Predictions 2014-2015

[J.E.]

There are too many better teams

In the east? How many significant East <-> West player movements were there? Love, Tyson Chandler, (Channing Frye)? Wasn't the East historically bad last year? Granted, their division is probably stronger now (CLE gets LBJ, Love; CHI gets Rose, MIL improves slightly, IND takes a hit), but still, winning 37 games in the current eastern conference isn't the toughest of tasks. Monroe, if not traded, is in a contract year which might affect his performance positively

What does RiRAPM mean/stand for?

"RAPM informed RAPM". Basically you use player ratings from year X-1 as priors when computing ratings for year X, until you arrive at the current season. The results are slightly different compared to simply doing multiyear RAPM and weighing older data (seasons) less heavily, but it doesn't make a huge difference

[Crow]

Vegas has Pistons as 10th best in east and 21st overall. Is a team at that rank going to win over 45%? I think no is more likely than yes. 10th in east last season won 32, 21st overall won 33.

[Crow]

The summer ESPN consensus average for Pistons was 33 wins.

[Mike G]

Pistons at 36.5? I'd favor the under.

They had 31 PWins last season. They went 21-29 under Cheeks, then 8-24 under Loyer. Van Gundy is probably a lot better than both of them, esp. Loyer. KCP and Drummond are young and should improve next season. I can definitely see why their O/U is so "high"

Wow. I had not noticed that extreme coaching 'split'. Under Cheeks they were heading to about 34 wins -- 42% -- with the other guy, they won 25%.

Last year, our 12 prognosticators all supposed the Pistons would improve by at least 6 wins over their 2013 Pythagorean. We were all off by at least 7, the average error being 11. And this doesn't account for the much easier schedule in the East, which none of us figured.

This brings up a broader question: Using a system like RAPM, do you add some expected wins to teams in the East?
Last year, there was an avg SOS diff >1.0 ppg between W and E. That translates to about 3 more wins if you're in the East.

[J.E.]

This brings up a broader question: Using a system like RAPM, do you add some expected wins to teams in the East?

If you just straight up multiply player RAPM with minutes to come up with a projected team win total, then yes, you should add wins to teams in the East.
Optimally, you'd simulate the entire season, taking opponent strength, back2backs etc. into account

[jimmytonic]

I added a "Rest Metric" to my projections that penalizes teams according to the number of 3IN4s (the third game of an X-X-O-X), B2B (second game of X-X), B2B-3IN4 (third game of an X-O-X-X), and 5IN4 (fifth game of X-X-O-X-X) that team and the opposing team have using stats pulled from NBAStuffer's rest schedule analysis. Each is worth -0.035/-0.07/-0.105/-0.14 wins for the team, and the absolute value when occurring for the opposing team. So for example, a team with 8 B2Bs that faces 10 teams on a B2B will have 0.14 wins added to their projected record. The penalties/bonuses are just an estimate based on the the 7% reduction in winrate for teams on B2Bs observed here https://rotogrinders.com/lessons/NBA_Sc ... cks-209256. It's a work in progress - I'm just guessing at what the reduction would be for the other 3 kinds of low-rest days, how they might extend off the 7% reduction from B2Bs. All other types of games include at least 2 nights of rest in the span of four days, and I decided not to factor additional rest in (i.e. a game after 3 days of rest worth the same as a game with 2 days of rest).

Anyway, I run this on each team's previous and upcoming seasons to observe the net shift in their rest values, and the Pistons come out in 22nd:

Thunder RV: -1.575
Clippers RV: -1.47
Hawks RV: -1.435
Pelicans RV: -1.33
76ers RV: -1.05
Warriors RV: -0.945
Spurs RV: -0.735
Raptors RV: -0.63
Pistons RV: -0.525
Cavaliers RV: -0.42
Bulls RV: -0.42
Magic RV: -0.385
Nuggets RV: -0.35
Rockets RV: -0.35
Knicks RV: -0.14
Trailblazers RV: -0.07
Grizzlies RV: 0.14
Celtics RV: 0.315
Nets RV: 0.315
Mavericks RV: 0.35
Timberwolves RV: 0.525
Hornets RV: 0.77
Bucks RV: 0.805
Jazz RV: 0.875
Pacers RV: 0.91
Lakers RV: 0.945
Suns RV: 0.98
Wizards RV: 1.295
Kings RV: 1.40
Heat RV: 2.205

Team's with positive RVs have easier schedules relative to their previous season, teams with negative RVs have harder schedules relative to their previous season. The RVs might not exactly translate to wins, but they should be fairly accurate with respect to the relative position of each team's shift rest schedule.

