2018-19 team win projection contest

[Crow]

Seems like there are more big surprises this season (plus or minus 7 games compared to projection average) so far than normal; but I haven't checked the past to try to compare.

[tarrazu]

Largest Errors vs. Average Entry (all teams about +/-7 or more):

Over-Estimated:
HOU: Harden/CP3 missed games early, lost Ast. Coach Bzdelik (initially), Ennis/Melo downgrade, teams adjusting to their switching scheme?
WAS: Internal turmoil, Porter missed games, not a lot of depth, Wall/Morris have struggled
CLE: Love out, Lue fired, Hill traded, youth movement started early but have stabilized of late
GSW: Curry/Green missed games, C by committee, small margin for error with highest win total
PHX: Booker missed games, playing a ton of youth/J-Crossover, Sarver leadership
UTA: Hardest team to pinpoint, difficult/road heavy schedule so far, Favors (and his lineups) have struggled

Under-Estimated:
SAC: Big jumps from Fox/WCS/Bogdan, Bjelica underrated signing, embracing fast-tempo after last year constantly fiddling with old/slow, young/fast lineups
MEM: Healthy Conley one of league’s top point guards, Gasol bounce-back, impact rookie JJJ
MIL: Coach Bud, Giannis, Jabari gone, Brook has fit in well
DAL: Impact rookies Doncic/Brunson, DAJ filled hole at 5, just more stability overall with this year’s roster

[Mike G]

Here are the top ten over- and under-achieving teams, relative to the average of the predictions.

Code: Select all

err  tm   avg  proj      err  tm   avg  prj
19   Sac   23   42      -15   Hou   57   41
10   Mil   48   58      -12   Was   45   33
+9   Dal   31   40      -11   GSW   64   52
+9   Mem   33   42      -11   Uta   54   43
+8   Orl   31   38       -7   Cle   32   25
+6   LAC   38   44       -6   Phl   52   46
+6   Den   49   55       -5   Mia   42   37
+5   Por   40   46       -5   Min   45   41
+4   Ind   46   50       -4   Chi   26   22
+4   Brk   33   37       -3   Atl   25   22
avg overs  37   45      avg unders  44   36

Does it seem we'd do better to regress everyone more toward .500 ?

12/22 update

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   5.33   6.65      bbst   5.99   7.27
cali   5.51   7.32      RyRi   5.99   7.37
vegas  5.71   6.98      sndi   6.07   7.81
shad   5.73   6.88      gold   6.21   7.67
emin   5.77   7.26      538_   6.53   8.04
Jinx   5.89   7.57      Crow   6.77   8.66
KPel   5.90   7.09      kmed   7.15   9.14
J-TF   5.90   8.75      Cica   7.71   9.73
AnJo   5.94   7.11      Rd11   7.83  10.11

[caliban]

30+ games in. Projections courtesy of

TeamRankings, Rankings & Projections Summary
FiveThirtyEight, NBA Predictions
bbref Playoff Probability Report

Merry Christmas

[Mike G]

Bad night for almost everyone. Average .022 worse than yesterday.

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   5.37   6.84      bbst   6.07   7.52
cali   5.56   7.52      sndi   6.23   8.17
shad   5.69   7.01      J-TF   6.33   8.78
emin   5.79   7.41      gold   6.38   7.96
vegas  5.83   7.15      538_   6.58   8.30
KPel   5.84   7.24      Crow   6.69   8.71
Jinx   5.88   7.82      kmed   7.32   9.37
AnJo   5.89   7.29      Rd11   7.95  10.35
RyRi   5.96   7.55      Cica   7.96   9.78

[Crow]

If anyone wants to expound on their methodology or what tweak they think they should have made / will make in future, that could be interesting, especially from the leaders but from anyone who feels they've learned something.

How is the blended average doing? Was a respectable 6th a week ago.

14 better than Vegas a week ago on MAE. Only 3 by RMSE then. Only 4 by RMSE now according to Mike G.

I should probably verify the data entry on my projections. Sometime before April.

[Mike G]

Blended avg of 18 entries would be about 8th, but in a crowded pack might be 5 to 10.
In RMSE, avg ranks 10th today, but is .10 from #7
Pretty much average.

I wonder if previous years' updates would show whether straight avg or RMSE better predicts the season-final rankings.

