Will we ever get to a point we can use analytics to judge players relatively accurately?
Moderators: trex_8063, PaulieWal, Doctor MJ, Clyde Frazier, penbeast0
Will we ever get to a point we can use analytics to judge players relatively accurately?
- Smoothbutta
- Freshman
- Posts: 61
- And1: 10
- Joined: Jan 31, 2011
- Location: California
Will we ever get to a point we can use analytics to judge players relatively accurately?
At this point in the databall era, can we use some combination of VORP, WS/48, and RAPM/PIPM or some other metrics to evaluate players not perfectly but to some reasonable extent? If not now, will we be able to at some point in the next decade?
Also why are there no great sources for RAPM or PIPM stats that are both updated and go back to 1996? Please share if they are somewhere reliable.
Also why are there no great sources for RAPM or PIPM stats that are both updated and go back to 1996? Please share if they are somewhere reliable.
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
-
- Lead Assistant
- Posts: 5,388
- And1: 4,969
- Joined: Mar 28, 2020
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
I don't think we'll ever get to a point where we have analytics that paint a 100% accurate picture but I do believe that using an aggregate of many different stats can already evaluate players to a reasonable extent. Personally I use composite stats like EPM and LEBRON the most and consult boxscore stats (BPM and WS), +- and on-off to look at the more isolated parts.
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
-
- RealGM
- Posts: 13,946
- And1: 10,666
- Joined: Jun 13, 2017
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
I think we got to that point 5-10 years ago tbh. It's just that you have to use a combination of them and some of the better ones are likely kept under wraps by teams and data only goes so far back. It's subjective though because any data people don't think meets their own idea of how good a player is tends to be rejected. That's human nature when it comes to which analytics people tend to use more. How do we know when we've found the perfect advanced metric? Do bells go off in the heavens or something? There's no way of knowing.
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
- Smoothbutta
- Freshman
- Posts: 61
- And1: 10
- Joined: Jan 31, 2011
- Location: California
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
Cavsfansince84 wrote:I think we got to that point 5-10 years ago tbh. It's just that you have to use a combination of them and some of the better ones are likely kept under wraps by teams and data only goes so far back. It's subjective though because any data people don't think meets their own idea of how good a player is tends to be rejected. That's human nature when it comes to which analytics people tend to use more. How do we know when we've found the perfect advanced metric? Do bells go off in the heavens or something? There's no way of knowing.
Yea but like, where is that data then? Is there any website that is still active that gives data like EPM/RAPM/LEBRON since 1996 and is available for us to use?
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
-
- Senior Mod - NBA Player Comparisons
- Posts: 29,038
- And1: 9,075
- Joined: Aug 14, 2004
- Location: South Florida
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
What do we mean by "relatively accurately?" If you mean better than by people's individual eye test, we reached that point on offense a long long time ago, though not on defense. Tracking data and the like make our defensive evaluations a lot better too. If you are looking at factors like locker room compatability, heart, ability to rise in the clutch, and even BBIQ, we have a tough time even figuring out what we want to measure before we even get to the issue of can we measure it. If you mean "perfectly," it will never happen.
“Most people use statistics like a drunk man uses a lamppost; more for support than illumination,” Andrew Lang.
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
-
- Lead Assistant
- Posts: 5,037
- And1: 3,303
- Joined: Jun 22, 2022
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
Smoothbutta wrote:At this point in the databall era, can we use some combination of VORP, WS/48, and RAPM/PIPM or some other metrics to evaluate players not perfectly but to some reasonable extent? If not now, will we be able to at some point in the next decade?
Also why are there no great sources for RAPM or PIPM stats that are both updated and go back to 1996? Please share if they are somewhere reliable.
Depends on what specifically you're looking at and your criteria.
