Billy Donovan gets contract extension

[2weekswithpay]

I'm fairly certain RAPM utilizes play-by-play data, which became available in 1997. For Jordan, only 4 seasons are included, his last two seasons as a Bull and two seasons with the Wizards. I think he grades out well despite having his prime excluded.

I don't think Chris Paul is the 4th best player since 1997, but Paul has made every single team he's played on better, so I'm not surprised he's regarded so highly by RAPM.

If the team performs better with a player on the court, RAPM will have a favorable view of them. I find this information valuable, even if there are outliers like Nene.

Watch on YouTube

[sco]

I'm fairly certain RAPM utilizes play-by-play data, which became available in 1997. For Jordan, only 4 seasons are included, his last two seasons as a Bull and two seasons with the Wizards. I think he grades out well despite having his prime excluded.

I don't think Chris Paul is the 4th best player since 1997, but Paul has made every single team he's played on better, so I'm not surprised he's regarded so highly by RAPM.

If the team performs better with a player on the court, RAPM will have a favorable view of them. I find this information valuable, even if there are outliers like Nene.

Watch on YouTube

How does RAPM account for the quality of the guy's backup?

[dougthonus]

How does RAPM account for the quality of the guy's backup?

It tries to use lineup data and regress for both quality of your teammates and opponents. Like anything where you are taking this data and regressing out stuff to try to make it more balanced, it will have flaws. I'd imagine varying backup quality is one of those things that's tricky to regress out well.

But that's why all these +/- driven stats are incredibly noisy. You are starting with correlation data to begin with, and you don't have large enough sample sizes before it changes radically.

[meekrab]

I don't think Chris Paul is the 4th best player since 1997, but Paul has made every single team he's played on better, so I'm not surprised he's regarded so highly by RAPM.

He's definitely one of the highest regular-season impact players of the past 20 years, he always gave 100% effort. His impact decreases in the playoffs because he runs up against bigger more athletic PGs also giving 100%.

[kodo]

The rest of the top RAPM players meets the eye test: Lebron, KG, Duncan, Steph, Kawhi, Dirk, Shaq, etc..
If the same process also puts CP3 in the mix, it's probably worth considering perceptions are wrong and CP3 is actually quite good.

The alternative is to say all those other players listed are also not that great.

[SalmonsSuperfan]

I'm fairly certain RAPM utilizes play-by-play data, which became available in 1997. For Jordan, only 4 seasons are included, his last two seasons as a Bull and two seasons with the Wizards. I think he grades out well despite having his prime excluded.

I don't think Chris Paul is the 4th best player since 1997, but Paul has made every single team he's played on better, so I'm not surprised he's regarded so highly by RAPM.

If the team performs better with a player on the court, RAPM will have a favorable view of them. I find this information valuable, even if there are outliers like Nene.

Watch on YouTube

Thanks, didn't realize it began in 1997. I'd just wonder if it tells you anything novel compared to traditional box score stats. Or reveals underlying patterns. I think it's a good attempt to capture the importance of coaching; but to me, Thibs is the best coach I've ever watched and impacts his teams the most, and my intuition suggests that Donovan is not a better coach than Thibs. That's an 'eye test' sort of opinion, not a statistical one. So I'd wanna question the methodology to see if it's actually revealing anything interesting. Or if you could just look at traditional +/- and maybe it says the same thing as the 'advanced stat' does.
Some of the 'spatial stats' I work with reveal interesting patterns (might call it 'hot spot analysis') but they aren't necessarily the truth and need to be investigated more rigorously using other methods. Such as "the eye test" lol. As an example, is the West Side of Chicago a very ecologically healthy region compared to, I dunno, Lake View or the Loop, or are there a lot of vacant lots that make the dataset for the region seem more 'green' than it actually is? You can walk around Garfield Park and see that it's not necessarily the case. Whereas the Loop compares pretty favorably to other big city downtowns. You can add in other data to the model, such as temperature or an index of industrial activity, of air quality, and it makes the model seemingly more effective, but then you worry about overfitting and if you're just using statistics to confirm your own "eye test" biases.

Putting all of the box score stats into one metric is a pretty good idea to me, but I don't think the NBA has figured it out yet, at least not publicly. I think WAR is the best one in sports, but I also think it overrates shortstops, second basemen and center fielders in a similar way that PER overrates big men who grab a lot of rebounds and put up high shooting percentages close to the basket. QBR seems like a good one, but I don't really know anything about football.