[Mike G]

Almost no changes in this update, but it's been a week, so:

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   5.29   6.69      bbst   6.11   7.48
cali   5.53   7.22      sndi   6.29   8.01
shad   5.71   6.81      J-TF   6.40   8.62
KPel   5.80   7.08      gold   6.46   7.93
emin   5.82   7.30      538_   6.65   8.15
vegas  5.85   6.97      Crow   6.72   8.55
AnJo   5.86   7.19      kmed   7.29   9.16
Jinx   5.90   7.59      Rd11   7.74  10.10
RyRi   6.07   7.37      Cica   7.90   9.43

Avg error remains 6.30

[Nathan]

This is mildly off topic, but I was thinking...it's well known that teams perform somewhat better when trailing, and it's common practice to account for this when calculating APM from play-by-play data. It would also make sense to account for this in calculating team ratings; for instance, a wire-to-wire 2-point win is, at least in theory, more impressive than a comeback 2-point win. Has anyone actually done this, and if so, does it make much of a difference?

[tarrazu]

This is mildly off topic, but I was thinking...it's well known that teams perform somewhat better when trailing, and it's common practice to account for this when calculating APM from play-by-play data. It would also make sense to account for this in calculating team ratings; for instance, a wire-to-wire 2-point win is, at least in theory, more impressive than a comeback 2-point win. Has anyone actually done this, and if so, does it make much of a difference?

Bart Torvik, who runs the CBB site T-Rank (http://www.barttorvik.com/) added +/- average lead/deficit over the course of a game as a statistic for each team's gamelog. Nathan, you can look at Duke's here for example: http://www.barttorvik.com/team.php?team=Duke.

He blogs a little bit more about it here:http://adamcwisports.blogspot.com/2018/ ... pdate.html and currently utilizes it in his rankings. Seems like an interesting topic to explore further.

[Nathan]

This is mildly off topic, but I was thinking...it's well known that teams perform somewhat better when trailing, and it's common practice to account for this when calculating APM from play-by-play data. It would also make sense to account for this in calculating team ratings; for instance, a wire-to-wire 2-point win is, at least in theory, more impressive than a comeback 2-point win. Has anyone actually done this, and if so, does it make much of a difference?

Bart Torvik, who runs the CBB site T-Rank (http://www.barttorvik.com/) added +/- average lead/deficit over the course of a game as a statistic for each team's gamelog. Nathan, you can look at Duke's here for example: http://www.barttorvik.com/team.php?team=Duke.

He blogs a little bit more about it here:http://adamcwisports.blogspot.com/2018/ ... pdate.html and currently utilizes it in his rankings. Seems like an interesting topic to explore further.

Very interesting! I agree with his thought that average lead is an important metric, with a larger average lead meaning a more impressive game holding margin of victory constant.

On a season level, I think it would actually be pretty straightforward to calculate "adjusted" net ratings that account for average lead. Basically, if a team's average lead for the season is, say, 5 points, then the team's raw net rating really describes how well they play when they're leading by 5 points. From that we can calculate what their net rating is in a tie game, which is in a sense their "real" net rating.

Now, it's possible that net rating and average lead are so strongly correlated that this correction doesn't actually yield useful information...but if anyone decides to investigate I'd be interested to hear the results. I mainly dabble in player stats, not team stats, so I don't have all the relevant information at my fingertips for this mini-project, but I'm sure some people here do.

[Mike G]

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   5.14   6.41      bbst   5.87   7.20
cali   5.48   6.95      sndi   6.13   7.73
KPel   5.56   6.87      J-TF   6.16   8.35
emin   5.57   6.94      gold   6.20   7.78
shad   5.60   6.62      Crow   6.24   8.15
vegas  5.70   6.75      538_   6.46   7.91
AnJo   5.72   7.02      kmed   7.07   8.80
Jinx   5.80   7.30      Cica   7.37   8.91
RyRi   5.83   7.13      Rd11   7.42   9.79

Once again, very little change in the order. Improvement overall, and tarrazu extends his lead.

[Mike G]

Most years, our composite average is a contender for 'best' guess. At present, that entry would rank 6th, but .53 off the pace set by tarrazu.
On the left are teams that trzu is currently beating the average; on the right are those where he looks worse than "avg" , relative to the b-r.com win projections.

