Using Westbrook lineups to determine rotation

[danfantastk32]

No no no....don't be fooled guys. We just need to dump this guy, and move on. This was a failure.

Yeah, maybe everyone's resigned to the season being what it is...and so Lebron's getting his stats, and so is Westbrook. Sure, Westbrook is prob looking alot better playing HIS game. But it's a loser in the long run.

I will take heart though, that if this sudden surge is tempting you guys, then perhaps we can find a team willing to kick the tires.

[aaron_gray]

I’d be curious to see how he does if the ball is completely taken out of his hands and he’s asked to set screens and make plays on the short roll when playing alongside Lebron.

Then let him run the 2nd unit with Dwight and three floor spacers like Melo, Monk and Ariza. We’ll lose the defensive matchup but if that line up can at least get a +3 against opposite beach units, that’s good enough.

I feel like this has been brought up as a way to maximize Westbrook's potential his entire career lol. I don't know which of the good analysts left are still on this train of possibility (Zach Lowe used to say this all the time).

[aaron_gray]

The number of variables and the data available make it impossible to do a on/off court evaluation on a player by player basis. Individual and team matchups overlaid with game situation metrics and on and on are needed to get a trend going.

I'm in the data analytics industry and a few friends and I several years ago got access to some data sets of pro sports and did a variable map and tried to find a meaningful query path to usable data. It's not there.

Most of what passes as advanced stats are intended to increase clicks and support the "fantasy industry"other than some individual metrics which are marginally useful IMHO. When you base them on flawed metrics (individual assists, rebounds, steal, blocks etc.) you get flawed results.

When you say pro sports, did you include big 4 american sports, or just pro basketball overseas?

Have you looked at in conjunction with other advanced stats? From an uneducated fan perspective, eyeballing adjusted on/off splits with other stats like TS% and win shares gives you a pretty good idea of where the contributions are.

[Landsberger]

The number of variables and the data available make it impossible to do a on/off court evaluation on a player by player basis. Individual and team matchups overlaid with game situation metrics and on and on are needed to get a trend going.

I'm in the data analytics industry and a few friends and I several years ago got access to some data sets of pro sports and did a variable map and tried to find a meaningful query path to usable data. It's not there.

Most of what passes as advanced stats are intended to increase clicks and support the "fantasy industry"other than some individual metrics which are marginally useful IMHO. When you base them on flawed metrics (individual assists, rebounds, steal, blocks etc.) you get flawed results.

When you say pro sports, did you include big 4 american sports, or just pro basketball overseas?

Have you looked at in conjunction with other advanced stats? From an uneducated fan perspective, eyeballing adjusted on/off splits with other stats like TS% and win shares gives you a pretty good idea of where the contributions are.

The underlying stats that those are created from are flawed, so by definition anything derived from them will be as well. Assists, individual rebounds, steals and blocked shots are all flawed stats when viewed either individually or collectively. For instance, the annual assist leader in the NBA is almost never getting to the finals and even more rare is winning a chip. Magic is one of the few exceptions. Rebounding by a team if infinitely more important than individual rebounding in the abstract. If the team is - 4 in team rebounding and an individual is second in the league it's of little importance. On the flip side if the starting 5 all have 5 boards a game averages while the team is leading the league in rebounding it's similarly not that important. Blocked shots are nothing more than touching an opponents shot. Steals are a tally of how many time you gamble and win.... not a percentage (which would be meaningful)..... and so on. The so called advanced stats all use these and other "tally" stats as the basis for the statistical extrapolations.

The dynamics of a team sport that could contribute to a true meaningful statistical analysis are infinite. From what the player ate for dinner the night before he got on the plane to the rotations to how much sleep he got to if he's having off the court issues to a cut on his hand. Whether or not he makes a great pass 10 times in a game and he teammates only convert 3 of them into baskets is not reflected in the Assist stat or the resultant ±. The type of defense you play against, the guy guarding you, the rotations, the venue, the foul situation, the game situation and so on are never accounted for in these stats.

I've done data modeling, coding for large complex data bases and simulations for the defense industry, did a couple stints in the financial security analysis industry doing quant modeling and data base development and am in the blockchain development currently. I've actually run in to a few data modelers who've worked as outsourced labor for some of the gaming industries statistical groups. Most of the stuff produced is targeted at creating debate and interest in fantasy sports and/or gaming. That's the target of the product.