I don't mean to say it isn't valuable, just that it needs work in a way that I don't really understand because there isn't a whole body of research on NBA stats -- I just think the x,y coordinate data that exists now (Courtvision I think it's called) would probably make these stats a whole lot more accurate.
And it also might be impossible in basketball, so much is happening on the court at once, which I think is a lot of the fun about the sport. The boomer who hate on analytics might actually be statistically correct, whether he knows it or not!

https://basetunnel.substack.com/p/the-boomers-were-right

I thought this was a pretty fun article to read about batting averages in baseball and it sorta validates my belief that all this stuff 'Fangraphs' is doing is kinda pseudo. Maybe Moneyball wasn't on to anything interesting.

[dougthonus]

The rest of the top RAPM players meets the eye test: Lebron, KG, Duncan, Steph, Kawhi, Dirk, Shaq, etc..
If the same process also puts CP3 in the mix, it's probably worth considering perceptions are wrong and CP3 is actually quite good.

The alternative is to say all those other players listed are also not that great.

You could find a ton of statistical models that tell you who the greatest players are. Figuring out the greatest players is super duper trivially easy.

+/- stats are exceptionally noisy and effectively do not measure what people think they measure. People use them as a "this player is this good" stat, but really it tells a story about the impact of what happened when a player was on the floor in the context of a particular team he was on in a given season.

There's also a ton of them, none are transparent about their formulas, and they all have wildly different numbers.

[2weekswithpay]

I'm fairly certain RAPM utilizes play-by-play data, which became available in 1997. For Jordan, only 4 seasons are included, his last two seasons as a Bull and two seasons with the Wizards. I think he grades out well despite having his prime excluded.

I don't think Chris Paul is the 4th best player since 1997, but Paul has made every single team he's played on better, so I'm not surprised he's regarded so highly by RAPM.

If the team performs better with a player on the court, RAPM will have a favorable view of them. I find this information valuable, even if there are outliers like Nene.

Watch on YouTube

Thanks, didn't realize it began in 1997. I'd just wonder if it tells you anything novel compared to traditional box score stats. Or reveals underlying patterns. I think it's a good attempt to capture the importance of coaching; but to me, Thibs is the best coach I've ever watched and impacts his teams the most, and my intuition suggests that Donovan is not a better coach than Thibs. That's an 'eye test' sort of opinion, not a statistical one. So I'd wanna question the methodology to see if it's actually revealing anything interesting. Or if you could just look at traditional +/- and maybe it says the same thing as the 'advanced stat' does.
Some of the 'spatial stats' I work with reveal interesting patterns (might call it 'hot spot analysis') but they aren't necessarily the truth and need to be investigated more rigorously using other methods. Such as "the eye test" lol. As an example, is the West Side of Chicago a very ecologically healthy region compared to, I dunno, Lake View or the Loop, or are there a lot of vacant lots that make the dataset for the region seem more 'green' than it actually is? You can walk around Garfield Park and see that it's not necessarily the case. Whereas the Loop compares pretty favorably to other big city downtowns. You can add in other data to the model, such as temperature or an index of industrial activity, of air quality, and it makes the model seemingly more effective, but then you worry about overfitting and if you're just using statistics to confirm your own "eye test" biases.

Putting all of the box score stats into one metric is a pretty good idea to me, but I don't think the NBA has figured it out yet, at least not publicly. I think WAR is the best one in sports, but I also think it overrates shortstops, second basemen and center fielders in a similar way that PER overrates big men who grab a lot of rebounds and put up high shooting percentages close to the basket. QBR seems like a good one, but I don't really know anything about football.

I don't mean to say it isn't valuable, just that it needs work in a way that I don't really understand because there isn't a whole body of research on NBA stats -- I just think the x,y coordinate data that exists now (Courtvision I think it's called) would probably make these stats a whole lot more accurate.
And it also might be impossible in basketball, so much is happening on the court at once, which I think is a lot of the fun about the sport. The boomer who hate on analytics might actually be statistically correct, whether he knows it or not!

https://basetunnel.substack.com/p/the-boomers-were-right

I thought this was a pretty fun article to read about batting averages in baseball and it sorta validates my belief that all this stuff 'Fangraphs' is doing is kinda pseudo. Maybe Moneyball wasn't on to anything interesting.

I suppose the novelty lies in the fact that traditional statistics aren't always reflective of a player's impact. Certain players are undervalued if you judge them by boxscore stats, Draymond being a good example. The eye test tends to be too unreliable IMO. Nothing against the eye test, but opinions and biases cloud things too much. I don't doubt actual film study, though.

I agree that Thibs is a better coach, but I focus less on the rankings and more on why RAPM thinks highly of Donovan. It's hard to separate coaching from results, and Donovan being the coach of a 40 win team consistently may skew people's view of him. RAPM likes Billy so much because it doesn't think highly of the players on the Bulls, so Donovan gets credit for the team overperforming.

This method isn't perfect, as there's no scheme or lineup analysis involved, but coaching is hard to evaluate, and it does provide interesting information even if the stat is unreliable.