Code: Select all

tm   avg  trzu  diff     tm   avg  trzu  diff
Sac   23   27   4.3      OKC   47   42   -5.4
Dal   31   35   4.0      Det   39   42   -3.4
Orl   31   34   3.4      Mia   42   44   -2.1
Phl   52   49   3.3      Ind   46   44   -1.8
LAC   38   41   3.2      Cle   32   34   -1.8
Chi   26   23   2.8      Den   49   48   -1.2
Mil   48   50   2.4      Min   45   46   -1.0
Was   45   43   2.2      Cha   38   39   -0.9
NOP   45   40   2.0      SAS   41   40   -0.8
Hou   57   55   1.7      Bos   54   55   -0.7
Por   40   42   1.6      Phx   25   21   -0.6
Mem   33   35   1.5      LAL   44   45   -0.6
GSW   64   62   1.5      Uta   54   54   -0.3
Brk   33   34   1.1      Atl   25   25   -0.1
NYK   25   24   1.1               
Tor   58   57   0.6               

An "overachieving" team on the left is revealed by trzu predicting more wins than avg. Where he predicted fewer, he correctly predicted the team is underperforming.

update Jan. 18, nice improvements all around:

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   4.82   6.08      RyRi   5.68   6.90
emin   5.14   6.80      gold   5.85   7.46
cali   5.16   6.67      sndi   5.92   7.45
AnJo   5.28   6.66      Crow   5.92   7.92
KPel   5.35   6.65      J-TF   6.07   8.37
shad   5.42   6.46      538_   6.22   7.57
vegas  5.44   6.61      kmed   6.78   8.53
bbst   5.47   6.88      Cica   7.00   8.60
Jinx   5.58   7.01      Rd11   7.21   9.50

Jan.28 update, on avg people have slid by .21

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   5.04   6.31      RyRi   5.87   7.14
cali   5.32   6.72      Crow   5.98   7.99
emin   5.49   7.03      gold   6.08   7.65
KPel   5.64   6.93      sndi   6.11   7.59
AnJo   5.69   7.02      538_   6.40   7.69
shad   5.72   6.80      J-TF   6.59   8.61
vegas  5.74   6.94      kmed   6.78   8.49
Jinx   5.75   7.08      Rd11   7.00   9.37
bbst   5.83   7.09      Cica   7.18   8.67

Previous report is in fact the best looking set of predictions this season, so far.
Only Rd made a jump in the meantime.

Jan. 30, suddenly a close race:

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   5.14   6.32      RyRi   5.81   7.10
cali   5.24   6.69      Crow   5.91   7.96
emin   5.52   7.02      gold   5.97   7.59
AnJo   5.62   6.97      sndi   6.07   7.51
Jinx   5.67   7.00      538_   6.26   7.63
shad   5.67   6.79      kmed   6.62   8.44
KPel   5.67   6.92      J-TF   6.65   8.69
vegas  5.70   6.95      Rd11   6.90   9.27
bbst   5.81   7.04      Cica   7.16   8.64

[Mike G]

After 24 teams playing yesterday, an abrupt turn for the worse:

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   5.38   6.58      bbst   6.15   7.31
cali   5.47   6.97      Crow   6.19   8.23
AnJo   5.81   7.20      gold   6.28   7.84
emin   5.86   7.32      sndi   6.39   7.73
KPel   5.88   7.23      538_   6.46   7.83
Jinx   5.90   7.31      J-TF   6.74   8.82
shad   5.94   7.03      kmed   6.80   8.60
vegas  6.01   7.15      Rd11   7.14   9.53
RyRi   6.07   7.32      Cica   7.51   8.95

The composite avg of 6.22 is worst in almost a month; the leader error is worst since Dec. 21

Feb. 13 update. Continued worsening errors all around.

Code: Select all

avg abs err   rmse      avg abs err   rmse
trzu   5.51   6.64      RyRi   6.25   7.50
cali   5.63   7.11      gold   6.39   7.87
AnJo   5.87   7.35      sndi   6.43   7.83
emin   5.95   7.47      Crow   6.47   8.46
KPel   6.07   7.48      538_   6.57   7.95
shad   6.12   7.27      kmed   6.73   8.63
vegas  6.13   7.36      J-TF   6.90   8.84
Jinx   6.16   7.57      Rd11   7.37   9.75
bbst   6.22   7.43      Cica   7.49   9.09

[caliban]

ASbreak . Projections courtesy of

TeamRankings, Rankings & Projections Summary
FiveThirtyEight, NBA Predictions
bbref Playoff Probability Report

Added a trade deadLine at the bottom to track how it might have impacted the hopeful teams.