Predictions 2013-14

[nbacouchside]

I think what's helped us a lot are the New York teams slowly climbing to 0.500 ball.

By the way, Arturo Galetti/Wins Produced prediction is at 8.4 (absolute error) and 10.6 RMSE. It would be nice to see them discuss why WP's usually underperforms compared to other predictions/metrics.


Now that a large portion of the season is gone and they're no longer a fluke, what did we miss with the Phoenix Suns? How could a prediction methodology change to incorporate the 2014 Suns?

WP usually underperforms because it's a garbage metric mostly.

[Mike G]

Their expected win - loss record is 24 -14 and yet they are 18-20. I don't recall seeing a greater imbalance especially for just half a season.

Indeed, the Wolves' Win% is some .172 less than it should be with their MOV (.646 - .474)
In the last 14 seasons ( 2000-2013) the biggest underperformers for a full season are the Sixers of 2012. With a MOV of 4.24, they should have won .641 but instead won only .530 of games.

On the most plus side, Utah'08 had -2.63 ppg but won half their games, a .088 overperformance. That is barely half the current Min shortfall, in Win%.

Is there a 'culture of winning' in some franchises, and a tendency to lose the close ones in others?
Average (W% - Pyth%) in this century, ranked:

Code: Select all

Tm     over         Tm     over
Dal    .023        Hou    -.017
LAL    .018        Orl    -.015
Atl    .012        Was    -.012
Cle    .011        Tor    -.011
Chi    .009        GSW    -.011
Uta    .009        Sac    -.008
Den    .008        Bos    -.008
Cha    .008        Mem    -.008
Por    .008        Mil    -.008
LAC    .006        SAS    -.007
Mia    .006        Min    -.007

Since one win in 82 is about .012, the Mavs have averaged 2 wins above their Pyth in the Nowitzki/Cuban era.

Is there a tendency for certain factors which contribute to MOV to be worth less (or more) in a close game? There do seem to be correlations. Factors correlated with (W% - Pyth%) -

Code: Select all

corr   factor         corr   factor
.067    Blk/G        -.124    Pace
.062    As/FG        -.100   OppTO%
.053   OppFT/FGA     -.073    PF/G
.032    DRtg         -.062    TO%
.019    ORb%         -.036    eFG%
.009   OppEFG%       -.031    3/fga
.006    DRb%         -.021    FT/FGA
.                    -.010    ORtg

So . . . If your competitive edge is based on high pace, turnover differential, and fouling a lot, you aren't really that good in a close game.
If your team loves to pass and block shots, it bodes well.
Based on 415 team-seasons.

[Crow]

Only 22 teams on that list. I was looking for OKC.


"Is there a tendency for certain factors which contribute to MOV to be worth less (or more) in a close game?"

Is this really a correlation in a close game or is one or both for the season (as your last line suggests)?


Own FG% 4 times more important than opponent FG%? How could this be? and yet def rtg is three times more important than offense rtg? -.031 3/fga??? Could you share more about the methodology used? Is it simple correlation for each and every listed stat separately or these and only these variables in a single run multi-variable regression?

[AcrossTheCourt]

Can you translate those coefficients into something useful? Like show an example calculation for Minnesota for the pace and blocked shots factors.

"So . . . If your competitive edge is based on high pace, turnover differential, and fouling a lot, you aren't really that good in a close game."

It would be good to know if those were statistically significant and what the magnitude really was.

[Bobbofitos]

So . . . If your competitive edge is based on high pace, turnover differential, and fouling a lot, you aren't really that good in a close game.
If your team loves to pass and block shots, it bodes well.
Based on 415 team-seasons.

This is a good summary. It speaks to the strength of playing in the open court (the top) vs the halfcourt (the bottom), 2 things that "matter" as far as end-game performance

[Mike G]

Wolves' pace is 97.7, which would be 7th highest in the sample.
To put it in the middle of a sample, we can look at the 13 highest paced teams of the century.
These teams, above 96.5 poss/G, are extremely average -- they avg +0.03 MOV, and they should have won .501 of their games; but they won .484 -- a shortfall of .017
The top 50 pace teams avg a (W% - Pyth%) of -.010

Minny averages 3.21 blocks per game. This is also 7th lowest in the last 415 team-seasons.
The bottom 13 in Blk/G have been mostly bad teams, totaling a .413 W%. Their pt-diff says they should have won .422 -- so they're .009 worse than that.