Will just link my approach to player eval(maybe a little out of date now, but includes all the various stats there):
https://forums.realgm.com/boards/viewtopic.php?f=64&t=2248282&start=300
Spoiler:
Some additional contextual considerations(and time machnine/direct player comparisons:
Spoiler:
[/spoiler]
Stat Sources:
-> Lineup-ratings -> PBP
-> WOWY/indirect/raw singlas -> Statmuse, Backpicks
-> RAPM (there are various, but the one that seems to be the most transparent is amhed cheema's)
-> Hybrids (you can find them on their given website)
-> On/off -> BBR
-> Box-stats -> depends on the stat, nba.com, bbr, synergy, bball-index, and of course there's sometimes value in creating your own while tracking film (lots of tracking has already been done here by other posters)
Generally I'd say "advanced stats" like LEBRON or whatever should be the last steps in most comparisons.
its my last message in this thread, but I just admit, that all the people, casual and analytical minds, more or less have consencus who has the weight of a rubberized duck. And its not JaivLLLL
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
- Smoothbutta
- Freshman
- Posts: 61
- And1: 10
- Joined: Jan 31, 2011
- Location: California
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
OhayoKD wrote:Stat Sources:
-> Lineup-ratings -> PBP
-> WOWY/indirect/raw singlas -> Statmuse, Backpicks
-> RAPM (there are various, but the one that seems to be the most transparent is amhed cheema's)
-> Hybrids (you can find them on their given website)
-> On/off -> BBR
-> Box-stats -> depends on the stat, nba.com, bbr, synergy, bball-index, and of course there's sometimes value in creating your own while tracking film (lots of tracking has already been done here by other posters)
Generally I'd say "advanced stats" like LEBRON or whatever should be the last steps in most comparisons.
Do you have a link to WOWY data on statmuse?
Do you have a link to Ahmed Cheema's RAPM data? I see some public charts that go to 2021 but nothing past that.
I never noticed the on/off tab on BBR, thank you
What are hybrids?
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
-
- Senior
- Posts: 733
- And1: 598
- Joined: May 21, 2022
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
Think its very difficult without stats based on player tracking and some sort of input for schemes and role. Too many variables compared to something like baseball, especially defensively.
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
-
- Starter
- Posts: 2,194
- And1: 2,196
- Joined: Jul 01, 2022
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
I second my guy capfan33 here
Mogspan wrote:I think they see the super rare combo of high IQ with freakish athleticism and overrate the former a bit, kind of like a hot girl who is rather articulate being thought of as “super smart.” I don’t know kind of a weird analogy, but you catch my drift.
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
-
- Analyst
- Posts: 3,170
- And1: 2,868
- Joined: Dec 25, 2019
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
We can.
Understanding these metrics without watching the game would give you better ability to estimate player impact than 99% of NBA fandom.
Understanding these metrics without watching the game would give you better ability to estimate player impact than 99% of NBA fandom.
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
-
- Lead Assistant
- Posts: 5,037
- And1: 3,303
- Joined: Jun 22, 2022
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
capfan33 wrote:Think its very difficult without stats based on player tracking and some sort of input for schemes and role. Too many variables compared to something like baseball, especially defensively.
Depends on your bar for "accurate"
its my last message in this thread, but I just admit, that all the people, casual and analytical minds, more or less have consencus who has the weight of a rubberized duck. And its not JaivLLLL
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
-
- Lead Assistant
- Posts: 5,037
- And1: 3,303
- Joined: Jun 22, 2022
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
Smoothbutta wrote:OhayoKD wrote:Stat Sources:
-> Lineup-ratings -> PBP
-> WOWY/indirect/raw singlas -> Statmuse, Backpicks
-> RAPM (there are various, but the one that seems to be the most transparent is amhed cheema's)
-> Hybrids (you can find them on their given website)
-> On/off -> BBR
-> Box-stats -> depends on the stat, nba.com, bbr, synergy, bball-index, and of course there's sometimes value in creating your own while tracking film (lots of tracking has already been done here by other posters)
Generally I'd say "advanced stats" like LEBRON or whatever should be the last steps in most comparisons.
Do you have a link to WOWY data on statmuse?
Do you have a link to Ahmed Cheema's RAPM data? I see some public charts that go to 2021 but nothing past that.
I never noticed the on/off tab on BBR, thank you
What are hybrids?
