2012 predictions

[EvanZ]

Crow is right. My last set of predictions is as updated as I'm going to make it (too late now, I guess, anyway). Here they are again:

Code: Select all

RK	TEAM	WIN%	PD	W	L
1	MIA	76.5%	9.38	51	15
2	BOS	65.1%	5.69	43	23
3	CHI	64.9%	5.63	43	23
4	OKC	64.2%	5.39	42	24
5	ORL	64.0%	5.32	42	24
6	LAL	62.9%	4.97	42	24
7	DAL	62.7%	4.90	41	25
8	LAC	57.8%	3.32	38	28
9	SAS	56.7%	2.97	37	29
10	MEM	54.5%	2.27	36	30
11	ATL	54.5%	2.25	36	30
12	POR	53.9%	2.08	36	30
13	DEN	53.7%	2.01	35	31
14	PHI	53.5%	1.92	35	31
15	NYK	53.2%	1.84	35	31
16	IND	48.8%	0.42	32	34
17	MIL	47.9%	0.14	32	34
18	PHO	47.2%	-0.10	31	35
19	HOU	45.2%	-0.73	30	36
20	GSW	43.1%	-1.43	28	38
21	UTA	39.2%	-2.69	26	40
22	SAC	39.2%	-2.70	26	40
23	NOH	38.7%	-2.83	26	40
24	NJN	37.6%	-3.21	25	41
25	CHA	36.0%	-3.71	24	42
26	DET	34.2%	-4.30	23	43
27	MIN	34.2%	-4.30	23	43
28	CLE	33.0%	-4.68	22	44
29	TOR	32.5%	-4.85	21	45
30	WAS	28.6%	-6.10	19	47

[Mike G]

Evan, do your wins sum to 990?
I've double checked, and it seems they add up to 980.

Also wondering about the Vegas lines: without LAL and NJ, there are 914 wins predicted; leaving 76 to split between those 2 teams.
But our averages are about 64 (39+25) wins between them.
Since all the numbers are something-and-a-half, I wonder if they should be rounded up. This would add 14 wins ( 0.5 * 28 ) and leave 62 wins between the Lakers and the Nets.
I seem to recall some such convention with their predictions.

[EvanZ]

Ok, I got it to 990 (just needed to change my multiplier to 0.961):

RAPM + ezPM blended

Code: Select all

RK	TEAM	WIN%	PD	W	L
1	MIA	77.4%	8.51	51	15
2	BOS	65.9%	4.93	43	23
3	CHI	65.7%	4.87	43	23
4	OKC	64.9%	4.64	43	23
5	ORL	64.7%	4.57	43	23
6	LAL	63.6%	4.23	42	24
7	DAL	63.4%	4.16	42	24
8	LAC	58.4%	2.62	39	27
9	SAS	57.4%	2.28	38	28
10	MEM	55.2%	1.60	36	30
11	ATL	55.1%	1.59	36	30
12	POR	54.6%	1.41	36	30
13	DEN	54.3%	1.35	36	30
14	PHI	54.1%	1.27	36	30
15	NYK	53.8%	1.18	36	30
16	IND	49.4%	-0.20	33	33
17	MIL	48.5%	-0.47	32	34
18	PHO	47.7%	-0.70	32	34
19	HOU	45.8%	-1.32	30	36
20	GSW	43.6%	-1.99	29	37
21	UTA	39.6%	-3.22	26	40
22	SAC	39.6%	-3.22	26	40
23	NOH	39.2%	-3.36	26	40
24	NJN	38.0%	-3.72	25	41
25	CHA	36.4%	-4.21	24	42
26	DET	34.6%	-4.78	23	43
27	MIN	34.6%	-4.79	23	43
28	CLE	33.4%	-5.15	22	44
29	TOR	32.9%	-5.32	22	44
30	WAS	28.9%	-6.53	19	47

[EvanZ]

Since Jerry didn't actually post "RAPM picks" here are my ratings using RAPM:

Code: Select all

RK	TEAM	WIN%	PD	W	L
1	MIA	72.5%	6.78	48	18
2	BOS	68.9%	5.62	45	21
3	DAL	64.3%	4.13	42	24
4	CHI	63.3%	3.80	42	24
5	LAL	62.3%	3.49	41	25
6	OKC	62.2%	3.44	41	25
7	ORL	60.6%	2.93	40	26
8	NYK	57.8%	2.03	38	28
9	MEM	57.0%	1.77	38	28
10	SAS	56.1%	1.49	37	29
11	DEN	56.1%	1.47	37	29
12	POR	55.6%	1.32	37	29
13	PHI	54.1%	0.84	36	30
14	ATL	53.3%	0.58	35	31
15	LAC	51.3%	-0.06	34	32
16	IND	49.8%	-0.56	33	33
17	MIL	49.7%	-0.59	33	33
18	HOU	43.6%	-2.56	29	37
19	GSW	43.0%	-2.73	28	38
20	PHO	42.5%	-2.92	28	38
21	UTA	41.6%	-3.20	27	39
22	NOH	41.1%	-3.37	27	39
23	NJN	39.0%	-4.03	26	40
24	CLE	38.9%	-4.08	26	40
25	DET	37.9%	-4.40	25	41
26	SAC	37.7%	-4.46	25	41
27	MIN	36.5%	-4.84	24	42
28	CHA	36.1%	-4.98	24	42
29	WAS	35.0%	-5.33	23	43
30	TOR	34.6%	-5.44	23	43

[Mike G]

I've averaged the predictions and ranked by the expected improvement over last season, using 2010-11 Pythagorean W% * 66 games.
That column (11Py) is included as a minimum 'to beat' number. Some years, it has done pretty well vs the field.
Evan, your blended entry is shown as EV, and the purely RAPM is shown as EV2. These don't actually identify the method, so let us know if you have more-preferred designations. Or anyone else, on behalf of anyone else.
For space and clarity, I've gone integer in this display. All of Vegas' numbers are shown 0.5 too high, as are about 1/3 of dis'.

The Clippers' expected improvement is more than twice anyone else's. And their (3) point guards totaled 96 minutes last night.
Then some perennial doormats expected to rise.

Code: Select all

delta  avg   tm   11Py  DSM   MG   EZ  Vegas JH   JE  EZ2  dis  Crow  KP
14.3  40.4   LAC   26    43   42   39   43   41   40   34   42   41   40
6.7   26.3   Min   20    25   33   23   23   29   23   24   28   24   33
6.0   25.5   NJN   20    27   24   25   28   28   23   26   22   24   29
5.2   20.3   Cle   15    21   21   22   17   18   19   26   20   18   22
4.0   34.5   Ind   31    34   38   33   37   36   31   33   35   37   33

3.5   21.4   Was   18    24   22   19   21   22   21   23   21   22   20
3.0   24.8   Sac   22    29   29   26   21   25   20   25   25   23   26
2.7   33.7   Atl   31    33   31   36   35   33   33   35   35   36   31
2.5   37.1   NYK   35    35   38   36   42   35   38   38   38   36   36
2.1   22.1   Tor   20    25   21   22   17   19   24   23   22   21   28

delta  avg   tm   11Py  DSM   MG   EZ  Vegas JH   JE  EZ2  dis  Crow  KP
2.1   42.9   Okl   41    41   39   43   48   45   44   41   44   45   40
1.3   32.1   Mil   31    33   30   32   32   34   32   33   31   33   32
1.1   37.5   Por   36    37   38   36   38   38   38   37   37   39   38
1.0   49.3   Mia   48    46   46   51   51   52   49   48   53   50   48
0.4   28.4   GSW   28    32   29   29   27   26   29   28   33   27   25

-0.4  35.8   Phi   36    35   33   36   36   37   39   36   34   37   36
-0.7  24.3   Det   25    28   24   23   23   22   27   25   23   23   26
-1.2  36.9   Mem   38    37   36   36   39   37   37   38   37   40   32
-1.5  40.6   Dal   42    38   42   42   44   39   43   42   41   40   37
-1.6  27.3   Uta   29    28   30   26   26   27   25   27   29   26   30

delta  avg   tm   11Py  DSM   MG   EZ  Vegas JH   JE  EZ2  dis  Crow  KP
-2.4  46.1   Chi   49    41   50   43   48   48   49   42   49   46   46
-3.2  20.6   Cha   24    21   22   24   17   13   18   24   22   18   28
-3.6  27.4   Phx   31    29   23   32   27   24   27   28   29   28   28
-4.1  40.8   Bos   45    37   42   43   40   43   43   45   40   41   36
-4.5  40.3   SAS   45    42   43   38   42   42   44   37   38   40   38