It doesn't look like a minus-.172 W-Pyth can be predicted from their particular strengths and weaknesses. But they may be a perfect storm of late-close-game vulnerabilities which exacerbate one another.
It's certainly an interesting case study. The odds of a team losing all 11 of its closest games is about 1/2048

[Mike G]

Only 22 teams on that list. I was looking for OKC.

OKC is at .001, that is to say no 14-yr tendency to over- or under-perform. However, they have had their ups and downs. Including Sonics years:

Code: Select all

.yr     W   L     W%     mov    PW%     W-P
2000   45   37   .549    .95   .532    .017
2001   44   38   .537    .02   .501    .036
2002   45   37   .549   3.02   .601   -.052
2003   40   42   .488   -.12   .496   -.008
2004   37   45   .451   -.63   .479   -.028
2005   52   30   .634   2.29   .576    .058
2006   35   47   .427  -3.02   .399    .028
2007   31   51   .378  -2.89   .404   -.026
2008   20   62   .244  -8.76   .208    .036
2009   23   59   .280  -6.10   .297   -.017
2010   50   32   .610   3.49   .616   -.006
2011   55   27   .671   3.79   .626    .045
2012   47   19   .712   6.12   .704    .008
2013   60   22   .732   9.21   .807   -.075

Last year's anomaly is the 7th biggest (negative) in the century. They were 5-10 in decisions of 5 pts or less.

"Is there a tendency for certain factors which contribute to MOV to be worth less (or more) in a close game?"

Is this really a correlation in a close game or is one or both for the season (as your last line suggests)?

We haven't seen a case of a team which loses too many games (relative to MOV) but wins the close ones. The Wolves' example isn't typical of anything, but it may be ultra-typical.

Own FG% 4 times more important than opponent FG%? How could this be? and yet def rtg is three times more important than offense rtg? -.031 3/fga??? Could you share more about the methodology used? Is it simple correlation for each and every listed stat separately or these and only these variables in a single run multi-variable regression?

I'm not sure why there are more negative than positive correlations. But remember that these are relative to a team's strength as defined by their MOV. A negative on 3/fga doesn't mean that 3fga are bad for your MOV, but that they're not as valuable in close games, for securing the win.

The top 20 teams in 3fga/fga have won .574 of their games. But their MOV says they should have won .586
It may be that they're around .500 in games of 5 points or less, and .600 in less-close games.

The top 20 eFG% should have won .701 of their games; but they've floundered around at .692 -- again, not too significant.

Correlations are just what Excel spits out.

[Mike G]

At the midpoint of the season, our avg error has shrunk to 7.36, slightly better than one week ago.
As Durant is charging toward his first MVP, Bobbo is also leaving us in the dust.

Code: Select all

Bobb    6.15        Yoop    7.76
jBro    6.43        deep    7.79
hDon    6.79        416x    7.88
ncs     6.88         eW     7.89
ATC     7.15        fpli    8.72
jank    7.55        13Py    8.76

[AcrossTheCourt]

The Nets are continuing to play much better with Garnett as a center and with a smallball four (before the season I thought this would happen, but with Kirilenko at PF.) The Clippers look great now. The Wolves continue to disappoint, and now injuries are occurring again. Memphis is coming back, slowly.

Portland and Phoenix really surprised everyone, but I think we're doing pretty well. There's a lot of the season left.

[Mike G]

At about 63% of the season, there's a race for 1st here.
Also close at 3-4 and at 5-6.

Code: Select all

Bobb    6.06         Yoop    7.66
jBro    6.10         416x    7.76
hDon    6.28         deep    7.92
ncs     6.31          eW     8.03
ATC     7.00         fpli    8.23
jank    7.01         13py    9.07

Avg error for the group (not including 2013 Pyth) is 7.12

[bbstats]

Too late for me to join? Here's my terrible projections:

http://thebasketballdistribution.blogsp ... acets.html

[Mike G]

Given that you are out in the boonies and had to walk some miles to deliver your predictions, you may join -- at the back of the bus.

Code: Select all

Bobb    6.10        Yoop    7.37
jBro    6.35        416x    8.09
hDon    6.57        fpli    8.14
ncs     6.59        deep    8.18
ATC     6.63         eW     8.21
jank    7.20        13Py    9.06

(late)
bbs     6.36

Avg error is now 7.22, relative to the b-r.com forecast
http://www.basketball-reference.com/fri ... f_prob.cgi

Relative to straight pythagorean point differential, we'd be off by an avg of 7.67

Code: Select all

Bobb    6.58        Yoop    7.70
bbs     7.06        deep    7.74
jBro    7.12        jank    7.87
hDon    7.37        416x    7.90
ATC     7.38         eW     8.29
ncs     7.56        fpli    8.91

Still sux to be me.