You go to statmuse.com and enter queries like "2023 hawks net rating with and without trae young"
Some WOWY has been compiled already actually:
https://forums.realgm.com/boards/viewtopic.php?t=2353834
https://forums.realgm.com/boards/viewtopic.php?t=2310915
Ben taylor's backpicks top 40 usually features health-adjusted srs for concentrated stretches
You can also go to bbr and check season to season srs swings
Here is a breakdown of cheema's rapm with links to the original data and explanation for methodology:
viewtopic.php?t=2301003
The tab for on/off in bbr is "play by play"
and here's a 5-year variant:
https://www.thespax.com/nba/quantifying-the-nbas-greatest-five-year-peaks-since-1997/
its my last message in this thread, but I just admit, that all the people, casual and analytical minds, more or less have consencus who has the weight of a rubberized duck. And its not JaivLLLL
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
- Smoothbutta
- Freshman
- Posts: 61
- And1: 10
- Joined: Jan 31, 2011
- Location: California
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
Yea unfortunately Cheema's RAPM data only goes till 2021. So I still haven't seen any complete data-set of RAPM from 1996 to 2024 unfortunately, I'm not complaining btw just trying to understand what is and what isn't available.
I started compiling some players single season peak On-Off and career On-Off from BBR and it's interesting that Draymond's from 2016 is higher than every single other 2000s and onward player in the top 50 or 60 all-time. I understand sample size and taking one data-set with a grain of salt but that is still really interesting.
I started compiling some players single season peak On-Off and career On-Off from BBR and it's interesting that Draymond's from 2016 is higher than every single other 2000s and onward player in the top 50 or 60 all-time. I understand sample size and taking one data-set with a grain of salt but that is still really interesting.
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
-
- Lead Assistant
- Posts: 5,037
- And1: 3,303
- Joined: Jun 22, 2022
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
Smoothbutta wrote:Yea unfortunately Cheema's RAPM data only goes till 2021. So I still haven't seen any complete data-set of RAPM from 1996 to 2024 unfortunately, I'm not complaining btw just trying to understand what is and what isn't available.
I started compiling some players single season peak On-Off and career On-Off from BBR and it's interesting that Draymond's from 2016 is higher than every single other 2000s and onward player in the top 50 or 60 all-time. I understand sample size and taking one data-set with a grain of salt but that is still really interesting.
some things to keep in mind with on/off and it's derivatives(this includes rapm)
-> lineup effects:
a. co-linearity, when the stars of a team play heavy minutes together, on/off tends to go higher. RAPM can mitigate this but it's still just an approximation (curry and draymond, jordan and pippen, 2024 jokic)
b. staggering, when the stars of a team play substantial minutes separate from each other, the on/off tends to go lower as the off is inflated and the on is suppressed (ex: 2024 Curry with cp3, luka until this trade deadline, Lebron in Miami with Wade)
c. minutes load, players who play much higher minutes than everyone else on their team tend to see their numbers suppressed as they play with the backups too (ex: Duncan). Opposite effect with low minute loads (drob)
These can affect all stats, but on/off, largely based on spot minutes without a player, is super sensitive to this. WOWY minimizes this sort of bias but is relatively variable. RAPM is stable but approximates and can't be used year to year in the same way. Nonetheless, checking on/off against both when possible is a good idea.
It's also a good move to check lineup ratings(pbp) and/or rotation sheets
its my last message in this thread, but I just admit, that all the people, casual and analytical minds, more or less have consencus who has the weight of a rubberized duck. And its not JaivLLLL
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
-
- Senior Mod
- Posts: 51,578
- And1: 20,317
- Joined: Mar 10, 2005
- Location: Cali
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
Smoothbutta wrote:At this point in the databall era, can we use some combination of VORP, WS/48, and RAPM/PIPM or some other metrics to evaluate players not perfectly but to some reasonable extent? If not now, will we be able to at some point in the next decade?
Also why are there no great sources for RAPM or PIPM stats that are both updated and go back to 1996? Please share if they are somewhere reliable.
Well, I would argue that the fact that we call this the "databall" era is due to NBA teams using analytics to judge players relatively accurately.
So then it's really a question of what the specific threshold on your mind is.
For me, the gold standard would be a stat that is based on feature-extracting skills based on player tracking data. I feel like it's only a matter of time before NBA teams are doing this for particular skills (some will be easier than others), if they're not doing it already, but I'm not what along these lines will be available to the public.
In terms of the lack of reliable stats here, it's 2 issues:
1. People who publish these sort of stats tend to get hired to do similar work, often by NBA teams, and then they at best leave the site static, but sometimes take down the site.
