Dutchball97 wrote:Any reason why people use RAPM and PIPM over BPM/VORP? RAPM and PIPM are terrible at giving a rough ranking of how well someone played unlike BPM/VORP but then there is also very little useful in depth analysis going on in RAPM/PIPM to offset that.
Is it just people enjoying their shiny new toys over the 'boring' alternatives?
So, I've noticed a trend recently where people assume that stats that aren't easy to find on major websites are newer than the ones on major websites, when the reality is closer to the opposite of this. Stuff bubbled up out of places like APBRmetrics & RealGM, the stuff that was easy to package for a mainstream audience got embraced by major sites, and now people tend to think that stuff came out of nowhere.
So, the mind behind bkref's BPM is Daniel Myers, who was a poster on APBRmetrics & RealGM before and after the time the version of that BPM came out in 2014. RAPM by contrast came out in 2010, and when Myers made that BPM he was also one of the many people who had done stuff with RAPM.
So no, the "shiny theory" is definitely not the answer.
You can find my explanation for why I'm drawn to +/- stats such as RAPM:
https://asubstituteforwar.wordpress.com/2011/03/26/the-nash-disequilibrium-or-why-i-use-statistics/But I'll specifically just show an image from that explanation with a bit of background:
There are two major axis to use to judge the confidence you should place in a stat: Validity & Reliability.
The box score tends to give you Reliability compared to +/- stats for a variety of reasons, but its Validity is inevitably biased toward the stuff that scorekeepers actually track. If they could record every meaningful action, as they can come far closer to in baseball, that would be all you needed. But basketball is 1) an open field sport with 2) continuous play. Scorekeepers will never come close to capturing it all. (It may be that AI eventually will come much closer to capturing it all.)
+/- stats give you strong Validity with iffy Reliability. They are valid because they are tied directly to the thing we're interested in - the score of the game - and thus can be argued to be un-biasable (though that depends on how you use the word "bias"). But there's noise mixed in with the signal - a bucket scored while a guy is on the floor doesn't mean he is a significant cause of that bucket.
What does that tell us about what we should use for evaluation of player impact?
1. If you've got something that gives you access to unbiased Validity, you should include it in your toolbox. You have to think about where noise comes into play as well as the fact that we don't get to see a player is all situations in all roles with all teammates against all opponents, and thus these stats cannot be used as "the answer", but where you have a good amount of data, if that data tells a story that deviates from what you would have expected, this tells you where you need to do more research. Why is the data what it is? The answer may not be directly related to a notion of how good the player is, but there were causes that caused the data to become what you eventually say, those causes are what you need to understand.
And note: If you're not using something with unbiased-validity among your main list of statistical tools just expect that you're being biased against smarter players. Expect that you'll overrate young guys who are thriving within particular regimented roles (young Andre Drummond was a classic example of this), and underrate veterans.
2. Based on a player's box score columns and your assessment of his holistic impact, you can really start asking yourself what was working and what wasn't. Add video analysis into this, and you can sometimes get to real concrete.
3. My personal holy grail is a more sophisticated form of BPM based on advanced player tracking which a) successfully emulates regression-based +/- stats in cases where we have sufficient sample size to have confidence while b) giving us access to an infrastructure of now-weighted player tracking action. You'd have access to good-as-it-gets holistic pegging of player impact yes, but the real power of the stat would be to have access granular level player actions in a form that could be argued to be both Valid and Reliable.
Make a stat like that, and coaches would be able to really, really figure out what to focus on to allow players to make the most progress in adding value to the team.
.
