APBRmetrics

Looking at talkingpractice’s RAPM and JE’s from last week I see some cases of divergent player estimates. I know that the methods are different (and the # games covered might be too) so they are not going to be identical but I think a comparison is still worthwhile.

A few extreme differences:

On TP’s Ray Allen was –1.6 on offense and 1.5 below average on defense too for a total of –3.2. On JE’s Allen is +1.5 on offense and –1.9 on defense for a total of -0.4.

Aldridge’s offensive estimates were different by 4 and defense by 6.5.
vince Carter’s, Blake Griffin’s and James Harden’s overall ratings varied by 4 pts. Seth Curry’s by 6 pts.

45% of the overall estimates varied by less than 1 pt. 28% varied by 1-2 pts. 27% by 2+ pts. The average absolute difference is 1.4. The average difference is not that big but it is something worth keeping in mind.

You can use one metric or the other or look at both or average them or maybe do something to make a more even comparison.

Crow wrote:Looking at talkingpractice’s RAPM and JE’s from last week I see some cases of divergent player estimates. I know that the methods are different (and the # games covered might be too) so they are not going to be identical but I think a comparison is still worthwhile.

A few extreme differences:

On TP’s Ray Allen was –1.6 on offense and 1.5 below average on defense too for a total of –3.2. On JE’s Allen is +1.5 on offense and –1.9 on defense for a total of -0.4.

Aldridge’s offensive estimates were different by 4 and defense by 6.5.
vince Carter’s, Blake Griffin’s and James Harden’s overall ratings varied by 4 pts. Seth Curry’s by 6 pts.

45% of the overall estimates varied by less than 1 pt. 28% varied by 1-2 pts. 27% by 2+ pts. The average absolute difference is 1.4. The average difference is not that big but it is something worth keeping in mind.

You can use one metric or the other or look at both or average them or maybe do something to make a more even comparison.

Crow, I'm not sure where the differences you're saying are coming from? I assume its me misunderstanding your post. For the players (LaMarcus, Carter, Blake, Harden) you mention, I'm seeing a net difference of 1-1.5 or so, and not 4. Are you talking about the sum of our differences on Off and Def?

I talked about overall for some players and just offense or defense on others.

It may be that I had some issue in the construction and manipulation of the comparison data thru I did doublecheck it when I saw high variances. As I said earlier the data was from a week or two ago and it has been superseded by now. I will look into it closer at a later time or just start over with fresh data.

Talking practice's and JE's RAPM data sets had different column organization and different ways of displaying defensive impact. In setting up my spreadsheet I was aware of this but somehow mishandled a column. The average absolution difference is only 1.1 points. 55% of player estimates are within 1 point. 34% between 1-2 pts variance and only 11% varied by more than 2 points. Of the players I talked about above only Vince Carter varied by more than 2 on overall rating and it was only by about 3 points.

Updated my rankings to reflect the new age and 'effect of leading' adjustment. You can reach it through the main page again by clicking the '2013-2014' link in the left bottom. For now, no BoxScore information is used to compute these rankings

Everything else has been updated with new data from this season, so when you click a team name on the main page it shows team data for 2013-2014, like here.
'Same 4 Plus Minus' has also been updated - you can click a player's name on the team page to get to that player's S4PM page

Most of these pages are probably going to be updated weekly

J.E. wrote:Updated my rankings to reflect the new age and 'effect of leading' adjustment. You can reach it through the main page again by clicking the '2013-2014' link in the left bottom. For now, no BoxScore information is used to compute these rankings

Everything else has been updated with new data from this season, so when you click a team name on the main page it shows team data for 2013-2014, like here.
'Same 4 Plus Minus' has also been updated - you can click a player's name on the team page to get to that player's S4PM page

Most of these pages are probably going to be updated weekly

What's the methodology for the 13-14 pure RAPM? Is that a prior informed or no prior informed RAPM? Boxscore or no boxscore? If its prior informed, whats the priors?

J.E. wrote:Updated my rankings to reflect the new age and 'effect of leading' adjustment. You can reach it through the main page again by clicking the '2013-2014' link in the left bottom. For now, no BoxScore information is used to compute these rankings

Everything else has been updated with new data from this season, so when you click a team name on the main page it shows team data for 2013-2014, like here.
'Same 4 Plus Minus' has also been updated - you can click a player's name on the team page to get to that player's S4PM page

Most of these pages are probably going to be updated weekly

The individual team pages still say xRAPM, but this is just pure RAPM now, correct?

