APBRmetrics

Separately, re: the discussion about the Basketball-Reference win projections; they look to be adding approximately 7-8 games of league average performance to each team's projected SRS the reason of the season via Bayesian padding.

Based on my testing, this number is well calibrated if you come into the season with no prior. I use their projections as my own sanity check baseline as well for this reason.

kmedved wrote: ↑Tue Dec 09, 2025 12:30 am It looks like this accidentally flipped the Charlotte and Chicago numbers for the Walrus Projections (they had Chicago at 36.17 and Charlotte at 32.55).

Ah, yet again. Great catch and a great time for it, as Cha and Chi are almost exactly equal in their b-r.com projections; so Walrus' current errors are unaffected.
This happens when submissions use CHO for the Hornets, reversing the alphabetic order with CHI.

That mistaken 36.2 for Cha was the only outlier hi/low guess for Walrus and their only high or low in the East.
In the West, they are high on Utah, low on Dal and LAC -- all of which look pretty darn good.

The Thunder are up to 70 wins in my CraftedNBA Projections now. BBREF has them at 66, which would mean they project them to finish the season 47-11.

My projections won't quit on the Clippers, despite all the evidence, and still see them finishing with 39 wins (BBREF 31), which would have them 33-25 in their remaining games.

How do you project player minutes for the rest of the season?
If Kawhi misses 42% of games (as he has so far) and Jones half of remaining games, that's worse than normal for them.

And JDub: does he return to his 87% career norm or project at 5/24 of future games?

B-Ref guesses the Clipps go 25-33 the rest of the way, just the inverse of your projection.
Best scenario of 10k simulations is .655 W%, and worst is .224 -- barely worse than their current .250

Best numbers of the year for almost everyone. https://www.basketball-reference.com/fr ... _prob.html

Could be interesting to compile the average error of "Vegas" year to year. Any trend? "Vegas" source has varied, so maybe it isn't a clean comparison. But bouncing around, getter "better" or "worse"? Better and worse average lines relative to average outcomes is one thing, by team would be another level of detail and what matters for the house and bettors. More "accurate" / "inaccurate" lines for which teams, bettors?

I looked up a few previous years' results and my memory was corrected. "Vegas" has recently (5 years?) done pretty well. Then again, we might expect them to, since it's their money.
What I thought was the case was that "we" almost always had a few who beat vegas, and maybe that was once the case. Whether they have gotten better or we have gotten (relatively) worse, I don't know.

Every team over .500 projected by BRef to have a lower win% from here, every team below. 500 projected to have a higher win%.

Likely to see variety, not that strict uniformity. Probably lots of similar win%s.


Thunder given 56% chance of winning title by BRef, with Knicks and Rockets over 19% and nobody else over 5% and only 4 more over 2%. Still giving 16 teams at least a tiny chance, down from 19 last check. Thunder chances might be too high. 56% chance of title is equivalent to 90% chance of making conference finals and 80% chances of winning that and finals. Are there any matchups with lower chances than that? My quick guess is yes.

Knicks with 5th best overall uear to year improvement on offense, 4th best on defense. Massive change in own 3pta rate and large increase in fta rate and blocks. Top 10 on both offensive and defensive efficiency with 3rd best offense. At factor level, it is 4 top 10s, 3 middle, 1 bottom 10.

Rockets also in top 5 in yr to yr improvement on both sides. Moderate decline in 3pta frequency down to last place, accuracy up to 3rd. Major increase in fta rate. But also major increase in TO rate. Major increase in offensive rebounding rate, modest decline on defensive glass.

Nuggets, tops on offensive improvement, barely above median on defensive improvement. Strong improvement from 3 and ft line. Small slippage on rebounding.

8 upcoming games against these 3, but none til Jan. 15.

Thunder #2 on offensive improvement, #1 on defensive improvement. A little slippage in 3pt frequency, masaive increase in fta rate. Major change in offensive rebounding. 7 top 10 factors and 3rd worst at offensive rebounding. Design, I guess but could bite some.

In 2 Thunder games against Knicks last season, the Thunder shot above 50% from 3 and Knicks barely hit 20%. Thunder won comfortably. But if 3pt fg% was more normal, Knicks would have won.

If Knicks make Finals, will the refereeing be fair, favorable or not favorable? 8th lowest on FT/FGA this season. 20th ranked last regular season on ft rate.... but 4th highest in the playoffs... But Thunder made almost the same transition last time. Who will get the more frequent whistle next time, deserved or otherwise?

Knicks with . . . large increase in fta rate and blocks

In 1.3 seasons post-Hartenstein, Knicks' team block rate is avg 30% higher when Mitchell Robinson is on the floor.
https://www.basketball-reference.com/te ... 25/on-off/
Add 30% to their next-to-worst 4.0 Blk/100 rate last year (MR hardly played) and it would be 2nd/3rd highest this year or last.

In 3 days since almost everyone had their smallest errors of the season, we've gotten .45 worse on avg. Damages range from (WShr) .24 to .67 (KPel) It may be that b-r.com has reconfigured and reduced the regression-to-.500 element, as last night's wins and losses bumped teams' projections up or down by as much as 3 wins.

Mike G wrote: ↑Mon Dec 15, 2025 3:36 pm

Knicks with . . . large increase in fta rate and blocks

In 1.3 seasons post-Hartenstein, Knicks' team block rate is avg 30% higher when Mitchell Robinson is on the floor.
https://www.basketball-reference.com/te ... 25/on-off/
Add 30% to their next-to-worst 4.0 Blk/100 rate last year (MR hardly played) and it would be 2nd/3rd highest this year or last.

In 3 days since almost everyone had their smallest errors of the season, we've gotten .45 worse on avg. Damages range from (WShr) .24 to .67 (KPel)

Code: Select all

.   avg err   rmse   r^2      .   avg err   rmse   r^2
vegas  6.28   7.71   .60      EExp   7.08   8.21   .53
DRKO   6.36   7.94   .56      WShr   7.40   8.82   .46
bpmW   6.57   8.03   .55      dtka   7.41   8.74   .47
Walr   6.66   7.96   .56      Crow   7.65   9.34   .43
knar   6.84   8.10   .54      medi   8.00   9.16   .47
Mgoo   6.92   8.48   .50      25pr   8.05   9.56   .38
avgA   7.04   8.32   .52      KPel   8.17   9.81   .38
eWin   7.05   8.62   .48      DQin   8.51  10.51   .35
perW   7.07   9.21   .42       

It may be that b-r.com has reconfigured and reduced the regression-to-.500 element, as last night's wins and losses bumped teams' projections up or down by as much as 3 wins.

I really don't think b-r.com has reconfigured anything here, they really havent redone any formulas in basketball for a long time.

Every year I've tracked these contests, they seem to make discrete adjustments to the amount of regression-to-mean in their projections. It goes smoothly for a few weeks, then abruptly registers some big changes.
Before the "NBA Cup", they don't even have team projections summing to 1230, but to something less; and I had to adjust for that. Now that the Cup is over, they show 1230 season wins; and coincidentally, big bumps in team win projections.
I conclude their projection formula does not smoothly adjust to number of games played vs yet-to-play.