-4.9  39.9   Orl   45    40   40   43   39   40   40   40   41   36   40
-5.6  40.2   LAL   46    40   39   42   49   40   42   41   36   41   34
-5.7  31.8   Hou   37    31   33   30   32   32   34   29   31   34   33
-5.9  36.6   Den   43    34   31   36   38   43   42   37   34   37   35
-9.8  25.2   NOH   35    25   21   26   25   22   25   27   24   27   31

0.0  990.3  totals      990  990  992  990  990  990  992  989  990  990

The Hornets got good returns for Paul, in my opinion. But nothing for West.
The Nuggets looked fairly lost in the playoffs last year. And then they lost half their team. Hollinger thinks they're just as good anyway, and I like them least.
The biggest discrepancy on the board is with Charlotte, whom Kevin P expects to rise to 28-29 wins, vs Vegas which expects 12-13.

Vegas supposes the Lakers improve by 3 games; to almost 7 wins more than what anyone else thinks.
(The Nets' Vegas wins are deduced, after I managed to find LA's. )

[EvanZ]

Thanks for putting those together.

[J.E.]

Deng looked, perhaps not surprisingly, like an All-World defender against LAL. But Rip Hamilton has not played well for the Bulls yet.
Amir Johnson seems to be starting for Toronto and played 33 minutes in the first game. Would be great if it stayed that way.
The Jazz are not starting Millsap, are they tanking already? That's quite a bit early..
Nowitzki admitted to being out of shape at the european championship. Maybe he was saying that to make the (otherwise horrible German) team look better, but if it's true one might question his dedication. Dallas looked really bad so far, but had two of the toughest opponents. I expect them to bounce back sooner rather than later.
At the start of the season Nowitzki had quite a bit of a lead for the prior informed RAPM, but already fell to 3rd place after 2 games, with LeBron leading the way now

Here's a post about predictions from the WP camp, but to me it's pretty much just an excuse for their bad prediction results last year. I'm fine with that, sometimes predictions can be off for different reasons. But when you take a jab at APM and PER at the end of the post for no reason, even though they outperform WP at retrodiction (which controls for some of the unforeseen circumstances), then that makes you look a little pathetic. Just sayin'..

[Mike G]

One reason my predictions seem to be rather devoid of super winners or losers -- all between 20 and 50 wins -- is that I don't expect any player to average over 36 minutes over the course of the season.
Kevin Pelton assumes all players miss at least 5 games (I think), and that limits a 39 mpg player to 36 over the full schedule.
LeBron James, who seems almost indestructible, is the only player I'm giving 36 mpg.
At 35 mpg -- actually, 2300 minutes/66 games -- Rondo, Rose, Deng, Griffin, Wade, Bosh, Durant, Dwight, Aldridge.
All these guys played more minutes last year, either per game or total. But most teams will have someone missing major minutes; and if they don't, they can expect to exceed expectations.

[DSMok1]

One reason my predictions seem to be rather devoid of super winners or losers -- all between 20 and 50 wins -- is that I don't expect any player to average over 36 minutes over the course of the season.
Kevin Pelton assumes all players miss at least 5 games (I think), and that limits a 39 mpg player to 36 over the full schedule.
LeBron James, who seems almost indestructible, is the only player I'm giving 36 mpg.
At 35 mpg -- actually, 2300 minutes/66 games -- Rondo, Rose, Deng, Griffin, Wade, Bosh, Durant, Dwight, Aldridge.
All these guys played more minutes last year, either per game or total. But most teams will have someone missing major minutes; and if they don't, they can expect to exceed expectations.

My minutes distribution is why my spread is so tight, also. I just used depth charts and the same time for all starters on each team--that could obviously be a lot better. I don't think I had anybody above even 33 MPG! (That said, the minutes distribution will probably be tighter this year because of rest issues.)

