Announcement

Collapse
No announcement yet.

Expected Performance and The Four Factors

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    DanH wrote: View Post
    As always, please feel free to critique, comment or question any part of this. Any suggestions on how to better use this data, or what limitations I might have overlooked, would be much appreciated.
    Dan, really nice effort. Personally, I LOVE this kind of stuff and pretty much can't get enough of it. (geek alert, lol). Anyways, the one thing I always point out is the importance of taking into account USG% or POSS% into the equation in offense. Not sure if you're calculations already do that, but I can see that you follow Dean Oliver. Most people DO overlook USG, but doing so always makes guys like Tyson Chandler and Amir look good (which they are), but you couldn't build an offense around guys like that because their usage is so limited. So the combination of high efficiency/low usage + less efficient/high usage guys can actually be beneficial when you can't get your hands on an elite offensive player (high efficiency/high usage). The unit as a whole must be looked at in this way, rather than the individual, because if you remove those high usage guys and force Amir to take a wider variety of shots (*cough*, game winning 3, *cough*), then he's going to start looking worse and worse.

    There is a great chapter in Dean Oliver's book, called "The Problem with Scorers" that I think is must reading for any stats geek that breaks down the importance of usage and the 'skill curve' on offense. The link is below...

    http://books.google.ca/books?id=jltv...corers&f=false

    Comment


    • #17
      golden wrote: View Post
      Dan, really nice effort. Personally, I LOVE this kind of stuff and pretty much can't get enough of it. (geek alert, lol). Anyways, the one thing I always point out is the importance of taking into account USG% or POSS% into the equation in offense. Not sure if you're calculations already do that, but I can see that you follow Dean Oliver. Most people DO overlook USG, but doing so always makes guys like Tyson Chandler and Amir look good (which they are), but you couldn't build an offense around guys like that because their usage is so limited. So the combination of high efficiency/low usage + less efficient/high usage guys can actually be beneficial when you can't get your hands on an elite offensive player (high efficiency/high usage). The unit as a whole must be looked at in this way, rather than the individual, because if you remove those high usage guys and force Amir to take a wider variety of shots (*cough*, game winning 3, *cough*), then he's going to start looking worse and worse.

      There is a great chapter in Dean Oliver's book, called "The Problem with Scorers" that I think is must reading for any stats geek that breaks down the importance of usage and the 'skill curve' on offense. The link is below...

      http://books.google.ca/books?id=jltv...corers&f=false
      I absolutely agree. Keep in mind that this analysis a) looks at the team's performance, not the player's, and b) is meant to encapsulate how well a player is playing within their role. You can't have a bunch of players playing the same role on a team, and this process wasn't meant to insinuate that at all.
      twitter.com/dhackett1565

      Comment


      • #18
        DanH wrote: View Post
        I absolutely agree. Keep in mind that this analysis a) looks at the team's performance, not the player's, and b) is meant to encapsulate how well a player is playing within their role. You can't have a bunch of players playing the same role on a team, and this process wasn't meant to insinuate that at all.
        If usage is not taken into account with ORTG, then perhaps one problem with your analysis might be that it would overvalue the impact on wins by role players vs. guys who shoulder a larger burden of responsibility for winning. Specifically, defensively competent players who are low usage/high efficiency on offense would seem to score well, like 3&D wings, or PnR centers. I think there has to be some adjustment to account for usage, since it's a huge part of the individual/team dynamic.

        Comment


        • #19
          golden wrote: View Post
          If usage is not taken into account with ORTG, then perhaps one problem with your analysis might be that it would overvalue the impact on wins by role players vs. guys who shoulder a larger burden of responsibility for winning. Specifically, defensively competent players who are low usage/high efficiency on offense would seem to score well, like 3&D wings, or PnR centers. I think there has to be some adjustment to account for usage, since it's a huge part of the individual/team dynamic.
          It depends on what ORTG we are using.

          BBallref uses points per 100 possessions used. NBA.com/82games is pts per 100 possession while on the floor.

          There is a fair case for BBallref's method of calculating ORTG, not so much in NBA.com/82games version since it is straight points scored while on the floor, and all impact of usage would be baked into the #

          I checked NBAwowy, and all I saw was stats compiled using publically available data, but no reference to whose data.


          Just as a note, I did a post here based on inpact of usage data (from sloan) on PPP, and usage really does not have a significant impact on efficiency. I'll see if I can round it up again, but I think it was 0.0025 to 0.0045 pts per possession per 1% increase in usage. Accounting for usage doesn't make an inefficient player efficient or even average (or vice versa). The impact is there definetely, but its not particulariy large.

