2020-21 team win projection contest

[Crow]

There is a difference between pre-season predictions and extrapolation of current season data, or there should be.

Ultimately the predictions will be measured against the actuals. If the in season projections are not well regarded, could just measure against simple season to date performances.


I am not good with formatting tables or graphs here.

[vzografos]

There is a difference between pre-season predictions and extrapolation of current season data, or there should be.

Ultimately the predictions will be measured against the actuals. If the in season projections are not well regarded, could just measure against simple season to date performances.

Sure there is a difference.
1) Pre-season predictions without any current season data
2) Remaining season predictions using all existing data (past and current)
3) Game to game prediction. So, not predicting the whole season but predicting the next game only.

I am only tracking the seasonal performance of 3) because 1) and 2) are not interesting for me and a lot of experts predict game by game. So, I can only quote the data I am looking at. However, I am certain that there is a correlation on the accuracy of 3), 2) and 1). And if a predictor is bad in 3), I cannot see how they will be good at 1).

At any rate, 538 has published their predictions for the whole season (this is 1) from above ) that you can download in a file and you can crunch the numbers. I am actually looking at the numbers they publish every day.

[liminal_space]

am late to do the spreadsheet. working on it now

[liminal_space]

here is v1 of the sheet: https://docs.google.com/spreadsheets/d/ ... sp=sharing

for the projection column I am just ensembling 538 RAPTOR and bbref. taking suggestions on how to make that number "better"

[Crow]

Thank you.

538 Raptor and Elo in lead, partly because of 538 data involvement in the measuring stick probably.

What did you use from BRef?
This page
https://www.basketball-reference.com/fr ... _prob.html
is from last season still. It appears you used this. It should be replaced for now. Will be available later.

I might use combo of current record projected as is and 538 Elo for now. BRef and ESPN projections could be added to mix later or replace the currently available options.

[liminal_space]

Thank you.

538 Raptor and Elo in lead, partly because of 538 data involvement in the measuring stick probably.

What did you use from BRef?
This page
https://www.basketball-reference.com/fr ... _prob.html
is from last season still. It appears you used this. It should be replaced for now. Will be available later.

I might use combo of current record projected as is and 538 Elo for now. BRef and ESPN projections could be added to mix later or replace the currently available options.

Thanks for pointing that out re: bballref. I have removed that from the projection.

The current formula is

Code: Select all

0.45 * raptor + 0.45 * teamrankings + 0.1 * avg of entrants

[Crow]

Sounds like a better mix.

I'm 0.2 off the MAE lead. Want Lakers to stay high, TWolves very low. Nuggets, Hawks, Rockets, Nets, Griz and Jazz higher than others projected. Bulls and Blazers lower. It would help if Pelicans got a lot better but I am not very hopeful on that one.

[Mike G]

here is v1 of the sheet: https://docs.google.com/spreadsheets/d/ ... sp=sharing

Is there an entry for what we have previously labeled "vegas"? Circa maybe?

[Mike G]

Seeing no contest spreadsheet last night, I went ahead and did one.
Now we have liminal's sheet, so I'm using his projections.
Relative to the average of 10 APBR-ists (incl. Pelton and bbstats), teams are projecting to over- or under-achieve after 72 games:

Code: Select all

over    tm        uner    tm
8.8    Uta       -15.5    Mia
8.3    NYK       -12.5    Min
6.8    Cha        -8.3    Dal
6.0    LAC        -8.3    Was
4.7    LAL        -6.6    Orl
4.6    SAS        -6.2    NOP
4.6    Phl        -4.7    Tor
4.4    Cle        -4.0    Por
3.3    Den        -2.5    Sac
3.2    Mem        -1.3    Hou
3.2    Atl        -0.2    GSW
3.0    Phx            
1.9    Mil            
1.6    Brk            
1.4    Chi            
0.9    OKC            
0.8    Ind            
0.3    Det            
0.1    Bos            

The Dubs were a vast uncertainty -- predictions ranging from 28 (caliban) to 43 (eminence). Yet they're right on pace to meet their 35-win avg call.
Heat and Wolves (as usual) total -28 from projected wins -- basically giving a win to every other team.
With the season almost 30% completed, I'd expect b-ref.com to have their Playoff Projections appearing.

[Crow]

20 with a variance of 5 or less. 9 within 2.

B-Ref says they'll publish when the second half schedule is released.

[Crow]

Second dataset may increase updates and facilitate more specialized analyses like the above.

[Crow]

I see liminal space's sheet is updating.


BPI the clear worst scored projections. 538 Elo is the next worst.

The Apbrmetrics average is doing well, with a small edge over the ESPN expert panel.

Yes, Circa must be the "Vegas" stand in. Circa Sports is a betting book.

Including teamrankings projections in the measuring stick was probably a good call. 538 Raptor might be suspect and perhaps should be reduced in weight or eliminated later.

[Mike G]

Thumbnail rankings update

Code: Select all

.     avg err   rmse          .   avg err   rmse
shad    4.23    5.67        cali    4.88    5.97
bbst    4.30    5.73        lisp    5.01    6.57
TeRa    4.41    5.90        KPel    5.08    6.66
Crow    4.43    5.88        538R    5.09    6.42
avg.    4.44    5.88        DSM1    5.10    6.46
EExp    4.45    5.93        emin    5.42    7.06
vegas   4.49    5.97        538E    5.94    7.46
trzu    4.55    6.12        EBPI    7.77    9.08
dtka    4.61    6.07                

avg refers to 10 apbr people. vegas = Circa

[eminence]

Hot start for team eminence it seems.

[Crow]

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.