APBRmetrics

DSMok1 wrote:

Statman wrote:Yep, very much so. My metric makes assumptions that if a guy has miserable production but plays a ton - especially on a good team - his rating goes up significantly (probably a great defender). A guy with great per minute production but plays little - especially on a bad team - his rating goes down significantly (probably a bad defender). All players on their teams are adjusted accordingly, so player ratings still compile correctly at the team level. BPM has minutes played as part of the regression I believe - but I'm not sure the value is tied inversely to production. THAT would make sense to me in terms of solid correlations - if it hasn't been tested, it should be.

That's a clever approach. Sort of a Bayesian prior methodology, using minutes and "box score production" to inform a defensive prior estimate. Good idea.

Thanks, been using that approach for years, and have never doubted that it was the right way to go in terms of how I approach box-score metrics.

bchaikin wrote:Javale McGee and Serge Ibaka had extremely high block rates, and don't rate that well in DRAPM.

are you saying serge ibaka is not a very good defender because DRAPM says so, or are you saying he is a very good defender but DRAPM is just not picking it up, and thus is not a good measure of a player's defense? here's why i ask...

I'm not going to pretend to speak for Daniel, but Serge Ibaka rates just fine in plus/minus land. In 2012, his peak Block% year, he was rated the 11th best defender in DRPM and the 12th best in J.E.'s xDRAPM, with similar estimates.

And if you take that as a reference point, subsequent changes in his Block% track changes in these two defensive measures. Then as for the 2012 baseline, Ibaka's defensive plus/minus estimate is rather well explained by conventional estimates of the contribution of defensive box score statistics (Steal%, PF%, and DefReb%) and his positional average.

They're really not much left to explain in his case.

if you separate "fake blocks" ie blocks in which the offense keeps possession, from "real blocks" ie blocks where the defense gets possession, then youll see that the former drop right out of everything.

then if you add in sportvu rim protection info, a lot of the value of "good blocks" falls right out, too.

Correlation of blocks to anything is going to be seriously weakened if you don't account for home scorekeeping inflation.
Since I make that correction, I get a value for blocks that is close to the value for steals.

All of you are right.

But the thing is BPM does not account for sportvu data or home bias. So blocks are underrated in the metric. It's simple as that.

But home/away splits are available back to 1985-86
http://www.basketball-reference.com/pla ... lits/1986/
Thus any boxscore metric could incorporate estimates of block and assist inflation.

permaximum wrote:All of you are right.

But the thing is BPM does not account for sportvu data or home bias. So blocks are underrated in the metric. It's simple as that.

I understand that you feel blocks are underrated, and perhaps they are. However, the coefficient simply comes from the process of the derivation, and that is the only place I can make changes. I welcome any comments that would help me revise and improve the methodology! As a reminder, it is all written up here: http://www.basketball-reference.com/about/bpm.html .

I can't tweak coefficients, but I can improve the derivation. I've already made one fix related to how I was weighting the players (I had been incorrectly been using sqrt(min) rather than min), but I welcome any further adjustments.

DSMok1 wrote: I welcome any comments that would help me revise and improve the methodology! As a reminder, it is all written up here: http://www.basketball-reference.com/about/bpm.html .
..

Yah right. You've misspelled a player's name right off the bat:
http://www.basketball-reference.com/pla ... eki01.html
I made the same mistake for years, before an apbr old-timer pointed it out.

Also, eWins always gets better correlation with pyth Wins when blocks and assists are adjusted for home bias.
This has the momentary effect of depressing assist rates league-wide; but that's offset by giving more weight to assists on 3-pointers.

In Steve Nash's heyday, both of these corrections work to boost his overall rating. His assists were actually undercounted in Phx (relative to what constitutes an Assist around the league), AND a lot of them created 3-pt possessions.
I was under-rating him at the time but since have gone back and fixed it.

Mike G wrote: Yah right. You've misspelled a player's name right off the bat:
http://www.basketball-reference.com/pla ... eki01.html
I made the same mistake for years, before an apbr old-timer pointed it out.

Also, eWins always gets better correlation with pyth Wins when blocks and assists are adjusted for home bias.
This has the momentary effect of depressing assist rates league-wide; but that's offset by giving more weight to assists on 3-pointers.

In Steve Nash's heyday, both of these corrections work to boost his overall rating. His assists were actually undercounted in Phx (relative to what constitutes an Assist around the league), AND a lot of them created 3-pt possessions.
I was under-rating him at the time but since have gone back and fixed it.

