This is Google's cache of viewtopic.php?t=2550&sid=14a5c8b3498faa94e18b35633e0b2c37. It is a snapshot of the page as it appeared on Feb 27, 2011 02:48:07 GMT. The current page could have changed in the meantime. Learn more

Text-only version
These search terms are highlighted: back2newbelf  
APBRmetrics :: View topic - 2010 playoff predictions
APBRmetrics Forum Index APBRmetrics
The statistical revolution will not be televised.
 
 FAQFAQ   SearchSearch   MemberlistMemberlist   UsergroupsUsergroups   RegisterRegister 
 ProfileProfile   Log in to check your private messagesLog in to check your private messages   Log inLog in 

2010 playoff predictions
Goto page 1, 2  Next
 
Post new topic   Reply to topic    APBRmetrics Forum Index -> General discussion
View previous topic :: View next topic  
Author Message
back2newbelf



Joined: 21 Jun 2005
Posts: 250

PostPosted: Thu Apr 15, 2010 5:49 am    Post subject: 2010 playoff predictions Reply with quote

Using SRS as indicator for team strength, then simulating 10000 7-game series with the WPyth% formula with an exponent of 9.5
-------------------------
-------------------------
Cleveland Chicago
41 29.58%
42 19.49%
40 19.36%
43 15.99%
34 6.22%
24 5.46%
14 2.65%
04 1.25%
--------------
84.42 % 15.58 %
-------------------------
-------------------------
Orlando Charlotte
41 24.46%
42 20.06%
43 18.08%
40 14.69%
24 8.64%
34 8.13%
14 3.89%
04 2.05%
--------------
77.29 % 22.71 %
-------------------------
-------------------------
Atlanta Milwaukee
41 19.21%
43 18.95%
42 16.48%
24 13.02%
34 11.97%
40 9.75%
14 6.93%
04 3.69%
--------------
64.39 % 35.61 %
-------------------------
-------------------------
Boston Miami
43 18.26%
41 17.82%
42 15.97%
24 14.93%
34 12.66%
14 8.49%
40 7.59%
04 4.28%
--------------
59.64 % 40.36 %
-------------------------
-------------------------
Los Angeles Oklahoma
43 18.46%
24 15.81%
41 15.77%
42 15.13%
34 13.1%
14 9.71%
40 6.7%
04 5.32%
--------------
56.06 % 43.94 %
-------------------------
-------------------------
Dallas San Antonio
24 20.97%
14 15.28%
43 15.1%
34 14.76%
42 10.52%
04 9.9%
41 9.78%
40 3.69%
--------------
39.09 % 60.91 %
-------------------------
-------------------------
Phoenix Portland
43 18.06%
24 16.5%
41 15.98%
42 14.69%
34 13.3%
14 9.9%
40 6.82%
04 4.75%
--------------
55.55 % 44.45 %
-------------------------
-------------------------
Denver Utah
24 19.77%
43 16.67%
34 14.2%
42 12.53%
14 12.52%
41 11.73%
04 7.54%
40 5.04%
--------------
45.97 % 54.03 %
-------------------------
-------------------------

There are some obvious questions and problems:
-Clevelands' rating would probably have been higher if LeBron didn't sit out the last couple of games
-How good is Milwaukee without Bogut? I think they're still pretty good
-Does Oklahoma really have a 44% chance to beat L.A.? The Lakers' SRS rating is just not very good. How much does the absence of Bynum have to do with it and when is he coming back? Can Kobe return to form?
-Is Dallas' true rating higher because of the trade? I doubt it. I really don't like Caron Butler
-How good is Portland really? Their SRS rating includes lots of games with Oden/Joel/without Batum. Can Roy play?

Things that people thought to be of importance but (I believe) were proven insignificant:
-experience, which many people will probably use to make a case for L.A. vs Oklahoma
-performance in regular season close games, which people might use to make a case for Dallas over San Antonio
-defense becoming more important in the playoffs, which would be a bad thing for Phoenix


We can make this a contest, but we will have to figure out a scoring system first


Last edited by back2newbelf on Thu Apr 15, 2010 9:34 am; edited 1 time in total
Back to top
View user's profile Send private message
DSMok1



Joined: 05 Aug 2009
Posts: 580
Location: Where the wind comes sweeping down the plains

PostPosted: Thu Apr 15, 2010 7:47 am    Post subject: Reply with quote

Can you do the same thing with Basketball Geek's efficiency differential ratings? I consider those the best, and since they adjust for both HCA and tempo inherently they are superior to SRS.