[Mike G]

I think I'm in last place now, by every measure.
Just wondering. Do the 'standings' here depart much from expected random scattering? Not expecting an answer, but if I ever win this thing, I hope to remember it was largely luck.

Anyway, there are a few teams that we all failed to see exploding or imploding. The one biggest differential between my own predictions and that of Bobbo (still leading) is Cleveland.

Here are my predictions; and their current rates and minutes, projected to 82 games.

Code: Select all

eWins        predicted        actual/projected
. Cavs   Min   e484   eW      Min   e484   eW
Irving    32   1.52   8.2      34   1.56   9.0
Varejao   22   1.84   6.9      24   1.21   4.9
Thompson  32   1.15   6.3      33    .98   5.5
Bynum     20   1.40   4.7       8   1.13   1.5
Jack      26   1.07   4.7      26    .50   2.2

Waiters   24    .79   3.2      23    .79   3.0
Miles     16    .90   2.5      16   1.06   2.9
Zeller    24    .60   2.5      12    .87   1.8
Clark     18    .76   2.3      11    .43   0.8
Gee       28    .43   2.0      12    .05   0.1

totals   242         43.2     198         31.6
                     
new guys         Deng,Luol     14    .86   2.0
.         Dellavedova,Matt     13    .24   0.5
.          Bennett,Anthony     10    .25   0.4
.            Hawes,Spencer      3   1.31   0.7

.            totals           238         35.3

Min = total minutes / 61 (games played so far)
Sims and Karasev a few mpg but not much in win production.

For the main guys, my predicted mpg are pretty close. Varejao goes 30 min and misses one game in 5, = 24 mpg.
But he and Thompson and Jack are seriously diminished from last year's rates. That's 3 of their top 4 who have flopped, and Bynum makes it 4 of 5.
Gee went back to being dirt, Clark has been weak, and Zeller has hardly played in spite of improving himself. His shooting is up .100 this year, he's a 14-10 guy (per36) after an 11-8 year, and his minutes are cut in half?

A deficit of 7 eWins is equivalent to 14 fewer wins on the year.
I predicted these young Cavs to win 43, and they're headed for 31.
Bobbo called them at 30 wins; he's off by 1, and I'm off by 12.
What in my calculations was so off? How does one predict that everyone on a team will underperform?

[Bobbofitos]

For Cleveland I used these minute projections:

Code: Select all

Kyrie Irving		2,520.00
Jarrett Jack		2,340.00
CJ Miles		1,420.00
Alonzo Gee		1,440.00
Andrew Bynum		1,131.00
Anderson Varejao		1,350.00
Earl Clark		1,872.00
Tristan Thompson 	1,900.00
Dion Waiters		2,160.00
		

Looks like my Irving projection is going to be pretty good. Same with Jack, Miles slightly off. Bynum obviously off by about double. I love Varejao but he's always hurt... I was pessimistic, but he's more or less validated that concern, playing just 1449 on the season. I thought Bynum/Bennett would cut into TT's minutes more than they have. I would've upped their total wins if you take away those minutes and replace them with TT... Got lucky in the sense this didn't happen, and they upgraded their team midseason. I was low on Bennett but basically just included a large replacement tag there, which soured me on them considerably.

Eyeballing our rate stats, I had Waiters/Miles/Clark as more damaging than ewins punishes them. You also were a bit higher on Bynum than I was. (Although I thought he'd produce at a better clip than he did for them)

[Mike G]

Before missing a year, Bynum had 5 consecutive years at >.180 WS/48. I downgraded him to .140-ish but figured him for 1640 minutes (82*20)
So he shoots about .150 below his career avg and basically can't play.
I did nail his 20 mpg -- but for just 24 games.

Miles is the one player who's been better than I expected. But oddly, his minutes are down.
Did you expect Jack to fall off? His shooting is off by .060. PER and WS/48 say he was a good player last year (15.9/.115) and terrible this year (10.9/.038). From mostly the best numbers of his career to the worst.

Meanwhile, if you thought Earl Clark was terrible, why would you expect him to get as many minutes as Thompson?
Thompson improved immensely from yr 1 to yr 2, and this year, he's stalled out.