If we assume the Pistons are closer to a 34 win team that tanked the second half of the season to try to salvage their first rounder, will have improved coaching and an improved roster (both offseason acquisitions and player development from a young roster), even after factoring in a harder schedule (as far as rest goes) I think taking the over at 36.5 is a decent proposition. They're definitely one of the more volatile teams though and I wouldn't bet either way. My schedule analysis only looks at rest, but the Pistons are also going to be hurt by playing in the Central division as the now have 4 games each against Cleveland, Chicago, and Milwaukee and presumably the total improvement of those teams will outweigh the regression of the Pacers.

[Mike G]

This is pretty interesting. Do you know if there's as much variation in teams' opponents' schedules? It seems that there would be, and probably a lot more work to include it.

...the Pistons are also going to be hurt by playing in the Central division as the now have 4 games each against Cleveland, Chicago, and Milwaukee and presumably the total improvement of those teams will outweigh the regression of the Pacers...

Pistons have 4 games against: Atl, Chi, Cha, Cle, Ind, Mil, NYK, Orl, Phl, Tor.
3 games vs: Bos, Brk, Mia, Was
Don't see much in the way of a divisional disadvantage.

I hope you will submit an entry to the prediction challenge.

[jimmytonic]

If you meant whether the metric includes whether the opponent has sub-standard rest (i.e. a 3IN4, B2B, 3IN4-B2B, or 4IN5) the game the team faces them, then that is included (NBAStuffer.com provides detailed information about both each team's rest schedule and how much rest their game-by-game opponents have). So, to clarify, each team gets a penalty for each game they have sub-standard rest and an equal, but opposite reward for each game their opponent has sub-standard rest. This is done for the previous season as well, so the rest value (which hypothetically directly translates to wins - the Heat with their RV of 2.205 would be expected to win an additional 2.205 games this season, isolating for all other changes) is just the net shift in rest schedule difficulty. A team could have a really hard rest schedule (i.e. a bunch of B2Bs and rarely facing any opponents on a B2B), but if they had an even more difficult rest schedule last season, their projected wins would increase.

I think I might separate own-rest from opponent-rest. I could see how an injury-prone or older team like the Heat (mostly just Wade) would be more affected by how much rest they have between games compared to how much rest their opponents have since - e.g. if the Heat could choose between shedding 20 B2Bs on their schedule or having an additional 20 games where their opponent is on a B2B, they'd probably choose the former since that's an additional 20 games Wade is going to be able to play (a much more consistent benefit than however a B2B might affect their opponent). Conversely, I could see how a young, high-tempo team like the Nuggets might not be particularly affected by playing with sub-standard rest, yet would benefit a lot from facing opponents on a B2B.

The ideal would be to create multiple minute distributions for each team, each reflecting the expected minutes for each player when having a certain amount of rest, and then going through each team's schedule on a game-by-game basis and applying both the team's and the opponent's minute distributions based on how much rest they have. And conceptually that seems like a very simple thing to program, and I'm sure many people and organizations already do something like that, but I don't have the willpower or time to do it myself, so the rest metric is just an attempt to simplify a very real factor in a team's performance. It has a ton of noise, but it's better than nothing.

Anyway, I am planning on submitted a prediction - right now my projection is just each team's averaged xRAPM and Wins Produced projection, then applying the rest metric, but I'm looking to replace Wins Produced with a different production-based metric. Probably WS, but I really dislike how defensive win shares are calculated.