EDIT: In fact, because Ryan includes the error associated with each rating, it would probably be better to use a Z-score based estimate of the odds in each game.

In other words: OKC vs. LAL. HCA for LAL: HCA (estimated) of 3.5 (Pts/100 Pos).

Game 1: Normsdist( (LAL-OKC+HCA)/sqrt(OKCerr^2 + LALerr^2) ) = 98%

That doesn't seem right... those errors are too low. Wait! That must be the error of the estimate, not the opponent adjusted standard deviation. duh. Forget this method, then.
Back to top
View user's profile Send private message Visit poster's website
back2newbelf



Joined: 21 Jun 2005
Posts: 250

PostPosted: Thu Apr 15, 2010 8:20 am    Post subject: Reply with quote

DSMok1 wrote:
Can you do the same thing with Basketball Geek's efficiency differential ratings? I consider those the best, and since they adjust for both HCA and tempo inherently they are superior to SRS.

I use my own version of SRS which adjusts for tempo and HCA. Since everybody has played 82 games now(41 at home, 41 away), the adjustment for HCA isn't even necessary
Back to top
View user's profile Send private message
DSMok1



Joined: 05 Aug 2009
Posts: 580
Location: Where the wind comes sweeping down the plains

PostPosted: Thu Apr 15, 2010 9:00 am    Post subject: Reply with quote

back2newbelf wrote:
DSMok1 wrote:
Can you do the same thing with Basketball Geek's efficiency differential ratings? I consider those the best, and since they adjust for both HCA and tempo inherently they are superior to SRS.

I use my own version of SRS which adjusts for tempo and HCA. Since everybody has played 82 games now(41 at home, 41 away), the adjustment for HCA isn't even necessary


Ryan does not use a uniform home court advantage for each team in calculating the ratings, IIRC. It is included as a part of the multilevel model predicting likelihood of 0, 1, 2, 3, or 4(?) pts in a possession as something to control for.

EDIT: full discussion here: http://www.basketballgeek.com/2009/11/26/power-rankings-ssr-updates-and-more/


Last edited by DSMok1 on Thu Apr 15, 2010 9:04 am; edited 1 time in total
Back to top
View user's profile Send private message Visit poster's website
schtevie



Joined: 18 Apr 2005
Posts: 405

PostPosted: Thu Apr 15, 2010 9:01 am    Post subject: Reply with quote

I should think that these simulation results are substantively incorrect if they are based upon the entire 2009-10 season. Just to take two examples, the Celtics started off 23-5, but this is not how they finished. Also, the beginning of the year saw the Shaq etc. integration problems in Cleveland. Why not try running simulations based on the back half of the season as a comparison. That should be more suggestive.

Better still, though it would be difficult to come up with a simple rule, would be to model anticipated lineup usage. Coaches tend to tighten these up and this surely influences outcomes on the margin.
Back to top
View user's profile Send private message
DSMok1



Joined: 05 Aug 2009
Posts: 580
Location: Where the wind comes sweeping down the plains

PostPosted: Thu Apr 15, 2010 9:22 am    Post subject: Reply with quote

schtevie wrote:
I should think that these simulation results are substantively incorrect if they are based upon the entire 2009-10 season. Just to take two examples, the Celtics started off 23-5, but this is not how they finished. Also, the beginning of the year saw the Shaq etc. integration problems in Cleveland. Why not try running simulations based on the back half of the season as a comparison. That should be more suggestive.

Better still, though it would be difficult to come up with a simple rule, would be to model anticipated lineup usage. Coaches tend to tighten these up and this surely influences outcomes on the margin.


A good rough way would be to estimate the lineup in the playoffs and use SPM's to model overall team efficiency. I have SPM's calculated to the end of the regular season that are adjusted according to Ryan J Parker's team ratings.

Here they are: Statistical Plus/Minus to end of regular season.

If we could estimate the lineups for each team, it wouldn't be too hard to create efficiency estimates.
Back to top
View user's profile Send private message Visit poster's website
back2newbelf



Joined: 21 Jun 2005
Posts: 250

PostPosted: Thu Apr 15, 2010 9:25 am    Post subject: Reply with quote

DSMok1 wrote:

Ryan does not use a uniform home court advantage for each team in calculating the ratings, IIRC.

