| View previous topic :: View next topic |
| Author |
Message |
basketballvalue
Joined: 07 Mar 2006
Posts: 208
|
 Posted: Mon Dec 15, 2008 11:42 pm Post subject: BasketballValue.com update: 2 Year Adjusted +/- for '08-'09 |
 |
|
All,
I'm happy to report that I have updated basketballvalue.com to show 2 year adjusted +/- data for the 2008-2009 season and future seasons as they occur. In line with our recent article at 82games, the current regular season is weighted at 5x, the previous playoffs is weighted at 2x, and the previous regular season is weighted at 1x. When the 2009 playoffs occur, they will be weighted at 10x. The code is currently set up so that all future seasons will have 2-year adjusted +/- calculated using this 10x/5x/2x/1x weighting.
Please keep in mind, no other statistics are calculated as "2-year" stats. As a result, the 2008-2009 unadjusted data (e.g. Team Offensive Rebounding Rate or Unadjusted +/-) is only for 2008-2009.
Additionally, I have fixed a small bug I found in counting possessions and reprocessed the data for the 2007 playoffs up to this season. I hope to reprocess the 2006-2007 regular season at some point so that I can show adjusted +/- for that season as well as the 2007 playoffs.
As always, I appreciate the feedback you've provided. Please let me know if you see any issues resulting from this change. You can post responses here or send me an email. Implementing this update has caused me to fall a little behind on my daily processing, so I will be working to get back to "approximately real-time" processing of games over the next couple of days. After that, I'm planning to bring back the team vs team pages as my next step for the website so you can see how different units or players matched up when teams played.
Thanks,
Aaron
_________________
www.basketballvalue.com
Follow on Twitter |
|
| Back to top |
|
 |
cherokee_ACB
Joined: 22 Mar 2006
Posts: 157
|
| Posted: Tue Dec 16, 2008 4:50 am Post subject: Re: BasketballValue.com update: 2 Year Adjusted +/- for '08- |
|
|
| basketballvalue wrote: | The code is currently set up so that all future seasons will have 2-year adjusted +/- calculated using this 10x/5x/2x/1x weighting.
|
Good work. Are you using that same weighting to set the minutes played thresholds? |
|
| Back to top |
|
|
basketballvalue
Joined: 07 Mar 2006
Posts: 208
|
| Posted: Tue Dec 16, 2008 6:45 am Post subject: Re: BasketballValue.com update: 2 Year Adjusted +/- for '08- |
|
|
| cherokee_ACB wrote: |
Good work. Are you using that same weighting to set the minutes played thresholds? |
Hmm, good point. I was not weighting the minutes. I had originally only included minutes from this year (and this playoffs if applicable), but I'm planning to switch it to cumulative minutes. I figure that people who played significant minutes last year but are retired this year should still be rated.
Thanks,
Aaron
_________________
www.basketballvalue.com
Follow on Twitter |
|
| Back to top |
|
|
Mike G
Joined: 14 Jan 2005
Posts: 3623
Location: Hendersonville, NC
|
| Posted: Tue Dec 16, 2008 4:45 pm Post subject: |
|
|
I see that Al Jefferson is no longer the worst player in the league. In fact, this year he's a net +9.25 .
This even makes his 2-yr rating a slight positive. Far out.
_________________
`
36% of all statistics are wrong |
|
| Back to top |
|
|
basketballvalue
Joined: 07 Mar 2006
Posts: 208
|
| Posted: Wed Dec 17, 2008 12:00 am Post subject: Re: BasketballValue.com update: 2 Year Adjusted +/- for '08- |
|
|
| cherokee_ACB wrote: | | basketballvalue wrote: | The code is currently set up so that all future seasons will have 2-year adjusted +/- calculated using this 10x/5x/2x/1x weighting.
|
Good work. Are you using that same weighting to set the minutes played thresholds? |
Cherokee,
I'm using the weightings to establish who is in the top 75% by weighted minutes and should be rated in the regression. However, in the display on the website I'm still filtering based on unweighted minutes played in the current season, since I only want to show players who are playing significant minutes in the selected season. For this season, they are intended to be 2008-2009 ratings with the older data incorporated to reduce the noise, so I only want to show 2008-2009 players who have played enough minutes.
Thanks,
Aaron
_________________
www.basketballvalue.com
Follow on Twitter |
|
| Back to top |
|
|
schtevie
Joined: 18 Apr 2005
Posts: 414
|
| Posted: Sat Dec 20, 2008 10:35 am Post subject: |
|
|
Aaron,
Could you give a short explanation as to why a previous season's data should be weighted at 1/5 the current season? I cannot think of a theoretical justification.
