Ingles    846   16.5   .190   3.9     12.5   .110   1.0
Favors    550   20.2   .150   2.3     18.9   .147   0.9
Niang     481    9.9   .081  -1.9     11.4   .085  -0.8

. Jazz   8209   16.7   .151   1.8     15.4   .114   0.9
.     x5               .756   9.1            .572   4.4

That bottom line suggests they 'should be' a 57% winning team, but instead are 75% (27-9 currently)
Also that they're winning by 9.1 ppg on avg (it's +8.8, but SRS = 9.5)