2. There are no objective standards for these stats. When you make a RAPM, there are decisions you have to make with no clear right answers, and so when different people run RAPM, they get different results. This has a number of negative consequences for wide spread adoption as you might imagine. In practice what this means is that when a new source of RAPM comes out, the community evaluates what they see with a skeptical eye. If they see some wacko results they won't allege fraud or anything like that, but also won't tend to keep using that source.
Getting ready for the RealGM 100 on the PC Board
Come join the WNBA Board if you're a fan!
Come join the WNBA Board if you're a fan!
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
-
- Lead Assistant
- Posts: 5,037
- And1: 3,303
- Joined: Jun 22, 2022
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
Doctor MJ wrote:Smoothbutta wrote:At this point in the databall era, can we use some combination of VORP, WS/48, and RAPM/PIPM or some other metrics to evaluate players not perfectly but to some reasonable extent? If not now, will we be able to at some point in the next decade?
Also why are there no great sources for RAPM or PIPM stats that are both updated and go back to 1996? Please share if they are somewhere reliable.
2. There are no objective standards for these stats. When you make a RAPM, there are decisions you have to make with no clear right answers, and so when different people run RAPM, they get different results. This has a number of negative consequences for wide spread adoption as you might imagine. In practice what this means is that when a new source of RAPM comes out, the community evaluates what they see with a skeptical eye. If they see some wacko results they won't allege fraud or anything like that, but also won't tend to keep using that source.
It's lightly damning that cheema is the only rapm source(setting aside scaled one-year from cryptbeam) that is transparent on methodology and formulas
its my last message in this thread, but I just admit, that all the people, casual and analytical minds, more or less have consencus who has the weight of a rubberized duck. And its not JaivLLLL
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
- andyhop
- Analyst
- Posts: 3,615
- And1: 1,309
- Joined: May 08, 2007
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
Doctor MJ wrote:
For me, the gold standard would be a stat that is based on feature-extracting skills based on player tracking data. I feel like it's only a matter of time before NBA teams are doing this for particular skills (some will be easier than others), if they're not doing it already, but I'm not what along these lines will be available to the public.
I wonder if anyone is modelling basketball like Liverpool model football.
Instead, he spent months building a model that calculates the chance each team had of scoring a goal before any given action – a pass, a missed shot, a slide tackle – and then what chance it had immediately after that action. Using his model, he can quantify how much each player affected his team’s chance of winning during the game.
https://www.afr.com/companies/sport/liverpool-show-moneyball-works-in-soccer-too-20190523-p51qlc ( sorry it's paywalled but very interesting if you can read it).
Seems like a similar approach would be possible for basketball as well. Liverpool use the data in their recruitment strategy , combining it with other information and the financial cost of acquiring the player to try and get the best bang for their buck.
"Football is not a matter of life and death...it's much more important than that."- Bill Shankley
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
-
- Sixth Man
- Posts: 1,972
- And1: 1,683
- Joined: Apr 13, 2013
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
My answer to the question of the thread is basically no, because of a combination of inherently limited sample sizes and the fact that players change in quality relatively rapidly over the years.
Overall, we want to get at impact, since there’s lots of little things on the court that matter but aren’t really specifically quantifiable. IMO, the best we can do basically is to take RAPM and layer on other data as a prior. That data can be box data that is weighted in a way that correlates with RAPM. Now that we have tracking data, it’d be even better to also include tracking data there as well. So basically, you take RAPM with a prior that uses box and tracking data that correlates well with RAPM. That will be a lot less noisy than raw RAPM (which is itself less noisy and more accurate than raw on-off). But even that sort of data is inherently inaccurate. Over relatively small samples (say, for instance, a single season), you’re still going to get a lot of noise because, even with a good prior, the RAPM itself is still super noisy in that sample. Using a prior reduces the noise, but you definitely don’t at all eliminate it. Theoretically, one could deal with that by doing that sort of analysis over a larger time horizon. So, for instance, what if we used RAPM with the same kind of prior, but did it over a 5-year time period, instead of a 1-year time period? That’d be less noisy, because 5-year RAPM is a lot less noisy than 1-year RAPM. But then you run into the problem that players are not the same players over those long periods of time—and that’s true both for the people whose data output you’re looking at but also the teammates and opponents that the RAPM model is aiming to correct for. So you lower the statistical noise but in exchange you get a bunch of substantive imprecision. And that’s not even mentioning the obvious fact that that’s just not an option at all if what you’re specifically wanting to do is compare individual seasons, nor is it mentioning specific methodological flaws or blind spots in a model.