[aaron_gray]

The number of variables and the data available make it impossible to do a on/off court evaluation on a player by player basis. Individual and team matchups overlaid with game situation metrics and on and on are needed to get a trend going.

I'm in the data analytics industry and a few friends and I several years ago got access to some data sets of pro sports and did a variable map and tried to find a meaningful query path to usable data. It's not there.

Most of what passes as advanced stats are intended to increase clicks and support the "fantasy industry"other than some individual metrics which are marginally useful IMHO. When you base them on flawed metrics (individual assists, rebounds, steal, blocks etc.) you get flawed results.

When you say pro sports, did you include big 4 american sports, or just pro basketball overseas?

Have you looked at in conjunction with other advanced stats? From an uneducated fan perspective, eyeballing adjusted on/off splits with other stats like TS% and win shares gives you a pretty good idea of where the contributions are.

The underlying stats that those are created from are flawed, so by definition anything derived from them will be as well. Assists, individual rebounds, steals and blocked shots are all flawed stats when viewed either individually or collectively. For instance, the annual assist leader in the NBA is almost never getting to the finals and even more rare is winning a chip. Magic is one of the few exceptions. Rebounding by a team if infinitely more important than individual rebounding in the abstract. If the team is - 4 in team rebounding and an individual is second in the league it's of little importance. On the flip side if the starting 5 all have 5 boards a game averages while the team is leading the league in rebounding it's similarly not that important. Blocked shots are nothing more than touching an opponents shot. Steals are a tally of how many time you gamble and win.... not a percentage (which would be meaningful)..... and so on. The so called advanced stats all use these and other "tally" stats as the basis for the statistical extrapolations.

The dynamics of a team sport that could contribute to a true meaningful statistical analysis are infinite. From what the player ate for dinner the night before he got on the plane to the rotations to how much sleep he got to if he's having off the court issues to a cut on his hand. Whether or not he makes a great pass 10 times in a game and he teammates only convert 3 of them into baskets is not reflected in the Assist stat or the resultant ±. The type of defense you play against, the guy guarding you, the rotations, the venue, the foul situation, the game situation and so on are never accounted for in these stats.

I've done data modeling, coding for large complex data bases and simulations for the defense industry, did a couple stints in the financial security analysis industry doing quant modeling and data base development and am in the blockchain development currently. I've actually run in to a few data modelers who've worked as outsourced labor for some of the gaming industries statistical groups. Most of the stuff produced is targeted at creating debate and interest in fantasy sports and/or gaming. That's the target of the product.

The examples you gave are almost all well known already, and accounted for with different advanced stats.

-Assist leaders have historically been the Rondo types that effectively make you play four on five. They get a lot of assists because they hog the ball and won't pass to you unless you shoot, killing all offensive movement just so they might get the assist. At the same time, just because the relationship isn't linear doesn't mean that a relationship doesn't exist. You can't deny that Lebron's ability to pass out of double teams and run the p/r effectively is captured in his assist numbers.

-On rebounding, this is a pretty well known fact that is typically emphasized with the Lopez brothers. While they don't often put up huge individual rebounding, their boxing out creates a massive team rebounding advantage. That's why people usually look at on/off impact on team rebounding as well as contested rebounds as opposed to the raw number someone collects (Westbrook is infamous for getting a bunch of uncontested rebounds to pad his triple doubles).

-Blocked shots are often gambles too - you gamble by going for the block instead of boxing out. That's why most people look at FG% allowed that's available on NBA.com, in conjunction with on/off rebounding. It is also very well known that steals leaders are frequently some of the worst defenders in the league (Westbrook, Wall, etc.).



-Also lastly, I can't speak for defense or blockchain, but quant modeling for finance is very different from roster construction analysis and is more suitable for sports gambling.

[Landsberger]

When you say pro sports, did you include big 4 american sports, or just pro basketball overseas?

Have you looked at in conjunction with other advanced stats? From an uneducated fan perspective, eyeballing adjusted on/off splits with other stats like TS% and win shares gives you a pretty good idea of where the contributions are.