[Crow]

Defensive winshares is not near as bad for a team win projection as it is for an individual evaluation. It will still have likelihood to distort to the extent which players dissimilar from avg. team defensive performance get more or less minutes thru rotation change or offseason movement. If player minutes distribution is 70, 80 or 90% similar to last season then most of the team's shot defense and gambling for steals and blocks and getting then or not will be included in the cumulative player data, in one place or another, one part of the measurement or another.

[nbacouchside]

Team Wins
OKC 63
SAS 60
CLE 59
GSW 58
LAC 58
CHI 55
DAL 55
MEM 53
PHX 50
HOU 50
TOR 50
MIA 47
WAS 47
POR 45
IND 45
ATL 38
BKN 38
MIN 37
CHA 36
NOP 36
DEN 35
DET 32
BOS 29
ORL 28
NYK 28
UTA 24
SAC 23
MIL 21
LAL 16
PHI 14

[jimmytonic]

Team Wins
OKC 63
SAS 60
CLE 59
GSW 58
LAC 58
CHI 55
DAL 55
MEM 53
PHX 50
HOU 50
TOR 50
MIA 47
WAS 47
POR 45
IND 45
ATL 38
BKN 38
MIN 37
CHA 36
NOP 36
DEN 35
DET 32
BOS 29
ORL 28
NYK 28
UTA 24
SAC 23
MIL 21
LAL 16
PHI 14

Is this a gut projection or based on something more methodical?

[Crow]

nbacouchside wrote:
Team Wins

LAL 16


Don't tell Kobe...

[Mike G]

Has there been major player movement lately? I have not been paying attention.
My latest reckoning (may be final if no changes to lineups)

Code: Select all

west   wins       east    wins
LAC    60.0        Cle    61.6
Okl    55.9        Was    48.1
SAS    55.1        Tor    46.3
Por    47.9        Atl    44.0
Phx    46.3        Cha    43.1

GSW    45.4        Chi    42.1
Hou    44.4        NYK    38.4
NOP    44.1        Det    37.8
Mem    43.3        Mia    36.5
Dal    43.2        Bos    34.1

Den    41.4        Brk    32.1
Min    37.4        Ind    31.1
Sac    34.8        Orl    28.2
LAL    34.8        Mil    27.1
Uta    32.5        Phl    12.8

avg    44.4        avg    37.5

[nbacouchside]

Team Wins
OKC 63
SAS 60
CLE 59
GSW 58
LAC 58
CHI 55
DAL 55
MEM 53
PHX 50
HOU 50
TOR 50
MIA 47
WAS 47
POR 45
IND 45
ATL 38
BKN 38
MIN 37
CHA 36
NOP 36
DEN 35
DET 32
BOS 29
ORL 28
NYK 28
UTA 24
SAC 23
MIL 21
LAL 16
PHI 14

Is this a gut projection or based on something more methodical?

It's RPM with an age curve, run through the schedule, with a mean regressed version of Jerry's Coach Rating added in. New coaches got no boost or reduction. I think it might have effected the results more than I like though, so I might make the mean regression harsher before the season starts when I settle on a final prediction.

[J.E.]

It's RPM with an age curve, run through the schedule, with a mean regressed version of Jerry's Coach Rating added in. New coaches got no boost or reduction. I think it might have effected the results more than I like though, so I might make the mean regression harsher before the season starts when I settle on a final prediction.

RPM is not coach adjusted, so esp. for those teams who didn't see alot of roster turnover you're double counting when accounting for the coach (because in RPM the credit, in the absence of coaches, gets shoveled onto the players)
I'm definitely not against using the coach adjustment in some way or form, but if the player metric isn't coach adjustment I'd use only sprinkles of it. I'd probably just completely ignore it if the roster is largely the same and the coach didn't change (TOR/SAS/OKC etc.)
You'd obviously want it to influence your predictions in case the coach changed (esp. if the RAPM of the old and new coach are very different, so BRK/DET etc.); and you'd want it to influence your predictions a little for teams with signficant roster additions (but same coach), in case the new players came from a coach with a way better/worse RAPM rating than the teams' current coach has (maybe CHI/IND..)

I know it's not the perfect answer because you can't just enter a factor into your code, but instead have to look at each teams' situation and then use some guesswork how strong you'd want the effect to be. Sorry