          Comment


          • #20
            DanH wrote: View Post
            Keep in mind, I'm not leaving them out entirely. I left out their win values due to the time it takes to compile that data, but every player on the team factors into those expected performance values for each player. As a note, if I included the win values, the best they could all be is 82 win players, the worst is 0, the average 41. But the likely max is more like 60, versus the likely min of 20. And I'm using the term "likely" very lightly here.

            So the 60 win deep bench equates to 41 total wins (instead of the predicted 39 wins), a terrible 20 win deep bench (which is what we most likely have, frankly) equates to 37 total wins, and a 41 win bench equates to no change to the current 39 win prediction. Any win value for the bench between 34.5 and 42.5 wins keeps the rounded win expectation at 39.

            So: first, those little guys that make all the difference ARE being counted, in the form of their contributions to the expected performances of their teammates. Second, even if they were counted like everyone else is in win totals, there is very little chance they would change anything.
            Clearly you do not believe in Jedi Knights.

            Comment


            • #21
              DanH wrote: View Post
              I definitely agree about doing this league wide. Unfortunately, I don't have direct access to the data and it was a bit of a slog to pull the numbers for the team. Don't have the time to do it league wide.
              I've down a couple of posts where I've drawn for multiple stats in the past, and I know how much of a pain the the @$$ it can be, and understand how it isn't feasible to do the whole league. Any chance you'd be able to do it for 5-10 players around the league? Maybe two guys from each position? That might be very helpful in pointing out potential problems in your methodology.

              I would recommend picking the best player in the position and then a middle of the road player

              Chris Paul, Jose Calderon
              James Harden, Thabo Sefaloosha
              Lebron James, Carmelo Anthony
              Marcus Aldrige, Paul Millsap
              Dwight Howard, Marcin Gortat

              Doing some targeted selection sampling would definitely help the validity.


              Either way, great article. I'm okay with a stat that reflects how good they are at their current role. Doesn't need to be reflective of how they would do in a different role to be useful IMO.




              [QUOTE=Craig;252626]Clearly you do not believe in Jedi Knights.

              "They're going to have to rename the whole conference after us: Toronto Raptors 2014-2015 Northern Conference Champions" ~ ezzbee Dec. 2014

              "I guess I got a little carried away there" ~ ezzbee Apr. 2015

              "We only have one rule on this team. What is that rule? E.L.E. That's right's, E.L.E, and what does E.L.E. stand for? EVERYBODY LOVE EVERYBODY. Right there up on the wall, because this isn't just a basketball team, this is a lifestyle. ~ Jackie Moon

              Comment


              • #22
                Craiger wrote: View Post
                It depends on what ORTG we are using.

                BBallref uses points per 100 possessions used. NBA.com/82games is pts per 100 possession while on the floor.

                There is a fair case for BBallref's method of calculating ORTG, not so much in NBA.com/82games version since it is straight points scored while on the floor, and all impact of usage would be baked into the #

                I checked NBAwowy, and all I saw was stats compiled using publically available data, but no reference to whose data.


                Just as a note, I did a post here based on inpact of usage data (from sloan) on PPP, and usage really does not have a significant impact on efficiency. I'll see if I can round it up again, but I think it was 0.0025 to 0.0045 pts per possession per 1% increase in usage. Accounting for usage doesn't make an inefficient player efficient or even average (or vice versa). The impact is there definetely, but its not particulariy large.
                Definitely using the "82games" version. What we're looking at here is how the team performs when the players are on the floor, not how the player performs. Bball ref has both, by the way, the individual is on the player stats page, while the "82games" version is in their on-off court numbers.
                twitter.com/dhackett1565

                Comment


                • #23
                  ezz_bee wrote: View Post
                  I've down a couple of posts where I've drawn for multiple stats in the past, and I know how much of a pain the the @$$ it can be, and understand how it isn't feasible to do the whole league. Any chance you'd be able to do it for 5-10 players around the league? Maybe two guys from each position? That might be very helpful in pointing out potential problems in your methodology.

                  I would recommend picking the best player in the position and then a middle of the road player

                  Chris Paul, Jose Calderon
                  James Harden, Thabo Sefaloosha
                  Lebron James, Carmelo Anthony
                  Marcus Aldrige, Paul Millsap
                  Dwight Howard, Marcin Gortat

                  Doing some targeted selection sampling would definitely help the validity.


                  Either way, great article. I'm okay with a stat that reflects how good they are at their current role. Doesn't need to be reflective of how they would do in a different role to be useful IMO.


                  Unfortunately, for each player, I have to pull a full team's worth of data. What is easier is going back a year and being able to use basketball-reference's lineup data - that allows me to pull an entire team much more quickly. I used my approach on last year's Heat and it showed that applying my deltas to the 100 RTG team is skewing the full team results towards .500. The values for each player are still valid within the team structure, but I'm working on a way to manipulate the numbers so they are comparable league wide.
                  twitter.com/dhackett1565

                  Comment

                  Working...
                  X