The underlying stats that those are created from are flawed, so by definition anything derived from them will be as well. Assists, individual rebounds, steals and blocked shots are all flawed stats when viewed either individually or collectively. For instance, the annual assist leader in the NBA is almost never getting to the finals and even more rare is winning a chip. Magic is one of the few exceptions. Rebounding by a team if infinitely more important than individual rebounding in the abstract. If the team is - 4 in team rebounding and an individual is second in the league it's of little importance. On the flip side if the starting 5 all have 5 boards a game averages while the team is leading the league in rebounding it's similarly not that important. Blocked shots are nothing more than touching an opponents shot. Steals are a tally of how many time you gamble and win.... not a percentage (which would be meaningful)..... and so on. The so called advanced stats all use these and other "tally" stats as the basis for the statistical extrapolations.

The dynamics of a team sport that could contribute to a true meaningful statistical analysis are infinite. From what the player ate for dinner the night before he got on the plane to the rotations to how much sleep he got to if he's having off the court issues to a cut on his hand. Whether or not he makes a great pass 10 times in a game and he teammates only convert 3 of them into baskets is not reflected in the Assist stat or the resultant ±. The type of defense you play against, the guy guarding you, the rotations, the venue, the foul situation, the game situation and so on are never accounted for in these stats.

I've done data modeling, coding for large complex data bases and simulations for the defense industry, did a couple stints in the financial security analysis industry doing quant modeling and data base development and am in the blockchain development currently. I've actually run in to a few data modelers who've worked as outsourced labor for some of the gaming industries statistical groups. Most of the stuff produced is targeted at creating debate and interest in fantasy sports and/or gaming. That's the target of the product.

The examples you gave are almost all well known already, and accounted for with different advanced stats.

-Assist leaders have historically been the Rondo types that effectively make you play four on five. They get a lot of assists because they hog the ball and won't pass to you unless you shoot, killing all offensive movement just so they might get the assist. At the same time, just because the relationship isn't linear doesn't mean that a relationship doesn't exist. You can't deny that Lebron's ability to pass out of double teams and run the p/r effectively is captured in his assist numbers.

-On rebounding, this is a pretty well known fact that is typically emphasized with the Lopez brothers. While they don't often put up huge individual rebounding, their boxing out creates a massive team rebounding advantage. That's why people usually look at on/off impact on team rebounding as well as contested rebounds as opposed to the raw number someone collects (Westbrook is infamous for getting a bunch of uncontested rebounds to pad his triple doubles).

-Blocked shots are often gambles too - you gamble by going for the block instead of boxing out. That's why most people look at FG% allowed that's available on NBA.com, in conjunction with on/off rebounding. It is also very well known that steals leaders are frequently some of the worst defenders in the league (Westbrook, Wall, etc.).



-Also lastly, I can't speak for defense or blockchain, but quant modeling for finance is very different from roster construction analysis and is more suitable for sports gambling.

So, In short.... you're right I'm wrong because you can't speak to what I'd done.

You did a pretty good job of showing why you have to eye test any advance stats on the way to making my point.

I'm curious as to why financial quant modeling is seen as different. BTW... the quant models I worked on were a mix of investor psychology models overlayed with trading patterns by the largest holders of a security and finally an overlay of economic indicators vs fundamentals. It wasn't the stuff of day trading technical analysis... Their models were our target.

Modeling random events in a controlled system is pretty much the same when you identify the meaningful variables... and more importantly... the meaningless variables.

[aaron_gray]

Nah, your point was that basic stats are flawed because of all the examples you gave where there wasn't a perfect linear relationship to wins. My point was that in the examples you gave, there either was a positive relationship, or people just didn't interpret them in the simple manner you described. Thus, those stats are not actually "flawed", or interpreted incorrectly.


The thinking you need to predict an event is somewhat different from creating an event from a roster building standpoint. The best example I can give is with soccer, where there's no salary cap and wages are determined by your club's commercial value and your owners' wealth. Many gambling models (as well as 538's predictive model) incorporate either wages or player market values as predictive factors of performance, given that richer teams win more in a very linear way. However, as Liverpool's Ian Graham put it, if you simply paid the same players 2x what they make, they're not going to magically win 2x more games.

This is getting away from what we're discussing though. My point on quant models had more to do with how to build a team, not how to evaluate player performance.

[Landsberger]

The thinking you need to predict an event is somewhat different from creating an event from a roster building standpoint.

I see no difference. It's a system with random components like any other. "creating" a roster and predicting how it will perform are semantics unless all you are interested in is creating the roster.

A flawed individual stat is a variable that has little use in predictive modeling other than a control.

You reference sports betting. You need to understand the source of the "modeling" and the purpose. House still wins 85% of the time if you throw a dart or use modeling.