2012 predictions

[Mike G]

Just 3 games last night changed things again for the better.
Errors relative to current wins plus Pyth% of remaining:

Code: Select all

5.65   JohnH      6.85    ezPM
6.25    J.E.      6.88    ez2
6.30   Crow       6.91   KevinP
6.30   MikeG      7.09    zRA
6.60    dis       7.30   11Pyth
6.66   Vegas      7.36   DSMok1

Hollinger's lead on #2 J.E. was .24 just 5 days ago; now it's .60

Compared to pure Pyth% of 66 games, ignoring current W-L :

Code: Select all

5.79   JohnH      6.74   Vegas
6.28    J.E.      6.74    ez2
6.31   MikeG      6.77    zRA
6.35    Crow      6.84    ezPM
6.50    dis       7.14   11Pyth
6.72   KevinP     7.29   DSMok1

The avg error is about .04 smaller here. The spread is also smaller (without a luck factor?)

[Crow]

John Hollinger is 4+ games closer than I am on the projections for Charlotte, Denver, Minnesota, New Orleans and Orlando. This is obviously sorting in my favor but if you just looked at the other 25 teams I'd be ahead of him. I am not surprised that I am not close on the first 4 teams, I don't claim to know them well. Orlando I was betting on a Howard trade disrupting or the trade talk disrupting. So far it hasn't.

[EvanZ]

Thinking about it now, did any of the WoW bloggers do win projections this season?

[Crow]

I haven't really checked comprehensively but I haven't run into any. There wasn't much analysis from them about all of them finishing in the bottom half of WiLQs report on last season predictions. There was the off-season change in the defensive rebounding credit but no response to my suggestion to recalculate last season's performance if that change had already been implemented and apparently not much interest to re-engage in the prediction contest this season, despite people being previously told that metric performance is important to know and important to use in evaluating metrics. Instead a certain author choose to emphasize that short tests of metric power aren't that reliable or interesting.

[xkonk]

Metric performance is (near?) impossible to compare going forward in time. Did you all use the same minute projections? Did Hollinger send his around? Crow has already said that he made his prediction for Orlando based on the idea that Howard would be gone at some point; I assume not everyone else did. If Hollinger then wins, does that mean PER is the best measure around or that he had better minute estimates (or used something else)? If Daniel ends up in last, will he stop using ASPM because it is obviously flawed? Making the predictions is a fun exercise, but I don't know how much knowledge you'll have gained at the end of the year.

[J.E.]

Metric performance is (near?) impossible to compare going forward in time. Did you all use the same minute projections? Did Hollinger send his around? Crow has already said that he made his prediction for Orlando based on the idea that Howard would be gone at some point; I assume not everyone else did. If Hollinger then wins, does that mean PER is the best measure around or that he had better minute estimates (or used something else)? If Daniel ends up in last, will he stop using ASPM because it is obviously flawed? Making the predictions is a fun exercise, but I don't know how much knowledge you'll have gained at the end of the year.

Completely agree. For metric comparison we need to do retrodiction with actual minutes

[Mike G]

The idea was floated to come up with minutes beforehand, but with rosters still taking shape -- not to mention rumors of players being traded, being lost overseas, etc -- this is fairly impractical.
Perhaps others besides myself also included projection methods based on age, multi-year trends, etc. These would be separate from player ratings, but they've been mixed in.

Yesterday, I worked up the standings but failed to post them:

Code: Select all

5.69   JohnH      6.74    ezPM
6.14   Crow       6.77    ez2
6.27    J.E.      6.83   KevinP
6.29   MikeG      6.90    zRA
6.32    dis       7.11   11Pyth
6.67   Vegas      7.29   DSMok1

Again, these standings are based on wins to date and point differential. If they included strength of schedule, past and future -- or if they were Pythagorean -- or both (!) -- they'd be different.
We could all be the winner!

[xkonk]

It looks like Vegas is on the move. Back on the 6th it was in last, now it's fifth.

[Mike G]

Not sure how this continual improvement can be sustained.

Code: Select all

5.22   JohnH      6.39    ez2
5.60    Crow      6.48   KevinP
5.77    J.E.      6.49    ezPM
5.83    dis       6.63    zRA
5.86   MikeG      6.74   11Pyth
6.02   Vegas      6.94   DSMok1

[EvanZ]

Is the rate of improvement slowing down, at least?

[Mike G]

After a night of upsets:

Code: Select all

4.92   JohnH      6.20   KevinP
5.35    Crow      6.26    ez2
5.52    J.E.      6.30    zRA
5.57   MikeG      6.36    ezPM
5.75    dis       6.49   11Pyth
5.83   Vegas      6.55   DSMok1

[J.E.]

When discussing effects of age on performance I posted this http://www.in-the-game.org/?p=17035 a couple of pages back.
Here's the same thing done for the NBA http://www.nba.com/hoop/are_bulls__thun ... 01_28.html with pretty much the same results. Doesn't exactly prove anything but I think it's one more argument for the peak basketball age being later than 26

Regarding season predictions: Orlando is slowly becoming the mess that most people figured they would become at some point, which led to overall lower win total predictions

[Mike G]

With 31% of the season completed, we've all slipped since the previous report.

Code: Select all

5.21   JohnH      6.32   KevinP
5.48    Crow      6.49    ez2
5.69   MikeG      6.64    ezPM
5.69    J.E.      6.71    zRA
5.96    dis       7.06   DSMok1
6.18   Vegas      7.17   11Pyth

Among the leaders, Hollinger got the most worse, proving that his early lead was pure luck.

[Mike G]

Regarding season predictions: Orlando is slowly becoming the mess that most people figured they would become at some point, which led to overall lower win total predictions

They're looking at 35.1 wins. Our lowest guess was (Crow) at 36. (High was ezPM at 45.) So they've already slipped below all our expectations.

When discussing effects of age on performance I posted this http://www.in-the-game.org/?p=17035 a couple of pages back.
Here's the same thing done for the NBA http://www.nba.com/hoop/are_bulls__thun ... 01_28.html with pretty much the same results. Doesn't exactly prove anything but I think it's one more argument for the peak basketball age being later than 26

I wonder if a typical player might still peak in the 24-28 range, but he may be more adaptable to a better team after that time.
In other words a LeBron or a Shaq might be more able to lead a team to titles when he's 24-28 years old; but his supporting cast might likely be older than NBA avg.

It may just be that hot young stars tend to want to have "their" team; and then with maturity they'd prefer to win a title -- manipulating their whereabouts to achieve this.
That doesn't mean they became "better" after 28, but that their priorities changed.

[Mike G]

At .331 of the season

Code: Select all

4.79   JohnH      6.37    ez2
5.13   Crow       6.47    ezPM
5.50    J.E.      6.50   KevinP
5.60   MikeG      6.53    zRA
5.77   Vegas      6.76   DSMok1
5.96    dis       7.08   11Pyth

In 2 days, almost everyone improved a little or a lot.
Kevin P got .018 worse, while John H whittled another .042 off his error.
Crow gained .035 and DSMok1 .030