Good catch on the name!

I agree that the home bias adjustment would be preferred and the best approach, but in the interest of making BPM as generally applicable as possible (many leagues/applications/eras don't have such splits easily available), I prefer not to include it.

I'm not sure the non-boxscore defensive adjustment should apply to all players. Big men typically have their defensive contributions show up in the box score. How many good defensive big men do you know that don't have a decent DRB%, blk%, or stl%? Most of the great defensive big men show up in those categories. So the adjustment wouldn't make sense for them. You should only adjust for perimeter players. That way the defensive contributions of guys like Allen, Dumars, Bowen, Battier who are great perimeter defenders but don't have steals get credit for their defensive contribution. Maybe a variable with height included should be added. The shorter you are, the more non-box score defensive credit you get.

You can also do this for offense. If a player has low offensive numbers but plays a lot, maybe you should include a 3 point rate variable to account for spacing. Players like Kyle Korver or JJ Redick get credit with this.

colts18 wrote:I'm not sure the non-boxscore defensive adjustment should apply to all players. Big men typically have their defensive contributions show up in the box score. How many good defensive big men do you know that don't have a decent DRB%, blk%, or stl%? Most of the great defensive big men show up in those categories. So the adjustment wouldn't make sense for them. You should only adjust for perimeter players. That way the defensive contributions of guys like Allen, Dumars, Bowen, Battier who are great perimeter defenders but don't have steals get credit for their defensive contribution. Maybe a variable with height included should be added. The shorter you are, the more non-box score defensive credit you get.

You can also do this for offense. If a player has low offensive numbers but plays a lot, maybe you should include a 3 point rate variable to account for spacing. Players like Kyle Korver or JJ Redick get credit with this.

I could try adding in a position proxy variable to the adjustment process, perhaps. Would be a bit tricky to implement.

Otherwise--this sounds rather difficult to implement in a robust, statistically valid way. Any more thoughts on that?

DSMok1 wrote:
I could try adding in a position proxy variable to the adjustment process, perhaps. Would be a bit tricky to implement.

Otherwise--this sounds rather difficult to implement in a robust, statistically valid way. Any more thoughts on that?

A position proxy would be difficult to implement. Try using height. Obviously its not perfect (6' 9 Magic Johnson is a PG but 6' 7 Ben Wallace is a center). But small defensive guys should get credit more than big men because their contributions will never show on a box score unlike big men.

colts18 wrote:A position proxy would be difficult to implement. Try using height. Obviously its not perfect (6' 9 Magic Johnson is a PG but 6' 7 Ben Wallace is a center). But small defensive guys should get credit more than big men because their contributions will never show on a box score unlike big men.

Looking at it further--I think this is already equivalently done. Box score stats can very easily nail down the "position" a player plays at. My regression includes the team adjustment, so when the regression is run, it should already be making that adjustment--simply tweaking the box score values such that the appropriate "positions" get a buff, with the given that the team adjustment is the same for all players.

When you look at the results, BPM is actually pretty good for guard defense; in fact, it may be better at nailing down guard defense than big man defense (where positioning and influencing shots is more important and not measured).

DSMok1 wrote:

Mike G wrote: Yah right. You've misspelled a player's name right off the bat:
http://www.basketball-reference.com/pla ... eki01.html
I made the same mistake for years, before an apbr old-timer pointed it out.

Also, eWins always gets better correlation with pyth Wins when blocks and assists are adjusted for home bias.
This has the momentary effect of depressing assist rates league-wide; but that's offset by giving more weight to assists on 3-pointers.

In Steve Nash's heyday, both of these corrections work to boost his overall rating. His assists were actually undercounted in Phx (relative to what constitutes an Assist around the league), AND a lot of them created 3-pt possessions.
I was under-rating him at the time but since have gone back and fixed it.

Good catch on the name!

I agree that the home bias adjustment would be preferred and the best approach, but in the interest of making BPM as generally applicable as possible (many leagues/applications/eras don't have such splits easily available), I prefer not to include it.

I know I keep bringing this up, but when players are compiled at the team level, the home court scorer bias kinda fades away for the most part. Players on a team will see their respective assist &/or block weight drop if the scorer is too liberal (like a pace adjustment).

DSMok1 wrote: When you look at the results, BPM is actually pretty good for guard defense; in fact, it may be better at nailing down guard defense than big man defense (where positioning and influencing shots is more important and not measured).

I agree.