2012 ASPM
All of the derivation of ASPM and VORP may be found on my ASPM and VORP page.
To see the underlying spreadsheet:
- Excel Spreadsheet, with macros for updating.
ASPM data last updated 4/27/2012 with year-end numbers
All of the derivation of ASPM and VORP may be found on my ASPM and VORP page.
To see the underlying spreadsheet:
ASPM data last updated 4/27/2012 with year-end numbers
Pretty neat, but just fyi sorting the table doesn’t work for me.
Working on it
Okay, fixed it. Had to sacrifice conditional formatting.
Cool. One other thing – I don’t see Kobe Bryant???
And Dwight Howard…
Got a bug… I’ll squash it…
Actually, it wasn’t a bug–the maximum value for the MP filter was not automatically setting to the highest value in the database. I just set that filter to be a “minimum” filter rather than a “range” filter to avoid confusion.
[...] Source: godismyjudgeok.com [...]
Hello, Daniel. Thank you for responding to my previous questions about ASPM. The following are some suggestions–ideas–on creating a more predictive value metric. I imagine that a lot of these ideas would take an incredibly long time to actually implement, so these are just suggestions for the nebulous future. Some of these ideas may be redundant–i.e. I looked at your list of things to do and it seems like some of these overlap. Also, considering that I’m presenting these ideas as for a predictive measure, you may want to consider them as going to a form of ASPM that would literally be denoted as purely predictive (Future ASPM, FASPM, PASPM?). The list: home court advantage–considered for each team, days of rest (are these what you mean by “temporally-locally weighted rankings”?, distance of travel, age-curves (which I see is on your list–could it be sensitive to height, wingspan, position, mileage? (which would require some defining)), possibly replacing the defensive end of the metric with DRAPM (or finding a way to blend them)(perhaps this defeats the purpose of a box-score stat though), regressing play by play data onto RAPM (I assume you’re using normal box-score data?), strength of schedule (is it possible to measure players at the possession level?) ***I’m sure others can think of a thousand other ways to chew up your time, so that we can all know how our favorite teams will perform. I’ve only recently discovered sports-skeptic, the city2, and your site–I’ve been stuck on wages of wins island. So it follows, I really appreciate your work, and what seems like an effort for transparency on your part, along with a willingness to take the time to politely respond to comments.
Oh, and to make it clear, I wasn’t trying to barb Wages of Wins. To be honest, they’ve for the most part responded to my comments and they also do a good job–I wasn’t trying to create a positive / negative dichotomy.
I certainly agree about the need to create a predictive metric. My thought is to emulate the framework used in baseball analytics: start by explaining what happened/measuring what happened accurately. That is what, I hope, ASPM does.
Then move toward prediction. To create a predictive ASPM model (which I already did for last year, but have not automated the process), basically regress toward a Bayesian prior to obtain a true talent estimate, then apply aging curves to and Bayesian weighting to obtain an estimated true-talent ASPM for year n+1. Fine tune the prior, aging curves, and weighting using the 30+ years of ASPM data I have compiled. The generic aging curve for ASPM is here; it was already calculated some time ago on the APBRmetrics Forum:
ASPM Aging Curve
For the current ASPM, home court advantage is taken care of when summing to each team’s adjusted efficiency differential; this also roughly accounts for strength of opposition. Read the full exposition of ASPM’s theory at http://godismyjudgeok.com/DStats/aspm-and-vorp/ .
Eventually I’d like to create fully-adjusted team efficiencies–adjusting for injuries, playing time, rest days (see my research on them here: http://www.apbr.org/metrics/viewtopic.php?f=2&t=56 ), etc.
Also, eventually, I’d like to create game-by-game ASPM estimates, which I have also done before but not automated.