I found a non-uniform home court advantage to be useless in past regular seasons. See thread linked in first post
schtevie wrote:

I should think that these simulation results are substantively incorrect if they are based upon the entire 2009-10 season. Just to take two examples, the Celtics started off 23-5, but this is not how they finished. Also, the beginning of the year saw the Shaq etc. integration problems in Cleveland. Why not try running simulations based on the back half of the season as a comparison. That should be more suggestive.

I also found weighting (or omitting) parts of the season to be (almost) useless, also see thread linked in first post.

The results posted are my final predictions and I'm relatively confident in them. This is the best I came up with in my limited research in predicting playoff series'. I will not put any more work into those. If you want predictions from another method, do them yourself and post them


Last edited by back2newbelf on Thu Apr 15, 2010 9:36 am; edited 1 time in total
Back to top
View user's profile Send private message
Ryan J. Parker



Joined: 23 Mar 2007
Posts: 707
Location: Raleigh, NC

PostPosted: Thu Apr 15, 2010 9:30 am    Post subject: Reply with quote

I use data from each possession, so it is still important (for me) to adjust for this, even after an 82 game season. Also, I don't have data for every game. Cool

DSMok1, the errors are for the mean estimate, so you won't be able to use those to estimate win probabilities.

Also, I fit a uniform per possession HCA. It would be really nice to fit this as a hierarchical model, but doing the computation required for this multinomial model is tough for me at this point.

This is part of my senior project, so I'll more in depth details will be available soon.
_________________
I am a basketball geek.
Back to top
View user's profile Send private message Visit poster's website
DSMok1



Joined: 05 Aug 2009
Posts: 580
Location: Where the wind comes sweeping down the plains

PostPosted: Thu Apr 15, 2010 10:27 am    Post subject: Reply with quote

Ryan J. Parker wrote:
I use data from each possession, so it is still important (for me) to adjust for this, even after an 82 game season. Also, I don't have data for every game. Cool

DSMok1, the errors are for the mean estimate, so you won't be able to use those to estimate win probabilities.

Also, I fit a uniform per possession HCA. It would be really nice to fit this as a hierarchical model, but doing the computation required for this multinomial model is tough for me at this point.

This is part of my senior project, so I'll more in depth details will be available soon.


Yeah, I figured that out pretty quick!

I'm going to post some estimated playoff team efficiencies, based purely upon playing time and SPM.
Back to top
View user's profile Send private message Visit poster's website
DSMok1



Joined: 05 Aug 2009
Posts: 580
Location: Where the wind comes sweeping down the plains

PostPosted: Thu Apr 15, 2010 10:59 am    Post subject: Reply with quote

Estimated team ratings based on Statistical Plus/Minus, adjusting for estimated playing time in the playoffs.

In general, I bumped up the players who normally get the most playing time for each team. The basic method was to take all players MPG^1.45 and then sum the team totals to 240. (Bench players who played in quite a few minutes in just a few games, like Sebastian Telfair, had their factor multiplied by GP/82). Of course, I also adjusted based on the injury prognosis for each player.

The results:

Western Conference:
1 LAL 9.13
8 OKC 6.22

4 DEN 6.86
5 UTA 7.97

3 PHX 6.61
6 POR 7.27 (with Roy at half of expected PT, 5.75 if he is out completely, 8.55 if he plays normally)

2 DAL 5.45
7 SAS 8.23 (The big 3 bumped up to 35 minutes a game)

Eastern Conference:
1 CLE 10.85
8 CHI -2.01

4 BOS 5.82
5 MIA 4.61

3 ATL 7.09
6 MIL -0.18

2 ORL 9.02
7 CHA 3.20
Back to top
View user's profile Send private message Visit poster's website
Neil Paine



Joined: 13 Oct 2005
Posts: 774
Location: Atlanta, GA

PostPosted: Fri Apr 16, 2010 12:25 pm    Post subject: Reply with quote

I got similar results when I did this yesterday, except I used "True SPM Skill", which is a predictor of next year's SPM, and then backed out the 2011 age adjustment:

Code:
team_id talent
ATL     4.69
BOS     6.49
CHA     2.92
CHI     -2.09
CLE     12.78
MIA     3.08
MIL     -0.81
ORL     9.54
DAL     6.10
DEN     4.56
LAL     8.11
OKC     1.49
PHO     4.37
POR     4.09
SAS     7.97
UTA     4.90


BTW, did you check out the latest SPM formula? I re-ran it yesterday, adding 2010 data and throwing out the 2007 Ilardi estimates (they weren't per-100 poss. and they included the playoffs).