[EvanZ]

Uh, so it dawned on me while reading Alex Konkel's interesting post today, that I never posted my pure ezPM predictions. Seems like a good thing to do. Obviously, these have not been made with any data from 2012 (using same minutes projections as the others). Mike, if you care to, you could add these as EZ3 or something to your collection.

Code: Select all

RK	TEAM	WIN%	PD	W	L
1	MIA	81.7%	10.23	54	12
2	ORL	68.3%	5.91	45	21
3	CHI	67.8%	5.74	45	21
4	OKC	67.3%	5.60	44	22
5	LAL	64.7%	4.75	43	23
6	LAC	64.6%	4.72	43	23
7	BOS	63.2%	4.26	42	24
8	DAL	62.6%	4.06	41	25
9	SAS	58.4%	2.71	39	27
10	ATL	56.6%	2.15	37	29
11	PHI	54.0%	1.30	36	30
12	POR	53.6%	1.17	35	31
13	MEM	53.5%	1.14	35	31
14	DEN	52.8%	0.90	35	31
15	PHO	52.3%	0.75	35	31
16	NYK	50.3%	0.09	33	33
17	IND	48.9%	-0.34	32	34
18	HOU	47.6%	-0.77	31	35
19	MIL	47.4%	-0.82	31	35
20	GSW	44.0%	-1.93	29	37
21	SAC	41.2%	-2.83	27	39
22	UTA	37.9%	-3.91	25	41
23	NOH	37.5%	-4.04	25	41
24	NJN	37.1%	-4.17	24	42
25	CHA	36.7%	-4.29	24	42
26	MIN	32.8%	-5.54	22	44
27	DET	31.7%	-5.91	21	45
28	TOR	31.3%	-6.05	21	45
29	CLE	28.6%	-6.91	19	47
30	WAS	23.6%	-8.51	16	50

[DSMok1]

Evan, do those PD's sum to an average of 0? They look skewed.

[EvanZ]

Yep, there was a snafu with my spreadsheet. Looks better now.

[Mike G]

The selection bias issue to be concerned about ..: at the tail end of careers, when players are marginal, only the "lucky" players get another chance. Then, the next year, they are "unlucky", and that is the end of their careers. Thus the YtY change is biased high--the players that were "unlucky" in year 1 never get a chance to be "lucky" in year 2, to counter balance the "lucky"->"unlucky" sequence. ..

Can't say that I follow this. Some players at any age will miss a year, or most of a year, and bounce back. Sometimes, the year(s) off may seem to prolong a career.

Meanwhile, those players who have showed up, in uniform, trying out for a team, should be considered to be back on the aging curve, shouldn't they? If they're past their prime, we can expect their minutes to drop by X% and their effectiveness to drop by Y% ?

Isn't it exactly the averages of continuing players from year(x) to year(x+1) that we want to know, at the beginning of the season?

[Mike G]

... I never posted my pure ezPM predictions. Seems like a good thing to do. Obviously, these have not been made with any data from 2012 (using same minutes projections as the others). Mike, if you care to, you could add these as EZ3 or something to your collection.

If this is straight ezPM, then I'm inclined to call it that.
The one that's the blend of 2 metrics = ez2
And the one that's RAPM, with your minutes allotment, zRA -- because there's nothing ez about it.

I was more interested in the pure ezPM, actually, than in the blend.
And in fact, we could blend any 2 or more metrics to find the best 'predictor', during or after the season.

Also, you've submitted integer wins along with more detailed pt-diffs. I've hand-entered your W predictions as integers, and this could conceivably matter in the 'standings'.
It could rock your boat or hand you a false victory.
In any case, they add up to 992 wins. Just so you know.

[EvanZ]

And in fact, we could blend any 2 or more metrics to find the best 'predictor', during or after the season.

Yeah, I think we should try to address that question, too. I'm not sure each individual metric is going to get significantly better, or that there will be some incredible new metric that gets it all right. But maybe an "ensemble" average of all metrics would beat the best individual metric. Just a matter of figuring out how to blend them.