Basically, the problem is that the data we are interested in and the rate at which players change over the years means that we are left with noisy samples and/or have to handwave serious changes in player quality over a larger sample. Both lead to inaccuracy.
That is then exacerbated massively by the playoffs. Playoffs matter more than the regular season, but the samples are even smaller. There’s essentially no way to get a genuinely adequate playoff sample for virtually any player, and there’s certainly no way to do it without pulling from so many years that you’re *definitely* handwaving away player changes over the years. So it’s virtually impossible to get genuinely accurate playoff impact data. And yet as fans we have a general sense that playoffs are what shows what a player is really made of, and that some players are genuinely better or worse in the playoffs than in the regular season. We basically can’t accurately measure any of that though.
So yeah, my answer is generally no. I think the best middle ground on all this is probably to take something like 3-year RAPM with a good box/tracking-data prior that correlates well with large-sample RAPM, and include both regular-season and playoff data but give playoff data some additional weight. I’m not aware of anything that does exactly that, but it’s theoretically possible and would probably be something I’d like. It would still have the types of flaws I identified above, though, and inevitably also some other methodological flaws or blind spots (for instance, no prior is perfect, so it’ll always unduly favor or penalize certain players).
We do actually have RAPM data that goes through 2024:
https://docs.google.com/spreadsheets/d/1bg8KxzagN7D0O16EmUO9_kCyXwthEUjKywlrWPQUQt8/edit#gid=0
Overall, we want to get at impact, since there’s lots of little things on the court that matter but aren’t really specifically quantifiable. IMO, the best we can do basically is to take RAPM and layer on other data as a prior. That data can be box data that is weighted in a way that correlates with RAPM. Now that we have tracking data, it’d be even better to also include tracking data there as well. So basically, you take RAPM with a prior that uses box and tracking data that correlates well with RAPM. That will be a lot less noisy than raw RAPM (which is itself less noisy and more accurate than raw on-off). But even that sort of data is inherently inaccurate. Over relatively small samples (say, for instance, a single season), you’re still going to get a lot of noise because, even with a good prior, the RAPM itself is still super noisy in that sample. Using a prior reduces the noise, but you definitely don’t at all eliminate it. Theoretically, one could deal with that by doing that sort of analysis over a larger time horizon. So, for instance, what if we used RAPM with the same kind of prior, but did it over a 5-year time period, instead of a 1-year time period? That’d be less noisy, because 5-year RAPM is a lot less noisy than 1-year RAPM. But then you run into the problem that players are not the same players over those long periods of time—and that’s true both for the people whose data output you’re looking at but also the teammates and opponents that the RAPM model is aiming to correct for. So you lower the statistical noise but in exchange you get a bunch of substantive imprecision. And that’s not even mentioning the obvious fact that that’s just not an option at all if what you’re specifically wanting to do is compare individual seasons, nor is it mentioning specific methodological flaws or blind spots in a model.
Basically, the problem is that the data we are interested in and the rate at which players change over the years means that we are left with noisy samples and/or have to handwave serious changes in player quality over a larger sample. Both lead to inaccuracy.
That is then exacerbated massively by the playoffs. Playoffs matter more than the regular season, but the samples are even smaller. There’s essentially no way to get a genuinely adequate playoff sample for virtually any player, and there’s certainly no way to do it without pulling from so many years that you’re *definitely* handwaving away player changes over the years. So it’s virtually impossible to get genuinely accurate playoff impact data. And yet as fans we have a general sense that playoffs are what shows what a player is really made of, and that some players are genuinely better or worse in the playoffs than in the regular season. We basically can’t accurately measure any of that though.
So yeah, my answer is generally no. I think the best middle ground on all this is probably to take something like 3-year RAPM with a good box/tracking-data prior that correlates well with large-sample RAPM, and include both regular-season and playoff data but give playoff data some additional weight. I’m not aware of anything that does exactly that, but it’s theoretically possible and would probably be something I’d like. It would still have the types of flaws I identified above, though, and inevitably also some other methodological flaws or blind spots (for instance, no prior is perfect, so it’ll always unduly favor or penalize certain players).
