An Unbiased Fan wrote:acrossthecourt wrote:An Unbiased Fan wrote:That's great, but I don't value RAPM as a metric for comparisons. All RAPM data tells me is that KG was in better rotations/lineups for his team than anyone else.
There has never been a connection shown between RAPM & Impact, so the over use of it just makes me disengage.It's like we're supposed to throw out performances, and just focus on an experimental metric that has never bore out as reliable. If you ask any player, coach, scout, analysis, fan...they would look at you funny if you said KG was a Top 5 player ever. He's arguably not even a Top 5 player of his era, but yet with RAPM, here we are. Top player of the 00's....that phrase will yield names like Kobe, Duncan, and Shaq, but not KG. But again, I guess we watched the games wrong and didn't realize KG's impact as he lost seven straight times in the 1st round, and then two years in a row to the Lakers despite having HCA, or missing the playoffs 3 straight years in his prime. Somehow we missed this GOAT impact that RAPM is trying to explain to us.
Sorry in advance, this post isn't directed at you, its just a spill over from the last thread.
That is completely false. Where are you getting that from? RAPM is heavily tested. I can find links if you want. I've tested it myself.
I think a lot of people agree with you too, sadly, but where are you getting that from? If RAPM is untested, then so is everything, and we all know the human mind has terrible biases and poor judgement, so what are we going to do? Just throw darts?
"Tested" it? I don't want this thread to turn into another RAPM debate, but post links that show RAPM correlates to impact. I really want to understand how Rashard Lewis had more impact than the DPOY in 2009. Or how Taj Gobson And Matt Bonner were the best players in 2012...oh wait, its NPI, so it has to be skewed even more with other data from previous seasons to "make it pass the eyetest". Seems legit.
You see my problem goes to the very notion that lineup data can be processed to display individual impact. it's simply not possible because everyone on the floor is lumped together. No matter how many models you throw at it, it will never work. You would have to have synergy type data, or a whole different methodology with the PbP data to yield something useful.
And I don't believe the box score says everything about a player. You can nitpick with Lewis and Gibson and others, but you can find similar outliers in every metric.
No one's solely using RAPM to rank legends. It's just a bit of evidence people are using, like championships, how teams do when a star is traded, PPG, etc.
Here's one link:
http://www.apbr.org/metrics/viewtopic.p ... 96&p=15343RAPM outperforms everything else in prediction. I think that's xRAPM though, a blend.
I did one too for 2013:
http://ascreamingcomesacrossthecourt.bl ... trics.htmlRAPM holds up well, and it does better when teams have more player personnel changes. That's a good sign.
Another:
http://sportskeptic.wordpress.com/2012/ ... the-goods/One note: if you predict next season's win totals with a metric, something like Win Shares does well because most teams stay intact and Win Shares is about explaining wins. PER doesn't even try on defense and is incomplete, while RAPM is reducing a different kind of error. Thus, it's better to look at how the metrics predict future wins or point differential two or three years out or with teams that have a high amount of changes.
xRAPM was built on out of sample testing and improving prediction. Saying it wasn't tested means you don't know the metric (I hope that doesn't come across as offensive; it's just the basic fact of how it was built.) You could use RAPM and compete with Vegas with a few tweaks. Yes it does align with impact. That's why it became so popular.
If you want a good blended metric, Real Plus/minus is pretty good and Talkingpractice's stuff, IPV, is probably even better, but they don't have historical results.
Something to consider: machines are getting better at predictions now, outperforming humans, but what does best is a human working together with a machine/computer. There was an anecdote about this in weather forecasting. So no, don't rank players on RAPM. But it's more evidence. Look for patterns. Use your human-powered pattern recognition and knowledge.
Poking at Rashard's DRAPM does nothing useful. It's one player. No metric is perfect. You don't sell a good car because you don't like the cup holders. Stare too long at stats and Dantley is better than Bird for most of the 80's, Robin Lopez is the next Bill Walton for his seismic shift in wins, and Stephon Marbury was a good point guard because of points and assists.