http://www.basketball-reference.com/blog/?page_id=4122
_________________
http://www.basketball-reference.com/blog/
Back to top
View user's profile Send private message Visit poster's website
DSMok1



Joined: 05 Aug 2009
Posts: 580
Location: Where the wind comes sweeping down the plains

PostPosted: Sat Apr 17, 2010 8:01 am    Post subject: Reply with quote

Neil Paine wrote:
BTW, did you check out the latest SPM formula? I re-ran it yesterday, adding 2010 data and throwing out the 2007 Ilardi estimates (they weren't per-100 poss. and they included the playoffs).

http://www.basketball-reference.com/blog/?page_id=4122


That's wonderful!!! Very Happy .

Okay, so I get excited about strange things...

I have a request to make: could you please run a SPM WITHOUT MPG also? Here's why: I'm using a Bayesian prior based on team strength and MPG (as a proxy for the coach's evaluation of the player). Because of the MPG dependence of the SPM regression, I cannot truly say the information from the 2 "models" is orthogonal/independent. So could you make a second SPM run without MPG as a predictor, please?

EDIT: And even better... could you run a regression on APM with only these predictors: MPG, MPG^2, MPG^3, Adjusted Team Eff, Age, Age^2, Age^3? Or something like that? That would create the Bayesian prior for my true talent estimations. I'm trying to get an estimate of what we should expect a player's talent to be, before applying our SPM (preferably w/o MPG, because MPG depends on team strength) in a Bayesian update. Do you see what I mean?

Neil Paine wrote:
I got similar results when I did this yesterday, except I used "True SPM Skill", which is a predictor of next year's SPM, and then backed out the 2011 age adjustment:

Code:
team_id talent
ATL     4.69
BOS     6.49
CHA     2.92
CHI     -2.09
CLE     12.78
MIA     3.08
MIL     -0.81
ORL     9.54
DAL     6.10
DEN     4.56
LAL     8.11
OKC     1.49
PHO     4.37
POR     4.09
SAS     7.97
UTA     4.90

I'm using just this year's SPM's, and I normalize to Ryan J Parker's team efficiencies (what efficiencies are you using?). I'm also adjusting for injuries and bumping up time for the top players on each team to better reflect "peak" team efficiency.


Last edited by DSMok1 on Sat Apr 17, 2010 8:45 am; edited 1 time in total
Back to top
View user's profile Send private message Visit poster's website
DSMok1



Joined: 05 Aug 2009
Posts: 580
Location: Where the wind comes sweeping down the plains

PostPosted: Sat Apr 17, 2010 8:39 am    Post subject: Reply with quote

Okay, using the new SPM regression, and updating for the latest injury reports.

The results:

Western Conference:
1 LAL 8.71
8 OKC 6.11

4 DEN 6.79
5 UTA 7.63

3 PHX 7.09
6 POR 6.02

2 DAL 5.73
7 SAS 7.97 (The big 3 bumped up to 35 minutes a game)

Eastern Conference:
1 CLE 10.38
8 CHI -0.71

4 BOS 5.60
5 MIA 4.61

3 ATL 7.39
6 MIL 0.25

2 ORL 8.93
7 CHA 3.09
Back to top
View user's profile Send private message Visit poster's website
DSMok1



Joined: 05 Aug 2009
Posts: 580
Location: Where the wind comes sweeping down the plains

PostPosted: Sat Apr 17, 2010 11:05 am    Post subject: Reply with quote

Okay, here are my playoff calculations, similar to what B2NB did. I actually calculated the outcomes rigorously (looking at the odds of each possible permutation and summing) rather than running a Monte Carlo sim.

These are based on the adjusted team efficiencies in the last post. I applied a home court advantage of 3.7 pts/100 poss, used an average efficiency of 107, and used a pythagorean exponent of 14.

