2015 NBA Draft

[ampersand5]

I experimented with varying numbers/blends but used RAPM-WS because I liked how that blend ranked the NBA's players. While we know that things like RAPM are more predictive that WS, there are some plus-minus darlings that I don't want a model to consider as an ultimate NBA success story (Amir Johnson, etc.). RAPM-WS provides a nice balance and also produces sell-able results.

I find it difficult to understand why you are intentionally producing results that you know are inferior.
If you're goal is to just produce sellable results, then what's the point?

[nrestifo]

I find it difficult to understand why you are intentionally producing results that you know are inferior.
If you're goal is to just produce sellable results, then what's the point?

Not necessarily inferior though right? I'll be the first to jump on the RAPM-is-the-most-predictive-stat train, but should I only be considering one type of stat with a draft model? And is that really what I'm looking to identify? Let's say I have two players, A and B, and both have good RAPM. If WS likes Player A too, and get's there from a totally different methodology, isn't Player A better in a vacuum than Player B who just has good RAPM, but so-so WS? Maybe sellable is a tough-to-swallow word, but I think at some point I have to decide what kind of end-point I want to mold the model on. Is it the one with Kevin Durant at #8 and Rodney Stuckey in the top 30, or the one with Kevin Durant at #2 and Rodney Stuckey 41 spots lower?

[J.E.]

Definitely would prefer WS+RAPM over pure RAPM

[ampersand5]

I find it difficult to understand why you are intentionally producing results that you know are inferior.
If you're goal is to just produce sellable results, then what's the point?

Not necessarily inferior though right? I'll be the first to jump on the RAPM-is-the-most-predictive-stat train, but should I only be considering one type of stat with a draft model? And is that really what I'm looking to identify? Let's say I have two players, A and B, and both have good RAPM. If WS likes Player A too, and get's there from a totally different methodology, isn't Player A better in a vacuum than Player B who just has good RAPM, but so-so WS? Maybe sellable is a tough-to-swallow word, but I think at some point I have to decide what kind of end-point I want to mold the model on. Is it the one with Kevin Durant at #8 and Rodney Stuckey in the top 30, or the one with Kevin Durant at #2 and Rodney Stuckey 41 spots lower?

To step back, I shouldn't say that WS-RAPM is worse than another blend (evidenced by JE's post right above mine) and If the results are in fact more predictive, then that's obviously great. But the thought process confuses me.

If you are trying to avoid quixotic results such as Amir placing so high in the league when he seems by boxscore metrics to be mediocre, is that not just denying the power of RAPM?

If Stuckey is 30 in the model, why fudge it just to make him look lower?

If you are adding WS because you think RAPM is lacking something, and its addition to the blend adds an important feature, that makes sense. But framing it under the guise of making it "look" better and to mitigate RAPM stars seems counterintuitive for me.

thoughts?

[DSMok1]

To step back, I shouldn't say that WS-RAPM is worse than another blend (evidenced by JE's post right above mine) and If the results are in fact more predictive, then that's obviously great. But the thought process confuses me.

If you are trying to avoid quixotic results such as Amir placing so high in the league when he seems by boxscore metrics to be mediocre, is that not just denying the power of RAPM?

If Stuckey is 30 in the model, why fudge it just to make him look lower?

If you are adding WS because you think RAPM is lacking something, and its addition to the blend adds an important feature, that makes sense. But framing it under the guise of making it "look" better and to mitigate RAPM stars seems counterintuitive for me.

thoughts?

Using WS+RAPM or BPM+RAPM is a proxy for using RPM, which incorporates box scores in the most coherent manner. It's just not available historically. Using both a box score and a lineup based metric should always be better than just using one or the other--their strengths balance out the other's weaknesses.