There is the argument that crunch time minutes should be weighted more, given that apparently players don't believe what is true that a point is a point is a point no matter when it is scored, and I suppose on similar grounds that play-off minutes are more indicative of true ability. But I don't understand the concept of a set seasonal weighting.
I do understand the idea of weighting for expected career productivity (a set of rising and falling weights - why not, as a first pass, use Ed's age profiles?) but not set seasonal weights.
Thanks. |
|
| Back to top |
|
|
basketballvalue
Joined: 07 Mar 2006
Posts: 208
|
| Posted: Sun Dec 21, 2008 1:39 pm Post subject: |
|
|
| schtevie wrote: | Aaron,
Could you give a short explanation as to why a previous season's data should be weighted at 1/5 the current season? I cannot think of a theoretical justification.
|
Schtevie,
Thanks for your question, sorry if I was unclear. I would not say that there's a theoretical justification for 1/5.
Stepping back, what I'm trying to do is produce an adjusted +/- rating for the season in question (e.g. 2008-2009). Ideally, this estimate would
1. only be based on 2008-2009 data, and
2. would perfectly estimate the true rating with 0 standard error
Of course, we know those two ideals are incompatible and that as we add more data the error term is reduced. Previously, I only was including the data from the current year (weight of 0 for the previous season), and people rightfully commented that the error term in the first few weeks and even now was too high for the estimates to be meaningful.
One approach to address the standard errors would be to weight last year and this year equally. This weighting would result in the greatest noise reduction (addressing ideal 2 as best as possible), but would be describing performance over both years. As a result, players on the decline would have a rating that is higher than their true performance for this year only, and players on the rise would have a rating that is lower.
So, choosing the weights is essentially trading off the two ideals. There's no right answer, it's a judgment call. If I were trying to describe their performance over the entire two years, I would weight them equally. Since I am trying to produce a 2008-2009 rating right now, I knew I did not want to weight them equally (Perhaps there's a better description I could use instead of "2-year"?) Choosing 1/5 felt like the right balance, but I could easily have used 1/4 (leading to a nice 1/2 to 1/4 progression for last years playoffs and regular season), or 1/8, or something else. I reserve the right to change it in the future if I identify a better choice during the implementation. 
Thanks,
Aaron
PS Sorry I'm a little behind in my daily processing, I'm still catching up from the break associated with upgrading to incorporating that 2nd year to reduce the noise.
_________________
www.basketballvalue.com
Follow on Twitter |
|
| Back to top |
|
|
deepak
Joined: 26 Apr 2006
Posts: 665
|
| Posted: Sun Dec 21, 2008 2:18 pm Post subject: |
|
|
Are you weighting the entire current season 5x more than all of last year, or are you weighting each minute of the current season 5x more than last year?
For example, Luther Head so far has played about 1/6th the total minutes he played a season ago. Does that mean his 2-year APM is mostly based on last year? |
|
| Back to top |
|
|
basketballvalue
Joined: 07 Mar 2006
Posts: 208
|
| Posted: Sun Dec 21, 2008 2:54 pm Post subject: |
|
|
| deepak_e wrote: | Are you weighting the entire current season 5x more than all of last year, or are you weighting each minute of the current season 5x more than last year?
For example, Luther Head so far has played about 1/6th the total minutes he played a season ago. Does that mean his 2-year APM is mostly based on last year? |
I don't think of it in terms of players, but of the whole league. At this point, I've analyzed the 12/12 games but haven't posted them. That's 329 games due to the broken play-by-plays. 329/1230 is 26.75% of the games, so if we were ignoring the playoffs the estimates would be weighting the season to date a little more than last season (5*26.7 = 134% for this season to 100% for last season = 57%-43%).
Thanks,
Aaron
_________________
www.basketballvalue.com
Follow on Twitter |
|
| Back to top |
|
|
Mountain
Joined: 13 Mar 2007
Posts: 1527
|
| Posted: Mon Dec 22, 2008 4:03 am Post subject: |
|
|
Maybe the weights could slide as the season progresses. Extra tweaking but up to 1/4 thru the current season I'd think the previous weight should be higher than 1/5th. Maybe even 1/2? Then in second quarter maybe it goes to 1/3 or 1/2.5. After midpoint of season maybe somewhere between 1/4th and 1/3. At end of season it would still be useful to see 1 year current season values standalone.
Just one approach for you consideration if it makes sense to you operationally and philosophically. |
|
| Back to top |
|
|
|
FAQ
Search
Memberlist
Usergroups
Register
Profile
Log in to check your private messages
Log in
