Curry 22' vs Harden 18'

[jalengreen]

Are you saying that you would have predicted the Warriors sans KD would win a series over Houston in 2019 before it happened? Because that would make you close to unique.

No, I'm sure I would have predicted the Rockets to win (assuming we're talking about the Warriors not having KD from the start of the series. I did actually think once GSW won G5 that the series was over... which may have been hyperbolic given that G6 was fairly close but still, gives you an idea of my thoughts at the time) - but I'm also sure that I would've maintained the possibility of the Warriors winning. At a glance I think my thoughts are comparable to what you were talking about earlier in another thread about the 2011 Mavs. (I don't want this to come off the wrong way, this conversation just instantly reminded me of reading your comment earlier [which I agreed with] and I think it explains my perspective well.)

First thing I'll say: I'm not going to say you're crazy for disagreeing with me about the Mavs. These are all champion teams worthy of respect, and in fact during the '10-11 season I was saying repeatedly that the Mavs when healthy had an extremely impressive regular season record. I didn't pick them in the series against the Lakers where they won in a sweep that left us gobsmacked, but I was saying ahead of time that Dallas could win, which most didn't believe.

Yet, I never argued that those Mavs were better than the Laker & Heat champs that surrounded them. Why? Hmm, well, some salient points:

1. The '10-11 Mavs played the Lakers on the downswing and the Heat on the upswing. I don't think you'll find anyone that believes the teams the Mavs beat were the best versions of those teams. Doesn't mean that '10-11 Mavs couldn't be even better, but we shouldn't forget this.

Similarly, I wouldn't have picked the Warriors to win a full series without KD against the 2019 Rockets, but I wouldn't have counted that team out completely. Further, I wouldn't argue that the series has much to do with what happened in 2018 because I don't believe the Warriors beat the best version of the Harden/CP3 Rockets. In either 2018 or 2019, that is.

You make good points and I was nodding along until the end when...

wait are you saying you think the Rockets were simply a better team than the Warriors period?

no, i was saying that i think both the rockets teams that the warriors faced werent the best version of the harden/cp3 rockets. because the best version of them was when healthy in 2018. in 2018 WCF cp3 gets hurt, and again in 2019 i think they were worse largely due to cp3's decline.

[jalengreen]

Yes, explainable machine learning is quite the discussion. I don't think there's a consensus on the matter - some consider it question-dependent (healthcare, criminal justice, military defense, etc - high-stakes decision making fields), some think that it's a worthy sacrifice for improved accuracy. Furthermore, there's the question of when a process reaches the point of being a black box, as there's no clear threshold. A deep learning algorithm with millions of parameters is more likely to be viewd as a black box, for instance.

But a ridge regression with a Bayesian prior? I'm not convinced that it qualifies. Again, that's a personal decision and other are certainly free to disagree and prefer other metrics. But the reason I feel this way is that you can go to the page with LEBRON data and see each player's offensive and defensive box priors. To quote the authors, "This data should not be used for comparisons and is less accurate than LEBRON data, but can be used to get a sense for if a player’s impact is coming from their box contributions or on/off data." I don't personally look at these metrics and think of the black box issue.

The issue is what you use for the prior.

When you use the box score as a prior, then you bias the stat towards production ("He's better in LEBRON and PER, that's impact and production both!"), and since production stats are readily available and part of what we use to do our analysis along with a stat like this, now you're double counting the production. Of course it's not so clean as actually "doubling" the production - if it were it would be easier to deal with. In reality, you can't just subtract some production factor, because it wasn't ever actually added in - in the arithmetic sense - the first place.

As an analyst, I can do a better job if the production and impact stats remain available separately. This is not to say I'm fundamentally against creating stats that mash them together, but I want to see what they look like separately too. And I'll say, I love that bball-index lists out stuff like that beyond LEBRON...I just wish they did more of it.

It is true that such metrics will be biased towards production as opposed to biased towards 0 (like vanilla RAPM). Which is why I prefer them - such incorporation of the prior improves the performance of the metric in predictive tests. The combination of both the RAPM component and the informed prior tends to reduce predictive error. The value of that fact is debatable.

And of course, it's fair to want to see what they look like separately - that's why it's nice that we still have single-season RAPM available, like at nbashotcharts.com.

[f4p]

and now you just sound like you are quoting chapter and verse from the harden-hater Bible. strip clubs? seriously. he might be fat now, but the guy was routinely at or near the top of the league in minutes while hard-carrying as much as anyone in the league. he appeared to be serious enough to do that.

I have to stop here. You're not going to learn anything if you insist on thinking of people from a "hater" lens.

You made the excuse of Harden being unable to take games off out of obligation to his team's needs as if that was something that was costing him in some way. Me bringing up that Harden wasn't behaving like a man who was concerned about the limitations of his body speaks directly to his mindset, and it doesn't match what you're trying to project into his brain as a motivation.

i said you sound like a harden-hater b/c it was starting to get in the psychology/narrative stuff and the typical "harden loves strip clubs" territory. he also apparently plays basketball about as much as anyone off the court. apparently he wakes up, plays basketball, and goes to strip clubs, rinse and repeat. maybe it is biting him now but it seems hard to argue it did back then because he's not a physical freak and yet could handle a tremendous workload with almost no games off. maybe he could have been more focused on taking time off, but when did he realistically have a chance to do it (plus i'm not sure what the original point even was about with taking time off, i think it was about SRS somehow)?

2018 would be the only season the rockets had any sort of reasonably assured seeding and he missed a career high 10 games anyway. 2015 we got the 2 seed by one game and otherwise would have been 6th i think, 2016 made the playoffs (stupidly b/c we lost a pick) by 1 game, 2019 we missed the 2 seed (and dodging the warriors in the 2nd round) by 1 game because paul george made a buzzer beating 3. 2020 i believe we got homecourt by 1 game.

it's not like we could just be sure that dwight howard and chris paul would turn it up so much in the playoffs like a team like the warriors could think about their loaded lineup, at least not to the point that our seeding was inconsequential. and we didn't play in the east with a lebron level player who could make seeding not matter.