Smoothbutta wrote:Yea unfortunately Cheema's RAPM data only goes till 2021. So I still haven't seen any complete data-set of RAPM from 1996 to 2024 unfortunately, I'm not complaining btw just trying to understand what is and what isn't available.
I started compiling some players single season peak On-Off and career On-Off from BBR and it's interesting that Draymond's from 2016 is higher than every single other 2000s and onward player in the top 50 or 60 all-time. I understand sample size and taking one data-set with a grain of salt but that is still really interesting.
We do actually have RAPM data that goes through 2024:
https://docs.google.com/spreadsheets/d/1bg8KxzagN7D0O16EmUO9_kCyXwthEUjKywlrWPQUQt8/edit#gid=0
OhayoKD wrote:Lebron contributes more to all the phases of play than Messi does. And he is of course a defensive anchor unlike messi.
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
-
- Sixth Man
- Posts: 1,972
- And1: 1,683
- Joined: Apr 13, 2013
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
andyhop wrote:Doctor MJ wrote:
For me, the gold standard would be a stat that is based on feature-extracting skills based on player tracking data. I feel like it's only a matter of time before NBA teams are doing this for particular skills (some will be easier than others), if they're not doing it already, but I'm not what along these lines will be available to the public.
I wonder if anyone is modelling basketball like Liverpool model football.Instead, he spent months building a model that calculates the chance each team had of scoring a goal before any given action – a pass, a missed shot, a slide tackle – and then what chance it had immediately after that action. Using his model, he can quantify how much each player affected his team’s chance of winning during the game.
https://www.afr.com/companies/sport/liverpool-show-moneyball-works-in-soccer-too-20190523-p51qlc ( sorry it's paywalled but very interesting if you can read it).
Seems like a similar approach would be possible for basketball as well. Liverpool use the data in their recruitment strategy , combining it with other information and the financial cost of acquiring the player to try and get the best bang for their buck.
It’s interesting to think about that sort of approach in basketball. Like, anytime a player gets the ball, you’d measure the expected points scored in the possession (based on player positions, shot clock situation, etc.) and then you subtract that from the expected (or actual) points after they score, pass, or turn the ball over. Similar to how this sort of analysis works in soccer, that’d give an expected value of a player’s on-ball actions.
I think there’s a couple issues with that though:
1. It doesn’t measure off-ball action. That’s a flaw in the soccer version of these sorts of stats as well.
2. It’s not really a type of model that could meaningfully measure most any individual defense. Again, that’s a flaw in the soccer version too, but is perhaps less important there because player roles are more siloed and so you have lots of attacking players in soccer for whom their defensive actions are much less important than any basketball player’s defense. In other words, the thing this sort of model misses doesn’t always matter all that much in soccer, but it would always matter a good bit in basketball.
In general, I actually think impact data in basketball is actually better than this. This sort of analysis is a bottom-up approach that tries to take discrete player actions and model out their value. But it still inherently can’t get at everything a player does. The beauty of impact data is that it is a top-down approach that is inherently looking to measure everything a player does. This is because it is not limited to valuing discrete player actions, but rather is asking what happens when they’re playing compared to when they aren’t, and what happens when they’re playing is affected by *everything* they do rather than just specific discrete actions that a model might look at. The reason you can’t really do this analysis in soccer is basically because players aren’t subbed out very much, so there’s just not much of an “off” sample for most players, which means that you can’t meaningfully do impact data. Without impact data, you’re kind of just stuck doing a bottom-up approach and trying to be as rigorous as possible about how you are valuing player actions. It’s probably the best that can be done in soccer, but I think we have basketball approaches that are better.
OhayoKD wrote:Lebron contributes more to all the phases of play than Messi does. And he is of course a defensive anchor unlike messi.
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
-
- Sixth Man
- Posts: 1,885
- And1: 1,352
- Joined: Sep 23, 2023
Re: Will we ever get to a point we can use analytics to judge players relatively accurately?
I think we are basically at that point now. You have to know which ones to use and how to weigh them, but you do get most of the picture with just advanced all in one stats