LAL vs OKC
4 to 0 10.4%
4 to 1 23.0%
4 to 2 16.6%
4 to 3 20.7%
3 to 4 9.1%
2 to 4 12.3%
1 to 4 5.1%
0 to 4 2.7%

LAL OKC
70.8% 29.2%

DEN vs UTA
4 to 0 4.4%
4 to 1 13.2%
4 to 2 11.4%
4 to 3 19.0%
3 to 4 13.1%
2 to 4 20.7%
1 to 4 11.3%
0 to 4 6.9%

DEN UTA
48.0% 52.0%

PHO vs POR
4 to 0 7.3%
4 to 1 18.5%
4 to 2 14.6%
4 to 3 20.8%
3 to 4 11.1%
2 to 4 16.1%
1 to 4 7.5%
0 to 4 4.2%

PHO POR
61.1% 38.9%

DAL vs SAS
4 to 0 3.0%
4 to 1 9.8%
4 to 2 9.0%
4 to 3 16.7%
3 to 4 13.8%
2 to 4 23.7%
1 to 4 14.5%
0 to 4 9.6%

DAL SAS
38.4% 61.6%

---------

CLE vs CHI
4 to 0 40.1%
4 to 1 36.8%
4 to 2 13.6%
4 to 3 7.0%
3 to 4 1.0%
2 to 4 1.1%
1 to 4 0.3%
0 to 4 0.1%

CLE CHI
97.5% 2.5%

BOS vs MIA
4 to 0 7.1%
4 to 1 18.2%
4 to 2 14.4%
4 to 3 20.8%
3 to 4 11.2%
2 to 4 16.3%
1 to 4 7.6%
0 to 4 4.3%

BOS MIA
60.6% 39.4%

ATL vs MIL
4 to 0 24.2%
4 to 1 34.2%
4 to 2 17.9%
4 to 3 14.3%
3 to 4 3.5%
2 to 4 4.0%
1 to 4 1.3%
0 to 4 0.6%

ATL MIL
90.6% 9.4%

ORL vs CHA
4 to 0 19.6%
4 to 1 31.7%
4 to 2 18.4%
4 to 3 16.8%
3 to 4 4.8%
2 to 4 5.8%
1 to 4 2.0%
0 to 4 0.9%

ORL CHA
86.5% 13.5%

Subsequent rounds in the next post:
Back to top
View user's profile Send private message Visit poster's website
DSMok1



Joined: 05 Aug 2009
Posts: 580
Location: Where the wind comes sweeping down the plains

PostPosted: Sat Apr 17, 2010 12:01 pm    Post subject: Reply with quote

Continued:

Round 2:

LAL vs UTA
4 to 0 7.1%
4 to 1 18.1%
4 to 2 14.4%
4 to 3 20.7%
3 to 4 11.3%
2 to 4 16.4%
1 to 4 7.7%
0 to 4 4.3%

LAL UTA
60.3% 39.7%

PHO vs SAS
4 to 0 4.4%
4 to 1 13.0%
4 to 2 11.3%
4 to 3 19.0%
3 to 4 13.1%
2 to 4 20.9%
1 to 4 11.4%
0 to 4 7.0%

PHO SAS
47.7% 52.3%

CLE vs BOS
4 to 0 16.2%
4 to 1 29.2%
4 to 2 18.2%
4 to 3 18.5%
3 to 4 6.1%
2 to 4 7.6%
1 to 4 2.8%
0 to 4 1.3%

CLE BOS
82.2% 17.8%

ORL vs ATL
4 to 0 8.2%
4 to 1 19.9%
4 to 2 15.3%
4 to 3 20.9%
3 to 4 10.5%
2 to 4 14.9%
1 to 4 6.7%
0 to 4 3.6%

ORL ATL
64.2% 35.8%

Semifinals:

LAL vs SAS
4 to 0 6.5%
4 to 1 17.2%
4 to 2 13.8%
4 to 3 20.5%
3 to 4 11.7%
2 to 4 17.2%
1 to 4 8.3%
0 to 4 4.7%

LAL SAS
58.0% 42.0%

CLE vs ORL
4 to 0 8.0%
4 to 1 19.6%
4 to 2 15.1%
4 to 3 20.9%
3 to 4 10.7%
2 to 4 15.1%
1 to 4 6.8%
0 to 4 3.7%

CLE ORL
63.6% 36.4%

Finals:
CLE vs LAL
4 to 0 8.7%
4 to 1 13.4%
4 to 2 22.8%
4 to 3 20.9%
3 to 4 10.2%
2 to 4 9.8%
1 to 4 10.8%
0 to 4 3.4%

CLE LAL
65.9% 34.1%
Back to top
View user's profile Send private message Visit poster's website
Display posts from previous:   
Post new topic   Reply to topic    APBRmetrics Forum Index -> General discussion All times are GMT - 5 Hours
Goto page 1, 2  Next
Page 1 of 2

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum


Powered by phpBB © 2001, 2005 phpBB Group