DSMok1 wrote:The individual team pages still say xRAPM, but this is just pure RAPM now, correct?

Yes. Good catch. I'll change the column header next time I update

colts18, it's the same as I described here viewtopic.php?p=18485#p18485
Multiyear, weighed, no priors other than 0, no BoxScore

J.E. wrote:

DSMok1 wrote:The individual team pages still say xRAPM, but this is just pure RAPM now, correct?

Yes. Good catch. I'll change the column header next time I update

colts18, it's the same as I described here viewtopic.php?p=18485#p18485
Multiyear, weighed, no priors other than 0, no BoxScore

So, wait--how does the aging get applied, then? Preprocessing step?

DSMok1 wrote:So, wait--how does the aging get applied, then? Preprocessing step?

Aging gets applied in a process that's extremely similar to using priors. So, for the sake of not confusing people I should probably say that it does use an age prior. Coach- and 'effect of leading'-adjustment are also built into this version

JE, can you explain what you mean by 'effect of leading' adjustment?

Afflalo (around –2 on RAPM estimate), did this factor into no one bidding aggressively for him or did Orlando overvalue him, not knowing or caring about this estimate?

Similar questio with Harrison Barnes. It is understandable they decided to wait at least 3-9 more months. They may ennd up with him having more or less value then. Do they really understand his game, potential and value? What evidence is there that they do?

Aaron Brooks estimated at –2.3 on last JE’s RAPM I looked at, -3.6 on Talkingpractice’s version. Houston appeared not to care about this metric or rating with respect to Brooks in signing him several times but they did manage to ship him out again. The Cavs must not have used RAPM as a significant part of their evaluation of Earl Clark pre-acquisition. He was one of the worst in the league last season and now he is much worse. But they did move him out. Are different factions in the front office using and not using RAPm or was these contradictory decisions decided on another basis or were these pieces acquired with a strong probability of being used as future trade pieces?

The new Kings owner is puffing up about believing in Cousins in the face of others wanting to trade him. He is paying near max or max and Cousins’ RAPM estimate is +0.3 to +0.4. He will probably never get him into the playoffs as the #1 guy.

I’ve often talked about blended metrics. Does anyone see problems with doing a run of RAPM where you strip out and assign half the value of a play’s outcome to the person who was credited for that end action in the boxscore according to SPM values and then let the RAPM regression assign the other half? How much would that change player values? It could. Wouldn’t be informative to see such a comparison, even once? Surely, a player deserves at least half the boxscore credit on average.

Jeff Green at better than –1 for first time of career.

Tim Hardaway Jr at near –4. Wait or sell?

For predicting RAPM in year 5 for a player what blended formula of prior year RAPM produces the best fit? Has anyone run an analysis of that?

Miles better on RAPM than Mason Plumlee. Realistic?

What % of players who end –3 or worse at end of first season ever get to positive by year 3 or 5? That would seem worth knowing. Anybody here run that or is willing to? How many teams know this information?

Crow wrote:
What % of players who end –3 or worse at end of first season ever get to positive by year 3 or 5? That would seem worth knowing. Anybody here run that or is willing to? How many teams know this information?

Kevin Durant was -4 or worse in both of his first 2 seasons according to J.E.'s RAPM. We know how that turned out.

FWIW, Monta is now +2. He was +1.2 in 2013. +0.8 in 2012. +0.5 in 2011. -1.0 in 2010. -0.8 in 2009.

Maybe the story on him need to be re-written (again).

colts18 wrote:

Crow wrote:
What % of players who end –3 or worse at end of first season ever get to positive by year 3 or 5? That would seem worth knowing. Anybody here run that or is willing to? How many teams know this information?

Kevin Durant was -4 or worse in both of his first 2 seasons according to J.E.'s RAPM. We know how that turned out.

I don't have the old RAPM files with me but by JE's current RAPM method, Durant was -3.2 in first year at SG under PJ Carelismo then improved to -1.5 the 2nd year.

It is better to judge a player by 2nd or 3rd year than first but it seemed like a question that would be worth checking (along with others) if there was enough time and interest.